Review ArticleNot Only Mutations Matter Molecular Picture of Acute MyeloidLeukemia Emerging from Transcriptome Studies

Luiza Handschuh

Institute of Bioorganic Chemistry Polish Academy of Sciences Poznan Poland

Correspondence should be addressed to Luiza Handschuh luizahanibchpoznanpl

Received 26 April 2019 Accepted 12 June 2019 Published 30 July 2019

Guest Editor Annalisa Lonetti

Copyright copy 2019 Luiza Handschuh This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The last two decades of genome-scale research revealed a complex molecular picture of acutemyeloid leukemia (AML) On the onehand a number of mutations were discovered and associated with AML diagnosis and prognosis some of them were introducedinto diagnostic tests On the other hand transcriptome studies which preceded AML exome and genome sequencing remainedpoorly translated into clinics Nevertheless gene expression studies significantly contributed to the elucidation ofAMLpathogenesisand indicated potential therapeutic directionsThe power of transcriptomic approach lies in its comprehensiveness we can observehow genome manifests its function in a particular type of cells and follow many genes in one test Moreover gene expressionmeasurement can be combinedwithmutation detection as high-impactmutations are often present in transcriptsThis review sumsup 20 years of transcriptome research devoted to AML Gene expression profiling (GEP) revealed signatures distinctive for selectedAML subtypes anduncovered the additional within-subtype heterogeneityThe resultswere particularly valuable in the case ofAMLwith normal karyotype which concerns up to 50 of AML cases With the use of GEP new classes of the disease were identifiedand prognostic predictors were proposed A plenty of genes were detected as overexpressed in AML when compared to healthycontrol including KIT BAALC ERG MN1 CDX2 WT1 PRAME and HOX genes High expression of these genes constitutesusually an unfavorable prognostic factor Upregulation of FLT3 and NPM1 genes independent on their mutation status was alsoreported in AML and correlatedwith poor outcomeHowever transcriptome is not limited to the protein-coding genes other typesof RNAmolecules exist in a cell and regulate genome function It was shown that microRNA (miRNA) profiles differentiated AMLgroups and predicted outcome not worse than protein-coding gene profiles For example upregulation of miR-10a miR-10b andmiR-196b and downregulation of miR-192 were found as typical of AML with NPM1 mutation whereas overexpression of miR-155was associatedwith FLT3-internal tandem duplication (FLT3-ITD) Development of high-throughput technologies andmicroarrayreplacement by next generation sequencing (RNA-seq) enabled uncovering a real variety of leukemic cell transcriptomes reflectedby gene fusions chimeric RNAs alternatively spliced transcripts miRNAs piRNAs long noncoding RNAs (lncRNAs) and theirspecial type circular RNAs Many of them can be considered as AML biomarkers and potential therapeutic targets The relationsbetween particular RNA puzzles and other components of leukemic cells and their microenvironment such as exosomes are nowunder investigation Hopefully the results of this research will shed the light on these aspects of AML pathogenesis which are stillnot completely understood

1 Introduction

Acute myeloid leukemia (AML) the most frequent leukemiain adults is a severe myeloproliferative disorder with thehigh risk of relapse and high mortality rate [1 2] Randomgenetic alterations sequentially acquired by hematopoieticstem and progenitor cells disrupt hematopoiesis by differen-tiation blockades uncontrolled growth and proliferation andinhibition of apoptosis [3] Immature partially differentiated

blast cells with self-renewal capacity first accumulate inbone marrow (BM) and then infiltrate peripheral blood (PB)and organs impairing their functions [4] Despite similarsymptoms blast morphology and clinical implications AMLis a very heterogeneous disease presenting a wide spectrumof subtypes with different molecular features and outcomes[5 6] A number of chromosomal rearrangements and smallmutations have been detected in AML and associated withthe pathogenesis diagnosis and prognosis of the disease

HindawiJournal of OncologyVolume 2019 Article ID 7239206 36 pageshttpsdoiorg10115520197239206

2 Journal of Oncology

AML

02000400060008000

100001200014000160001800020000

leukemiabreast cancerlung cancer

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

2014

2016

2018

(a)

First microarrayfor gene expression

analysis

First automatic sequencerGenome Analyzer (Solexa)

AML amp gene expression

0

50

100

150

200

250

300

AML amp transcriptomeAML amp microRNAAML amp non-coding RNAAML amp genomeAML amp exome

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

2014

2016

2018

(b)

First microarray-based AML study

First AML genome sequenced

microarrayNGS

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

01000200030004000500060007000

(c)

Figure 1 The number of publications found in PubMed devoted to (a) AML leukemia and the two most common human cancers (b)transcriptome and genome-based AML studies (c) two the most common high-throughput technologies used in AML studies microarraysand next generation sequencing (NGS) The search terms and exact numbers of publications are noted in Supplementary Table 1

[5 7 8] AML heterogeneity is also reflected in its clas-sification first established in 1976 by French-American-British (FAB) cooperative groups based on morphologicaland cytochemical criteria [9] and revised nine years later[10] In 2001 an alternative classification system combiningthe leukemic cell lineage and maturation stage with geneticaberrations was proposed by World Health Organization(WHO) and improved in 2008 and 2016 [11ndash13]

Although AML accounts for not more than 1 of allcancer diseases it belongs to one of the most extensivelystudied human tumors which has been confirmed by theever-growing number of scientific reports (Figure 1(a)) In theCancer Genome Atlas (TCGA) a landmark cancer genomicsprogram (httpswwwcancergovtcga) AML is one of 33cancer types collected hitherto being represented by 200cases Availability of the tumor cells which can be easily andin extensive amounts extracted from BM aspirates or evenPB makes AML a perfect model for cancer studies In AMLthe existence of a cancer stem cell was first demonstratedproving the rightness of the tumor stem-cell concept [14ndash16]Since then our knowledge about cancer stem cells startedto increase [6 17] Progress in the development of high-throughputmethods such asmicroarrays and next generationsequencing (NGS) advanced our understanding of AML and

other cancers [18 19] (Figures 1(b) and 1(c)) Figure 2 presentsthe timeline and milestones of AML research intertwinedwith the milestones of the Human Genome Project (HGP)The first application of global gene expression profiling(GEP) for cancer classification was demonstrated in 1999 onthe example of two acute leukemias arisen from differentlineages myeloid (AML) and lymphoid (acute lymphoblasticleukemia ALL) [20] The first cancer genomes sequencedderived fromAML patients [21 22] In 2013 TCGA ResearchNetwork published the sequences of 50 whole genomesand 150 exomes of AML patients [23] Three years latertargeted resequencing of 111 genes in 1540 AML patientsrevealed more than 5 thousands of driver mutations [24]In 2018 functional genomic landscape was drawn based onthe exome sequencing gene expression and the analysisof ex vivo drug sensitivity in a cohort of over 500 AMLpatients [25] Genome-wide studies revealed that the numberof driver mutations in AML (on average 13 somatic variantsper patient) is lower than in solid tumors [23 25] NewAML entities of diagnostic and prognostic significance havebeen identified and potential therapeutic targets have beenindicated [26 27] Despite the tremendous effort put intoresearch AML (except for acute promyelocytic leukemiaAPL) still lacks effective medical treatment [28 29] However

Journal of Oncology 3

Figure 2 The milestones in genomic and transcriptomic research of acute myeloid leukemia A symbolic mRNA molecule serves as atimeline on which the most important papers and events are marked starting from the first FAB classification of AML in 1976 [9] andits revised version published in 1985 [10] The microscopic images of M1-M7 FAB AML types come from the private collection of Prof JohnM Bennett and were used thanks to the courtesy of the Professor The original pictures from the following publications were used withthe permission of the authors and magazine publishers Schena et al PNAS 1996 [47] (Copyright 1996 National Academy of Sciences)Golub et al Science 1999 [20] (reprinted with permission of AAAS) Lu et al Nature 2005 [51] (reprinted by permission from SpringerNature Nature Copyright 2005) Falini et al NEJM 2005 [52] (Copyright 2005 Massachusetts Medical Society reprinted with permissionfrom Massachusetts Medical Society) Haferlach et al Blood 2005 [53] (reprinted by permission from American Society of HematologyCopyright 2005) Mi et al PNAS 2007 [54] (Copyright 2007 National Academy of Sciences) TCGA paper from NEJM 2013 [23] (Copyright2013MassachusettsMedical Society reprintedwith permission fromMassachusettsMedical Society) Tyner et al Nature 2018 [25] (reprintedby permission from Springer Nature Nature Copyright 2018) Hoadley et al Cell 2018 [55] (reprinted by permission from Elsevier CellCopyright 2018) Two Nature journal covers were reprinted by permission from Springer Nature Nature Copyright 2001 and 2008 TwoScience journal covers from 1999 and 2001 were reprinted with permission of AAAS WHO publication cover images were reproducedwith permission from Jaffe ES Harris NL Stein H Vardiman JW Eds WHO Classification of Tumours Pathology and Geneticsof Tumours of Haematopoietic and Lymphoid Tissues IARC Lyon 2001 [11] Swerdlow SH Campo E Harris NL Jaffe ES Pileri SAStein H Thiele J Vardiman JW World Health Organization Classification of Tumours of Haematopoietic and Lymphoid Tissues IARCLyon 2008 Swerdlow SH Campo E Harris NL Jaffe ES Pileri SA Stein H Thiele J Arber DA Hasserjian RP Le Beau MM OraziA Siebert R World Health Organization Classification of Tumours of Haematopoietic and Lymphoid Tissues revised 4th edition IARCLyon 2017 The photograph of Francis Collins and Craig Venter made by Chuck Kennedy in 2000 (krtphotos001229) was used with thelicense of Newscom (httpswwwnewscomcom) Vitruvian man image was downloaded from Wikimedia Commons under a free license(httpscommonswikimediaorgwikiFileVitruvian manjpg) Graphics representing the HumanGenome Project (HGP) were used thanksto the courtesy of National Human Genome Research Institute ((NHGRI httpswwwgenomegov) The TCGA logo was used with thepermission of National Cancer Institute (NCI httpswwwcancergov)

some promising therapeutic strategies are currently underinvestigation [30]

Despite the factmuch attention has recently been devotedto genetic alterations occurring in AML it should be

remembered that the state of the cell is largely reflected byits transcriptome which is the product of genomic activityIn fact transcriptome-level analyses preceded whole genomesequencing and still serve as supplementary approaches in

Journal of Oncology 3

Figure 2 The milestones in genomic and transcriptomic research of acute myeloid leukemia A symbolic mRNA molecule serves as atimeline on which the most important papers and events are marked starting from the first FAB classification of AML in 1976 [9] andits revised version published in 1985 [10] The microscopic images of M1-M7 FAB AML types come from the private collection of Prof JohnM Bennett and were used thanks to the courtesy of the Professor The original pictures from the following publications were used withthe permission of the authors and magazine publishers Schena et al PNAS 1996 [47] (Copyright 1996 National Academy of Sciences)Golub et al Science 1999 [20] (reprinted with permission of AAAS) Lu et al Nature 2005 [51] (reprinted by permission from SpringerNature Nature Copyright 2005) Falini et al NEJM 2005 [52] (Copyright 2005 Massachusetts Medical Society reprinted with permissionfrom Massachusetts Medical Society) Haferlach et al Blood 2005 [53] (reprinted by permission from American Society of HematologyCopyright 2005) Mi et al PNAS 2007 [54] (Copyright 2007 National Academy of Sciences) TCGA paper from NEJM 2013 [23] (Copyright2013MassachusettsMedical Society reprintedwith permission fromMassachusettsMedical Society) Tyner et al Nature 2018 [25] (reprintedby permission from Springer Nature Nature Copyright 2018) Hoadley et al Cell 2018 [55] (reprinted by permission from Elsevier CellCopyright 2018) Two Nature journal covers were reprinted by permission from Springer Nature Nature Copyright 2001 and 2008 TwoScience journal covers from 1999 and 2001 were reprinted with permission of AAAS WHO publication cover images were reproducedwith permission from Jaffe ES Harris NL Stein H Vardiman JW Eds WHO Classification of Tumours Pathology and Geneticsof Tumours of Haematopoietic and Lymphoid Tissues IARC Lyon 2001 [11] Swerdlow SH Campo E Harris NL Jaffe ES Pileri SAStein H Thiele J Vardiman JW World Health Organization Classification of Tumours of Haematopoietic and Lymphoid Tissues IARCLyon 2008 Swerdlow SH Campo E Harris NL Jaffe ES Pileri SA Stein H Thiele J Arber DA Hasserjian RP Le Beau MM OraziA Siebert R World Health Organization Classification of Tumours of Haematopoietic and Lymphoid Tissues revised 4th edition IARCLyon 2017 The photograph of Francis Collins and Craig Venter made by Chuck Kennedy in 2000 (krtphotos001229) was used with thelicense of Newscom (httpswwwnewscomcom) Vitruvian man image was downloaded from Wikimedia Commons under a free license(httpscommonswikimediaorgwikiFileVitruvian manjpg) Graphics representing the HumanGenome Project (HGP) were used thanksto the courtesy of National Human Genome Research Institute ((NHGRI httpswwwgenomegov) The TCGA logo was used with thepermission of National Cancer Institute (NCI httpswwwcancergov)

some promising therapeutic strategies are currently underinvestigation [30]

Despite the factmuch attention has recently been devotedto genetic alterations occurring in AML it should be

remembered that the state of the cell is largely reflected byits transcriptome which is the product of genomic activityIn fact transcriptome-level analyses preceded whole genomesequencing and still serve as supplementary approaches in

4 Journal of Oncology

AML studies In this review I tried to show to what extentdeeper insight into AML transcriptomes helped to unravelthe mysteries of the disease

2 The Early Beginnings Studies of SingleProtein-Coding Genes

At the turn of the 1980s and 1990s more and more reportsdocumenting gene expression in AML started to appearExpression of several protooncogenes encoding transcrip-tion factors (MYC MYB and FOS) and tyrosine kinases(ABL1 FES KIT and PIM) with essential roles in theregulation of hematopoiesis cell proliferation differentia-tion cell cycle and apoptosis was demonstrated in AMLcells extracted from patients [31ndash34] Erythroid progenitorsand HEL erythroleukemia cells presented amplification ofanother transcription factor involved in cell proliferationE2F1 [35] Overexpression of RUNX1 gene previously knownas AML1 or CBF2A regulator of hematopoiesis particularlyof myeloid lineage suppressed granulocytic differentiationand stimulated cell proliferation in murine cells [36] Carowet al [37] showed that expression of the hematopoieticgrowth factor receptor-encoding FLT3 gene was not limitedto normal stemprogenitor cells but was even elevated inleukemic blasts In AML M2 with t(821) high percentageof CD34+ cells was correlated with high expression ofCD34 gene [38] BMI1 member of the polycomb compleximplicated inmaintenance of normal and leukemic stem cellswas found to be expressed in AML M0 at a higher level thanin other AML subtypes [39] High expression of some geneswas associated with adverse AML prognosis eg WT1 [40]MN1 [41] BAALC [42] ERG [43] and EVI1 (ecotropic viralintegration site 1 at present known as MECOM from MDS1and EVI1 complex locus) [44] Genes which were renamedwithin the last years are listed in Table 1 together with themost commonly used abbreviations

3 Microarray-Based Gene ExpressionProfiling A Tool for Disease Classification

Technological progress enabled transition from single geneanalysis to whole transcriptome scale at the end of theprevious millennium [45] The milestone was the inventionof the microarray chip-format tool which allowed for simul-taneous analysis of thousands of genes in one test [46 47]Golub et al [20] and Alizadeh et al [48] were the firstwho showed that global gene expression profiling (GEP)could be a tool for cancer research and classification Eachgroup used a different type of microarrays Golub et alused commercially available high-density GeneChips madeof short oligonucleotides synthesized in situ (Affymetrix)whereas Alizadeh et al [48] constructed their own cDNAarray Lymphochip dedicated to analysis of gene expressionin normal and malignant lymphocytes Golub et al [20]proved that the distinction between two acute leukemiasAML and ALL could be performed in a single test Out of6817 human genes measured expression of 50 genes wasselected as the most closely correlated with AML-ALL class

distinction The 50-gene predictor was successfully validatedin an independent collection including samples from PBinstead of BM samples from childhood AML patients andsamples collected by various laboratories Among the mostinformative genes overexpressed in AML were known genesencoding cell surface proteins eg CD33 and CD11c (cur-rently ITGAX) and transcription factors including HOXA9oncogene whose high expression level was noted in AMLpatients with poor outcome In fact HOXA9 seemed to be asingle gene capable of predicting treatment failure in AMLThe study revealed also novel AML markers such as genecoding for leptin receptor with antiapoptotic activity or zyxingene encoding protein with cell adhesion functionThe aboveresults showed the power of GEP in disease classificationand class discovery and encouraged other investigators toimplement DNAmicroarray technology in their laboratories

4 Protein-Coding Transcriptomes ofCytogenetically Defined AML Subtypes

Distinction between AML and other hematologic disordersseemed to be trivial The question appeared whether geneexpression profiles could successfully differentiate AML sub-groups with cytogenetic and genetic abnormalities StartingfromAMLwith themost common chromosomal aberrationsdifferent authors demonstrated that each of AML subgroupspossessed its own gene expression signature and could beeasily distinguished from one another Below the exemplarystudies are described in more detail

Virtaneva et al [49] comparedAMLwith isolated trisomy8 (+8) to cytogenetically normal AML (CN-AML) andrevealed fundamental biological differences between thesetwo types of the disease Common feature of both AMLtypes was downregulation of genes encoding hematopoietictranscription factors (STAT4 FUS MCM3 and MCM5) andmyeloid markers (ELANE and MPO) in comparison to nor-mal CD34+ bone marrow cells Genes encoding complementfactor D (CFD) proteins involved in cell growth and differ-entiation (NDRG1 and BTG1) transcription factorsKLF6 andATF3 and transcription coactivator TAF10 were upregulatedin both AMLs in relation to control CD34+ fraction Two themost differentiating genes between AML +8 and CN-AMLwere MLLT2 (present AFF1 (AF4FMR2 Family Member 1)upregulated in CN-AML) encoding regulator of transcriptionand chromatin remodeling and FABP5 (upregulated in AML+8) encoding protein involved in fatty acid metabolismUnsurprisingly AML +8 blasts presented general overex-pression of genes encoded on chromosome 8 The effect ofgenomic gains and losses on expression levels of genes locatedwithin the affected regions was further confirmed in a studyof AML with trisomy 8 11 13 monosomy 7 and deletion5q [50] Surprisingly in a study of Virtaneva et al [49]protooncogene MYC also encoded on chromosome 8 wasdownregulated in AML +8 On the one hand in comparisontoCN-AML decreased expression level of proapoptotic genes(eg CRADD BAD) was noted for AML +8 whereas TP53gene encoding tumor suppressor and also apoptosis inducerwas increased in AML +8 On the other hand only CN-AML

Journal of Oncology 5

Table 1 The list of the genes renamed within the last years and the most commonly used abbreviations

Renamed GenesPrevious Name Previous Description Current Name Current Description

AML1 AcuteMyeloid Leukemia 1 RUNX1 Runt Related TranscriptionFactor 1

Ang-1 Angiopoietin-1 ANGPT1 Angiopoietin 1BRN3A Brain-3A POU4F1 POU Class 4 Homeobox 1CD11c CD11c Antigen ITGAX Integrin Subunit Alpha XCD133 CD133 Antigen PROM1 Prominin 1

ELA2 Elastase-2 ELANE Elastase NeutrophilExpressed

ETO Protein ETO RUNX1T1 RUNX1 TranslocationPartner 1

EVI1 Ecotropic Viral Integration Site 1 MECOM MDS1 And EVI1 ComplexLocus

FLJ14054 Homo sapiens cDNA FLJ14054 fis cloneHEMBB1000240 NPR3 Natriuretic Peptide

Receptor 3

IL8 Interleukin-8 CXCL8 C-X-CMotif ChemokineLigand 8

MADH1 Mothers Against Decapentaplegic Homolog 1 SMAD1 SMAD Family Member 1

MDR1 Multidrug Resistance Protein 1 ABCB1 ATP Binding CassetteSubfamily B Member 1

MEL1 MDS1EVI1-Like Gene 1 PRDM16 PRSET Domain 16

MLL Mixed Lineage Leukemia KMT2A Lysine Methyltransferase2A

MLLT2 MyeloidLymphoid Or Mixed-Lineage Leukemia(Trithorax (Drosophila) Homolog) Translocated To 2 AFF1 AF4FMR2 Family

Member 1

MLLT4 MyeloidLymphoid Or Mixed-Lineage LeukemiaTranslocated To 4 AFDN Afadin Adherens Junction

Formation Factor

NICAL NEDD9-Interacting Protein With Calponin HomologyAnd LIM Domains MICAL1

Microtubule AssociatedMonooxygenase Calponin

And LIM DomainContaining 1

OPN Osteopontin SPP1 Secreted Phosphoprotein 1

PTRF Polymerase I And Transcript Release Factor CAVIN1 Caveolae AssociatedProtein 1

PU1 Hematopoietic Transcription Factor PU1 SPI1 Spi-1 Proto-OncogeneAbbreviationsAML acute myeloid leukemiaALL acute lymphoblastic leukemiaAPL acute promyelocytic leukemiaBM bone marrowCBF core binding factorcircRNAs circular RNAsCLL chronic lymphocytic leukemiaCML chronic myeloid leukemiaCN-AML cytogenetically normal AMLCR complete remissionDEGs differentially expressed genesDFS disease-free survivalFLT3-ITD FLT3-internal tandem duplicationFAB French-American-British (classification system)

6 Journal of Oncology

Table 1 Continued

GEP gene expression profilingHH HedgehogHSCs hematopoietic stem cellsHSPCs hematopoietic stem-progenitor cellslncRNAs long noncoding RNAsLSCs leukemic stem cellsMDS myelodysplastic syndromesMRD minimal residual diseaseMSC mesenchymal stem cellsNGS next generation sequencingNK-AML normal karyotype AMLNPMc+ NPM-cytoplasmic positiveOS overall survivalPB peripheral bloodPBMCs peripheral blood mononuclear cellsSAGE serial analysis of gene expressionsnoRNAs small nucleolar RNAsTCGA the Cancer Genome AtlasTEs transposable elementsWBC white blood cellWHO World Health Organization

showed upregulation of antiapoptotic gene DAD1 Thereforethe authors suggested different mechanisms of cell deathescape for the two studied leukemia types and associated itwith AML +8 resistance to cytarabine-based chemotherapywhich should induce apoptosis [49]

AML subtypes with three reciprocal rearrangementst(821)(q22q22) inv(16)(p13q22) and (1517)(q22q12) cor-responding to the morphological FAB subtypes M2 M4eoand M3M3v respectively were the subject of research ofSchoch et al [56] Principal component analysis (PCA)of microarray data performed with the use of 1000 mostinformative genes clearly separated AML samples accordingto chromosomal aberration The minimal set of 13 genes(PRKAR1B GNAI1 PRODH CD52 KRT18 CLIP3 CLUPTGDS HOXB2 CLEC2B CTSW S100A9 and MYH11) wassufficient to distinguish one AML subtype from another onthe basis of gene expression solely Expression levels of 36genes enabled accurate classification of all three studied AMLsubtypes Another set of 82 genes allowed for distinction ofM3 and M3v two phenotypically different AML types witht(1517) In addition the study showed that AMLs with alter-ations involved core binding factor (CBF) complex t(821)and inv(16) were more related to each other than to AMLwith t(1517) The authors explained the overexpression ofMYH11 in inv(16) and RUNX1T1 (former ETO) in t(821) as aconsequence of high expression of fusion transcripts affectingthese genes A new marker of t(821) was identified byDebernardi et al [57] who found that a level of transcriptionfactor-coding gene POU4F1 (former BRN3A) was 43-foldhigher in t(821) AML than in other AML samples

Verhaak et al [58] by gene expression analysis intwo independent cohorts of AML patients under 60 eachexceeding 200 cases perfectly distinguished three favorablecytogenetic AML subtypes t(821) t(1517) and inv(16) ForAML with NPM1 or CEBPAmutations GEP-based classifierswere less accurate The distinction of AML with othermutations (eg FLT3 and RAS) and aberrations (11q23 -55q- -77q- abn3q and t(922)) was not possible with the use ofGEP Nevertheless for abn3q the most discriminative genewas MECOM encoding an oncogenic transcription factoroften involved in 3q26 abnormalities and in 7(q) almost alldifferentially decreased genes were located on chromosome7

One of the most challenging tasks was to find uniquefeatures characterizing cytogenetically normal AML (CN-AML or NK-AML from normal karyotype AML) whichaccounts for 40-50 of all AML cases Debernardi et al[57] were the first who attempted to do that Althoughthe sample size was not large (28 adult AML samplesincluding 10 NK-AML) and NK-AML revealed higher vari-ability than AML with translocations the authors found NK-AML could be separated from AML samples with chro-mosomal rearrangements based on the expression levelsof certain members of the class I homeobox A and Bgene families which were low or undetectable in AMLwith (t(821) t(1517) and inv(16)) In NK-AML expres-sion level of 10 genes was extremely increased HOXA4HOXA5 HOXA9 HOXB2 HOXB3 HOXB5 HOXB6 andHOXB7 and two members of TALE family MEIS1 andPBX3 While overexpression of HOXB genes was unique for

Journal of Oncology 7

NK-AML the upregulation of the remaining 5 genes wasshared with 11q23 AMLwhereMLL (mixed lineage leukemia)gene now renamed to KMT2A (Lysine Methyltransferase2A) was fused with different partners High expressionof some homeobox genes (eg HOXA3 and HOXB6) waslater found as typical of hematopoietic stem cells (HSCs)[59]

In 2004 two remarkable papers identifying not onlyknown but also newmolecular AML subtypes through globalGEP were published in the same issue of the New EnglandJournal of Medicine [60 61] Bullinger et al (2004) [60]who analyzed 116 adult AML samples with the use of cDNAmicroarrays found that hierarchical clustering with over6 thousands of the most varied genes divided all AMLsamples into two main clusters Out of the cytogeneticgroups only t(1517) (APL) generated one condense sub-cluster To enable biological insight into AML pathogenesisgroup-specific gene expression signatures were establishedand functionally characterized Signature specific for APLincluded genes related to hemostasis (PLAU SERPING1ANXA8 and PLAUR) resistance to apoptosis (TNFRSF4AVEN and BIRC5) impairment of retinoic acid-stimulatedcell differentiation (TBL1X CALR and RARRES3) and resis-tance to chemotherapy (CYP2E1 EPHX1 and a group of met-allothionein (MT) genes) MLLT4 (present AFDN Afadin)one of KMT2A fusion partners was among the genes withunique expression profile in t(821) which suggested similarmechanism of pathogenesis with t(611) High expression ofNT5E observed in inv(16) was correlated with resistanceto cytarabine Interestingly high expression of homeoboxgenes (HOXA4 HOXA9 HOXA10 PBX3 and MEIS1) wasdetected in AML specimens with not only normal butalso complex karyotypes Within NK-AML Bullinger et al[60] distinguished two distinct groups one where FLT3aberrations and FAB subtypes M1 and M2 prevailed andthe second one where FAB M4 and M5 subtypes weremore common Of note patients classified to those groupshad different outcomes For AML with complex karyotypeAML with KMT2A partial tandem duplications and AML+8 it was impossible to find statistically significant uniquegene expression signatures Valk et al [61] who analyzed285 AML patients using Affymetrix GeneChips identified16 AML groups with distinct gene expression profiles Someof them were composed of AML samples with known cyto-genetic aberrations t(1821) t(1517) and inv(16) RUNX1T1gene which is a RUNX1 fusion partner was the mostdiscriminative gene for AML with t(821) Overexpressionof MYH11 was the most discriminative feature of inv(16)which produces CBFB-MYH11 fusion gene Simultaneousdownregulation of CBFB observed in this subtype couldbe explained by eg negative regulation of a wild-type(wt) CBFB allele by the fusion transcript For APL growthfactor-coding genes were the most discriminative (hepato-cyte growth factor (HGF) macrophage-stimulating 1 growthfactor (MST1) and fibroblast growth factor 13 (FGF13))However AMLs with 11q23 were segregated into two separateclusters and partially scattered among all samples studiedAlsoNK-AML sampleswere divided into several clustersTheobserved heterogeneity could be at least partially explained

by the presence of particular mutations and different out-comes

Further AML transcriptome studies performed on inde-pendent patient cohorts usually confirmed earlier researchreporting partially overlapping gene expression signaturesHowever each study delivered a portion of new informationwhich deepened our knowledge about AML pathogenesisGutierrez et al [62] performed hierarchical clustering ofBM samples from 43 adult AML patients based on theexpression of over 5 thousand genes Four distinct clustersthey obtained corresponded to AMLwith inv(16) monocyticAML APL and other AML samples which included NK-AML The authors developed a minimal 21-gene predictorwhich classified each sample to appropriate group with100 accuracy Its efficiency was then confirmed with anindependent AML sample set APL samples which formedthemost condense group among all samples studied revealedhigh expression of several growth factors and other sig-naling proteins eg HGF FGF13 MST1 VEGFA IGFBP2and FGFR1 Contrary overexpression of HOX family mem-bers (A5 A6 A7 A9 A10 B2 B5 and B7) includinggenes encoding TALE proteins (MEIS1 PBX3) and histoneproteins was shared by all non-APL leukemias Increasedlevel of MYH11 expression and downregulation of CBFBand RUNX3 genes were noted specifically for AML withinv(16) In monocytic leukemia CSPG2 other adhesionmolecules such as the lectins CLECSF6 CLECSF12 SIGLEC7and FCN1 were upregulated compared to remaining AMLsThe remaining AML samples presented more heterogeneitywhich was reflected by the existence of two subclustersone with overexpression of genes encoding hematopoieticserine proteases present in azurophil granules of neutrophilicpolymorphonuclear leukocytes (AZU1 azurocidin 1 ELANE(previous ELA2) elastase PRTN3 proteinase 3 and CTSGcathepsin G) second with upregulation of CD34 antigenreflecting an early maturation arrest and lack of granulocyticdifferentiation

AMLM3was extensively studied by Payton et al [63] whocompared the malignant promyelocytes from APL patientsto leukemic cells collected from other AML subtypes andto promyelocytes neutrophils and CD34+ cells extractedfrom healthy bone marrow donors The identified ldquoM3-specific dysregulomerdquo was composed of 510 genes and manyof them exhibited dramatic differences in expression levelcomparing to other AML subtypes or normal promyelocytesFor example GABRE FGF13 HGF ANXA8 and PGBD5were the most overexpressed genes whereasVNN1MS4A6AP2RY13 HK3 and S100A9 the most underexpressed genesin M3 vs other AMLs 33 genes selected from the identifiedsignature were validated by another high-throughput digitaltechnology (nCounter NanoString) capable of detecting aslittle as 05 fM of a specific mRNA and measuring up to500 genes in a multiplex reaction The authors demonstratednCounter reproducibility and applicability as a tool forbiomarker analysis when limited amounts of clinical materialare available 33 genes validated byNanoString assaywere alsoenriched in an independent AML dataset of 325 samples andAPL mouse model but notably not in a cell line expressingPML-RARA fusion gene

8 Journal of Oncology

One of the most impressive microarray-based studies wasthat of Haferlach et al [53] who analyzed almost 900 patientswith leukemia including 620 with AML with the use ofAffymetrix GeneChips and support vector machine (SVM)model The authors identified 13 separate leukemia typesincluding 6withinAML Someof them eg AMLwith t(821)and with t(1517) could be classified with 100 specificityand 100 sensitivity based on the expression profile of 100genes per groupThe overall prediction accuracy of 951wasachievedThemisclassification occurredmainly in subgroupswith a low sample number or high intragroup heterogeneityThe largest and the most heterogeneous subgroup was AMLwith normal karyotype which cosegregated with AML withless common cytogenetic aberrations classified as ldquootherrdquoHere AML samples with different fusion partners of KMT2Agene were included Haferlach et al [64] showed that APLwas not only distinct from other AML subtypes in the matterof gene expression but two M3 phenotypes one with heavygranulation and bundles of Auer rods and AML M3 variant(M3v) with non- or hypogranular cytoplasm and a bilobednucleus could be discriminated based on gene expressionsignatures

The largest GEP study in hematology and oncology wasconducted thanks to international collaboration within theEuropean Leukemia Net In 2010 Gene Expression ProfilingWorking Group directed by Torsten Haferlach published theresults of analysis of 3334 samples collected from leukemia(including 542 AMLs) and MDS patients by 11 laboratoriesacross three continents [65] Apart from European oneslaboratories from the United States and one from Singa-pore joined the program The main conclusion was GEPwas a robust technology for the diagnosis of hematologicmalignancies with high accuracy According to the authorsGEP had invaluable application potential and was vulnerableto standardization outperforming more subjective methodssuch as cytomorphology and metaphase cytogenetics Toenable better molecular understanding of leukemias theauthors deposited the collected data into a publicly availabledomain

In 2012 de la Bletiere et al [66] proved that AML cyto-genetic subtypes could be successfully determined with theuse of GEP even in samples with low leukemic blast contentor poor quality With the use of Illumina Expression Bead-Chips the authors first classified 71 good quality samplesfrom a training set representing APL t(821)-AML inv(16)-AML or NK-AML with at least 60 percent of leukemicblasts The optimal 40-marker gene classifier (10 markersper class including previously described as well as newlydiscovered genes) was applied to 111 suboptimalAML sampleswith low leukemic blast load (from 2 to 59) andor poorquality control criteria The overall error rate was 36 AllAPL and t(821) samples were correctly classified even thosecontaining as low as 2 percent blasts The worst result wasachieved for inv(16) Surprisingly poor sample quality didnot affect classification By the way de la Bletiere et al[66] demonstrated reliability robustness and sensitivity ofIllumina bead-based technology which seemed to be notworse than other commercially or academically developedmicroarray platforms used before

5 Between AML and ALL Acute Leukemiawith KMT2A Rearrangements

In 2002 Armstrong et al [67] showed that ALL withtranslocations involving the KMT2A gene (previously knownasMLL) presented a unique gene expression profile differentfrom ALL and AML without KMT2A abnormalities Thecore of this unique gene expression signature consisted ofmultilineage markers of early hematopoietic progenitors andHOX genes which corresponds with the fact that KMT2Agene encodes histone lysine methyltransferase a transcrip-tional coactivator regulating expression of genes (includ-ing HOX) during early development and hematopoiesisTherefore the authors proposed to distinguish a distinctleukemia entity termed ldquoMLLrdquo Then a common geneexpression signature enriched in homeobox genes (MEIS1HOXA4 HOXA5 HOXA7 HOXA9 and HIOXA10) wasdetermined for all acute leukemias with KMT2A fusionirrespectively of their lineage (myeloid or lymphoid) byRoss et al [68] Similarly Andersson et al [69] associatedchildhood acute leukemias with KMT2A rearrangementswith upregulation of homeobox genes (HOXA10 HOXA4MEIS1 and PBX3) In their study KMT2A-positive AMLswere also enriched in genes involved in cell communica-tion and adhesion whereas some antiapoptotic genes (ega tumor necrosis factor receptor TNFRSF21) and tumorsuppressor genes (BRCA1 DLC1) were downregulated inthis AML subtype Hierarchical clustering with a subset ofgenes encoding transcription factors showed that leukemicsamples with KMT2A rearrangements grouped togetherindependently on lineage Although KMT2A translocationsare prevalent in infant and treatment-related leukemiasthey also occur in adult leukemias that were studied byKohlmann et al (2005) [70] who wondered how the differingKMT2A partner genes influenced the global gene expres-sion signature and whether pathways could be identified toexplain the molecular determination of KMT2A leukemiasof both lineages The data analysis in both types of acuteleukemias revealed t(11q23)KMT2A-positive samples thatwere evidently distinct from other subtypes of the samelineage As in the case of childhood leukemia adult KMT2A-AML and KMT2A-ALL despite a shared common geneprofile revealed also lineage-specific expression markerssufficient to segregate them according to their lineageswith no respect to the KMT2A fusion partner The com-monly overexpressed genes were obviously the homeoboxgenes and their regulators (HOXA9 MEIS1 HOXA10 PBX3HOXA3 HOXA4 HOXA5 HOXA7) NICAL gene (presentMICAL encoding Microtubule Associated Monooxygenase)RUNX2 transcription factor and FLT3 gene The commondownregulated genes included TNF-receptor superfamilymembers (TNFRSF10A and TNFRSF10D) transcription fac-tor POU4F1 tumor suppressor ST18 or MADH1 (presentSMAD1) encoding a signal transducer and transcriptionalmodulator Comparing to KMT2A-ALL overexpression ofCEBPB CEBPA KITMADH2MITF FES and SPI1 (formerPU1) oncogenes and MNDA encoding the myeloid cellnuclear differentiation was noted in KMT2A-AML Sum-marizing their results the authors concluded AML with

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

Journal of Oncology 5

Table 1 The list of the genes renamed within the last years and the most commonly used abbreviations

Renamed GenesPrevious Name Previous Description Current Name Current Description

AML1 AcuteMyeloid Leukemia 1 RUNX1 Runt Related TranscriptionFactor 1

Ang-1 Angiopoietin-1 ANGPT1 Angiopoietin 1BRN3A Brain-3A POU4F1 POU Class 4 Homeobox 1CD11c CD11c Antigen ITGAX Integrin Subunit Alpha XCD133 CD133 Antigen PROM1 Prominin 1

ELA2 Elastase-2 ELANE Elastase NeutrophilExpressed

ETO Protein ETO RUNX1T1 RUNX1 TranslocationPartner 1

EVI1 Ecotropic Viral Integration Site 1 MECOM MDS1 And EVI1 ComplexLocus

FLJ14054 Homo sapiens cDNA FLJ14054 fis cloneHEMBB1000240 NPR3 Natriuretic Peptide

Receptor 3

IL8 Interleukin-8 CXCL8 C-X-CMotif ChemokineLigand 8

MADH1 Mothers Against Decapentaplegic Homolog 1 SMAD1 SMAD Family Member 1

MDR1 Multidrug Resistance Protein 1 ABCB1 ATP Binding CassetteSubfamily B Member 1

MEL1 MDS1EVI1-Like Gene 1 PRDM16 PRSET Domain 16

MLL Mixed Lineage Leukemia KMT2A Lysine Methyltransferase2A

MLLT2 MyeloidLymphoid Or Mixed-Lineage Leukemia(Trithorax (Drosophila) Homolog) Translocated To 2 AFF1 AF4FMR2 Family

Member 1

MLLT4 MyeloidLymphoid Or Mixed-Lineage LeukemiaTranslocated To 4 AFDN Afadin Adherens Junction

Formation Factor

NICAL NEDD9-Interacting Protein With Calponin HomologyAnd LIM Domains MICAL1

Microtubule AssociatedMonooxygenase Calponin

And LIM DomainContaining 1

OPN Osteopontin SPP1 Secreted Phosphoprotein 1

PTRF Polymerase I And Transcript Release Factor CAVIN1 Caveolae AssociatedProtein 1

PU1 Hematopoietic Transcription Factor PU1 SPI1 Spi-1 Proto-OncogeneAbbreviationsAML acute myeloid leukemiaALL acute lymphoblastic leukemiaAPL acute promyelocytic leukemiaBM bone marrowCBF core binding factorcircRNAs circular RNAsCLL chronic lymphocytic leukemiaCML chronic myeloid leukemiaCN-AML cytogenetically normal AMLCR complete remissionDEGs differentially expressed genesDFS disease-free survivalFLT3-ITD FLT3-internal tandem duplicationFAB French-American-British (classification system)

6 Journal of Oncology

Table 1 Continued

GEP gene expression profilingHH HedgehogHSCs hematopoietic stem cellsHSPCs hematopoietic stem-progenitor cellslncRNAs long noncoding RNAsLSCs leukemic stem cellsMDS myelodysplastic syndromesMRD minimal residual diseaseMSC mesenchymal stem cellsNGS next generation sequencingNK-AML normal karyotype AMLNPMc+ NPM-cytoplasmic positiveOS overall survivalPB peripheral bloodPBMCs peripheral blood mononuclear cellsSAGE serial analysis of gene expressionsnoRNAs small nucleolar RNAsTCGA the Cancer Genome AtlasTEs transposable elementsWBC white blood cellWHO World Health Organization

showed upregulation of antiapoptotic gene DAD1 Thereforethe authors suggested different mechanisms of cell deathescape for the two studied leukemia types and associated itwith AML +8 resistance to cytarabine-based chemotherapywhich should induce apoptosis [49]

AML subtypes with three reciprocal rearrangementst(821)(q22q22) inv(16)(p13q22) and (1517)(q22q12) cor-responding to the morphological FAB subtypes M2 M4eoand M3M3v respectively were the subject of research ofSchoch et al [56] Principal component analysis (PCA)of microarray data performed with the use of 1000 mostinformative genes clearly separated AML samples accordingto chromosomal aberration The minimal set of 13 genes(PRKAR1B GNAI1 PRODH CD52 KRT18 CLIP3 CLUPTGDS HOXB2 CLEC2B CTSW S100A9 and MYH11) wassufficient to distinguish one AML subtype from another onthe basis of gene expression solely Expression levels of 36genes enabled accurate classification of all three studied AMLsubtypes Another set of 82 genes allowed for distinction ofM3 and M3v two phenotypically different AML types witht(1517) In addition the study showed that AMLs with alter-ations involved core binding factor (CBF) complex t(821)and inv(16) were more related to each other than to AMLwith t(1517) The authors explained the overexpression ofMYH11 in inv(16) and RUNX1T1 (former ETO) in t(821) as aconsequence of high expression of fusion transcripts affectingthese genes A new marker of t(821) was identified byDebernardi et al [57] who found that a level of transcriptionfactor-coding gene POU4F1 (former BRN3A) was 43-foldhigher in t(821) AML than in other AML samples

Verhaak et al [58] by gene expression analysis intwo independent cohorts of AML patients under 60 eachexceeding 200 cases perfectly distinguished three favorablecytogenetic AML subtypes t(821) t(1517) and inv(16) ForAML with NPM1 or CEBPAmutations GEP-based classifierswere less accurate The distinction of AML with othermutations (eg FLT3 and RAS) and aberrations (11q23 -55q- -77q- abn3q and t(922)) was not possible with the use ofGEP Nevertheless for abn3q the most discriminative genewas MECOM encoding an oncogenic transcription factoroften involved in 3q26 abnormalities and in 7(q) almost alldifferentially decreased genes were located on chromosome7

One of the most challenging tasks was to find uniquefeatures characterizing cytogenetically normal AML (CN-AML or NK-AML from normal karyotype AML) whichaccounts for 40-50 of all AML cases Debernardi et al[57] were the first who attempted to do that Althoughthe sample size was not large (28 adult AML samplesincluding 10 NK-AML) and NK-AML revealed higher vari-ability than AML with translocations the authors found NK-AML could be separated from AML samples with chro-mosomal rearrangements based on the expression levelsof certain members of the class I homeobox A and Bgene families which were low or undetectable in AMLwith (t(821) t(1517) and inv(16)) In NK-AML expres-sion level of 10 genes was extremely increased HOXA4HOXA5 HOXA9 HOXB2 HOXB3 HOXB5 HOXB6 andHOXB7 and two members of TALE family MEIS1 andPBX3 While overexpression of HOXB genes was unique for

Journal of Oncology 7

NK-AML the upregulation of the remaining 5 genes wasshared with 11q23 AMLwhereMLL (mixed lineage leukemia)gene now renamed to KMT2A (Lysine Methyltransferase2A) was fused with different partners High expressionof some homeobox genes (eg HOXA3 and HOXB6) waslater found as typical of hematopoietic stem cells (HSCs)[59]

In 2004 two remarkable papers identifying not onlyknown but also newmolecular AML subtypes through globalGEP were published in the same issue of the New EnglandJournal of Medicine [60 61] Bullinger et al (2004) [60]who analyzed 116 adult AML samples with the use of cDNAmicroarrays found that hierarchical clustering with over6 thousands of the most varied genes divided all AMLsamples into two main clusters Out of the cytogeneticgroups only t(1517) (APL) generated one condense sub-cluster To enable biological insight into AML pathogenesisgroup-specific gene expression signatures were establishedand functionally characterized Signature specific for APLincluded genes related to hemostasis (PLAU SERPING1ANXA8 and PLAUR) resistance to apoptosis (TNFRSF4AVEN and BIRC5) impairment of retinoic acid-stimulatedcell differentiation (TBL1X CALR and RARRES3) and resis-tance to chemotherapy (CYP2E1 EPHX1 and a group of met-allothionein (MT) genes) MLLT4 (present AFDN Afadin)one of KMT2A fusion partners was among the genes withunique expression profile in t(821) which suggested similarmechanism of pathogenesis with t(611) High expression ofNT5E observed in inv(16) was correlated with resistanceto cytarabine Interestingly high expression of homeoboxgenes (HOXA4 HOXA9 HOXA10 PBX3 and MEIS1) wasdetected in AML specimens with not only normal butalso complex karyotypes Within NK-AML Bullinger et al[60] distinguished two distinct groups one where FLT3aberrations and FAB subtypes M1 and M2 prevailed andthe second one where FAB M4 and M5 subtypes weremore common Of note patients classified to those groupshad different outcomes For AML with complex karyotypeAML with KMT2A partial tandem duplications and AML+8 it was impossible to find statistically significant uniquegene expression signatures Valk et al [61] who analyzed285 AML patients using Affymetrix GeneChips identified16 AML groups with distinct gene expression profiles Someof them were composed of AML samples with known cyto-genetic aberrations t(1821) t(1517) and inv(16) RUNX1T1gene which is a RUNX1 fusion partner was the mostdiscriminative gene for AML with t(821) Overexpressionof MYH11 was the most discriminative feature of inv(16)which produces CBFB-MYH11 fusion gene Simultaneousdownregulation of CBFB observed in this subtype couldbe explained by eg negative regulation of a wild-type(wt) CBFB allele by the fusion transcript For APL growthfactor-coding genes were the most discriminative (hepato-cyte growth factor (HGF) macrophage-stimulating 1 growthfactor (MST1) and fibroblast growth factor 13 (FGF13))However AMLs with 11q23 were segregated into two separateclusters and partially scattered among all samples studiedAlsoNK-AML sampleswere divided into several clustersTheobserved heterogeneity could be at least partially explained

by the presence of particular mutations and different out-comes

Further AML transcriptome studies performed on inde-pendent patient cohorts usually confirmed earlier researchreporting partially overlapping gene expression signaturesHowever each study delivered a portion of new informationwhich deepened our knowledge about AML pathogenesisGutierrez et al [62] performed hierarchical clustering ofBM samples from 43 adult AML patients based on theexpression of over 5 thousand genes Four distinct clustersthey obtained corresponded to AMLwith inv(16) monocyticAML APL and other AML samples which included NK-AML The authors developed a minimal 21-gene predictorwhich classified each sample to appropriate group with100 accuracy Its efficiency was then confirmed with anindependent AML sample set APL samples which formedthemost condense group among all samples studied revealedhigh expression of several growth factors and other sig-naling proteins eg HGF FGF13 MST1 VEGFA IGFBP2and FGFR1 Contrary overexpression of HOX family mem-bers (A5 A6 A7 A9 A10 B2 B5 and B7) includinggenes encoding TALE proteins (MEIS1 PBX3) and histoneproteins was shared by all non-APL leukemias Increasedlevel of MYH11 expression and downregulation of CBFBand RUNX3 genes were noted specifically for AML withinv(16) In monocytic leukemia CSPG2 other adhesionmolecules such as the lectins CLECSF6 CLECSF12 SIGLEC7and FCN1 were upregulated compared to remaining AMLsThe remaining AML samples presented more heterogeneitywhich was reflected by the existence of two subclustersone with overexpression of genes encoding hematopoieticserine proteases present in azurophil granules of neutrophilicpolymorphonuclear leukocytes (AZU1 azurocidin 1 ELANE(previous ELA2) elastase PRTN3 proteinase 3 and CTSGcathepsin G) second with upregulation of CD34 antigenreflecting an early maturation arrest and lack of granulocyticdifferentiation

AMLM3was extensively studied by Payton et al [63] whocompared the malignant promyelocytes from APL patientsto leukemic cells collected from other AML subtypes andto promyelocytes neutrophils and CD34+ cells extractedfrom healthy bone marrow donors The identified ldquoM3-specific dysregulomerdquo was composed of 510 genes and manyof them exhibited dramatic differences in expression levelcomparing to other AML subtypes or normal promyelocytesFor example GABRE FGF13 HGF ANXA8 and PGBD5were the most overexpressed genes whereasVNN1MS4A6AP2RY13 HK3 and S100A9 the most underexpressed genesin M3 vs other AMLs 33 genes selected from the identifiedsignature were validated by another high-throughput digitaltechnology (nCounter NanoString) capable of detecting aslittle as 05 fM of a specific mRNA and measuring up to500 genes in a multiplex reaction The authors demonstratednCounter reproducibility and applicability as a tool forbiomarker analysis when limited amounts of clinical materialare available 33 genes validated byNanoString assaywere alsoenriched in an independent AML dataset of 325 samples andAPL mouse model but notably not in a cell line expressingPML-RARA fusion gene

8 Journal of Oncology

One of the most impressive microarray-based studies wasthat of Haferlach et al [53] who analyzed almost 900 patientswith leukemia including 620 with AML with the use ofAffymetrix GeneChips and support vector machine (SVM)model The authors identified 13 separate leukemia typesincluding 6withinAML Someof them eg AMLwith t(821)and with t(1517) could be classified with 100 specificityand 100 sensitivity based on the expression profile of 100genes per groupThe overall prediction accuracy of 951wasachievedThemisclassification occurredmainly in subgroupswith a low sample number or high intragroup heterogeneityThe largest and the most heterogeneous subgroup was AMLwith normal karyotype which cosegregated with AML withless common cytogenetic aberrations classified as ldquootherrdquoHere AML samples with different fusion partners of KMT2Agene were included Haferlach et al [64] showed that APLwas not only distinct from other AML subtypes in the matterof gene expression but two M3 phenotypes one with heavygranulation and bundles of Auer rods and AML M3 variant(M3v) with non- or hypogranular cytoplasm and a bilobednucleus could be discriminated based on gene expressionsignatures

The largest GEP study in hematology and oncology wasconducted thanks to international collaboration within theEuropean Leukemia Net In 2010 Gene Expression ProfilingWorking Group directed by Torsten Haferlach published theresults of analysis of 3334 samples collected from leukemia(including 542 AMLs) and MDS patients by 11 laboratoriesacross three continents [65] Apart from European oneslaboratories from the United States and one from Singa-pore joined the program The main conclusion was GEPwas a robust technology for the diagnosis of hematologicmalignancies with high accuracy According to the authorsGEP had invaluable application potential and was vulnerableto standardization outperforming more subjective methodssuch as cytomorphology and metaphase cytogenetics Toenable better molecular understanding of leukemias theauthors deposited the collected data into a publicly availabledomain

In 2012 de la Bletiere et al [66] proved that AML cyto-genetic subtypes could be successfully determined with theuse of GEP even in samples with low leukemic blast contentor poor quality With the use of Illumina Expression Bead-Chips the authors first classified 71 good quality samplesfrom a training set representing APL t(821)-AML inv(16)-AML or NK-AML with at least 60 percent of leukemicblasts The optimal 40-marker gene classifier (10 markersper class including previously described as well as newlydiscovered genes) was applied to 111 suboptimalAML sampleswith low leukemic blast load (from 2 to 59) andor poorquality control criteria The overall error rate was 36 AllAPL and t(821) samples were correctly classified even thosecontaining as low as 2 percent blasts The worst result wasachieved for inv(16) Surprisingly poor sample quality didnot affect classification By the way de la Bletiere et al[66] demonstrated reliability robustness and sensitivity ofIllumina bead-based technology which seemed to be notworse than other commercially or academically developedmicroarray platforms used before

5 Between AML and ALL Acute Leukemiawith KMT2A Rearrangements

In 2002 Armstrong et al [67] showed that ALL withtranslocations involving the KMT2A gene (previously knownasMLL) presented a unique gene expression profile differentfrom ALL and AML without KMT2A abnormalities Thecore of this unique gene expression signature consisted ofmultilineage markers of early hematopoietic progenitors andHOX genes which corresponds with the fact that KMT2Agene encodes histone lysine methyltransferase a transcrip-tional coactivator regulating expression of genes (includ-ing HOX) during early development and hematopoiesisTherefore the authors proposed to distinguish a distinctleukemia entity termed ldquoMLLrdquo Then a common geneexpression signature enriched in homeobox genes (MEIS1HOXA4 HOXA5 HOXA7 HOXA9 and HIOXA10) wasdetermined for all acute leukemias with KMT2A fusionirrespectively of their lineage (myeloid or lymphoid) byRoss et al [68] Similarly Andersson et al [69] associatedchildhood acute leukemias with KMT2A rearrangementswith upregulation of homeobox genes (HOXA10 HOXA4MEIS1 and PBX3) In their study KMT2A-positive AMLswere also enriched in genes involved in cell communica-tion and adhesion whereas some antiapoptotic genes (ega tumor necrosis factor receptor TNFRSF21) and tumorsuppressor genes (BRCA1 DLC1) were downregulated inthis AML subtype Hierarchical clustering with a subset ofgenes encoding transcription factors showed that leukemicsamples with KMT2A rearrangements grouped togetherindependently on lineage Although KMT2A translocationsare prevalent in infant and treatment-related leukemiasthey also occur in adult leukemias that were studied byKohlmann et al (2005) [70] who wondered how the differingKMT2A partner genes influenced the global gene expres-sion signature and whether pathways could be identified toexplain the molecular determination of KMT2A leukemiasof both lineages The data analysis in both types of acuteleukemias revealed t(11q23)KMT2A-positive samples thatwere evidently distinct from other subtypes of the samelineage As in the case of childhood leukemia adult KMT2A-AML and KMT2A-ALL despite a shared common geneprofile revealed also lineage-specific expression markerssufficient to segregate them according to their lineageswith no respect to the KMT2A fusion partner The com-monly overexpressed genes were obviously the homeoboxgenes and their regulators (HOXA9 MEIS1 HOXA10 PBX3HOXA3 HOXA4 HOXA5 HOXA7) NICAL gene (presentMICAL encoding Microtubule Associated Monooxygenase)RUNX2 transcription factor and FLT3 gene The commondownregulated genes included TNF-receptor superfamilymembers (TNFRSF10A and TNFRSF10D) transcription fac-tor POU4F1 tumor suppressor ST18 or MADH1 (presentSMAD1) encoding a signal transducer and transcriptionalmodulator Comparing to KMT2A-ALL overexpression ofCEBPB CEBPA KITMADH2MITF FES and SPI1 (formerPU1) oncogenes and MNDA encoding the myeloid cellnuclear differentiation was noted in KMT2A-AML Sum-marizing their results the authors concluded AML with

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

6 Journal of Oncology

Table 1 Continued

GEP gene expression profilingHH HedgehogHSCs hematopoietic stem cellsHSPCs hematopoietic stem-progenitor cellslncRNAs long noncoding RNAsLSCs leukemic stem cellsMDS myelodysplastic syndromesMRD minimal residual diseaseMSC mesenchymal stem cellsNGS next generation sequencingNK-AML normal karyotype AMLNPMc+ NPM-cytoplasmic positiveOS overall survivalPB peripheral bloodPBMCs peripheral blood mononuclear cellsSAGE serial analysis of gene expressionsnoRNAs small nucleolar RNAsTCGA the Cancer Genome AtlasTEs transposable elementsWBC white blood cellWHO World Health Organization

showed upregulation of antiapoptotic gene DAD1 Thereforethe authors suggested different mechanisms of cell deathescape for the two studied leukemia types and associated itwith AML +8 resistance to cytarabine-based chemotherapywhich should induce apoptosis [49]

AML subtypes with three reciprocal rearrangementst(821)(q22q22) inv(16)(p13q22) and (1517)(q22q12) cor-responding to the morphological FAB subtypes M2 M4eoand M3M3v respectively were the subject of research ofSchoch et al [56] Principal component analysis (PCA)of microarray data performed with the use of 1000 mostinformative genes clearly separated AML samples accordingto chromosomal aberration The minimal set of 13 genes(PRKAR1B GNAI1 PRODH CD52 KRT18 CLIP3 CLUPTGDS HOXB2 CLEC2B CTSW S100A9 and MYH11) wassufficient to distinguish one AML subtype from another onthe basis of gene expression solely Expression levels of 36genes enabled accurate classification of all three studied AMLsubtypes Another set of 82 genes allowed for distinction ofM3 and M3v two phenotypically different AML types witht(1517) In addition the study showed that AMLs with alter-ations involved core binding factor (CBF) complex t(821)and inv(16) were more related to each other than to AMLwith t(1517) The authors explained the overexpression ofMYH11 in inv(16) and RUNX1T1 (former ETO) in t(821) as aconsequence of high expression of fusion transcripts affectingthese genes A new marker of t(821) was identified byDebernardi et al [57] who found that a level of transcriptionfactor-coding gene POU4F1 (former BRN3A) was 43-foldhigher in t(821) AML than in other AML samples

Verhaak et al [58] by gene expression analysis intwo independent cohorts of AML patients under 60 eachexceeding 200 cases perfectly distinguished three favorablecytogenetic AML subtypes t(821) t(1517) and inv(16) ForAML with NPM1 or CEBPAmutations GEP-based classifierswere less accurate The distinction of AML with othermutations (eg FLT3 and RAS) and aberrations (11q23 -55q- -77q- abn3q and t(922)) was not possible with the use ofGEP Nevertheless for abn3q the most discriminative genewas MECOM encoding an oncogenic transcription factoroften involved in 3q26 abnormalities and in 7(q) almost alldifferentially decreased genes were located on chromosome7

One of the most challenging tasks was to find uniquefeatures characterizing cytogenetically normal AML (CN-AML or NK-AML from normal karyotype AML) whichaccounts for 40-50 of all AML cases Debernardi et al[57] were the first who attempted to do that Althoughthe sample size was not large (28 adult AML samplesincluding 10 NK-AML) and NK-AML revealed higher vari-ability than AML with translocations the authors found NK-AML could be separated from AML samples with chro-mosomal rearrangements based on the expression levelsof certain members of the class I homeobox A and Bgene families which were low or undetectable in AMLwith (t(821) t(1517) and inv(16)) In NK-AML expres-sion level of 10 genes was extremely increased HOXA4HOXA5 HOXA9 HOXB2 HOXB3 HOXB5 HOXB6 andHOXB7 and two members of TALE family MEIS1 andPBX3 While overexpression of HOXB genes was unique for

Journal of Oncology 7

NK-AML the upregulation of the remaining 5 genes wasshared with 11q23 AMLwhereMLL (mixed lineage leukemia)gene now renamed to KMT2A (Lysine Methyltransferase2A) was fused with different partners High expressionof some homeobox genes (eg HOXA3 and HOXB6) waslater found as typical of hematopoietic stem cells (HSCs)[59]

In 2004 two remarkable papers identifying not onlyknown but also newmolecular AML subtypes through globalGEP were published in the same issue of the New EnglandJournal of Medicine [60 61] Bullinger et al (2004) [60]who analyzed 116 adult AML samples with the use of cDNAmicroarrays found that hierarchical clustering with over6 thousands of the most varied genes divided all AMLsamples into two main clusters Out of the cytogeneticgroups only t(1517) (APL) generated one condense sub-cluster To enable biological insight into AML pathogenesisgroup-specific gene expression signatures were establishedand functionally characterized Signature specific for APLincluded genes related to hemostasis (PLAU SERPING1ANXA8 and PLAUR) resistance to apoptosis (TNFRSF4AVEN and BIRC5) impairment of retinoic acid-stimulatedcell differentiation (TBL1X CALR and RARRES3) and resis-tance to chemotherapy (CYP2E1 EPHX1 and a group of met-allothionein (MT) genes) MLLT4 (present AFDN Afadin)one of KMT2A fusion partners was among the genes withunique expression profile in t(821) which suggested similarmechanism of pathogenesis with t(611) High expression ofNT5E observed in inv(16) was correlated with resistanceto cytarabine Interestingly high expression of homeoboxgenes (HOXA4 HOXA9 HOXA10 PBX3 and MEIS1) wasdetected in AML specimens with not only normal butalso complex karyotypes Within NK-AML Bullinger et al[60] distinguished two distinct groups one where FLT3aberrations and FAB subtypes M1 and M2 prevailed andthe second one where FAB M4 and M5 subtypes weremore common Of note patients classified to those groupshad different outcomes For AML with complex karyotypeAML with KMT2A partial tandem duplications and AML+8 it was impossible to find statistically significant uniquegene expression signatures Valk et al [61] who analyzed285 AML patients using Affymetrix GeneChips identified16 AML groups with distinct gene expression profiles Someof them were composed of AML samples with known cyto-genetic aberrations t(1821) t(1517) and inv(16) RUNX1T1gene which is a RUNX1 fusion partner was the mostdiscriminative gene for AML with t(821) Overexpressionof MYH11 was the most discriminative feature of inv(16)which produces CBFB-MYH11 fusion gene Simultaneousdownregulation of CBFB observed in this subtype couldbe explained by eg negative regulation of a wild-type(wt) CBFB allele by the fusion transcript For APL growthfactor-coding genes were the most discriminative (hepato-cyte growth factor (HGF) macrophage-stimulating 1 growthfactor (MST1) and fibroblast growth factor 13 (FGF13))However AMLs with 11q23 were segregated into two separateclusters and partially scattered among all samples studiedAlsoNK-AML sampleswere divided into several clustersTheobserved heterogeneity could be at least partially explained

by the presence of particular mutations and different out-comes

Further AML transcriptome studies performed on inde-pendent patient cohorts usually confirmed earlier researchreporting partially overlapping gene expression signaturesHowever each study delivered a portion of new informationwhich deepened our knowledge about AML pathogenesisGutierrez et al [62] performed hierarchical clustering ofBM samples from 43 adult AML patients based on theexpression of over 5 thousand genes Four distinct clustersthey obtained corresponded to AMLwith inv(16) monocyticAML APL and other AML samples which included NK-AML The authors developed a minimal 21-gene predictorwhich classified each sample to appropriate group with100 accuracy Its efficiency was then confirmed with anindependent AML sample set APL samples which formedthemost condense group among all samples studied revealedhigh expression of several growth factors and other sig-naling proteins eg HGF FGF13 MST1 VEGFA IGFBP2and FGFR1 Contrary overexpression of HOX family mem-bers (A5 A6 A7 A9 A10 B2 B5 and B7) includinggenes encoding TALE proteins (MEIS1 PBX3) and histoneproteins was shared by all non-APL leukemias Increasedlevel of MYH11 expression and downregulation of CBFBand RUNX3 genes were noted specifically for AML withinv(16) In monocytic leukemia CSPG2 other adhesionmolecules such as the lectins CLECSF6 CLECSF12 SIGLEC7and FCN1 were upregulated compared to remaining AMLsThe remaining AML samples presented more heterogeneitywhich was reflected by the existence of two subclustersone with overexpression of genes encoding hematopoieticserine proteases present in azurophil granules of neutrophilicpolymorphonuclear leukocytes (AZU1 azurocidin 1 ELANE(previous ELA2) elastase PRTN3 proteinase 3 and CTSGcathepsin G) second with upregulation of CD34 antigenreflecting an early maturation arrest and lack of granulocyticdifferentiation

AMLM3was extensively studied by Payton et al [63] whocompared the malignant promyelocytes from APL patientsto leukemic cells collected from other AML subtypes andto promyelocytes neutrophils and CD34+ cells extractedfrom healthy bone marrow donors The identified ldquoM3-specific dysregulomerdquo was composed of 510 genes and manyof them exhibited dramatic differences in expression levelcomparing to other AML subtypes or normal promyelocytesFor example GABRE FGF13 HGF ANXA8 and PGBD5were the most overexpressed genes whereasVNN1MS4A6AP2RY13 HK3 and S100A9 the most underexpressed genesin M3 vs other AMLs 33 genes selected from the identifiedsignature were validated by another high-throughput digitaltechnology (nCounter NanoString) capable of detecting aslittle as 05 fM of a specific mRNA and measuring up to500 genes in a multiplex reaction The authors demonstratednCounter reproducibility and applicability as a tool forbiomarker analysis when limited amounts of clinical materialare available 33 genes validated byNanoString assaywere alsoenriched in an independent AML dataset of 325 samples andAPL mouse model but notably not in a cell line expressingPML-RARA fusion gene

8 Journal of Oncology

One of the most impressive microarray-based studies wasthat of Haferlach et al [53] who analyzed almost 900 patientswith leukemia including 620 with AML with the use ofAffymetrix GeneChips and support vector machine (SVM)model The authors identified 13 separate leukemia typesincluding 6withinAML Someof them eg AMLwith t(821)and with t(1517) could be classified with 100 specificityand 100 sensitivity based on the expression profile of 100genes per groupThe overall prediction accuracy of 951wasachievedThemisclassification occurredmainly in subgroupswith a low sample number or high intragroup heterogeneityThe largest and the most heterogeneous subgroup was AMLwith normal karyotype which cosegregated with AML withless common cytogenetic aberrations classified as ldquootherrdquoHere AML samples with different fusion partners of KMT2Agene were included Haferlach et al [64] showed that APLwas not only distinct from other AML subtypes in the matterof gene expression but two M3 phenotypes one with heavygranulation and bundles of Auer rods and AML M3 variant(M3v) with non- or hypogranular cytoplasm and a bilobednucleus could be discriminated based on gene expressionsignatures

The largest GEP study in hematology and oncology wasconducted thanks to international collaboration within theEuropean Leukemia Net In 2010 Gene Expression ProfilingWorking Group directed by Torsten Haferlach published theresults of analysis of 3334 samples collected from leukemia(including 542 AMLs) and MDS patients by 11 laboratoriesacross three continents [65] Apart from European oneslaboratories from the United States and one from Singa-pore joined the program The main conclusion was GEPwas a robust technology for the diagnosis of hematologicmalignancies with high accuracy According to the authorsGEP had invaluable application potential and was vulnerableto standardization outperforming more subjective methodssuch as cytomorphology and metaphase cytogenetics Toenable better molecular understanding of leukemias theauthors deposited the collected data into a publicly availabledomain

In 2012 de la Bletiere et al [66] proved that AML cyto-genetic subtypes could be successfully determined with theuse of GEP even in samples with low leukemic blast contentor poor quality With the use of Illumina Expression Bead-Chips the authors first classified 71 good quality samplesfrom a training set representing APL t(821)-AML inv(16)-AML or NK-AML with at least 60 percent of leukemicblasts The optimal 40-marker gene classifier (10 markersper class including previously described as well as newlydiscovered genes) was applied to 111 suboptimalAML sampleswith low leukemic blast load (from 2 to 59) andor poorquality control criteria The overall error rate was 36 AllAPL and t(821) samples were correctly classified even thosecontaining as low as 2 percent blasts The worst result wasachieved for inv(16) Surprisingly poor sample quality didnot affect classification By the way de la Bletiere et al[66] demonstrated reliability robustness and sensitivity ofIllumina bead-based technology which seemed to be notworse than other commercially or academically developedmicroarray platforms used before

5 Between AML and ALL Acute Leukemiawith KMT2A Rearrangements

In 2002 Armstrong et al [67] showed that ALL withtranslocations involving the KMT2A gene (previously knownasMLL) presented a unique gene expression profile differentfrom ALL and AML without KMT2A abnormalities Thecore of this unique gene expression signature consisted ofmultilineage markers of early hematopoietic progenitors andHOX genes which corresponds with the fact that KMT2Agene encodes histone lysine methyltransferase a transcrip-tional coactivator regulating expression of genes (includ-ing HOX) during early development and hematopoiesisTherefore the authors proposed to distinguish a distinctleukemia entity termed ldquoMLLrdquo Then a common geneexpression signature enriched in homeobox genes (MEIS1HOXA4 HOXA5 HOXA7 HOXA9 and HIOXA10) wasdetermined for all acute leukemias with KMT2A fusionirrespectively of their lineage (myeloid or lymphoid) byRoss et al [68] Similarly Andersson et al [69] associatedchildhood acute leukemias with KMT2A rearrangementswith upregulation of homeobox genes (HOXA10 HOXA4MEIS1 and PBX3) In their study KMT2A-positive AMLswere also enriched in genes involved in cell communica-tion and adhesion whereas some antiapoptotic genes (ega tumor necrosis factor receptor TNFRSF21) and tumorsuppressor genes (BRCA1 DLC1) were downregulated inthis AML subtype Hierarchical clustering with a subset ofgenes encoding transcription factors showed that leukemicsamples with KMT2A rearrangements grouped togetherindependently on lineage Although KMT2A translocationsare prevalent in infant and treatment-related leukemiasthey also occur in adult leukemias that were studied byKohlmann et al (2005) [70] who wondered how the differingKMT2A partner genes influenced the global gene expres-sion signature and whether pathways could be identified toexplain the molecular determination of KMT2A leukemiasof both lineages The data analysis in both types of acuteleukemias revealed t(11q23)KMT2A-positive samples thatwere evidently distinct from other subtypes of the samelineage As in the case of childhood leukemia adult KMT2A-AML and KMT2A-ALL despite a shared common geneprofile revealed also lineage-specific expression markerssufficient to segregate them according to their lineageswith no respect to the KMT2A fusion partner The com-monly overexpressed genes were obviously the homeoboxgenes and their regulators (HOXA9 MEIS1 HOXA10 PBX3HOXA3 HOXA4 HOXA5 HOXA7) NICAL gene (presentMICAL encoding Microtubule Associated Monooxygenase)RUNX2 transcription factor and FLT3 gene The commondownregulated genes included TNF-receptor superfamilymembers (TNFRSF10A and TNFRSF10D) transcription fac-tor POU4F1 tumor suppressor ST18 or MADH1 (presentSMAD1) encoding a signal transducer and transcriptionalmodulator Comparing to KMT2A-ALL overexpression ofCEBPB CEBPA KITMADH2MITF FES and SPI1 (formerPU1) oncogenes and MNDA encoding the myeloid cellnuclear differentiation was noted in KMT2A-AML Sum-marizing their results the authors concluded AML with

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

Journal of Oncology 7

NK-AML the upregulation of the remaining 5 genes wasshared with 11q23 AMLwhereMLL (mixed lineage leukemia)gene now renamed to KMT2A (Lysine Methyltransferase2A) was fused with different partners High expressionof some homeobox genes (eg HOXA3 and HOXB6) waslater found as typical of hematopoietic stem cells (HSCs)[59]

In 2004 two remarkable papers identifying not onlyknown but also newmolecular AML subtypes through globalGEP were published in the same issue of the New EnglandJournal of Medicine [60 61] Bullinger et al (2004) [60]who analyzed 116 adult AML samples with the use of cDNAmicroarrays found that hierarchical clustering with over6 thousands of the most varied genes divided all AMLsamples into two main clusters Out of the cytogeneticgroups only t(1517) (APL) generated one condense sub-cluster To enable biological insight into AML pathogenesisgroup-specific gene expression signatures were establishedand functionally characterized Signature specific for APLincluded genes related to hemostasis (PLAU SERPING1ANXA8 and PLAUR) resistance to apoptosis (TNFRSF4AVEN and BIRC5) impairment of retinoic acid-stimulatedcell differentiation (TBL1X CALR and RARRES3) and resis-tance to chemotherapy (CYP2E1 EPHX1 and a group of met-allothionein (MT) genes) MLLT4 (present AFDN Afadin)one of KMT2A fusion partners was among the genes withunique expression profile in t(821) which suggested similarmechanism of pathogenesis with t(611) High expression ofNT5E observed in inv(16) was correlated with resistanceto cytarabine Interestingly high expression of homeoboxgenes (HOXA4 HOXA9 HOXA10 PBX3 and MEIS1) wasdetected in AML specimens with not only normal butalso complex karyotypes Within NK-AML Bullinger et al[60] distinguished two distinct groups one where FLT3aberrations and FAB subtypes M1 and M2 prevailed andthe second one where FAB M4 and M5 subtypes weremore common Of note patients classified to those groupshad different outcomes For AML with complex karyotypeAML with KMT2A partial tandem duplications and AML+8 it was impossible to find statistically significant uniquegene expression signatures Valk et al [61] who analyzed285 AML patients using Affymetrix GeneChips identified16 AML groups with distinct gene expression profiles Someof them were composed of AML samples with known cyto-genetic aberrations t(1821) t(1517) and inv(16) RUNX1T1gene which is a RUNX1 fusion partner was the mostdiscriminative gene for AML with t(821) Overexpressionof MYH11 was the most discriminative feature of inv(16)which produces CBFB-MYH11 fusion gene Simultaneousdownregulation of CBFB observed in this subtype couldbe explained by eg negative regulation of a wild-type(wt) CBFB allele by the fusion transcript For APL growthfactor-coding genes were the most discriminative (hepato-cyte growth factor (HGF) macrophage-stimulating 1 growthfactor (MST1) and fibroblast growth factor 13 (FGF13))However AMLs with 11q23 were segregated into two separateclusters and partially scattered among all samples studiedAlsoNK-AML sampleswere divided into several clustersTheobserved heterogeneity could be at least partially explained

by the presence of particular mutations and different out-comes

Further AML transcriptome studies performed on inde-pendent patient cohorts usually confirmed earlier researchreporting partially overlapping gene expression signaturesHowever each study delivered a portion of new informationwhich deepened our knowledge about AML pathogenesisGutierrez et al [62] performed hierarchical clustering ofBM samples from 43 adult AML patients based on theexpression of over 5 thousand genes Four distinct clustersthey obtained corresponded to AMLwith inv(16) monocyticAML APL and other AML samples which included NK-AML The authors developed a minimal 21-gene predictorwhich classified each sample to appropriate group with100 accuracy Its efficiency was then confirmed with anindependent AML sample set APL samples which formedthemost condense group among all samples studied revealedhigh expression of several growth factors and other sig-naling proteins eg HGF FGF13 MST1 VEGFA IGFBP2and FGFR1 Contrary overexpression of HOX family mem-bers (A5 A6 A7 A9 A10 B2 B5 and B7) includinggenes encoding TALE proteins (MEIS1 PBX3) and histoneproteins was shared by all non-APL leukemias Increasedlevel of MYH11 expression and downregulation of CBFBand RUNX3 genes were noted specifically for AML withinv(16) In monocytic leukemia CSPG2 other adhesionmolecules such as the lectins CLECSF6 CLECSF12 SIGLEC7and FCN1 were upregulated compared to remaining AMLsThe remaining AML samples presented more heterogeneitywhich was reflected by the existence of two subclustersone with overexpression of genes encoding hematopoieticserine proteases present in azurophil granules of neutrophilicpolymorphonuclear leukocytes (AZU1 azurocidin 1 ELANE(previous ELA2) elastase PRTN3 proteinase 3 and CTSGcathepsin G) second with upregulation of CD34 antigenreflecting an early maturation arrest and lack of granulocyticdifferentiation

AMLM3was extensively studied by Payton et al [63] whocompared the malignant promyelocytes from APL patientsto leukemic cells collected from other AML subtypes andto promyelocytes neutrophils and CD34+ cells extractedfrom healthy bone marrow donors The identified ldquoM3-specific dysregulomerdquo was composed of 510 genes and manyof them exhibited dramatic differences in expression levelcomparing to other AML subtypes or normal promyelocytesFor example GABRE FGF13 HGF ANXA8 and PGBD5were the most overexpressed genes whereasVNN1MS4A6AP2RY13 HK3 and S100A9 the most underexpressed genesin M3 vs other AMLs 33 genes selected from the identifiedsignature were validated by another high-throughput digitaltechnology (nCounter NanoString) capable of detecting aslittle as 05 fM of a specific mRNA and measuring up to500 genes in a multiplex reaction The authors demonstratednCounter reproducibility and applicability as a tool forbiomarker analysis when limited amounts of clinical materialare available 33 genes validated byNanoString assaywere alsoenriched in an independent AML dataset of 325 samples andAPL mouse model but notably not in a cell line expressingPML-RARA fusion gene

8 Journal of Oncology

One of the most impressive microarray-based studies wasthat of Haferlach et al [53] who analyzed almost 900 patientswith leukemia including 620 with AML with the use ofAffymetrix GeneChips and support vector machine (SVM)model The authors identified 13 separate leukemia typesincluding 6withinAML Someof them eg AMLwith t(821)and with t(1517) could be classified with 100 specificityand 100 sensitivity based on the expression profile of 100genes per groupThe overall prediction accuracy of 951wasachievedThemisclassification occurredmainly in subgroupswith a low sample number or high intragroup heterogeneityThe largest and the most heterogeneous subgroup was AMLwith normal karyotype which cosegregated with AML withless common cytogenetic aberrations classified as ldquootherrdquoHere AML samples with different fusion partners of KMT2Agene were included Haferlach et al [64] showed that APLwas not only distinct from other AML subtypes in the matterof gene expression but two M3 phenotypes one with heavygranulation and bundles of Auer rods and AML M3 variant(M3v) with non- or hypogranular cytoplasm and a bilobednucleus could be discriminated based on gene expressionsignatures

The largest GEP study in hematology and oncology wasconducted thanks to international collaboration within theEuropean Leukemia Net In 2010 Gene Expression ProfilingWorking Group directed by Torsten Haferlach published theresults of analysis of 3334 samples collected from leukemia(including 542 AMLs) and MDS patients by 11 laboratoriesacross three continents [65] Apart from European oneslaboratories from the United States and one from Singa-pore joined the program The main conclusion was GEPwas a robust technology for the diagnosis of hematologicmalignancies with high accuracy According to the authorsGEP had invaluable application potential and was vulnerableto standardization outperforming more subjective methodssuch as cytomorphology and metaphase cytogenetics Toenable better molecular understanding of leukemias theauthors deposited the collected data into a publicly availabledomain

In 2012 de la Bletiere et al [66] proved that AML cyto-genetic subtypes could be successfully determined with theuse of GEP even in samples with low leukemic blast contentor poor quality With the use of Illumina Expression Bead-Chips the authors first classified 71 good quality samplesfrom a training set representing APL t(821)-AML inv(16)-AML or NK-AML with at least 60 percent of leukemicblasts The optimal 40-marker gene classifier (10 markersper class including previously described as well as newlydiscovered genes) was applied to 111 suboptimalAML sampleswith low leukemic blast load (from 2 to 59) andor poorquality control criteria The overall error rate was 36 AllAPL and t(821) samples were correctly classified even thosecontaining as low as 2 percent blasts The worst result wasachieved for inv(16) Surprisingly poor sample quality didnot affect classification By the way de la Bletiere et al[66] demonstrated reliability robustness and sensitivity ofIllumina bead-based technology which seemed to be notworse than other commercially or academically developedmicroarray platforms used before

5 Between AML and ALL Acute Leukemiawith KMT2A Rearrangements

In 2002 Armstrong et al [67] showed that ALL withtranslocations involving the KMT2A gene (previously knownasMLL) presented a unique gene expression profile differentfrom ALL and AML without KMT2A abnormalities Thecore of this unique gene expression signature consisted ofmultilineage markers of early hematopoietic progenitors andHOX genes which corresponds with the fact that KMT2Agene encodes histone lysine methyltransferase a transcrip-tional coactivator regulating expression of genes (includ-ing HOX) during early development and hematopoiesisTherefore the authors proposed to distinguish a distinctleukemia entity termed ldquoMLLrdquo Then a common geneexpression signature enriched in homeobox genes (MEIS1HOXA4 HOXA5 HOXA7 HOXA9 and HIOXA10) wasdetermined for all acute leukemias with KMT2A fusionirrespectively of their lineage (myeloid or lymphoid) byRoss et al [68] Similarly Andersson et al [69] associatedchildhood acute leukemias with KMT2A rearrangementswith upregulation of homeobox genes (HOXA10 HOXA4MEIS1 and PBX3) In their study KMT2A-positive AMLswere also enriched in genes involved in cell communica-tion and adhesion whereas some antiapoptotic genes (ega tumor necrosis factor receptor TNFRSF21) and tumorsuppressor genes (BRCA1 DLC1) were downregulated inthis AML subtype Hierarchical clustering with a subset ofgenes encoding transcription factors showed that leukemicsamples with KMT2A rearrangements grouped togetherindependently on lineage Although KMT2A translocationsare prevalent in infant and treatment-related leukemiasthey also occur in adult leukemias that were studied byKohlmann et al (2005) [70] who wondered how the differingKMT2A partner genes influenced the global gene expres-sion signature and whether pathways could be identified toexplain the molecular determination of KMT2A leukemiasof both lineages The data analysis in both types of acuteleukemias revealed t(11q23)KMT2A-positive samples thatwere evidently distinct from other subtypes of the samelineage As in the case of childhood leukemia adult KMT2A-AML and KMT2A-ALL despite a shared common geneprofile revealed also lineage-specific expression markerssufficient to segregate them according to their lineageswith no respect to the KMT2A fusion partner The com-monly overexpressed genes were obviously the homeoboxgenes and their regulators (HOXA9 MEIS1 HOXA10 PBX3HOXA3 HOXA4 HOXA5 HOXA7) NICAL gene (presentMICAL encoding Microtubule Associated Monooxygenase)RUNX2 transcription factor and FLT3 gene The commondownregulated genes included TNF-receptor superfamilymembers (TNFRSF10A and TNFRSF10D) transcription fac-tor POU4F1 tumor suppressor ST18 or MADH1 (presentSMAD1) encoding a signal transducer and transcriptionalmodulator Comparing to KMT2A-ALL overexpression ofCEBPB CEBPA KITMADH2MITF FES and SPI1 (formerPU1) oncogenes and MNDA encoding the myeloid cellnuclear differentiation was noted in KMT2A-AML Sum-marizing their results the authors concluded AML with

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

8 Journal of Oncology

One of the most impressive microarray-based studies wasthat of Haferlach et al [53] who analyzed almost 900 patientswith leukemia including 620 with AML with the use ofAffymetrix GeneChips and support vector machine (SVM)model The authors identified 13 separate leukemia typesincluding 6withinAML Someof them eg AMLwith t(821)and with t(1517) could be classified with 100 specificityand 100 sensitivity based on the expression profile of 100genes per groupThe overall prediction accuracy of 951wasachievedThemisclassification occurredmainly in subgroupswith a low sample number or high intragroup heterogeneityThe largest and the most heterogeneous subgroup was AMLwith normal karyotype which cosegregated with AML withless common cytogenetic aberrations classified as ldquootherrdquoHere AML samples with different fusion partners of KMT2Agene were included Haferlach et al [64] showed that APLwas not only distinct from other AML subtypes in the matterof gene expression but two M3 phenotypes one with heavygranulation and bundles of Auer rods and AML M3 variant(M3v) with non- or hypogranular cytoplasm and a bilobednucleus could be discriminated based on gene expressionsignatures

The largest GEP study in hematology and oncology wasconducted thanks to international collaboration within theEuropean Leukemia Net In 2010 Gene Expression ProfilingWorking Group directed by Torsten Haferlach published theresults of analysis of 3334 samples collected from leukemia(including 542 AMLs) and MDS patients by 11 laboratoriesacross three continents [65] Apart from European oneslaboratories from the United States and one from Singa-pore joined the program The main conclusion was GEPwas a robust technology for the diagnosis of hematologicmalignancies with high accuracy According to the authorsGEP had invaluable application potential and was vulnerableto standardization outperforming more subjective methodssuch as cytomorphology and metaphase cytogenetics Toenable better molecular understanding of leukemias theauthors deposited the collected data into a publicly availabledomain

In 2012 de la Bletiere et al [66] proved that AML cyto-genetic subtypes could be successfully determined with theuse of GEP even in samples with low leukemic blast contentor poor quality With the use of Illumina Expression Bead-Chips the authors first classified 71 good quality samplesfrom a training set representing APL t(821)-AML inv(16)-AML or NK-AML with at least 60 percent of leukemicblasts The optimal 40-marker gene classifier (10 markersper class including previously described as well as newlydiscovered genes) was applied to 111 suboptimalAML sampleswith low leukemic blast load (from 2 to 59) andor poorquality control criteria The overall error rate was 36 AllAPL and t(821) samples were correctly classified even thosecontaining as low as 2 percent blasts The worst result wasachieved for inv(16) Surprisingly poor sample quality didnot affect classification By the way de la Bletiere et al[66] demonstrated reliability robustness and sensitivity ofIllumina bead-based technology which seemed to be notworse than other commercially or academically developedmicroarray platforms used before

5 Between AML and ALL Acute Leukemiawith KMT2A Rearrangements

In 2002 Armstrong et al [67] showed that ALL withtranslocations involving the KMT2A gene (previously knownasMLL) presented a unique gene expression profile differentfrom ALL and AML without KMT2A abnormalities Thecore of this unique gene expression signature consisted ofmultilineage markers of early hematopoietic progenitors andHOX genes which corresponds with the fact that KMT2Agene encodes histone lysine methyltransferase a transcrip-tional coactivator regulating expression of genes (includ-ing HOX) during early development and hematopoiesisTherefore the authors proposed to distinguish a distinctleukemia entity termed ldquoMLLrdquo Then a common geneexpression signature enriched in homeobox genes (MEIS1HOXA4 HOXA5 HOXA7 HOXA9 and HIOXA10) wasdetermined for all acute leukemias with KMT2A fusionirrespectively of their lineage (myeloid or lymphoid) byRoss et al [68] Similarly Andersson et al [69] associatedchildhood acute leukemias with KMT2A rearrangementswith upregulation of homeobox genes (HOXA10 HOXA4MEIS1 and PBX3) In their study KMT2A-positive AMLswere also enriched in genes involved in cell communica-tion and adhesion whereas some antiapoptotic genes (ega tumor necrosis factor receptor TNFRSF21) and tumorsuppressor genes (BRCA1 DLC1) were downregulated inthis AML subtype Hierarchical clustering with a subset ofgenes encoding transcription factors showed that leukemicsamples with KMT2A rearrangements grouped togetherindependently on lineage Although KMT2A translocationsare prevalent in infant and treatment-related leukemiasthey also occur in adult leukemias that were studied byKohlmann et al (2005) [70] who wondered how the differingKMT2A partner genes influenced the global gene expres-sion signature and whether pathways could be identified toexplain the molecular determination of KMT2A leukemiasof both lineages The data analysis in both types of acuteleukemias revealed t(11q23)KMT2A-positive samples thatwere evidently distinct from other subtypes of the samelineage As in the case of childhood leukemia adult KMT2A-AML and KMT2A-ALL despite a shared common geneprofile revealed also lineage-specific expression markerssufficient to segregate them according to their lineageswith no respect to the KMT2A fusion partner The com-monly overexpressed genes were obviously the homeoboxgenes and their regulators (HOXA9 MEIS1 HOXA10 PBX3HOXA3 HOXA4 HOXA5 HOXA7) NICAL gene (presentMICAL encoding Microtubule Associated Monooxygenase)RUNX2 transcription factor and FLT3 gene The commondownregulated genes included TNF-receptor superfamilymembers (TNFRSF10A and TNFRSF10D) transcription fac-tor POU4F1 tumor suppressor ST18 or MADH1 (presentSMAD1) encoding a signal transducer and transcriptionalmodulator Comparing to KMT2A-ALL overexpression ofCEBPB CEBPA KITMADH2MITF FES and SPI1 (formerPU1) oncogenes and MNDA encoding the myeloid cellnuclear differentiation was noted in KMT2A-AML Sum-marizing their results the authors concluded AML with

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

Journal of Oncology 9

t(11q23)KMT2A and ALL with t(11q23)KMT2A are ratherdistinct entities

6 AML Risk Classification andOutcome Prediction

AML chemotherapy does not always lead to complete remis-sion (CR) 20-50 AML patients are primarily resistant toinduction therapy Having this information at the time ofdiagnosis would facilitate treatment decision making Takinginto account the success of GEP in AML diagnosis andclassification to particular disease subtypes its application inprognosis predictionwas only amatter of time Correlation ofHOXA9 upregulation with poor AML outcome was reportedby Golub et al in their first microarray paper devoted toALL and AML classification [20] Later Andreeff et al [71]demonstrated that many AML cases with intermediate andadverse prognosis presented HOX expression levels similarto the levels observed in normal CD34+ InterestinglyHOXAgenes could distinguish favorable vs unfavorable cases butonlyHOXB genes effectively distinguished intermediate fromunfavorable AMLs Despite the high coordination in HOXgene family expression HOXA9 seemed to be the best singlepredictor of overall survival (OS) disease-free survival (DFS)and response to therapy confirming earlier results of Golubet al [20]

AML outcome prediction was often matched with AMLclassification Analysis of AML GEP-based clusters definedby Valk et al [61] in the context of prognosis showed thatthree clusters overlapping with inv(16) t(1517) and t(1821)were associated with good outcome Patients classified to thecluster that common feature was MECOM overexpressionhad clearly worse outcome

Two prognostically different NK-AML subgroups werealso identified by Bullinger et al [60] A group with predom-inance of FLT3 aberrations and FAB subtypes M1 and M2presented shorter OS High expression of GATA2 NOTCH1DNMT3A andDNMT3B in this group suggested pathologicalimpact of aberrant methylation Genes deregulated in thesecond group where FAB M4 and M5 subtypes were morecommon were associated with granulocytic and monocyticdifferentiation immune response and hematopoietic stem-cell survival (VEGF) Analysis of prognostically relevantgenes led to the identification of 133-gene based prognosticsignature FOXO1A encoding transcription factor involvedin cell cycle arrest and apoptosis regulation was one of thegenes correlated with favorable outcome Poor outcome wasdetermined by overexpression of HOX genes and FLT3 genePrognostic gene expression signature proposed by Bullingeret al [60] was 2 years later applied to an independent cohortof 64 NK-AML patients below the age of 60 by Radmacheret al (2006) [72] GEP of the new sample set performedwith Affymertix GeneChips different array technology thanone that was used to establish the prognostic signatureallowed segregation of patients into 2 clusters with signifi-cantly different OS and DFS Strong association between theoutcome classification and FLT3-ITD status was observed67 patients with poor outcome were FLT3-ITD-positive

However FLT3-ITD was present in almost 20 of patientsfrom the good-outcome class which indicated contributionof other prognostic determinants Nevertheless Radmacheret al [72] not only validated the previous prognostic sig-nature but also developed a well-defined classifier whichmight be applied to individual patients with best accuracyto patients with normal cytogenetics and wt FLT3

Application of gene expressionmicroarrays for predictionof patient sensitivity to therapy was also demonstrated byHeuser et al [73] and Tagliafico et al [74] Heuser et al[73] identified gene expression profile that distinguishedAML M0-M5 (excluding M3) patients with good or poorresponses Hierarchical clustering performed on a trainingset of 33 AML samples divided good responders into twoclusters which suggested the effect of determinants otherthan treatment Interestingly samples with the lowest levelof myeloid cell maturation corresponding to FAB subtypesM0 and M1 were equally distributed between clusters rep-resenting good and poor response Over 30 of poor-response-associated genes eg MN1 FHL1 CD34 RBPMSLPAR6 and FLJ14054 gene (currently known as NPR3) wereearlier described as overexpressed in hematopoietic stemor progenitor cells particularly in the populations with thehighest self-renewing capacity Application of the identifiedgene expression signature to the test set of independent 104AML samples enabled dividing them into two prognosticsubgroups which correlated with the different response toinduction chemotherapy The accuracy of prediction was80 Tagliafico et al [74] conducted a similar analysis buttheir training set included 10 blast cell populations collectedform AML patients and 6 AML cell lines with determinedsensitivity to differentiation therapyThe identified predictionset containing such genes as MEIS1 and MS4A3 was thentested on the GEP datasets published by Valk et al [61]and Bullinger et al [60] Despite a significant overlap inprognosis prediction Tagliafico et al [23] distinguishedwithin the poor outcome groups described in original papersa subgroup of patients (20-40) which revealed sensitivityto maturation induction From the practical point of viewit suggested that these patients could benefit from a dif-ferentiation therapy even though the initial prognosis wasunfavorable

Gene expression analysis of samples from 170 olderAML patients (median age 65 years all FAB subtypesexcept for M3) presented by Wilson et al [75] showedthe problem of response to therapy as even more complexHierarchical clustering divided patients into 6 groups withdifferent rates of resistant disease complete response andDFS Distribution of FAB subtypes and NPM1 (but not FLT3-ITD) mutation differed significantly between clusters butin only two clusters particular subtypes prevailed eg onecluster almost exclusively consisted of monocytic leukemias(M4 and M5) Poor-risk clusters had lower WBC and blastcounts whereas cluster with the best DFS and OS contained75 of NK-AML and 78 samples with NPM1 mutationsEach cluster was defined by a specific expression profileof the 50 most discriminating genes For example in acluster with the poorest outcome the authors observedupregulation of multidrug resistance genes (ABCG2 and

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

10 Journal of Oncology

ABCB1 former MDR1) homeobox gene PBX1 which pre-vents myeloid differentiation and STK17 gene encodingapoptosis regulator Another poor outcome cluster revealedoverexpression of genes connected with immunity (IRF4IL10RA and MALT1) The most favorable outcome clus-ter was characterized by overexpression of genes encodingproteins implicated in cell signaling (IL12A) promotingapoptosis (CASP3 and LTBP1) and leukemic transformation(MEIS1 WT1 and FOXC1) and downregulation of genesencoding major histocompatibility complex (MHC) proteinsof class II

Based on gene expression data from a training cohort of163 AML patients collected by the German AML CoopertiveGroup Metzeler et al [76] elaborated 66 gene expressionsignatures to predict OS in CN-AMLThen the signature wasvalidated in two independent cohorts of 79 and 64 CN-AMLpatients from Europe and the United States respectivelyIn all three cohorts patients with low gene expression riskscore had better outcome Moreover in multivariate analysesof validation cohort the gene expression score proved tobe a stronger prognostic factor than age presence of FLT3-ITD and NPM1 mutation The genes from the identifiedsignature partially overlapped with the results of previousstudies eg TCF4 FHL1 CD109 and SPARC genes hadbeen earlier associated with poor response to chemotherapy[73]

7 Looking for New Therapeutics

Transcriptome as well as proteome reflects the currentcell status that dynamically evolves under the influence ofvarious stimuli eg therapeutic agents GEP is a sensitivetool to detect changes in genome activity therefore it canbe applied to monitor minimal residual disease (MRD) andcancer cell reaction to novel compounds AML treatment ischallenging because resistance to therapy is quite commonand even those patients who achieve CR are prone to relapseGEP was widely applied for analysis of resistance mecha-nisms and efficiency of potential drugs Kinase inhibitorsin the treatment of AML with FLT3-ITD correlated withnegative prognosis have been studied for a long time InApril 2017 staurosporine derivative PKC412 (midostaurin)a multikinase inhibitor was approved by the US FDA forthe treatment of newly diagnosed FLT3-mutant AML incombination with chemotherapy [77] Activity of this com-pound was analyzed inter alia in human myelomonoblasticcell line MV4-11 carrying FLT3-ITD by Stolzel et al [78]with the use of gene expression microarrays Two versionsof MV4-11 cells sensitive and resistant were comparedprior to and after treatment Significant downregulation ofTP53 and upregulation of JAG1 was observed in resistantcells before and after treatment MCL1 and KIT geneswere upregulated in resistant MV4-11 cells after incuba-tion with PKC412 The authors concluded that resistanceagainst PKC412 was mediated by antiapoptotic gene acti-vation and proapoptotic signal decrease with contributionof deregulation of genes involved in normal and malignanthematopoiesis

Tavor et al [79] studied gene expression responseof the AML cell line U937 under treatment with theCXCR4-antagonist AMD3100 CXCR4 a receptor for SDF-1 chemokine secreted by stromal cells participates inthe interactions of leukemic stem cells with the BMmicroenvironment necessary for cell migration and dis-ease progression In addition the role of elastase neu-trophil serine protease synthesized during the transition ofmyeloblast to promyelocyte was investigated The authorsdid not observe changes in gene expression after treat-ment with anti-CXCR4 antibody or elastase inhibitor butfound AMD3100-induced suppression of the SDF-1CXCR4axis or elastase inhibited leukemic cell proliferation aswell as activated genes involved in myeloid differentia-tion

Other candidates for target therapeutics in AML treat-ment are in clinical trials One example is pinometostat(EPZ-5676) a small-molecule inhibitor of DOT1L (his-tone methyltransferase disrupter of telomeric silencing 1-like) Pinometostat considered for combination therapiesof acute leukemias with KMT2A gene rearrangements wasproved to target DOT1L and reduce H3K79 methylation inadult AML patients with 11q23 translocations [80] Anotherpromising therapeutic strategy is directed against mem-bers of Hedgehog (HH) signaling pathway which playsa role in embryonic cell development as well as in pro-liferation and maintenance of adult stem cells includ-ing cancer stem cells [81 82] Comparing chemotherapy-sensitive and resistant cell lines Queiroz et al [83] indi-cated HH pathway as an essential component of myeloidleukemia MRD Overexpression of HH pathway effectorsGLI1 and PTCH1 followed by constitutive activation ofHH signaling was correlated with chemoresistant pheno-type The efficacy of a HH pathway inhibitor glasdegibwhich targets a smoothened protein (SMO) a G protein-coupled receptor interacting with PTCH1 was evaluatedby Cortes et al in AML and high-risk MDS patients whowere not eligible for intensive chemotherapy [81] At theend of 2018 glasdegib has been approved in the USAunder the name DAURISMO for use in combinationwith low-dose cytarabine for the treatment of newly diag-nosed AML patients excluded from intensive inductionchemotherapy due to age or comorbidities [84] Anothercompound GANT61 the inhibitor of GLI family proteinswas shown to specifically target the CBFA2T3-GLIS2 fusiongene in pediatric AML [82] The authors demonstrated thatGANT61 treatment significantly reduced the expression levelof GLIS2 and a gene encoding bone morphogenic protein(BMP2) Posttreatment microarray-based gene expressionanalysis revealed downregulation of CBFA2T3-GLIS2 tar-get genes as well as genes required for cell cycle pro-gression cell proliferation and epigenetic regulation NewAML therapies are still being elaborated Currently the USNational Cancer Institute (NCI) supports 75 clinical trials foradult AML treatment (httpswwwcancergovabout-can-certreatmentclinical-trialsdiseaseadult-amltreatment) Itis impossible to provide even a brief summary of all of themin this work

Journal of Oncology 11

8 Pediatric AML Distinct but Similar

AML in children is less frequent than in adults but revealssimilar level of heterogeneity In both age groups similarchromosomal aberrations and mutations occur though withdifferent proportions In children CN-AML concerns onlyabout 20 of all AMLs and the frequency of mutationsis generally lower [85] The power of GEP demonstratedon adult AML samples triggered the research of childhoodAML First the 35-gene expression signature was shown topredict prognosis in pediatric AML [86] Genes encodingcyclins and cyclin-dependent kinases required for cell cycleprogression (CDK6 CCND1 and CDC25A) and TRAF2gene encoding a signal transducer activating NFKB1 showedhigher expression in patientswith poor outcomeThe levels ofNFKBIA encoding NFKB1 inhibitor and STK17B encodingserinethreonine protein kinase inducing apoptosis werelower in patients with poor outcome STK17B downregula-tion and NFKB1 enhancement might explain why patientswith adverse prognosis escaped from chemotherapy-inducedapoptosis An additional reason for the poor outcome couldbe increased cell cycle progression Comparing pediatricAML patients with different FAB subtypes the authorsselected 213 probe set representing genes whose expressioncorrelated with FAB subtype Both signatures prognostic anddiagnostic shared only three genes (TYMP STK17B andATP6V0B)

Ross et al [68] compared gene expression in 130 pediatricand 20 adult AML samples with Affymetrix GeneChipsSomeAML groups namely t(1517) t(821) and FABM7morefrequent in children than adults were clearly distinguishedwhereas AML with CBFBMYH11 fusion gene (inv(16)) andKMT2A chimeric fusion genes revealed more heterogeneityindicating the existence of additional subgroups Biology ofthe disease seemed to be similar independently on age asonly minimal differences were observed in gene expressionprofiles between pediatric and adult AML samples con-taining the same lesions The authors identified a set ofclass discriminating genes which included genes specificallyoverexpressed in particular AML FAB types eg AML M2was characterized by increased expression of genes codingfor cell surface antigens (CD34 CD19) proteins regulatingdevelopmental processes (ROBO1 TWSG1 and PELI2) andtranscription factor POU4F1 Genes upregulated in AMLM3M4Eo andM7 encoded proteins reflecting particular stages ofmyeloid differentiation or lineage for example HGF MPOand CPA3 in M3 CD52 and CHI3L1 in M4Eo GP1BB andITGA2B in M7 The results concerning AML with KMT2Arearrangements were described above What is interestingRoss et al [68] tested on their dataset the 35-gene prognosticsignature described by Yagi et al [86] and did not confirmits correlation with patient outcome Instead Ross et al [68]selected another small set of genes whose high expressioncorrelated with poor outcome This shows gene expressionprofile dependence on sample set sample size protocols andlaboratory However another study of childhood leukemiapublished by Andersson et al [69] confirmed the resultsobtained by Ross et al [68] presenting 77ndash86 overlapbetween the differentially expressed genes (DEGs)

Analysis of 237 pediatric AML cases with gene expres-sion microarrays and double loop cross-validation methodallowed for the selection of 75 probe sets representing 59unique genes able to classify AML with the five most preva-lent cytogenetic subtypes constituting about 40 of pediatricleukemia [85] Among the most discriminative genes wereWHAMMP3 and ITM2A (encoding membrane associatedproteins) for KMT2A-rearranged RUNX1T1 IL5RA andPOU4F1 for t(821) MYH11 LPAR1 and NT5E for inv(16)HGF STAB1 and FAM19A5 for t(1517) TP53BP2 (codingfor p53-binding protein) and DNAAF4 (encoding proteininteracting with the estrogen receptors and the heat shockproteins) for t(712) The accuracy of the classifier validatedon two independent cohorts of patients was equal to 92 and99 However GEP had limited predictive value for AMLcases withNPM1CEBPAKMT2A(-PTD) FLT3(-ITD)KITPTPN11 and NK-RASmutations perhaps because of gener-ally lower frequency of mutations in children than in adults

9 AML in the Elderly

AML is a disease of older adults Within age not only theincidence of illness increases elderly AML patients usuallypresent worse outcome and weaker response to therapy Raoet al [87] reanalyzed clinically annotated GEP data from425 de novo AML patients in the context of age From thisdataset two age-related cohorts were selected 175 young(ltor= 45 years) patients and 144 elderly (gtor= 55 years)patients Indeed both cohorts significantly differed in OSand DFS This difference could be explained by uniquepattern of deregulated signaling pathway found for olderAML patients who had a lower probability of E2F and PI3-kinase pathway activation but a higher probability of RASTNF SRC and EPI pathway activation Thus the authorsconcluded AML in the elderly represents a distinct biologicentity The same conclusion was made by de Jonge et al [88]who discovered the downregulation of the tumor suppressorgene CDKN2A in older AML patients with intermediate-and unfavorable prognosis CDKN2A gene encodes a cyclin-dependent kinase inhibitor known as p16(INK4A) proteinwhose amount increases with physiologic ageing The authorsshowed that p16-INK4A besides cytogenetic risk groups wasan independent OS prognostic parameter in older patientsThe conclusion was that in the elderly oncogenesis might befacilitated by a suppression of defense mechanisms whichusually protect older cells against cell and DNA damage [89]

10 Between MDS and AML

Myelodysplastic syndromes (MDS) are a group of clonalheterogenous hematologic malignancies frequent in theelderly characterized by progenitor cell dysplasia ineffectivehematopoiesis and a high rate of transformation to AML[90] Due to the not clearly defined boundaries betweenMDSand other myeloid disorders establishing MDS diagnosiswith conventional method is often problematic Looking fora novel diagnostic strategy Miyazato et al [91] comparedthe transcriptomes of MDS with de novo AML and other

12 Journal of Oncology

bonemarrow diseases with the use of custom-made oligonu-cleotide microarrays The hematopoietic stem-cell fractionswere purified based on the expression of the surface markerPROM1 previously named AC133 or CD133 The authorsidentified a small set of genes preferentially expressed inMDS(eg DLK1 TEC and ITPR1) or AML (eg genes encodingsolute carrier (SLC) family members opioid receptor delta 1(OPRD1) and leptin receptor (LEPR))

AML with dysplasia which has a poor outcome withconventional chemotherapy was studied by Tsustumi et al[92] The authors analyzed three AML subcategories withdysplastic morphology AML with multilineage dysplasia(AML-MLD)MDS-related AML (MDS-AML) and therapy-related leukemia (TRL) and compared them with de novoAML without dysplasia As in the study of Miazato etal fractions of BM hematopoietic stem cells presentingCD133 antigen were selected for microarray-based tran-scriptome analysis 56 genes displayed different expressionlevels between AML-MLD and MDS-AML The genes pref-erentially expressed in AML-MLD comprise many genesencoding nuclear proteins ubiquitination-related proteinsandPF4 gene encoding platelet factor 4 a chemokine secretedby platelets and influencing BM environment The same genewas overexpressed in AML-MLD also when compared toAML without dysplasia suggesting the correlation of PF4expression with AML-MLDDistinction betweenMDS-AMLand AML without dysplasia was possible with the use of28 genes including 9 shared within the 56-gene signaturedifferentiating AML-MLD and MDS-AML One of themLAPTM5 gene encoding lysosomal transmembrane proteinwas clearly upregulated in MDS-AML being a candidatefor novel marker for MDS-related leukemia However thegene signatures determined by Tsustumi et al [92] were notperfect which showed global gene expression analysis maybe not adequate for AML subgroups with high intragroupheterogeneity and subtle intergroup differences

11 Bone Marrow Microenvironment

The main attention of AML investigators was focused onleukemic blasts However it is well known that other factorssuch as tumor microenvironment contribute to diseaseprogression In hematological malignancies the interplay ofcancer cells and surrounding stroma is particularly impor-tant BM microenvironment consists of a heterogeneouspopulation of cells directly involved in hematopoiesis orsupporting hematopoietic cell function migration adhesionmetabolism and differentiation eg by production of ligandsand cell adhesion molecules [93] The role of BM nichein AML has not been fully elucidated and is currentlyintensely studied [4 93] Experiments with themousemodelsindicated that the BM microenvironment not only facilitatesthe leukemic cell growth but can even initiate leukemogenesisin healthy cells [94] The expansion of a single dominanthematopoietic progenitor clone is favored in the aged BMmicroenvironment which causes monoclonality and maycontribute to higher rates of leukemia incidence with age[95] Moreover BM niche protects quiescent LSCs being

responsible for MRD and relapse On the other hand BMstromal cells reveal high level of plasticity and can also beaffected by malignant cells [93 96] Therefore the diseaseprogression depends on the leukemia-microenvironmentcrosstalk One of the best recognized interactions betweenleukemic blasts and stroma is directed by a transmembranechemokine receptor CXCR4 highly expressed by leukemiccells and CXCL12 protein secreted by BM stromal cellsCXCR4-CXCL12 binding promotes the homing residenceand survival of leukemic cells in the BM [4] Another inter-action between the integrin VLA-4 expressed by leukemiccells fibronectin present in the extracellular matrix andVCAM-1 on BM stroma contributes to chemoresistance [4]

In 2018 Kumar et al described how AML blasts trans-form the BM niche into a leukemia-promoting and nor-mal hematopoiesis-suppressive microenvironment through asecretion of exosomes small vesicles mediating cell-to-cellcommunication [96] The authors demonstrated that AML-derived exosomes target stromal and endothelial cells inthe BM niche Using human-to-mouse AML graft modelsthey proved AML-derived exosomes caused changes in micesimilar to those induced by AML cells ie reshaped theBM niche cell composition and modulated gene expressionin stromal cells Genes required for normal hematopoiesisand bone development eg CXCL12 KITL and IGF1 weredownregulated whereas a hematopoiesis and osteogenesissuppressor DKK1 was upregulated Reduction of exosomesecretion in AML cells delayed the disease progression

One of the recent studies used a unique ex vivo modelof growing leukemic cells on patientsrsquo own stroma (POS)derived in diagnosis (Dx) remission (Rm) and relapse(Rx) [97] Compared to healthy mesenchymal stromal cells(MSCs) POS presented different morphology larger cell sizereduced proliferation rate slower expansion and poor cell-cell contact Coculture cross experiments revealed that POSpreferentially supported proliferation of the same patientrsquosAML cells irrespective of the disease state POS was obtainedin The unique crosstalk between POS and AML cells wasmediated by cytokines and chemokines angiopoietin 1secreted phosphoprotein 1 and SDF-1 encoded by ANGPT1(former Ang-1) SPP1 (former OPN) and CXCL12 genesrespectively Compared to healthy MSCs SPP1 expressionwas higher in DxRx and Rm POS whereas ANGPT1 expres-sion was upregulated in DxRx POS and increased in thepresence of AML cell In contrast CXCL12 was decreased inDxRx and Rm POS which was associated by the authorswith interruption in the CXCL12-CXCR4 signaling anda consequent loss of hematopoietic progenitor quiescenceand induced proliferation Interestingly POS demonstratedsimilar features in remission as in the active disease whichindicates the critical role of BMniche in relapse and treatmentfailure

BM microenvironment-mediated protection of FLT3-ITDAML from tyrosine kinase inhibitors (TKIs) was recentlyreported by Chang et al [98] Drug resistance was a resultof elevated expression of genes encoding cytochrome P450enzymes in particular CYP3A4 by BM stromal cells BecauseCYP3A4 inhibitor reversed the protective effects of BMnichethe authors proposed a combination of FLT3 TKIs with

Journal of Oncology 13

CYP3A4 inhibitors as a novel strategy to treat FLT3-ITDAML

Passaro et al [99] who studied the BM vasculaturein AML using intravital two-photon microscopy associatedincreased vascular permeability in the BM microenviron-ment with disease progression and poor treatment responseTranscriptome analysis of BM-derived endothelial cells viaRNA-seq identified deregulation of genes involved in vascu-lature development angiogenesis and response to hypoxiaNox4 gene encoding NAPDH oxidase responsible for pro-duction of reactive oxygen species (ROS) activation of nitricoxide synthase 3 (NOS3) and release of nitric oxide (NO)was particularly upregulated Increased NO level contributedto the vascular leakiness in AML-engrafted mice and wasassociated with poor prognosis in AML patients Appli-cation of NO synthase inhibitors combined with standardchemotherapy restored normal vasculature and improved thetreatment response demonstrating the efficacy of combinedleukemia-niche therapies

The role of lymphocytes and other blood cells has longbeen neglected in the studies of AML However Le Dieu etal found that the absolute number of T-cells circulating in PBof de novo AML patients not belonging to malignant cloneswas increased compared to healthy controls [100] Activationof T-cells might reflect a response to growth signals present ina local microenvironment GEP of CD4+ and CD8+ T-cellsfrom AML patients and healthy volunteers revealed globaldifferences in transcription pattern with little similaritiesto T-cells of CLL patients Particularly genes associatedwith the actin cytoskeleton and cellular polarization werederegulated in AML T-cells According to the authors T-cellaberrant activation leads to their dysfunction and impairedimmune response which is not a sufficient weapon againstthe leukemic blasts Here a rationale to apply immunomod-ulatory drugs appears

12 Discovering the Power of Small MoleculesmiRNA Expression Profiling

The discovery of regulatory role of RNA in cell and organ-ism development completely changed our understanding ofbiology and at least partially explained the paradox that is thelarge size of mammalian genomes of which only a small per-centage are the protein-coding genes [101ndash103] Among smallregulatory RNAsmicroRNAs (miRNAs) termed due to theirsmall size (18-23 nt) are best recognized [104 105] Thefunction of miRNAs in gene expression regulation (usuallyrepression) controlling cell fate and normal developmentalprocesses as well as oncogenesis is well-established [106ndash109] As one miRNA targets multiple transcripts [110] dys-function of miRNA may result in a wide-scale deregulationof gene expression often triggering a cascade of eventsleading to pathogenesis In 2002 Calin et al demonstratedmiR-15 and miR-16 are located at chromosome 13q14 regionfrequently deleted in B-cell chronic lymphocytic leukemias(B-CLL) [111] Then more than half miRNA genes werelinked with cancer-associated genomic regions or fragilesites and their amplification or deletion in human cancers

supported miRNA role in malignant transformation [112]Since 2005 when Lu et al classified multiple human can-cers including AML based on miRNA expression profilesexclusively and proved general downregulation of miRNAsin tumors compared to normal tissues [51] miRNAs startedto be widely investigated in cancers and other diseases

MicroRNAs were also described as regulators of mam-malian hematopoiesis [113 114] In 2004 three microRNAswhichmodulatemouse hematopoietic lineage differentiationwere found byChen et al [113]miR-181 a positive regulator ofB-lymphoid cell differentiation miR-223 nearly exclusivelyexpressed in BM and myeloid cells and miR-142 found athighest levels in B-lymphoid and myeloid lineages Geor-gantas et al identified 33 miRNAs specifically expressed inCD34+ hematopoietic stem-progenitor cells (HSPCs) [115]The identified miRNA signature included miRNA-17 -24 -146 -155 -128 and -181 holding early hematopoietic cellsat a stem-progenitor stage and blocking their maturationmiRNA-16 -103 and -107 responsible for block differentiationof later progenitor cells and miRNA-221 -222 and -223controlling terminal stages of hematopoietic differentiationSome miRNAs indeed presented lineage-specific expres-sion which suggested the limitation of function to eglymphoid (miRNA-146) erythroid (miRNA-221 and -222)or granulocytic (miRNA-223) development Inhibition oferythropoiesis and erythroleukemic cell growth through KITgene suppression by miR-221 and -222 was also reportedby Felli et al [116] whereas granulopoiesis regulation by aminicircuitry involving miR-223 NFIA and CEBPA by Faziet al [117] Other miRNAs were able to control differentprocesses eg myelopoiesis and erythropoiesis as miRNA-155 [115] Contrary to the results of Chen et al [113] whostudied hematopoiesis on murine modelmiR-142 expressionwas not detected in human hematopoietic cells byGeorgantaset al [115] Later Bissels et al deepened the knowledge aboutmiRNA-regulated hematopoiesis by combining analyses ofmicroRNA and mRNA profiles in CD133+ and CD34+hematopoietic stem and progenitor cells [118] In both typesof cells 25 highest expressed miRNAs accounted for 73-74 of the total miRNA pool However the most abundantmiRNAs were rather common for both progenitor cell typesexcept for miR-142-3p which was upregulated in CD34+cells to the level of up to 5000 copies per cell Remarkablyone of miR-142-3p targets seemed to be CD133 gene Theauthors found 18 miRNAs expressed differentially betweentheCD133+ (ancestral) andCD34+CD133- (later progenitor)cells miR-10a -99a -125a and b and miR-146a and bexpressed at highest level in CD133+ cells were postulatedto maintain the stem-cell character whereas miR-484 andother miRNAs upregulated in CD34+ cells probably blockedcell differentiation at a later stage Generally differentiallyexpressed miRNAs were involved in inhibition of differenti-ation prevention of apoptosis and cytoskeletal remodeling

In 2007Mi et al [54] showed that discrimination of AMLfrom ALL is possible through miRNA expression profilingAmong 27 miRNAs differentially expressed between AMLand ALL four were sufficient to distinguish these two typesof acute leukemia let-7b and miR-223 were significantlyupregulated in AML whereas miR-128a and miR-128b were

14 Journal of Oncology

downregulated in AML comparing to ALL [54] In 2010Wang et al [119] conducted similar research on Chinesecohort of 85 patients and found 16 miRNAs differentiallyexpressed between AML and ALL Half of them were previ-ously reported by Mi et al [54] (miR-23a miR-27ab miR-128a miR-128b miR-221 miR-222 miR-223 and let-7b) buteight have not identified previously in this context (miR-17 miR-20a miR-29ac miR-29b miR-146a miR-150 miR-155 and miR-196b) In addition prognostically relevant sig-natures were determined for ALL AML and non-M3-AMLOne miRNA miR-146a was strongly inversely correlatedwith OS of both acute leukemias in two independent patientcohorts

Garzon et al studied miRNA expression in APL cellstreated with all-trans-retinoic acid (ATRA) [120] and foundupregulation of miR-15a -15b -16-1 let-7a-3 let-7c let-7dmiR-223 -342 and -107 and downregulation of miR-181bThe observed ATRA modulation of NFIA RAS and BCL2gene expression corresponded with the fact the mentionedgenes were targets of miR-107 let-7a and miR-15amiR-16-1respectively Then Garzon et al [121] evaluated the miRNAexpression in 122 newly diagnosed AML cases comparingto CD34+ cells from 10 healthy donors and found 26differentially expressed miRNAs all downregulated in AMLeg miR-126 -130a -135 -93 -146 -106b and -125a Expres-sion level of some miRNAs was variable within AML andcorrelated with AML cytogenetics prognosis and clinical fea-tures For example miR-181 was downregulated particularlyin AML with multilineage dysplasia whereas miR-155 andmiR-181b positively correlated with WBC (white blood cellcount) In AML with balanced 11q23 translocations manytumor suppressormiRNAs targeting knownoncogenes weredownregulated eg miR-34b (targeting CDK4 and CCNE2)miR-15a (targeting BCL2) let-7 family (targeting RAS genes)and miR-196 (targeting HOX genes) In trisomy 8 onlyupregulated miRNAswere identified including those locatedat chromosome 8 eg miR-124a whose known target isCEBPA NK-AML was the most heterogeneous thereforethe identified miRNA signature was not predictive of NK-AML However five miRNAs overexpressed in AML (miR-199a and b miR-191 miR-25 and miR-20a) were associatedwith adverse patient outcome

Debernardi et al [122] showed strong correlation ofmiR-181a expression with the AML FAB subtypes (elevatedin M1 and M2) and with the expression of its predictedtargets Half of them eg BCL2L11 KLF3 MAP2K1 werenegatively correlated with miR-181a expression Havelangeet al [123] observed two other mRNA-miRNA interactionsnegative correlation between miR-181a and miR-181b miR-155 and miR-146 expression with that of genes involved inimmunity and inflammation (IRF7 and TLR4) and positivecorrelation between miR-23a miR-26a miR-128a and miR-145 expression level with that of proapoptotic genes (BIMand PTEN) Association of the last three miRNA withapoptosis was experimentally validated Lineage-associationswere showed for miR-23a and miR-196a (positive correla-tion with myeloid differentiation) miR-191 miR-222 andmiR-17 (negative correlation with erythroid differentiation)Interaction analysis induced the authors to conclude that

a small group of miRNAs coordinately regulates protein-coding transcriptome influencing the same group of genes(presumably the key players) within the pathway

In 2008 distinctive patterns of miRNA expression asso-ciated with cytogenetic and genetic AML subtypes weredetermined by Dixon-McIver et al[124] Jongen-Lavrencicet al [125] and Li et al [126] Dixon-McIver et al [124]measured the expression level of 157 miRNAs in 100 AMLpatients and two AML cell lines with the use of bead-based flow cytometric miRNA expression assay and found33 miRNAs with differential expression level between AMLand normal BM 17 upregulated (let-7e miR-27a -30d -142-5p -155 -181a -181b -181c -195 -221 -222 -324-5p -326 -328 -331 -340 -374) and 16 downregulated (miR-9lowast-15b -26a -30a-3p -34c -103 -147 -151 -182 -184 -199a-302blowast -302d -325 -367 -372) Moreover they associatedt(1517) translocation with upregulation of miRNAs locatedin the 14q32 imprinted domain eg miR-127 miR-154 miR-154lowast miR-299 miR-323 miR-368 and miR-370 In AMLwith inv(16) high level of miR-99a miR-100 and miR-224expression was observed whereas t(821) AML presentedhigh expression of miR-146a and a decrease of miR-133aHigh degree of variability across samples was noted for miR-10a and miR-125b Jongen-Lavrencic et al [125] found a setof strongly upregulated microRNAs (miR-382 -134 -376a -127 -299ndash5p and -323) in t(1517) partially overlapping withthe APL signature defined by Dixon-McIver et al [124]Clustering of AML cases with miRNA expression revealedthat inv(16) were sometimes mixed with t(821) and shareda part of miRNA signature which is not unexpected as theseboth AML subtypes belong to CBF AMLs Predictors of mostAML subclasses containing from a few to several dozenmiRNAs were built for AML with NPM1mutation and evenfor AMLwith FLT3-ITD or FLT3-TKDmutationswhichwerenot separated from other AMLs as a result of global miRNAexpression-based clustering

Li et al [126] found miRNA signatures composed of 2-24miRNAs able to distinguish AMLwith KMT2A rearrange-ment t(1517) t(821) plus inv(16) t(821) inv(16) and normalcontrols They noted that miR-126126lowast were specificallyoverexpressed in both t(821) and inv(16) AMLs while miR-224 miR-368 and miR-382 in t(1517) In KMT2A-AMLsignificant overexpression of miRNAs from polycistronicmiRNA cluster mir-17-92 was observed A minimal class-predictor contained only seven miRNAs miR-126 -126lowast -224 -368 -382 17-5p and -20a Interestingly differentialexpression of miR-126126lowast was not associated with DNAduplication nor mutation but probably resulted from epi-genetic regulation Gain- and loss-of-function experimentsrevealed that high expression of miR-126 inhibits apoptosisand increases cell viability and proliferation synergisticallywith the fusion gene RUNX1-RUNX1T1 From 674 predictedmiR-126 targets the authors empirically tested 12 genes andconfirmed that only one of them PLK2 was indeed regulatedbymiR-126

MicroRNA expression pattern correspondence with FABclassification was shown by Wang et al [127] who noted thatM1 M2 M3 and M4 tended to depart from each other moreeffectively thanM5 Apart frommiRNAs reported previously

Journal of Oncology 15

the authors identified a spectrum of new miRNAs whoseexpression strongly correlated with particular AML FABtypes eg miR-1300 miR-1180 miR-297 miR-610 and miR-650 overexpressed exclusively in AML M1 High expressionof some miRNAs was common in a few FAB types egmiR-181a-d miR-221 and miR-222 by M1 M2 and M3 Themost distinct miRNA expression pattern was shown in AMLM3 with 36 miRNAs strongly and exclusively upregulatedeg miR-370 miR-224 miR-382 miR-154 described earlierand miR-100 miR-195 miR-452 miR-654-3p not reportedpreviously Comparing to normal PBMCs all AML samplesdisplayed downregulation ofmiR-29a and miR-142-3p whichwere proposed by the authors as AML diagnostic biomarkers

Analysis of miRNA expression data in CN-AML per-formed by Marcucci et al [128] allowed identification ofmiRNA signature of prognostic relevance Upregulation ofmiR-181a and b was associated with low risk whereas over-expression of miR-124 -128-1 -194 219-5p 220a and -320with increased risk of failure to achieve CR relapse or deathIncreased miRNA levels were correlated with increasedexpression of genes involved in innate immunity encodingtoll-like receptors interleukins and caspases Some of themwere putative targets ofmiR-181

13 Mutation-Defined AML Subtypes

Progressive accumulation of transcriptomic data regardingboth mRNA and miRNA expression allowed more preciselycharacterize AMLs with the recurrent mutations

131 AML with Mutated CEBPA Valk et al [61] first triedto determine gene expression profiles specific for AML withparticular mutations Having a set of nearly 300 samplesthe authors were able to distinguish a unique gene expres-sion signature for AML with CEBPA mutations CEBPAgene mutated in 5 to 15 of all AML cases encodes acritical regulator of hematopoietic stem-cell maintenanceand myeloid differentiation therefore unique gene expres-sion pattern was not unexpected for samples with theloss-of-function CEBPA mutation Valk et al [61] showedthat the most prominent features of CEBPA-mutated AMLwere CD7 overexpression and downregulation of CTNNA1TUBB and NDFIP1 Interestingly hierarchical clustering ofall AML samples segregated AMLs with CEBPA mutationsinto 2 different clusters and one of them included alsosamples without any known mutations or chromosomalaberrations In their subsequent paper the authors usingbisulfite genomic sequencing revealed that this previouslyunidentified subset of AML was represented by sampleswhere CEBPA gene promoter was hypermethylated [129]In fact within this mysterious AML cluster identified byValk et al [61] CEBPA levels were very high in AML caseswith CEBPA mutations whereas in AML with wt variantCEBPA expression was minimal or undetectable due to itsepigenetic silencing [129] Detailed characteristics ofCEBPA-silenced AML samples showed that they expressed bothmyeloid markers (CD13 CD33 and MPO) and T-lymphoidmarkers (eg CD7 mentioned above) In addition high

expression of the myeloid oncogene TRIB2 and NOTCH1gene encoding a membrane receptor and transcriptionalregulator of T-cell development was noted In a part of thoseAML patients activating NOTCH1 mutation was identifiedMoreover TRIB2 was determined to be a direct target ofNOTCH1 signaling Later the authors found that CEBPAmethylation was accompanied by aberrant hypermethylationof many genes compared to CEBPA-mutated AMLs or withnormal CD34+ hematopoietic progenitor cells [130] Thiscould explain an observed in vitro decreased response ofCEBPA-silenced AML to myeloid growth factors and makesthis AML subtype susceptible to dynamically developingtreatment with demethylating agents Interestingly compari-son of genome-wide methylation pattern with GEP revealedonly a minimal overlap (12 unique genes including CEBPA)between the differentially expressed and differentially methy-lated genes This suggested that gene expression and genomemethylation are biologically independent processes

132 AMLwith Mutated NPM1 and the Paradox of HOXGeneExpression Since the time a 4-nucleotide insertion in NPM1gene and its significance in AML was discovered by Faliniet al in 2005 [52] much attention was paid to unveilingthe mechanism of AML triggered by NPM1mutation NPM1encodes a multifunctional protein involved in ribosomebiosynthesis and transport DNAreplication and repair chro-matin remodeling protein chaperoning regulation of cellcycle embryogenesis and oncogenesis [131 132] AlthoughNPM1 protein localizes mainly in the nucleolus it constantlyshuttles between nucleus and cytoplasm The 4-nucleotideinsertion in the last exon of NPM1 gene results in aberrantcytoplasmic accumulation of a protein and consequentlyaffects its functions NPM1 gene is mutated in about 30 ofAML and in 50 to 60 cases of adult NK-AML Given thatNPM-cytoplasmic positive (NPMc+) AML reveals uniquemolecular and clinical features [133] it was introduced intoWHO classification as a separate entity [13] An invaluablecontribution to cognition of AML promotion by NPM1mutation was made by GEP

Alcalay et al [134] first claimedNPMc+AML representeda distinct entity which can be easily distinguished from otherAML samples regardless of the karyotype They comparedglobal gene expression of 58 AML NPMc+ samples withprevalence of NK-AML and frequent occurrence of FLT3mutations to the group of 20NK-AMLswithoutNPM1muta-tions and lower occurrence of FLT3mutations Unsupervisedapproach showed NPM1 mutation status was the strongestclustering parameter A selected 369-gene-predictor effi-ciently segregated NPMc+ from NPMc- patients Interest-ingly NPM1 transcript level did not differ between thesetwo groups indicating the lack of NPM1 mutation influenceon NPM1 gene expression I confirmed this observation byanalysis of NPM1 alternative transcripts with droplet digitalPCR (ddPCR) [135] In theNPMc+patientsCD34 andCD133antigens as well as POU4F1 and CDKN2C were suppressedwhereas a number of homeodomain-containing transcrip-tion factors including HOX and TALE genes were activated[134] Because several HOX genes are highly expressed in

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

12 Journal of Oncology

bonemarrow diseases with the use of custom-made oligonu-cleotide microarrays The hematopoietic stem-cell fractionswere purified based on the expression of the surface markerPROM1 previously named AC133 or CD133 The authorsidentified a small set of genes preferentially expressed inMDS(eg DLK1 TEC and ITPR1) or AML (eg genes encodingsolute carrier (SLC) family members opioid receptor delta 1(OPRD1) and leptin receptor (LEPR))

AML with dysplasia which has a poor outcome withconventional chemotherapy was studied by Tsustumi et al[92] The authors analyzed three AML subcategories withdysplastic morphology AML with multilineage dysplasia(AML-MLD)MDS-related AML (MDS-AML) and therapy-related leukemia (TRL) and compared them with de novoAML without dysplasia As in the study of Miazato etal fractions of BM hematopoietic stem cells presentingCD133 antigen were selected for microarray-based tran-scriptome analysis 56 genes displayed different expressionlevels between AML-MLD and MDS-AML The genes pref-erentially expressed in AML-MLD comprise many genesencoding nuclear proteins ubiquitination-related proteinsandPF4 gene encoding platelet factor 4 a chemokine secretedby platelets and influencing BM environment The same genewas overexpressed in AML-MLD also when compared toAML without dysplasia suggesting the correlation of PF4expression with AML-MLDDistinction betweenMDS-AMLand AML without dysplasia was possible with the use of28 genes including 9 shared within the 56-gene signaturedifferentiating AML-MLD and MDS-AML One of themLAPTM5 gene encoding lysosomal transmembrane proteinwas clearly upregulated in MDS-AML being a candidatefor novel marker for MDS-related leukemia However thegene signatures determined by Tsustumi et al [92] were notperfect which showed global gene expression analysis maybe not adequate for AML subgroups with high intragroupheterogeneity and subtle intergroup differences

11 Bone Marrow Microenvironment

The main attention of AML investigators was focused onleukemic blasts However it is well known that other factorssuch as tumor microenvironment contribute to diseaseprogression In hematological malignancies the interplay ofcancer cells and surrounding stroma is particularly impor-tant BM microenvironment consists of a heterogeneouspopulation of cells directly involved in hematopoiesis orsupporting hematopoietic cell function migration adhesionmetabolism and differentiation eg by production of ligandsand cell adhesion molecules [93] The role of BM nichein AML has not been fully elucidated and is currentlyintensely studied [4 93] Experiments with themousemodelsindicated that the BM microenvironment not only facilitatesthe leukemic cell growth but can even initiate leukemogenesisin healthy cells [94] The expansion of a single dominanthematopoietic progenitor clone is favored in the aged BMmicroenvironment which causes monoclonality and maycontribute to higher rates of leukemia incidence with age[95] Moreover BM niche protects quiescent LSCs being

responsible for MRD and relapse On the other hand BMstromal cells reveal high level of plasticity and can also beaffected by malignant cells [93 96] Therefore the diseaseprogression depends on the leukemia-microenvironmentcrosstalk One of the best recognized interactions betweenleukemic blasts and stroma is directed by a transmembranechemokine receptor CXCR4 highly expressed by leukemiccells and CXCL12 protein secreted by BM stromal cellsCXCR4-CXCL12 binding promotes the homing residenceand survival of leukemic cells in the BM [4] Another inter-action between the integrin VLA-4 expressed by leukemiccells fibronectin present in the extracellular matrix andVCAM-1 on BM stroma contributes to chemoresistance [4]

In 2018 Kumar et al described how AML blasts trans-form the BM niche into a leukemia-promoting and nor-mal hematopoiesis-suppressive microenvironment through asecretion of exosomes small vesicles mediating cell-to-cellcommunication [96] The authors demonstrated that AML-derived exosomes target stromal and endothelial cells inthe BM niche Using human-to-mouse AML graft modelsthey proved AML-derived exosomes caused changes in micesimilar to those induced by AML cells ie reshaped theBM niche cell composition and modulated gene expressionin stromal cells Genes required for normal hematopoiesisand bone development eg CXCL12 KITL and IGF1 weredownregulated whereas a hematopoiesis and osteogenesissuppressor DKK1 was upregulated Reduction of exosomesecretion in AML cells delayed the disease progression

One of the recent studies used a unique ex vivo modelof growing leukemic cells on patientsrsquo own stroma (POS)derived in diagnosis (Dx) remission (Rm) and relapse(Rx) [97] Compared to healthy mesenchymal stromal cells(MSCs) POS presented different morphology larger cell sizereduced proliferation rate slower expansion and poor cell-cell contact Coculture cross experiments revealed that POSpreferentially supported proliferation of the same patientrsquosAML cells irrespective of the disease state POS was obtainedin The unique crosstalk between POS and AML cells wasmediated by cytokines and chemokines angiopoietin 1secreted phosphoprotein 1 and SDF-1 encoded by ANGPT1(former Ang-1) SPP1 (former OPN) and CXCL12 genesrespectively Compared to healthy MSCs SPP1 expressionwas higher in DxRx and Rm POS whereas ANGPT1 expres-sion was upregulated in DxRx POS and increased in thepresence of AML cell In contrast CXCL12 was decreased inDxRx and Rm POS which was associated by the authorswith interruption in the CXCL12-CXCR4 signaling anda consequent loss of hematopoietic progenitor quiescenceand induced proliferation Interestingly POS demonstratedsimilar features in remission as in the active disease whichindicates the critical role of BMniche in relapse and treatmentfailure

BM microenvironment-mediated protection of FLT3-ITDAML from tyrosine kinase inhibitors (TKIs) was recentlyreported by Chang et al [98] Drug resistance was a resultof elevated expression of genes encoding cytochrome P450enzymes in particular CYP3A4 by BM stromal cells BecauseCYP3A4 inhibitor reversed the protective effects of BMnichethe authors proposed a combination of FLT3 TKIs with

Journal of Oncology 13

CYP3A4 inhibitors as a novel strategy to treat FLT3-ITDAML

Passaro et al [99] who studied the BM vasculaturein AML using intravital two-photon microscopy associatedincreased vascular permeability in the BM microenviron-ment with disease progression and poor treatment responseTranscriptome analysis of BM-derived endothelial cells viaRNA-seq identified deregulation of genes involved in vascu-lature development angiogenesis and response to hypoxiaNox4 gene encoding NAPDH oxidase responsible for pro-duction of reactive oxygen species (ROS) activation of nitricoxide synthase 3 (NOS3) and release of nitric oxide (NO)was particularly upregulated Increased NO level contributedto the vascular leakiness in AML-engrafted mice and wasassociated with poor prognosis in AML patients Appli-cation of NO synthase inhibitors combined with standardchemotherapy restored normal vasculature and improved thetreatment response demonstrating the efficacy of combinedleukemia-niche therapies

The role of lymphocytes and other blood cells has longbeen neglected in the studies of AML However Le Dieu etal found that the absolute number of T-cells circulating in PBof de novo AML patients not belonging to malignant cloneswas increased compared to healthy controls [100] Activationof T-cells might reflect a response to growth signals present ina local microenvironment GEP of CD4+ and CD8+ T-cellsfrom AML patients and healthy volunteers revealed globaldifferences in transcription pattern with little similaritiesto T-cells of CLL patients Particularly genes associatedwith the actin cytoskeleton and cellular polarization werederegulated in AML T-cells According to the authors T-cellaberrant activation leads to their dysfunction and impairedimmune response which is not a sufficient weapon againstthe leukemic blasts Here a rationale to apply immunomod-ulatory drugs appears

12 Discovering the Power of Small MoleculesmiRNA Expression Profiling

The discovery of regulatory role of RNA in cell and organ-ism development completely changed our understanding ofbiology and at least partially explained the paradox that is thelarge size of mammalian genomes of which only a small per-centage are the protein-coding genes [101ndash103] Among smallregulatory RNAsmicroRNAs (miRNAs) termed due to theirsmall size (18-23 nt) are best recognized [104 105] Thefunction of miRNAs in gene expression regulation (usuallyrepression) controlling cell fate and normal developmentalprocesses as well as oncogenesis is well-established [106ndash109] As one miRNA targets multiple transcripts [110] dys-function of miRNA may result in a wide-scale deregulationof gene expression often triggering a cascade of eventsleading to pathogenesis In 2002 Calin et al demonstratedmiR-15 and miR-16 are located at chromosome 13q14 regionfrequently deleted in B-cell chronic lymphocytic leukemias(B-CLL) [111] Then more than half miRNA genes werelinked with cancer-associated genomic regions or fragilesites and their amplification or deletion in human cancers

supported miRNA role in malignant transformation [112]Since 2005 when Lu et al classified multiple human can-cers including AML based on miRNA expression profilesexclusively and proved general downregulation of miRNAsin tumors compared to normal tissues [51] miRNAs startedto be widely investigated in cancers and other diseases

MicroRNAs were also described as regulators of mam-malian hematopoiesis [113 114] In 2004 three microRNAswhichmodulatemouse hematopoietic lineage differentiationwere found byChen et al [113]miR-181 a positive regulator ofB-lymphoid cell differentiation miR-223 nearly exclusivelyexpressed in BM and myeloid cells and miR-142 found athighest levels in B-lymphoid and myeloid lineages Geor-gantas et al identified 33 miRNAs specifically expressed inCD34+ hematopoietic stem-progenitor cells (HSPCs) [115]The identified miRNA signature included miRNA-17 -24 -146 -155 -128 and -181 holding early hematopoietic cellsat a stem-progenitor stage and blocking their maturationmiRNA-16 -103 and -107 responsible for block differentiationof later progenitor cells and miRNA-221 -222 and -223controlling terminal stages of hematopoietic differentiationSome miRNAs indeed presented lineage-specific expres-sion which suggested the limitation of function to eglymphoid (miRNA-146) erythroid (miRNA-221 and -222)or granulocytic (miRNA-223) development Inhibition oferythropoiesis and erythroleukemic cell growth through KITgene suppression by miR-221 and -222 was also reportedby Felli et al [116] whereas granulopoiesis regulation by aminicircuitry involving miR-223 NFIA and CEBPA by Faziet al [117] Other miRNAs were able to control differentprocesses eg myelopoiesis and erythropoiesis as miRNA-155 [115] Contrary to the results of Chen et al [113] whostudied hematopoiesis on murine modelmiR-142 expressionwas not detected in human hematopoietic cells byGeorgantaset al [115] Later Bissels et al deepened the knowledge aboutmiRNA-regulated hematopoiesis by combining analyses ofmicroRNA and mRNA profiles in CD133+ and CD34+hematopoietic stem and progenitor cells [118] In both typesof cells 25 highest expressed miRNAs accounted for 73-74 of the total miRNA pool However the most abundantmiRNAs were rather common for both progenitor cell typesexcept for miR-142-3p which was upregulated in CD34+cells to the level of up to 5000 copies per cell Remarkablyone of miR-142-3p targets seemed to be CD133 gene Theauthors found 18 miRNAs expressed differentially betweentheCD133+ (ancestral) andCD34+CD133- (later progenitor)cells miR-10a -99a -125a and b and miR-146a and bexpressed at highest level in CD133+ cells were postulatedto maintain the stem-cell character whereas miR-484 andother miRNAs upregulated in CD34+ cells probably blockedcell differentiation at a later stage Generally differentiallyexpressed miRNAs were involved in inhibition of differenti-ation prevention of apoptosis and cytoskeletal remodeling

In 2007Mi et al [54] showed that discrimination of AMLfrom ALL is possible through miRNA expression profilingAmong 27 miRNAs differentially expressed between AMLand ALL four were sufficient to distinguish these two typesof acute leukemia let-7b and miR-223 were significantlyupregulated in AML whereas miR-128a and miR-128b were

14 Journal of Oncology

downregulated in AML comparing to ALL [54] In 2010Wang et al [119] conducted similar research on Chinesecohort of 85 patients and found 16 miRNAs differentiallyexpressed between AML and ALL Half of them were previ-ously reported by Mi et al [54] (miR-23a miR-27ab miR-128a miR-128b miR-221 miR-222 miR-223 and let-7b) buteight have not identified previously in this context (miR-17 miR-20a miR-29ac miR-29b miR-146a miR-150 miR-155 and miR-196b) In addition prognostically relevant sig-natures were determined for ALL AML and non-M3-AMLOne miRNA miR-146a was strongly inversely correlatedwith OS of both acute leukemias in two independent patientcohorts

Garzon et al studied miRNA expression in APL cellstreated with all-trans-retinoic acid (ATRA) [120] and foundupregulation of miR-15a -15b -16-1 let-7a-3 let-7c let-7dmiR-223 -342 and -107 and downregulation of miR-181bThe observed ATRA modulation of NFIA RAS and BCL2gene expression corresponded with the fact the mentionedgenes were targets of miR-107 let-7a and miR-15amiR-16-1respectively Then Garzon et al [121] evaluated the miRNAexpression in 122 newly diagnosed AML cases comparingto CD34+ cells from 10 healthy donors and found 26differentially expressed miRNAs all downregulated in AMLeg miR-126 -130a -135 -93 -146 -106b and -125a Expres-sion level of some miRNAs was variable within AML andcorrelated with AML cytogenetics prognosis and clinical fea-tures For example miR-181 was downregulated particularlyin AML with multilineage dysplasia whereas miR-155 andmiR-181b positively correlated with WBC (white blood cellcount) In AML with balanced 11q23 translocations manytumor suppressormiRNAs targeting knownoncogenes weredownregulated eg miR-34b (targeting CDK4 and CCNE2)miR-15a (targeting BCL2) let-7 family (targeting RAS genes)and miR-196 (targeting HOX genes) In trisomy 8 onlyupregulated miRNAswere identified including those locatedat chromosome 8 eg miR-124a whose known target isCEBPA NK-AML was the most heterogeneous thereforethe identified miRNA signature was not predictive of NK-AML However five miRNAs overexpressed in AML (miR-199a and b miR-191 miR-25 and miR-20a) were associatedwith adverse patient outcome

Debernardi et al [122] showed strong correlation ofmiR-181a expression with the AML FAB subtypes (elevatedin M1 and M2) and with the expression of its predictedtargets Half of them eg BCL2L11 KLF3 MAP2K1 werenegatively correlated with miR-181a expression Havelangeet al [123] observed two other mRNA-miRNA interactionsnegative correlation between miR-181a and miR-181b miR-155 and miR-146 expression with that of genes involved inimmunity and inflammation (IRF7 and TLR4) and positivecorrelation between miR-23a miR-26a miR-128a and miR-145 expression level with that of proapoptotic genes (BIMand PTEN) Association of the last three miRNA withapoptosis was experimentally validated Lineage-associationswere showed for miR-23a and miR-196a (positive correla-tion with myeloid differentiation) miR-191 miR-222 andmiR-17 (negative correlation with erythroid differentiation)Interaction analysis induced the authors to conclude that

a small group of miRNAs coordinately regulates protein-coding transcriptome influencing the same group of genes(presumably the key players) within the pathway

In 2008 distinctive patterns of miRNA expression asso-ciated with cytogenetic and genetic AML subtypes weredetermined by Dixon-McIver et al[124] Jongen-Lavrencicet al [125] and Li et al [126] Dixon-McIver et al [124]measured the expression level of 157 miRNAs in 100 AMLpatients and two AML cell lines with the use of bead-based flow cytometric miRNA expression assay and found33 miRNAs with differential expression level between AMLand normal BM 17 upregulated (let-7e miR-27a -30d -142-5p -155 -181a -181b -181c -195 -221 -222 -324-5p -326 -328 -331 -340 -374) and 16 downregulated (miR-9lowast-15b -26a -30a-3p -34c -103 -147 -151 -182 -184 -199a-302blowast -302d -325 -367 -372) Moreover they associatedt(1517) translocation with upregulation of miRNAs locatedin the 14q32 imprinted domain eg miR-127 miR-154 miR-154lowast miR-299 miR-323 miR-368 and miR-370 In AMLwith inv(16) high level of miR-99a miR-100 and miR-224expression was observed whereas t(821) AML presentedhigh expression of miR-146a and a decrease of miR-133aHigh degree of variability across samples was noted for miR-10a and miR-125b Jongen-Lavrencic et al [125] found a setof strongly upregulated microRNAs (miR-382 -134 -376a -127 -299ndash5p and -323) in t(1517) partially overlapping withthe APL signature defined by Dixon-McIver et al [124]Clustering of AML cases with miRNA expression revealedthat inv(16) were sometimes mixed with t(821) and shareda part of miRNA signature which is not unexpected as theseboth AML subtypes belong to CBF AMLs Predictors of mostAML subclasses containing from a few to several dozenmiRNAs were built for AML with NPM1mutation and evenfor AMLwith FLT3-ITD or FLT3-TKDmutationswhichwerenot separated from other AMLs as a result of global miRNAexpression-based clustering

Li et al [126] found miRNA signatures composed of 2-24miRNAs able to distinguish AMLwith KMT2A rearrange-ment t(1517) t(821) plus inv(16) t(821) inv(16) and normalcontrols They noted that miR-126126lowast were specificallyoverexpressed in both t(821) and inv(16) AMLs while miR-224 miR-368 and miR-382 in t(1517) In KMT2A-AMLsignificant overexpression of miRNAs from polycistronicmiRNA cluster mir-17-92 was observed A minimal class-predictor contained only seven miRNAs miR-126 -126lowast -224 -368 -382 17-5p and -20a Interestingly differentialexpression of miR-126126lowast was not associated with DNAduplication nor mutation but probably resulted from epi-genetic regulation Gain- and loss-of-function experimentsrevealed that high expression of miR-126 inhibits apoptosisand increases cell viability and proliferation synergisticallywith the fusion gene RUNX1-RUNX1T1 From 674 predictedmiR-126 targets the authors empirically tested 12 genes andconfirmed that only one of them PLK2 was indeed regulatedbymiR-126

MicroRNA expression pattern correspondence with FABclassification was shown by Wang et al [127] who noted thatM1 M2 M3 and M4 tended to depart from each other moreeffectively thanM5 Apart frommiRNAs reported previously

Journal of Oncology 15

the authors identified a spectrum of new miRNAs whoseexpression strongly correlated with particular AML FABtypes eg miR-1300 miR-1180 miR-297 miR-610 and miR-650 overexpressed exclusively in AML M1 High expressionof some miRNAs was common in a few FAB types egmiR-181a-d miR-221 and miR-222 by M1 M2 and M3 Themost distinct miRNA expression pattern was shown in AMLM3 with 36 miRNAs strongly and exclusively upregulatedeg miR-370 miR-224 miR-382 miR-154 described earlierand miR-100 miR-195 miR-452 miR-654-3p not reportedpreviously Comparing to normal PBMCs all AML samplesdisplayed downregulation ofmiR-29a and miR-142-3p whichwere proposed by the authors as AML diagnostic biomarkers

Analysis of miRNA expression data in CN-AML per-formed by Marcucci et al [128] allowed identification ofmiRNA signature of prognostic relevance Upregulation ofmiR-181a and b was associated with low risk whereas over-expression of miR-124 -128-1 -194 219-5p 220a and -320with increased risk of failure to achieve CR relapse or deathIncreased miRNA levels were correlated with increasedexpression of genes involved in innate immunity encodingtoll-like receptors interleukins and caspases Some of themwere putative targets ofmiR-181

13 Mutation-Defined AML Subtypes

Progressive accumulation of transcriptomic data regardingboth mRNA and miRNA expression allowed more preciselycharacterize AMLs with the recurrent mutations

131 AML with Mutated CEBPA Valk et al [61] first triedto determine gene expression profiles specific for AML withparticular mutations Having a set of nearly 300 samplesthe authors were able to distinguish a unique gene expres-sion signature for AML with CEBPA mutations CEBPAgene mutated in 5 to 15 of all AML cases encodes acritical regulator of hematopoietic stem-cell maintenanceand myeloid differentiation therefore unique gene expres-sion pattern was not unexpected for samples with theloss-of-function CEBPA mutation Valk et al [61] showedthat the most prominent features of CEBPA-mutated AMLwere CD7 overexpression and downregulation of CTNNA1TUBB and NDFIP1 Interestingly hierarchical clustering ofall AML samples segregated AMLs with CEBPA mutationsinto 2 different clusters and one of them included alsosamples without any known mutations or chromosomalaberrations In their subsequent paper the authors usingbisulfite genomic sequencing revealed that this previouslyunidentified subset of AML was represented by sampleswhere CEBPA gene promoter was hypermethylated [129]In fact within this mysterious AML cluster identified byValk et al [61] CEBPA levels were very high in AML caseswith CEBPA mutations whereas in AML with wt variantCEBPA expression was minimal or undetectable due to itsepigenetic silencing [129] Detailed characteristics ofCEBPA-silenced AML samples showed that they expressed bothmyeloid markers (CD13 CD33 and MPO) and T-lymphoidmarkers (eg CD7 mentioned above) In addition high

expression of the myeloid oncogene TRIB2 and NOTCH1gene encoding a membrane receptor and transcriptionalregulator of T-cell development was noted In a part of thoseAML patients activating NOTCH1 mutation was identifiedMoreover TRIB2 was determined to be a direct target ofNOTCH1 signaling Later the authors found that CEBPAmethylation was accompanied by aberrant hypermethylationof many genes compared to CEBPA-mutated AMLs or withnormal CD34+ hematopoietic progenitor cells [130] Thiscould explain an observed in vitro decreased response ofCEBPA-silenced AML to myeloid growth factors and makesthis AML subtype susceptible to dynamically developingtreatment with demethylating agents Interestingly compari-son of genome-wide methylation pattern with GEP revealedonly a minimal overlap (12 unique genes including CEBPA)between the differentially expressed and differentially methy-lated genes This suggested that gene expression and genomemethylation are biologically independent processes

132 AMLwith Mutated NPM1 and the Paradox of HOXGeneExpression Since the time a 4-nucleotide insertion in NPM1gene and its significance in AML was discovered by Faliniet al in 2005 [52] much attention was paid to unveilingthe mechanism of AML triggered by NPM1mutation NPM1encodes a multifunctional protein involved in ribosomebiosynthesis and transport DNAreplication and repair chro-matin remodeling protein chaperoning regulation of cellcycle embryogenesis and oncogenesis [131 132] AlthoughNPM1 protein localizes mainly in the nucleolus it constantlyshuttles between nucleus and cytoplasm The 4-nucleotideinsertion in the last exon of NPM1 gene results in aberrantcytoplasmic accumulation of a protein and consequentlyaffects its functions NPM1 gene is mutated in about 30 ofAML and in 50 to 60 cases of adult NK-AML Given thatNPM-cytoplasmic positive (NPMc+) AML reveals uniquemolecular and clinical features [133] it was introduced intoWHO classification as a separate entity [13] An invaluablecontribution to cognition of AML promotion by NPM1mutation was made by GEP

Alcalay et al [134] first claimedNPMc+AML representeda distinct entity which can be easily distinguished from otherAML samples regardless of the karyotype They comparedglobal gene expression of 58 AML NPMc+ samples withprevalence of NK-AML and frequent occurrence of FLT3mutations to the group of 20NK-AMLswithoutNPM1muta-tions and lower occurrence of FLT3mutations Unsupervisedapproach showed NPM1 mutation status was the strongestclustering parameter A selected 369-gene-predictor effi-ciently segregated NPMc+ from NPMc- patients Interest-ingly NPM1 transcript level did not differ between thesetwo groups indicating the lack of NPM1 mutation influenceon NPM1 gene expression I confirmed this observation byanalysis of NPM1 alternative transcripts with droplet digitalPCR (ddPCR) [135] In theNPMc+patientsCD34 andCD133antigens as well as POU4F1 and CDKN2C were suppressedwhereas a number of homeodomain-containing transcrip-tion factors including HOX and TALE genes were activated[134] Because several HOX genes are highly expressed in

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 13

CYP3A4 inhibitors as a novel strategy to treat FLT3-ITDAML

Passaro et al [99] who studied the BM vasculaturein AML using intravital two-photon microscopy associatedincreased vascular permeability in the BM microenviron-ment with disease progression and poor treatment responseTranscriptome analysis of BM-derived endothelial cells viaRNA-seq identified deregulation of genes involved in vascu-lature development angiogenesis and response to hypoxiaNox4 gene encoding NAPDH oxidase responsible for pro-duction of reactive oxygen species (ROS) activation of nitricoxide synthase 3 (NOS3) and release of nitric oxide (NO)was particularly upregulated Increased NO level contributedto the vascular leakiness in AML-engrafted mice and wasassociated with poor prognosis in AML patients Appli-cation of NO synthase inhibitors combined with standardchemotherapy restored normal vasculature and improved thetreatment response demonstrating the efficacy of combinedleukemia-niche therapies

The role of lymphocytes and other blood cells has longbeen neglected in the studies of AML However Le Dieu etal found that the absolute number of T-cells circulating in PBof de novo AML patients not belonging to malignant cloneswas increased compared to healthy controls [100] Activationof T-cells might reflect a response to growth signals present ina local microenvironment GEP of CD4+ and CD8+ T-cellsfrom AML patients and healthy volunteers revealed globaldifferences in transcription pattern with little similaritiesto T-cells of CLL patients Particularly genes associatedwith the actin cytoskeleton and cellular polarization werederegulated in AML T-cells According to the authors T-cellaberrant activation leads to their dysfunction and impairedimmune response which is not a sufficient weapon againstthe leukemic blasts Here a rationale to apply immunomod-ulatory drugs appears

12 Discovering the Power of Small MoleculesmiRNA Expression Profiling

The discovery of regulatory role of RNA in cell and organ-ism development completely changed our understanding ofbiology and at least partially explained the paradox that is thelarge size of mammalian genomes of which only a small per-centage are the protein-coding genes [101ndash103] Among smallregulatory RNAsmicroRNAs (miRNAs) termed due to theirsmall size (18-23 nt) are best recognized [104 105] Thefunction of miRNAs in gene expression regulation (usuallyrepression) controlling cell fate and normal developmentalprocesses as well as oncogenesis is well-established [106ndash109] As one miRNA targets multiple transcripts [110] dys-function of miRNA may result in a wide-scale deregulationof gene expression often triggering a cascade of eventsleading to pathogenesis In 2002 Calin et al demonstratedmiR-15 and miR-16 are located at chromosome 13q14 regionfrequently deleted in B-cell chronic lymphocytic leukemias(B-CLL) [111] Then more than half miRNA genes werelinked with cancer-associated genomic regions or fragilesites and their amplification or deletion in human cancers

supported miRNA role in malignant transformation [112]Since 2005 when Lu et al classified multiple human can-cers including AML based on miRNA expression profilesexclusively and proved general downregulation of miRNAsin tumors compared to normal tissues [51] miRNAs startedto be widely investigated in cancers and other diseases

MicroRNAs were also described as regulators of mam-malian hematopoiesis [113 114] In 2004 three microRNAswhichmodulatemouse hematopoietic lineage differentiationwere found byChen et al [113]miR-181 a positive regulator ofB-lymphoid cell differentiation miR-223 nearly exclusivelyexpressed in BM and myeloid cells and miR-142 found athighest levels in B-lymphoid and myeloid lineages Geor-gantas et al identified 33 miRNAs specifically expressed inCD34+ hematopoietic stem-progenitor cells (HSPCs) [115]The identified miRNA signature included miRNA-17 -24 -146 -155 -128 and -181 holding early hematopoietic cellsat a stem-progenitor stage and blocking their maturationmiRNA-16 -103 and -107 responsible for block differentiationof later progenitor cells and miRNA-221 -222 and -223controlling terminal stages of hematopoietic differentiationSome miRNAs indeed presented lineage-specific expres-sion which suggested the limitation of function to eglymphoid (miRNA-146) erythroid (miRNA-221 and -222)or granulocytic (miRNA-223) development Inhibition oferythropoiesis and erythroleukemic cell growth through KITgene suppression by miR-221 and -222 was also reportedby Felli et al [116] whereas granulopoiesis regulation by aminicircuitry involving miR-223 NFIA and CEBPA by Faziet al [117] Other miRNAs were able to control differentprocesses eg myelopoiesis and erythropoiesis as miRNA-155 [115] Contrary to the results of Chen et al [113] whostudied hematopoiesis on murine modelmiR-142 expressionwas not detected in human hematopoietic cells byGeorgantaset al [115] Later Bissels et al deepened the knowledge aboutmiRNA-regulated hematopoiesis by combining analyses ofmicroRNA and mRNA profiles in CD133+ and CD34+hematopoietic stem and progenitor cells [118] In both typesof cells 25 highest expressed miRNAs accounted for 73-74 of the total miRNA pool However the most abundantmiRNAs were rather common for both progenitor cell typesexcept for miR-142-3p which was upregulated in CD34+cells to the level of up to 5000 copies per cell Remarkablyone of miR-142-3p targets seemed to be CD133 gene Theauthors found 18 miRNAs expressed differentially betweentheCD133+ (ancestral) andCD34+CD133- (later progenitor)cells miR-10a -99a -125a and b and miR-146a and bexpressed at highest level in CD133+ cells were postulatedto maintain the stem-cell character whereas miR-484 andother miRNAs upregulated in CD34+ cells probably blockedcell differentiation at a later stage Generally differentiallyexpressed miRNAs were involved in inhibition of differenti-ation prevention of apoptosis and cytoskeletal remodeling

In 2007Mi et al [54] showed that discrimination of AMLfrom ALL is possible through miRNA expression profilingAmong 27 miRNAs differentially expressed between AMLand ALL four were sufficient to distinguish these two typesof acute leukemia let-7b and miR-223 were significantlyupregulated in AML whereas miR-128a and miR-128b were

14 Journal of Oncology

downregulated in AML comparing to ALL [54] In 2010Wang et al [119] conducted similar research on Chinesecohort of 85 patients and found 16 miRNAs differentiallyexpressed between AML and ALL Half of them were previ-ously reported by Mi et al [54] (miR-23a miR-27ab miR-128a miR-128b miR-221 miR-222 miR-223 and let-7b) buteight have not identified previously in this context (miR-17 miR-20a miR-29ac miR-29b miR-146a miR-150 miR-155 and miR-196b) In addition prognostically relevant sig-natures were determined for ALL AML and non-M3-AMLOne miRNA miR-146a was strongly inversely correlatedwith OS of both acute leukemias in two independent patientcohorts

Garzon et al studied miRNA expression in APL cellstreated with all-trans-retinoic acid (ATRA) [120] and foundupregulation of miR-15a -15b -16-1 let-7a-3 let-7c let-7dmiR-223 -342 and -107 and downregulation of miR-181bThe observed ATRA modulation of NFIA RAS and BCL2gene expression corresponded with the fact the mentionedgenes were targets of miR-107 let-7a and miR-15amiR-16-1respectively Then Garzon et al [121] evaluated the miRNAexpression in 122 newly diagnosed AML cases comparingto CD34+ cells from 10 healthy donors and found 26differentially expressed miRNAs all downregulated in AMLeg miR-126 -130a -135 -93 -146 -106b and -125a Expres-sion level of some miRNAs was variable within AML andcorrelated with AML cytogenetics prognosis and clinical fea-tures For example miR-181 was downregulated particularlyin AML with multilineage dysplasia whereas miR-155 andmiR-181b positively correlated with WBC (white blood cellcount) In AML with balanced 11q23 translocations manytumor suppressormiRNAs targeting knownoncogenes weredownregulated eg miR-34b (targeting CDK4 and CCNE2)miR-15a (targeting BCL2) let-7 family (targeting RAS genes)and miR-196 (targeting HOX genes) In trisomy 8 onlyupregulated miRNAswere identified including those locatedat chromosome 8 eg miR-124a whose known target isCEBPA NK-AML was the most heterogeneous thereforethe identified miRNA signature was not predictive of NK-AML However five miRNAs overexpressed in AML (miR-199a and b miR-191 miR-25 and miR-20a) were associatedwith adverse patient outcome

Debernardi et al [122] showed strong correlation ofmiR-181a expression with the AML FAB subtypes (elevatedin M1 and M2) and with the expression of its predictedtargets Half of them eg BCL2L11 KLF3 MAP2K1 werenegatively correlated with miR-181a expression Havelangeet al [123] observed two other mRNA-miRNA interactionsnegative correlation between miR-181a and miR-181b miR-155 and miR-146 expression with that of genes involved inimmunity and inflammation (IRF7 and TLR4) and positivecorrelation between miR-23a miR-26a miR-128a and miR-145 expression level with that of proapoptotic genes (BIMand PTEN) Association of the last three miRNA withapoptosis was experimentally validated Lineage-associationswere showed for miR-23a and miR-196a (positive correla-tion with myeloid differentiation) miR-191 miR-222 andmiR-17 (negative correlation with erythroid differentiation)Interaction analysis induced the authors to conclude that

a small group of miRNAs coordinately regulates protein-coding transcriptome influencing the same group of genes(presumably the key players) within the pathway

In 2008 distinctive patterns of miRNA expression asso-ciated with cytogenetic and genetic AML subtypes weredetermined by Dixon-McIver et al[124] Jongen-Lavrencicet al [125] and Li et al [126] Dixon-McIver et al [124]measured the expression level of 157 miRNAs in 100 AMLpatients and two AML cell lines with the use of bead-based flow cytometric miRNA expression assay and found33 miRNAs with differential expression level between AMLand normal BM 17 upregulated (let-7e miR-27a -30d -142-5p -155 -181a -181b -181c -195 -221 -222 -324-5p -326 -328 -331 -340 -374) and 16 downregulated (miR-9lowast-15b -26a -30a-3p -34c -103 -147 -151 -182 -184 -199a-302blowast -302d -325 -367 -372) Moreover they associatedt(1517) translocation with upregulation of miRNAs locatedin the 14q32 imprinted domain eg miR-127 miR-154 miR-154lowast miR-299 miR-323 miR-368 and miR-370 In AMLwith inv(16) high level of miR-99a miR-100 and miR-224expression was observed whereas t(821) AML presentedhigh expression of miR-146a and a decrease of miR-133aHigh degree of variability across samples was noted for miR-10a and miR-125b Jongen-Lavrencic et al [125] found a setof strongly upregulated microRNAs (miR-382 -134 -376a -127 -299ndash5p and -323) in t(1517) partially overlapping withthe APL signature defined by Dixon-McIver et al [124]Clustering of AML cases with miRNA expression revealedthat inv(16) were sometimes mixed with t(821) and shareda part of miRNA signature which is not unexpected as theseboth AML subtypes belong to CBF AMLs Predictors of mostAML subclasses containing from a few to several dozenmiRNAs were built for AML with NPM1mutation and evenfor AMLwith FLT3-ITD or FLT3-TKDmutationswhichwerenot separated from other AMLs as a result of global miRNAexpression-based clustering

Li et al [126] found miRNA signatures composed of 2-24miRNAs able to distinguish AMLwith KMT2A rearrange-ment t(1517) t(821) plus inv(16) t(821) inv(16) and normalcontrols They noted that miR-126126lowast were specificallyoverexpressed in both t(821) and inv(16) AMLs while miR-224 miR-368 and miR-382 in t(1517) In KMT2A-AMLsignificant overexpression of miRNAs from polycistronicmiRNA cluster mir-17-92 was observed A minimal class-predictor contained only seven miRNAs miR-126 -126lowast -224 -368 -382 17-5p and -20a Interestingly differentialexpression of miR-126126lowast was not associated with DNAduplication nor mutation but probably resulted from epi-genetic regulation Gain- and loss-of-function experimentsrevealed that high expression of miR-126 inhibits apoptosisand increases cell viability and proliferation synergisticallywith the fusion gene RUNX1-RUNX1T1 From 674 predictedmiR-126 targets the authors empirically tested 12 genes andconfirmed that only one of them PLK2 was indeed regulatedbymiR-126

MicroRNA expression pattern correspondence with FABclassification was shown by Wang et al [127] who noted thatM1 M2 M3 and M4 tended to depart from each other moreeffectively thanM5 Apart frommiRNAs reported previously

Journal of Oncology 15

the authors identified a spectrum of new miRNAs whoseexpression strongly correlated with particular AML FABtypes eg miR-1300 miR-1180 miR-297 miR-610 and miR-650 overexpressed exclusively in AML M1 High expressionof some miRNAs was common in a few FAB types egmiR-181a-d miR-221 and miR-222 by M1 M2 and M3 Themost distinct miRNA expression pattern was shown in AMLM3 with 36 miRNAs strongly and exclusively upregulatedeg miR-370 miR-224 miR-382 miR-154 described earlierand miR-100 miR-195 miR-452 miR-654-3p not reportedpreviously Comparing to normal PBMCs all AML samplesdisplayed downregulation ofmiR-29a and miR-142-3p whichwere proposed by the authors as AML diagnostic biomarkers

Analysis of miRNA expression data in CN-AML per-formed by Marcucci et al [128] allowed identification ofmiRNA signature of prognostic relevance Upregulation ofmiR-181a and b was associated with low risk whereas over-expression of miR-124 -128-1 -194 219-5p 220a and -320with increased risk of failure to achieve CR relapse or deathIncreased miRNA levels were correlated with increasedexpression of genes involved in innate immunity encodingtoll-like receptors interleukins and caspases Some of themwere putative targets ofmiR-181

13 Mutation-Defined AML Subtypes

Progressive accumulation of transcriptomic data regardingboth mRNA and miRNA expression allowed more preciselycharacterize AMLs with the recurrent mutations

131 AML with Mutated CEBPA Valk et al [61] first triedto determine gene expression profiles specific for AML withparticular mutations Having a set of nearly 300 samplesthe authors were able to distinguish a unique gene expres-sion signature for AML with CEBPA mutations CEBPAgene mutated in 5 to 15 of all AML cases encodes acritical regulator of hematopoietic stem-cell maintenanceand myeloid differentiation therefore unique gene expres-sion pattern was not unexpected for samples with theloss-of-function CEBPA mutation Valk et al [61] showedthat the most prominent features of CEBPA-mutated AMLwere CD7 overexpression and downregulation of CTNNA1TUBB and NDFIP1 Interestingly hierarchical clustering ofall AML samples segregated AMLs with CEBPA mutationsinto 2 different clusters and one of them included alsosamples without any known mutations or chromosomalaberrations In their subsequent paper the authors usingbisulfite genomic sequencing revealed that this previouslyunidentified subset of AML was represented by sampleswhere CEBPA gene promoter was hypermethylated [129]In fact within this mysterious AML cluster identified byValk et al [61] CEBPA levels were very high in AML caseswith CEBPA mutations whereas in AML with wt variantCEBPA expression was minimal or undetectable due to itsepigenetic silencing [129] Detailed characteristics ofCEBPA-silenced AML samples showed that they expressed bothmyeloid markers (CD13 CD33 and MPO) and T-lymphoidmarkers (eg CD7 mentioned above) In addition high

expression of the myeloid oncogene TRIB2 and NOTCH1gene encoding a membrane receptor and transcriptionalregulator of T-cell development was noted In a part of thoseAML patients activating NOTCH1 mutation was identifiedMoreover TRIB2 was determined to be a direct target ofNOTCH1 signaling Later the authors found that CEBPAmethylation was accompanied by aberrant hypermethylationof many genes compared to CEBPA-mutated AMLs or withnormal CD34+ hematopoietic progenitor cells [130] Thiscould explain an observed in vitro decreased response ofCEBPA-silenced AML to myeloid growth factors and makesthis AML subtype susceptible to dynamically developingtreatment with demethylating agents Interestingly compari-son of genome-wide methylation pattern with GEP revealedonly a minimal overlap (12 unique genes including CEBPA)between the differentially expressed and differentially methy-lated genes This suggested that gene expression and genomemethylation are biologically independent processes

132 AMLwith Mutated NPM1 and the Paradox of HOXGeneExpression Since the time a 4-nucleotide insertion in NPM1gene and its significance in AML was discovered by Faliniet al in 2005 [52] much attention was paid to unveilingthe mechanism of AML triggered by NPM1mutation NPM1encodes a multifunctional protein involved in ribosomebiosynthesis and transport DNAreplication and repair chro-matin remodeling protein chaperoning regulation of cellcycle embryogenesis and oncogenesis [131 132] AlthoughNPM1 protein localizes mainly in the nucleolus it constantlyshuttles between nucleus and cytoplasm The 4-nucleotideinsertion in the last exon of NPM1 gene results in aberrantcytoplasmic accumulation of a protein and consequentlyaffects its functions NPM1 gene is mutated in about 30 ofAML and in 50 to 60 cases of adult NK-AML Given thatNPM-cytoplasmic positive (NPMc+) AML reveals uniquemolecular and clinical features [133] it was introduced intoWHO classification as a separate entity [13] An invaluablecontribution to cognition of AML promotion by NPM1mutation was made by GEP

Alcalay et al [134] first claimedNPMc+AML representeda distinct entity which can be easily distinguished from otherAML samples regardless of the karyotype They comparedglobal gene expression of 58 AML NPMc+ samples withprevalence of NK-AML and frequent occurrence of FLT3mutations to the group of 20NK-AMLswithoutNPM1muta-tions and lower occurrence of FLT3mutations Unsupervisedapproach showed NPM1 mutation status was the strongestclustering parameter A selected 369-gene-predictor effi-ciently segregated NPMc+ from NPMc- patients Interest-ingly NPM1 transcript level did not differ between thesetwo groups indicating the lack of NPM1 mutation influenceon NPM1 gene expression I confirmed this observation byanalysis of NPM1 alternative transcripts with droplet digitalPCR (ddPCR) [135] In theNPMc+patientsCD34 andCD133antigens as well as POU4F1 and CDKN2C were suppressedwhereas a number of homeodomain-containing transcrip-tion factors including HOX and TALE genes were activated[134] Because several HOX genes are highly expressed in

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

14 Journal of Oncology

downregulated in AML comparing to ALL [54] In 2010Wang et al [119] conducted similar research on Chinesecohort of 85 patients and found 16 miRNAs differentiallyexpressed between AML and ALL Half of them were previ-ously reported by Mi et al [54] (miR-23a miR-27ab miR-128a miR-128b miR-221 miR-222 miR-223 and let-7b) buteight have not identified previously in this context (miR-17 miR-20a miR-29ac miR-29b miR-146a miR-150 miR-155 and miR-196b) In addition prognostically relevant sig-natures were determined for ALL AML and non-M3-AMLOne miRNA miR-146a was strongly inversely correlatedwith OS of both acute leukemias in two independent patientcohorts

Garzon et al studied miRNA expression in APL cellstreated with all-trans-retinoic acid (ATRA) [120] and foundupregulation of miR-15a -15b -16-1 let-7a-3 let-7c let-7dmiR-223 -342 and -107 and downregulation of miR-181bThe observed ATRA modulation of NFIA RAS and BCL2gene expression corresponded with the fact the mentionedgenes were targets of miR-107 let-7a and miR-15amiR-16-1respectively Then Garzon et al [121] evaluated the miRNAexpression in 122 newly diagnosed AML cases comparingto CD34+ cells from 10 healthy donors and found 26differentially expressed miRNAs all downregulated in AMLeg miR-126 -130a -135 -93 -146 -106b and -125a Expres-sion level of some miRNAs was variable within AML andcorrelated with AML cytogenetics prognosis and clinical fea-tures For example miR-181 was downregulated particularlyin AML with multilineage dysplasia whereas miR-155 andmiR-181b positively correlated with WBC (white blood cellcount) In AML with balanced 11q23 translocations manytumor suppressormiRNAs targeting knownoncogenes weredownregulated eg miR-34b (targeting CDK4 and CCNE2)miR-15a (targeting BCL2) let-7 family (targeting RAS genes)and miR-196 (targeting HOX genes) In trisomy 8 onlyupregulated miRNAswere identified including those locatedat chromosome 8 eg miR-124a whose known target isCEBPA NK-AML was the most heterogeneous thereforethe identified miRNA signature was not predictive of NK-AML However five miRNAs overexpressed in AML (miR-199a and b miR-191 miR-25 and miR-20a) were associatedwith adverse patient outcome

Debernardi et al [122] showed strong correlation ofmiR-181a expression with the AML FAB subtypes (elevatedin M1 and M2) and with the expression of its predictedtargets Half of them eg BCL2L11 KLF3 MAP2K1 werenegatively correlated with miR-181a expression Havelangeet al [123] observed two other mRNA-miRNA interactionsnegative correlation between miR-181a and miR-181b miR-155 and miR-146 expression with that of genes involved inimmunity and inflammation (IRF7 and TLR4) and positivecorrelation between miR-23a miR-26a miR-128a and miR-145 expression level with that of proapoptotic genes (BIMand PTEN) Association of the last three miRNA withapoptosis was experimentally validated Lineage-associationswere showed for miR-23a and miR-196a (positive correla-tion with myeloid differentiation) miR-191 miR-222 andmiR-17 (negative correlation with erythroid differentiation)Interaction analysis induced the authors to conclude that

a small group of miRNAs coordinately regulates protein-coding transcriptome influencing the same group of genes(presumably the key players) within the pathway

In 2008 distinctive patterns of miRNA expression asso-ciated with cytogenetic and genetic AML subtypes weredetermined by Dixon-McIver et al[124] Jongen-Lavrencicet al [125] and Li et al [126] Dixon-McIver et al [124]measured the expression level of 157 miRNAs in 100 AMLpatients and two AML cell lines with the use of bead-based flow cytometric miRNA expression assay and found33 miRNAs with differential expression level between AMLand normal BM 17 upregulated (let-7e miR-27a -30d -142-5p -155 -181a -181b -181c -195 -221 -222 -324-5p -326 -328 -331 -340 -374) and 16 downregulated (miR-9lowast-15b -26a -30a-3p -34c -103 -147 -151 -182 -184 -199a-302blowast -302d -325 -367 -372) Moreover they associatedt(1517) translocation with upregulation of miRNAs locatedin the 14q32 imprinted domain eg miR-127 miR-154 miR-154lowast miR-299 miR-323 miR-368 and miR-370 In AMLwith inv(16) high level of miR-99a miR-100 and miR-224expression was observed whereas t(821) AML presentedhigh expression of miR-146a and a decrease of miR-133aHigh degree of variability across samples was noted for miR-10a and miR-125b Jongen-Lavrencic et al [125] found a setof strongly upregulated microRNAs (miR-382 -134 -376a -127 -299ndash5p and -323) in t(1517) partially overlapping withthe APL signature defined by Dixon-McIver et al [124]Clustering of AML cases with miRNA expression revealedthat inv(16) were sometimes mixed with t(821) and shareda part of miRNA signature which is not unexpected as theseboth AML subtypes belong to CBF AMLs Predictors of mostAML subclasses containing from a few to several dozenmiRNAs were built for AML with NPM1mutation and evenfor AMLwith FLT3-ITD or FLT3-TKDmutationswhichwerenot separated from other AMLs as a result of global miRNAexpression-based clustering

Li et al [126] found miRNA signatures composed of 2-24miRNAs able to distinguish AMLwith KMT2A rearrange-ment t(1517) t(821) plus inv(16) t(821) inv(16) and normalcontrols They noted that miR-126126lowast were specificallyoverexpressed in both t(821) and inv(16) AMLs while miR-224 miR-368 and miR-382 in t(1517) In KMT2A-AMLsignificant overexpression of miRNAs from polycistronicmiRNA cluster mir-17-92 was observed A minimal class-predictor contained only seven miRNAs miR-126 -126lowast -224 -368 -382 17-5p and -20a Interestingly differentialexpression of miR-126126lowast was not associated with DNAduplication nor mutation but probably resulted from epi-genetic regulation Gain- and loss-of-function experimentsrevealed that high expression of miR-126 inhibits apoptosisand increases cell viability and proliferation synergisticallywith the fusion gene RUNX1-RUNX1T1 From 674 predictedmiR-126 targets the authors empirically tested 12 genes andconfirmed that only one of them PLK2 was indeed regulatedbymiR-126

MicroRNA expression pattern correspondence with FABclassification was shown by Wang et al [127] who noted thatM1 M2 M3 and M4 tended to depart from each other moreeffectively thanM5 Apart frommiRNAs reported previously

Journal of Oncology 15

the authors identified a spectrum of new miRNAs whoseexpression strongly correlated with particular AML FABtypes eg miR-1300 miR-1180 miR-297 miR-610 and miR-650 overexpressed exclusively in AML M1 High expressionof some miRNAs was common in a few FAB types egmiR-181a-d miR-221 and miR-222 by M1 M2 and M3 Themost distinct miRNA expression pattern was shown in AMLM3 with 36 miRNAs strongly and exclusively upregulatedeg miR-370 miR-224 miR-382 miR-154 described earlierand miR-100 miR-195 miR-452 miR-654-3p not reportedpreviously Comparing to normal PBMCs all AML samplesdisplayed downregulation ofmiR-29a and miR-142-3p whichwere proposed by the authors as AML diagnostic biomarkers

Analysis of miRNA expression data in CN-AML per-formed by Marcucci et al [128] allowed identification ofmiRNA signature of prognostic relevance Upregulation ofmiR-181a and b was associated with low risk whereas over-expression of miR-124 -128-1 -194 219-5p 220a and -320with increased risk of failure to achieve CR relapse or deathIncreased miRNA levels were correlated with increasedexpression of genes involved in innate immunity encodingtoll-like receptors interleukins and caspases Some of themwere putative targets ofmiR-181

13 Mutation-Defined AML Subtypes

Progressive accumulation of transcriptomic data regardingboth mRNA and miRNA expression allowed more preciselycharacterize AMLs with the recurrent mutations

131 AML with Mutated CEBPA Valk et al [61] first triedto determine gene expression profiles specific for AML withparticular mutations Having a set of nearly 300 samplesthe authors were able to distinguish a unique gene expres-sion signature for AML with CEBPA mutations CEBPAgene mutated in 5 to 15 of all AML cases encodes acritical regulator of hematopoietic stem-cell maintenanceand myeloid differentiation therefore unique gene expres-sion pattern was not unexpected for samples with theloss-of-function CEBPA mutation Valk et al [61] showedthat the most prominent features of CEBPA-mutated AMLwere CD7 overexpression and downregulation of CTNNA1TUBB and NDFIP1 Interestingly hierarchical clustering ofall AML samples segregated AMLs with CEBPA mutationsinto 2 different clusters and one of them included alsosamples without any known mutations or chromosomalaberrations In their subsequent paper the authors usingbisulfite genomic sequencing revealed that this previouslyunidentified subset of AML was represented by sampleswhere CEBPA gene promoter was hypermethylated [129]In fact within this mysterious AML cluster identified byValk et al [61] CEBPA levels were very high in AML caseswith CEBPA mutations whereas in AML with wt variantCEBPA expression was minimal or undetectable due to itsepigenetic silencing [129] Detailed characteristics ofCEBPA-silenced AML samples showed that they expressed bothmyeloid markers (CD13 CD33 and MPO) and T-lymphoidmarkers (eg CD7 mentioned above) In addition high

expression of the myeloid oncogene TRIB2 and NOTCH1gene encoding a membrane receptor and transcriptionalregulator of T-cell development was noted In a part of thoseAML patients activating NOTCH1 mutation was identifiedMoreover TRIB2 was determined to be a direct target ofNOTCH1 signaling Later the authors found that CEBPAmethylation was accompanied by aberrant hypermethylationof many genes compared to CEBPA-mutated AMLs or withnormal CD34+ hematopoietic progenitor cells [130] Thiscould explain an observed in vitro decreased response ofCEBPA-silenced AML to myeloid growth factors and makesthis AML subtype susceptible to dynamically developingtreatment with demethylating agents Interestingly compari-son of genome-wide methylation pattern with GEP revealedonly a minimal overlap (12 unique genes including CEBPA)between the differentially expressed and differentially methy-lated genes This suggested that gene expression and genomemethylation are biologically independent processes

132 AMLwith Mutated NPM1 and the Paradox of HOXGeneExpression Since the time a 4-nucleotide insertion in NPM1gene and its significance in AML was discovered by Faliniet al in 2005 [52] much attention was paid to unveilingthe mechanism of AML triggered by NPM1mutation NPM1encodes a multifunctional protein involved in ribosomebiosynthesis and transport DNAreplication and repair chro-matin remodeling protein chaperoning regulation of cellcycle embryogenesis and oncogenesis [131 132] AlthoughNPM1 protein localizes mainly in the nucleolus it constantlyshuttles between nucleus and cytoplasm The 4-nucleotideinsertion in the last exon of NPM1 gene results in aberrantcytoplasmic accumulation of a protein and consequentlyaffects its functions NPM1 gene is mutated in about 30 ofAML and in 50 to 60 cases of adult NK-AML Given thatNPM-cytoplasmic positive (NPMc+) AML reveals uniquemolecular and clinical features [133] it was introduced intoWHO classification as a separate entity [13] An invaluablecontribution to cognition of AML promotion by NPM1mutation was made by GEP

Alcalay et al [134] first claimedNPMc+AML representeda distinct entity which can be easily distinguished from otherAML samples regardless of the karyotype They comparedglobal gene expression of 58 AML NPMc+ samples withprevalence of NK-AML and frequent occurrence of FLT3mutations to the group of 20NK-AMLswithoutNPM1muta-tions and lower occurrence of FLT3mutations Unsupervisedapproach showed NPM1 mutation status was the strongestclustering parameter A selected 369-gene-predictor effi-ciently segregated NPMc+ from NPMc- patients Interest-ingly NPM1 transcript level did not differ between thesetwo groups indicating the lack of NPM1 mutation influenceon NPM1 gene expression I confirmed this observation byanalysis of NPM1 alternative transcripts with droplet digitalPCR (ddPCR) [135] In theNPMc+patientsCD34 andCD133antigens as well as POU4F1 and CDKN2C were suppressedwhereas a number of homeodomain-containing transcrip-tion factors including HOX and TALE genes were activated[134] Because several HOX genes are highly expressed in

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 15

the authors identified a spectrum of new miRNAs whoseexpression strongly correlated with particular AML FABtypes eg miR-1300 miR-1180 miR-297 miR-610 and miR-650 overexpressed exclusively in AML M1 High expressionof some miRNAs was common in a few FAB types egmiR-181a-d miR-221 and miR-222 by M1 M2 and M3 Themost distinct miRNA expression pattern was shown in AMLM3 with 36 miRNAs strongly and exclusively upregulatedeg miR-370 miR-224 miR-382 miR-154 described earlierand miR-100 miR-195 miR-452 miR-654-3p not reportedpreviously Comparing to normal PBMCs all AML samplesdisplayed downregulation ofmiR-29a and miR-142-3p whichwere proposed by the authors as AML diagnostic biomarkers

Analysis of miRNA expression data in CN-AML per-formed by Marcucci et al [128] allowed identification ofmiRNA signature of prognostic relevance Upregulation ofmiR-181a and b was associated with low risk whereas over-expression of miR-124 -128-1 -194 219-5p 220a and -320with increased risk of failure to achieve CR relapse or deathIncreased miRNA levels were correlated with increasedexpression of genes involved in innate immunity encodingtoll-like receptors interleukins and caspases Some of themwere putative targets ofmiR-181

13 Mutation-Defined AML Subtypes

Progressive accumulation of transcriptomic data regardingboth mRNA and miRNA expression allowed more preciselycharacterize AMLs with the recurrent mutations

131 AML with Mutated CEBPA Valk et al [61] first triedto determine gene expression profiles specific for AML withparticular mutations Having a set of nearly 300 samplesthe authors were able to distinguish a unique gene expres-sion signature for AML with CEBPA mutations CEBPAgene mutated in 5 to 15 of all AML cases encodes acritical regulator of hematopoietic stem-cell maintenanceand myeloid differentiation therefore unique gene expres-sion pattern was not unexpected for samples with theloss-of-function CEBPA mutation Valk et al [61] showedthat the most prominent features of CEBPA-mutated AMLwere CD7 overexpression and downregulation of CTNNA1TUBB and NDFIP1 Interestingly hierarchical clustering ofall AML samples segregated AMLs with CEBPA mutationsinto 2 different clusters and one of them included alsosamples without any known mutations or chromosomalaberrations In their subsequent paper the authors usingbisulfite genomic sequencing revealed that this previouslyunidentified subset of AML was represented by sampleswhere CEBPA gene promoter was hypermethylated [129]In fact within this mysterious AML cluster identified byValk et al [61] CEBPA levels were very high in AML caseswith CEBPA mutations whereas in AML with wt variantCEBPA expression was minimal or undetectable due to itsepigenetic silencing [129] Detailed characteristics ofCEBPA-silenced AML samples showed that they expressed bothmyeloid markers (CD13 CD33 and MPO) and T-lymphoidmarkers (eg CD7 mentioned above) In addition high

expression of the myeloid oncogene TRIB2 and NOTCH1gene encoding a membrane receptor and transcriptionalregulator of T-cell development was noted In a part of thoseAML patients activating NOTCH1 mutation was identifiedMoreover TRIB2 was determined to be a direct target ofNOTCH1 signaling Later the authors found that CEBPAmethylation was accompanied by aberrant hypermethylationof many genes compared to CEBPA-mutated AMLs or withnormal CD34+ hematopoietic progenitor cells [130] Thiscould explain an observed in vitro decreased response ofCEBPA-silenced AML to myeloid growth factors and makesthis AML subtype susceptible to dynamically developingtreatment with demethylating agents Interestingly compari-son of genome-wide methylation pattern with GEP revealedonly a minimal overlap (12 unique genes including CEBPA)between the differentially expressed and differentially methy-lated genes This suggested that gene expression and genomemethylation are biologically independent processes

132 AMLwith Mutated NPM1 and the Paradox of HOXGeneExpression Since the time a 4-nucleotide insertion in NPM1gene and its significance in AML was discovered by Faliniet al in 2005 [52] much attention was paid to unveilingthe mechanism of AML triggered by NPM1mutation NPM1encodes a multifunctional protein involved in ribosomebiosynthesis and transport DNAreplication and repair chro-matin remodeling protein chaperoning regulation of cellcycle embryogenesis and oncogenesis [131 132] AlthoughNPM1 protein localizes mainly in the nucleolus it constantlyshuttles between nucleus and cytoplasm The 4-nucleotideinsertion in the last exon of NPM1 gene results in aberrantcytoplasmic accumulation of a protein and consequentlyaffects its functions NPM1 gene is mutated in about 30 ofAML and in 50 to 60 cases of adult NK-AML Given thatNPM-cytoplasmic positive (NPMc+) AML reveals uniquemolecular and clinical features [133] it was introduced intoWHO classification as a separate entity [13] An invaluablecontribution to cognition of AML promotion by NPM1mutation was made by GEP

Alcalay et al [134] first claimedNPMc+AML representeda distinct entity which can be easily distinguished from otherAML samples regardless of the karyotype They comparedglobal gene expression of 58 AML NPMc+ samples withprevalence of NK-AML and frequent occurrence of FLT3mutations to the group of 20NK-AMLswithoutNPM1muta-tions and lower occurrence of FLT3mutations Unsupervisedapproach showed NPM1 mutation status was the strongestclustering parameter A selected 369-gene-predictor effi-ciently segregated NPMc+ from NPMc- patients Interest-ingly NPM1 transcript level did not differ between thesetwo groups indicating the lack of NPM1 mutation influenceon NPM1 gene expression I confirmed this observation byanalysis of NPM1 alternative transcripts with droplet digitalPCR (ddPCR) [135] In theNPMc+patientsCD34 andCD133antigens as well as POU4F1 and CDKN2C were suppressedwhereas a number of homeodomain-containing transcrip-tion factors including HOX and TALE genes were activated[134] Because several HOX genes are highly expressed in

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

16 Journal of Oncology

HSCs and their expression decreases within cell differentia-tion the authors concluded HOX gene activation is a mech-anism of stem-cell phenotype maintenance utilized by AMLblasts The obtained results explained the overexpression ofHOX genes described earlier in an uncharacterized subgroupof NK-AML samples [57] which probably in significantproportion carried NPM1mutation

A miRNA-based expression signature of AML withmutated NPM1was later published by Garzon et al [136]Thesignature consisted of upregulated miR-10a and b membersof let-7 and miR-29 families and miR-15a-16-1 and miR-17-18a-19a-20a clusters Contrary miR-204 and miR-128apredicted to target HOX genes were downregulated inNPM1-mutated AML what is consistent with the observedupregulation ofHOX genes in this AML subtypeThe authorsproved positive correlation between miR-10a and HOXB4expression and confirmed that miR-204 targets HOXA10and MEIS1 genes Similarly other authors showed threemiRNAs located in intergenic regions in the HOX clustersmiR-10a miR-10b and miR-196a-1 were highly positivelycorrelated with HOX gene expression [122 123] Jongen-Lavrencic et al [125] observed in AML with NPM1mutationnot only overexpression of miR-10a and miR-10b but alsooverexpression ofmiR-196a and miR-196b

From the other side Verhaak et al [137] found thatHOX-gene-based discriminative signature was not limitedto AML with mutated NPM1 They showed HOX-basedclassification produced high percentage of false positiveresults including AML cases with 11q23 abnormalities andKMT2A gene rearrangements which corresponded withthe results described above One of possible explanationsis the fact that NPM1 mutation in AML is not exclusiveMore insight into HOX gene phenomenon was given byAndreeff et al [71] who measured using real-time RT-PCR technique expression of 39 HOX genes in 115 denovo AMLs representing various cytogenetic types Whilein normal CD34+ cells homogeneous expression of HOXgenes was observed AML samples were very heterogeneousin the matter of HOX expression As previously reportedlow levels of HOXA and HOXB expression was noted infavorable cytogenetic AMLs Overexpression of HOX geneswas detected in AMLs with intermediate cytogenetics and inAMLswithNPM1mutation usually associatedwith favorableprognosis Considering impact of FLT3-ITD the authorsobserved higher HOX expression in AML samples with bothmutations NPM1 and FLT3-ITD than in AMLwith exclusiveFLT3-ITD

Biological significance of NPM1mutation with concomi-tant FLT3-ITD andNPM1mutation with concomitant NRASmutation was also verified with the use of mouse knock-in models [138] Overexpression of HOX genes enhancedself-renewal expansion of hematopoietic progenitors andmyeloid differentiation bias were common for both combi-nations which indicated the persistence of transcriptionalsignature specific for NPM1 mutation in hematopoietic pro-genitors of both double-mutants Comparing to wtmice dra-matically altered gene expression profile was only observed inNPM1-FLT3-ITD mutants which also had higher leukocytecounts early depletion of common lymphoid progenitors

and a monocytic bias presenting more acute course of thedisease NPM1 and Nras-mutants characterized by granulo-cytic bias developed AML with a longer latency and a moremature phenotype Moreover additional somatic mutationswere required for AML progression The molecular-levelresults including GEP underpinned the higher frequencyand significantly worse prognosis of AML with simultaneousNPM1 and FLT3-ITD mutations

Kuhn et al [139] explained the phenomenon of HOXand FLT3 gene upregulation in NPM1-mutated AML as aresult of the activity of chromatin regulators KMT2A andDOT1L Earlier KMT2AandDOT1Lmethyltransferaseswereknown to positively regulateHOX gene expression in normalhematopoiesis and AML with KMT2A rearrangements [140]Many KMT2A fusion partners interact with DOT1L Withthe use of CRISPR-Cas9 genome editing small-moleculeinhibition and RNA-seq the authors proved that KMT2Aand DOT1L control the expression ofHOXMEIS1 and FLT3(which is a downstream target of MEIS1) and also celldifferentiation in NPM1-mutated leukemia despite the lackof KMT2A rearrangement [139] Interaction of wt KMT2Awith another protein called menin and their associationwith HOX and MEIS1 promoters were required for HOXMEIS1 and FLT3 upregulation DOT1L showed a synergis-tic effect Combinatorial inhibition of the menin-KMT2Ainteraction and DOT1L more profoundly suppressed HOXMEIS1 and FLT3 expression and induced differentiation ofNPM1-mutated AMLTherefore novel and possibly less toxictherapeutic strategies emerged for the acute leukemias withNPM1mutation and concomitant FLT3-ITD

133 AML with FLT3-ITD FLT3 gene encoding FMS-liketyrosine kinase 3 is mutated in one-third of AML patientsUsually the consequence is constitutive activation of thekinase receptorwhat impairs hematopoietic cell signaling anddisturbs hematopoiesis The presence of FLT3-ITD withoutconcomitant NPM1 mutation is well-established marker ofpoor AML prognosis In contrast to AML with mutatedNPM1 or CEBPA gene expression signature specific for AMLwith FLT3-ITD was not found for a long time probablydue to the cooccurrence of other mutations For examplein the study of Valk et al [61] samples with FLT3-ITDwere segregated into three clusters Considering miRNAexpression Garzon et al [113] reportedmiR-155 miR-10a and-10b were upregulated in AML with FLT3-ITD

Two classifiers based on the expression of 10 or 34genes predicting FLT3-ITD inNPM1-mutated CN-AMLweredetermined by Huang et al [141] by analysis of two inde-pendent AML patient cohorts each with over 100 CN-AMLpatients Among the 6 genes common for both classifiersone was downregulated (MIR155HGmiR-155 host gene andnoncoding oncogene) and 5 were upregulated encodingmembrane proteins (TMEM273 and STON2) ectonucleotidepyrophosphatasephosphodiesterase (ENPP2) matrix met-allopeptidase (MMP2) and cytokine signaling suppressor(SOCS2)

In 2017 Zhu et al [142] analyzed four microarray datasetsand identified 22 DEGs between FLT3-ITD-positive and

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 17

negative AMLs shared by all four datasets Reactome pathwayanalysis revealed correlation of the identified genes withhemopoiesis hemoglobin metabolic process hematopoieticor lymphoid organ development immune system develop-ment and myeloid cell differentiation Expression levels ofAHSP EPB42 GYPC and HEMGN genes were negativelycorrelated with FLT3 expression High expression of thesefour genes in NK-AML with FLT3-ITD was associated withbetter prognosis In concordance with the fact that HEMGNis a direct transcriptional target of HOXB4 a negativecorrelation between HEMGN and HOXB4 expression wasfound Based on the data collected the authors concludedthat FLT3-ITD might influence AML prognosis by decreas-ing the expression of AHSP EPB42 GYPC and HEMGNgenes

Wellbrock et al [143] associated FLT3 mutation withthe expression of HH pathway downstream effector GLI2Just as the presence of FLT3 mutation GLI2 expressionsignificantly decreased event-free survival (EFS) relapse-free survival (RFS) and OS Because GLI2 was coexpressedwith SMO (Smoothened) and GLI1 one can conclude FLT3mutation is generally associated with HH pathway activa-tion In fact the analysis of an independent patient cohortrevealed the expression ofGLI2 andGLI1 and FLT3mutationcould serve as independent risk factors for the survival ofAML patients Interestingly the expression of three HHpathway ligands Sonic Hedgehog (SHH) Desert Hedgehog(DHH) and Indian Hedgehog (IHH) undetectable in AMLblasts was detected in primary BM stromal cells Thus BMmicroenvironment seemed to sustain activation of HH path-way supporting leukemia progression and mediating AMLresistance to conventional chemotherapy [143] TargetingHH pathway emerges as an alternative or complimentarytherapeutic strategy against FLT3-mutated AML

134 AML with IDH Mutations Mutations in IDH1 andIDH2 genes encoding two isoforms of the nicotinamide ade-nine dinucleotide phosphate (NADP)-dependent isocitratedehydrogenases cytosolic and mitochondrial respectivelyoccur in 33 of CN-AML patients and confer unfavorableprognosis [144] Marcucci et al [144] identified a novelsubset of CN-AML with R172 IDH2 mutation which wasmutually exclusive with other known prognostic mutationsassociated with lower CR rates and presented distinctive geneand miRNA expression profiles Comparing to IDH1IDH2-wt patients AML with R172 IDH2 mutation revealed higherexpression of APP CXCL12 PAWR CDC42BPA and SPARCgenes and decreased expression of KYNU SUCLG2 CD93LY86 LIST1 and PTHR2 As far as the above-mentionedoverexpressed genes were more or less directly related toAML and cancer none of the downregulated genes had pre-viously been associated with AML In the miRNA expressionsignature specific for R172 IDH2-AML members of miR-125 family (including miR-125b which targets the tumorsuppressor gene TP53 and inhibits myeloid differentiation)and two microRNAs not associated with cancer but involvedin embryonal stem-cell differentiation miR-1 and miR-133were upregulated None of the downregulated miRNAs (eg

mir-194-1 miR-526 miR-520a-3p and mir-548b) had beenassociated with normal hematopoiesis or AML

135 AMLwithRUNX1Mutations Apart from translocationsand fusion transcripts small mutations were also found inrunt-related transcription factor 1 (RUNX1) gene in 6 [145]to more than 30 of AML patients [146] In older AMLpatients the frequency of RUNX1mutation was twice as highas in younger patients [147] Presence ofRUNX1mutationwasalso associatedwith the resistance to induction chemotherapy[145]

Gaidzik et al [145] found 148 genes differentiallyexpressed between RUNX1-mutated AML and AML with wtRUNX1 However the identified gene expression signaturewas not exclusive for RUNX1mutation but shared with AMLwith monosomy 7 and MECOM rearrangements and AMLwith complex karyotypes both deprived of RUNX1mutationA key feature of RUNX1-mutated AML was deregulation ofapoptotic pathway supported by an increased expression ofBCL2-like gene BCL2L1

Association of RUNX1 mutations with CN-AML pooroutcome and distinct gene and miRNA expression wasconfirmed by Mendler et al [147] In older (gt 60 years) CN-AML patients with RUNX1 mutation and wt NPM1 genesnormally expressed in primitive hematopoietic cells (egBAALC CD109 GNAI1 HGF and FHL1) and early lymphoidprecursors B-cell progenitors (eg DNTT BLNK FOXO1and FLT3) were upregulated whereas myelopoiesis promot-ers such as CEBPA components of neutrophil granules(AZU1 MPO and CTSG) were downregulated RegardingmiRNA profile miR-223 and two members of the let-7tumor suppressor familywere decreased inAMLwithRUNX1mutations Three other miRNAs of unknown functionsin leukemogenesis miR-211 miR-220 and miR-595 wereupregulated in RUNX1-mutated blasts

The collected data indicated definitely different biologyof RUNX1-mutated AML than for example NPM1-mutatedAML and contributed to the distinction of AML withmutated RUNX1 as a provisional entity in the revised WHOclassification of AML [13]

14 AML with Overexpression of ParticularProtein-Coding Genes

Valk et al [61] distinguished a compact cluster of AMLsamples with overexpression of MECOM transcriptionalregulator and oncoprotein involved in hematopoiesis apop-tosis development cell differentiation and proliferationHigh expression of BAALC gene postulated marker of earlyhematopoietic progenitor cells was earlier established asan independent poor prognostic factor in CN-AML [42]Langer et al [148] proved younger (lt60 years) CN-AMLpatients with BAALC overexpression presented distinct geneexpression signature with upregulation of genes earlierassociated with poor outcome (eg HGF MN1 CD200)genes involved in drug resistance (eg ABCB1 alias MDR1)and hematopoietic stem-cell markers (PROM1 alias CD133CD34 KIT) CD133 was the most upregulated gene in high

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

18 Journal of Oncology

BAALC expressers Interestingly no differences were foundin global microRNA expression but an inverse correlationbetween the expression levels of miR-148a and BAALC wasobserved suggesting that miR-148a might act as a negativeregulator for BAALC Later the same authors focused on themeningioma 1 (MN1) gene encoding a member of gene tran-scription regulator complex with the nuclear receptor RAR-RXR or the vitamin D receptor [149] They associated highMN1 expression with the lack of NPM1 mutation increasedBAALC expression less extramedullary involvement andworse outcome Gene- and microRNA-expression patternsdetermined from highMN1 expressers had common featureswith high BAALC expressers (upregulation of PROM1 CD34FZD6 CRYGD CD200 and ABCB1 genes) and patients withwt NPM1 (low levels of HOX genes) Positive correlationwas also found between the expression of MN1 gene and thehsa-miR-126 family contributing to proangiogenic activityof VEGF and formation of new blood vessels and hsa-miR-424 which regulates monocyte and macrophage differenti-ation Apoptosis-related hsa-miR-16 and miRNAs involvedin malignant transformation (eg hsa-miR-19a and hsa-miR-20a members of the miR-17-92 polycistron) as well as hsa-miR-100 and hsa-miR-196a were downregulated in AMLsamples with higherMN1 expression

Metzeler et al [150] who analyzed the expression levelsof ERG BAALC and MN1 in over 200 CN-AML patientswith the use of oligonucleotide microarrays confirmed theassociation of high level of expression of the studied geneswith inferior OS and a lower rate of CR Indeed theexpression levels of all three genes were highly correlatedHowever in multivariate analyses high ERG expressionsimilarly as FLT3-ITD seemed to be an independent andstrongest predictor of negative prognosis in younger andolder CN-AMLpatients The results suggested the prognosticvalue ofERG BAALC andMN1 genesmight partially overlapand highERG expression togetherwith the presence of FLT3-ITDmight be a sufficient combination of factors for high-riskstratification in CN-AML

15 Time for Meta-Analyses

After a significant amount of gene expression data hadbeen collected papers reporting meta-analysis started toappear For example in 2010 Miller et al [151] systematicallyanalyzed the results of 25 AML studies published between1999 and 2008 In total close to 16 thousand expressionfeatures corresponding to 5 thousand unique genes wereavailable from 2744 patient samples analyzed with 10 differ-ent microarray platforms One-third of genes were reportedinmore than one study Several genes egVCAN and PGDSwere identified only in AML cell lines 25 genes including 7HOX family members POU4F1 TSPAN7MYH11RUNX1T1RUNX3 CD34 and MN1 were reported as AML-specificby at least 8 independent studies HOXTALE expressionwas increased in AML with normal cytogenetics NPM1and FLT3 mutations and 11q23 abnormalities involving theKMT2A gene Decreased expression of these genes wastypical of CD34+ cells AML with CEBPA mutations and

AML with cytogenetic aberrations Considering prognosis-relevant signatures the authors found only a minority ofgenes (96) were reported by at least two studies Amongthese genes BCL11A TBXAS1 HOXB5 HOXA10 CD34MN1 NME1 FLT3 were upregulated whereas genes such asEML4 C3AR1 SMG1 FOXO1 AZU1 were downregulatedin AML samples with poor prognosis In AML with NPM1mutations increased expression of SMC4 gene was reportedby 5 different studies Apart from the selection of genesand pathways shared by different AML studies meta-analysismade by Miller et al [151] enabled identifying novel markergenes and potential therapeutic targetswhichwere skipped bysingle studiesThe exampleswere two geneswhose expressioncorrelated with response to therapy namely TBXAS1 andSEMA3F increased in AML samples with poor and goodprognosis respectively This evidently shows that reanalysisof collected transcriptomic data and combining the resultsfrom different studies may be an underestimated source ofnew AML-relevant information

16 Custom-Made Microarrays An Alternativeto Global GEP

Since the time microarray technology was established mul-tiple types and applications of microarrays were developed[45] To benefit from this dynamically developing tech-nology guidelines for microarray gene expression analysesin leukemia were formulated by three European leukemianetworks in 2006 [152] Among all microarray platforms usedfor gene expression analysis of AML commercially availableGeneChips of Affymetrix predominated (see Supplemen-tary Table 1) However a few prominent AML papers werepublished based on cDNAmicroarrays developed at the Stan-ford University [60 73] or Lund University [69] All of theabove microarray platforms were generated to study globalgene expression Alternatively small custom-made microar-rays dedicated to analysis of a selected subset of genes wereharnessed to AML studies IntelliGene Human Cancer CHIPcDNA microarray from Takara Biomedicals as well as twokinds of custom oligonucleotide microarrays covering 2304genes mainly encoding transcription factors membraneproteins growth factors and proteins involved in redoxregulation were used byMiyazato et al [91] to identify MDS-specific genes In-house microarray was applied by Park etal [153] to study the expression of about 300 prognosis-related genes in 4 clinical AML samples The genes wereselected based on globalGEPofAML cell lines and previouslypublished data Taking advantage of our own experiencewith custom microarrays we also designed and generated aboutique microarray dedicated to gene expression analysisof AML [154] Our AML-array was composed of about 900oligonucleotide probes complementary to genes selected bythe literature search proven and postulated acute leukemiabiomarkers general oncogenes genes specifically involved inleukemic transformation genes related to immune responseand a set of positive (housekeeping human genes) andnegative (plant and bacterial) control genes AML-array wasused to analyze gene expression in 33 AML patients without

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 19

or with minimal maturation (FAB M1 and M2 subtypes) and15 healthy volunteers (HV) Based on 83-gene classifier wewere able to perfectly distinguishAML fromHVsamplesThegenes overexpressed in AML included well-established AMLmarkers such asKITMYH11MYC CEBPAMN1MPO SETand HOXA10 but also genes rarely discussed in the contextof AML pathogenesis eg STMN1 (the most discriminativegene in our analysis) CDK6 ANGPT1 or ENO1 The roleof genes determined in our analysis as underexpressed inAML eg IFITM1 FCN1 S100A9 LTB LYZ FCER1G waseven less clear and demand further research We found thatthe upregulation of CPA3 gene was specific for AML withmutated NPM1 and FLT3 genes Although we observed somegene expression trends we were not able to find any geneswith statistically significant differences between AML sub-groups divided according to FAB subtype mutation statusor response to therapy This may be due to too small samplesize high homogeneity level within the study group limitedto M1 and M2 FAB subtypes which are not dramaticallydifferent too high technical bias or a preselection of geneswhich could skipmore discriminative genesNevertheless weshowed applicability of a small custom array to AML geneexpression analysis Following optimization it could servefor example as a first-line diagnostic tool With the use ofcomplementary quantitative RT-PCRmethods we identifiedthree genes (S100A9 ANXA3 and WT1) whose expressionlevels can be used to distinguish between M1 and M2 FABsubtypes We showed relationship between STMN1 and ABL1expression level and FLT3 and NPM1 mutation status Wehave also found correlation between positive response totreatment and high CAT expression and lowWT1 expression[154]

17 SAGE Alternative to Microarrays beforeMassive Sequencing Era

Apart from microarrays serial analysis of gene expression(SAGE) technique was also applied to AML gene expressionprofiling [59 155 156] Although thismethod did not demandprior gene sequence knowledge produced more quantitativeresults and was described as very sensitive it was definitelyless common than microarrays and finally was ousted byNGS

With the use of SAGE 22 AML samples with four mostcommon translocations t(821) t(1517) inv(16) and t(911)were compared to normal myeloid progenitor cells [155]Over 26 thousand transcripts were abnormally expressedAltered expression of 56 genes was shared by all AML sam-ples egNUBPL TRAM2 PTRF (presentCAVIN1) (upregu-lated) FCN1 LCN2 and FASN (downregulated) Other geneswere differentially expressed in one or some of the transloca-tions studied In all translocations except t(821) more than23 DEGs were underexpressed Of note only a small part ofthe SAGE results corresponded with the results of publishedmicroarray-based experiments For example Lee et al [155]did not observeMYH11 overexpression in inv(16) nor RUNX1and RUNX1T1 overexpression in t(821) In subsequent paperthe authors compared SAGE results obtained for three pooled

primary AMLs with t(911)(p22q23) with SAGE-based GEPof Mono Mac 6 (MM6) cell line representing AML withthis particular translocation Despite generally similar geneexpression profile the authors identified 884 alternativelyexpressed transcripts corresponding to 83 known genesmainly related to biosynthetic andmetabolic processes Inter-estingly HRAS with well-established role in leukemogenesisand three other genes fromERK1ERK2MAPKpathway gov-erning cell growth proliferation differentiation and survivalwere overexpressed exclusively in MM6

18 Next Generation SequencingUnlimited Perspectives

Microarray boom lasted about 15 years Since 2006 whena first high-throughput automatic sequencer Genome Ana-lyzer was launched by Solexa DNA microarrays were beinggradually replaced by the NGS termed also massive parallelsequencing (MPS) At the beginning the costs of NGSoutbalanced the costs of microarray experiment but theywere soon compensated While microarrays are still appliedfor genotyping due to their simplicity compared to thewhole genome sequence analysis in the field of transcrip-tome research NGS is incomparably better Transcriptomesequencing called RNA-seq is able to detect all types oftranscripts present in a cell including noncoding RNAsproducts of gene fusions and alternative splicing In addi-tion transcriptome sequencing is often combined with awhole genome exome or targeted resequencing which allowscompletely characterizing the studied object In 2013 suchcomprehensive study of AML was published by The CancerGenomeAtlas Research Network in theNewEngland Journalof Medicine [23] A total number of 200 cases representeddifferent AML subtypes were sequenced (50 whole genomesand 150 exomes) For the same individuals analyses of globalmRNA and miRNA expression and DNA methylation wereperformed In several cases RNA-seq revealed increasedor exclusive expression of the mutant DNMT3A RUNX1PHF6 and TP53 genes Gene fusions including 15 new fusionevents with maintained open reading frame were detected inalmost half of AML patients Hierarchical clustering of geneexpression data enabled distinguishing seven AML groupsbased onmRNAexpression and five groups based onmiRNAexpression Similarly as in microarray data analysis the iden-tified groups were highly correlated with AML FAB subtypesdifferentiation stage presence of the recurrentmutations andpatient outcomes Integration of gene expression and DNAmethylation data led to the discovery of a small RNA setwithin an imprinted locus on chromosome 14 These smallRNAs were specifically dysregulated in APL Patients withPML-RARA fusions had generally very distinct mRNA andmiRNA signatures that were strongly correlated with eachother and with a specific DNA methylation signature AMLwith RUNX1-RUNX1T1 AML with some KMT2A fusionsand AML with three mutations (in NPM1 DNMT3A andFLT3) together were also associated with mRNA and miRNAexpression signatures In compliance with previous researchthe most discriminatory miRNAs for the triple-mutant AML

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

20 Journal of Oncology

were miR-10a miR-424 miR-196b miR-130a and let-7bOther transcription factor fusions were correlated with onlymRNA expression signatures

At the end of 2018 functional genomic landscape of acutemyeloid leukemia was highlighted by Tyner et al [25] Thestudy within the frame of Beat AML program included halfthousand AML patients for whom whole-exome sequenc-ing and RNA sequencing data were integrated with theanalyses of ex vivo drug sensitivity Clustering of the 2000most variably expressed genes allowed to distinguish geneexpression signatures associated with genetic and cytogeneticAML groups Mutations in several genes eg TP53 andASXL1 seemed to be responsible for a broad pattern ofdrug resistance Specific gene expression signatures were alsoidentified for 78 out of 119 testable drugs For example 17 geneexpression-based signature predicting sensitivity to ibrutinibwas determined Multivariate modelling was harnessed toestimate contributions of both mutation and gene expressionpatterns in drug response prediction

Apart from two most prominent studies mentionedabove a significant number of NGS-based AML papers werepublished within the last several years Since referring to allof them is impossible in one review a few selected examplesare highlighted below

19 Better Tools Better Characterization

Like the microarrays RNA-seq was used to better char-acterize particular AML subgroups and cell lines Forexample transcriptome sequencing of three basic myeloidleukemia cell lines K562 HL-60 and THP1 representingchronic myeloid leukemia (CML) APL and acute monocyticleukemia respectively was conducted by Wang et al [157]They found ERKMAPK and JAK-STAT signaling pathwayswere more highly activated in K562 than in HL-60 cellsContrary PI3KPKB pathway induced by oncogene KITor FLT3 as well as PML and RARA genes which arefusion partners in APL were upregulated in HL-60 Genesrelated to cell cycle cell division and chemokine signalingpathway were also overexpressed in HL-60 cells Genesupregulated in THP1 cells were enriched in immune defenseinflammatory response and other processes connected withmonocyte functions (eg LYZ MPO HLA-B IL8 presentCXCL8 PRG2 SPI1 former PU1 and TFRC) Based onGEP the authors concluded K562 cells are a good model tostudy erythroid differentiation HL-60 cells chemotaxis andphagocytosis and THP1 inflammatory response Gosse etal [158] described a novel NK-AML cell line termed CG-SH A whole genome sequencing revealed the absence ofrecurrent mutations but novel small alterations were foundin several genes including GATA2 and EZH2 Compar-ing genome and transcriptome data showed allele-specificexpression of GATA2 gene which resulted from epigeneticsilencing Although the mutation was heterozygous only amutated variant was transcribed Interestingly genes whichare frequently mutated in AML but not mutated in CG-SH (eg NPM1 GATA2 IDH2 RUNX1 and TP53) wereupregulated in the studied cell line however their levels

of expression remained within the ranges observed for 55AML patients Differential expression of genes implicatedin proliferation apoptosis and differentiation was noted forCG-SH cells following cytokine treatment

Two subtypes of pediatric CBF AML t(821) and inv(16)were compared with the use of RNA-seq by Hsu et al[159] Although both CBF leukemias revealed many commonfeatures the authors were able to discover two hundreds ofDEGs In t(821) samples the most upregulated gene wasRUNX1T1 fusion partner gene whereas the most underex-pressed was RFX8 Overexpression of matrix metallopep-tidase gene MMP14 and downregulation of collagen geneCOL23A1 was typical of inv(16) Compared to NK-AMLsamples HOX gene family including MEIS1 and NKX2-3 transcription factors were downregulated in both CBFAMLs Within NK-AML two subgroups with and withoutFLT3-ITD were not able to distinguish based on GEP Intotal 287 fusion transcripts were identified 16 of them werenovel including three involving NUP98 gene In the wholecohort of 64 patients alternative splicing events (ASEs)differentially expressed across all subtypes were also detectedThe predominant alternative splicing events were skippedexon (SE) mutually exclusive exons (MXE) and retainedintron (RI)

Singh et al [160] compared genome-wide DNA bind-ing sites and transcriptome data associated with RUNX1-RUNX1T1 CBFB-MYH11 and PML-RARA oncofusion pro-tein expression and found many target genes pathways andacetylation patterns are shared between these three fusiontranscription factors In the case of RUNX1-RUNX1T1 andPML-RARA the percentage of common target genes reached40 Gene expression analysis revealed both common andunique signatures for each translocation The unique DEGsincluded genes described earlier as specific for particulartranslocation namely TRH POU4F1 PRAME and RUNX1T1genes for t(821) VCAN MN1 and S100A12 for inv(16) andCTSG and PTGDS for t(1517) However even these uniquegenes were members of the similar pathways in particularlinked to cell proliferation (eg TGFB signaling pathway)and apoptosis Therefore the authors hypothesized the threedifferentAML subtypes despite distinctmolecular properties(binding sites mechanisms of action) exploit common pro-grams of malignant cell transformation

Eisfeld et al [161] studied AML with a sole monosomyof chromosome 7 (-7 AML) the most frequent autosomalmonosomy associated with poor outcome In over 30 casesanalyzed the authors not only identified the most frequentAML mutations but found different mRNA and miRNAexpression profiles compared to AML with both copies ofchromosome 7 AmongDEGs downregulated prevailed with94 genes mapping to chromosome 7 affirming dosageeffectThemost overexpressed genes were PTPRM encodinga protein tyrosine phosphatase receptor a regulator of cellgrowth differentiation and oncogenic transformation ID1a downstream target of oncogenic tyrosine kinases andMECOM coding for a transcriptional regulator and onco-protein implicated in hematopoiesis apoptosis developmentcell differentiation and proliferation Out of 16 differen-tially expressed miRNAs 6 were significantly downregulated

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 21

including 5 from chromosome 7 and miR-9-1 from chromo-some 1 UpregulatedmiRNAs came from two clusters locatedon chromosome X (miR-20b miR-363 and miR-106a) andchromosome 19 (miR-99b miR-125a and miR-let7e)

RNA-seq data were also collected from 13 patients withdeletions of the long arm of chromosome 9 [del(9q)] arare aberration occurring in about 2 of all AML cases asa sole abnormality or accompanied by t(821) t(1517) orother cytogenetic aberration [162] Transcriptome of del(9q)combined with the exome and target amplicon sequencingwas compared with the transcriptomes of 454 AML patientswith normal karyotype or various cytogenetic aberrationsCharacteristic features of del(9q) AML were mutations inNPM1 DNMT3A and WT1 more frequent than in otherAML subtypes and downregulation of TLE4 gene

Mixed-phenotype acute leukemia (MPAL) a rare type ofprogenitor leukemia with ambiguous expression of myeloidand lymphoid lineage markers was studied by Pallavajjalaet al (2018) [163] RNA-seq combined with whole genomicsequencing (WGS) enabled identifying mutations in 70genes different translocations residing mainly in the non-coding regions of the genome and describing gene expressionprofiles in samples from four patients with TMyeloid MPALFor two patients with matched diagnostic and remissionsamples enriched pathway analysis allowed for associationof genes which were upregulated at diagnosis with pathwaysinvolving nucleosome and chromatin assembly and organi-zation

20 Discovery of New Fusion Transcripts

Being a useful tool for fusion transcript detection RNA-seq was applied to identify a complex 3-way translocationt(81221)(q22p11q22) in an individual AML M2 patient[164] In addition to RUNX1ndashRUNX1T1 fusion typical oft(821) AML the patient harbored two additional transloca-tions with the contribution of VPS13B gene a causative geneof Cohen syndrome encoding vacuolar protein sorting 13forming TM7SF3ndashVPS13B andVPS13BndashRUNX1 fusion genes

With the use of a whole transcriptome sequencing 88new fusion transcripts were discovered in AML by Wen etal [165] In total 134 fusion transcripts were detected in45 AML samples including 29 NK-AMLs The fusions werepredominantly formed between the genes adjacent in thesame chromosome in different orientations and distributedat different frequencies in the AML cases regardless of thekaryotype While comparing to other tumors the authorsfound only 5 common fusions all shared with only one tumortype (prostate cancer) It underpins the AML-specificity ofthe discovered fusions Out of 114 fusions identified in NK-AML seven were unique for this AML subtype MoreoverCIITA-DEXI fusion transcript occurring in three isoformswas found in 48 of NK-AML cases Of note the maximalnumber of fusion transcripts found in one NK-AML case was57 Although some fusions were generated posttranscription-ally these results suggest that genome-level changes are notso rare in AMLs with normal karyotypes The significance ofparticular fusions remained to be elucidated

Also in pediatric CN-AML novel fusion transcriptswere identified with RNA-seq eg NUP98-PHF23 [166] orCBFA2T3-GLIS2 [167 168] occurring with 26 and 43-84 frequency respectively More frequent CBFA2T3-GLIS2fusion resulted from a cryptic inversion of chromosome 16andwas correlatedwith high risk of relapse and poor outcome[167 168] Schuback et al [167] demonstrated the fusionwas most prevalent in the youngest patients (lt5 years) andabsent in adults (gt20 years) In another work Masetti et alidentified another fusion transcript in 40 of the CBFA2T3-GLIS2-positive patients [169] The novel fusion derived froma member of Hedgehog signaling pathway Desert Hedgehog(DHH) andRasHomologueEnrich inBrain Like 1 (RHEBL1)gene coding for a small GTPase of the Ras family DHH-RHEBL1ndashpositive patients exhibited a specific gene expres-sion pattern with upregulation of FLT3 BEX1 MUC4 andAFAP1L2 genes The outcome of these patients was evenworse than that of the patients with exclusive CBFA2T3-GLIS2-rearrangement Notably targeted treatments againstAML with CBFA2T3-GLIS2 are under evaluation GANT61the most potent inhibitor of GLI family proteins which arethe final effectors of Hedgehog pathway seems to be efficientalso against GLIS2 chimeric proteins [82]

21 Alternative Transcripts Another Source ofTranscriptome Variability

Therole of alternative splicing (AS) in AMLpathogenesis wasfirst highlighted by Tanaka et al (1995) [36] who analyzed twoof three previously identified alternative isoforms of AML1gene at present termed RUNX1 Both transcripts AML1aand AML1b shared a runt homology domain responsiblefor DNA binding whereas transcriptional activation domainwas present only in AML1b The authors found that the twoAS products regulated hematopoietic myeloid cell differen-tiation in an antagonistic way presumably via competingfor the binding to CBF2B gene encoding transcriptionalactivator While AML1a inhibited granulocytic differentia-tion and induced cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment concomitant AML1boverexpression recovered the granulocytic differentiation

Development of NGS contributed to the progress in ASresearch Now the role of splicing abnormalities in AMLprogression and drug resistance is incontestable [170] Whileoverexpression of SRSF1 was associated with solid tumorpromotion mutations in genes encoding splice factors ieSF3B1 SRSF2 U2AF1 are considered as important driversof hematological disorders such as MDS and AML [170]Contribution to AML pathogenesis was assigned to splicevariants of FLT3 aberrant splicing of BCL2 gene linkedwith drug resistance and overexpression of WT1 and E2F1genes which encode transcription factors taking part in ASregulation [170] We recently demonstrated that alternativetranscripts ofNPM1 gene are upregulated in AML and ALL atdiagnosis decrease in CR and increase again at relapse [135]High expression of twoNPM1 gene isoforms was significantlyassociated with shorter overall and disease-free survival Thissuggested that not only mutation but also expression level of

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

22 Journal of Oncology

NPM1 gene affects patient outcome Aberrant proportions ofparticular NPM1 splice variants could be linked to abnormalexpression of genes encoding alternative splicing factors

RNA-seq of two samples collected from an individualAML M2 patient at diagnosis and remission were exploredin the context of AS by Li et al (2014) [171] and Gao et al(2014) [172] In both studies a few dozens of differentiallysplicing events were detected in differentially splicing genesassociated with RNA processing cellular macromoleculecatabolic process and DNA binding

Shirai et al (2015) investigated the consequences ofthe most common U2AF1 mutation in a transgenic mousemodel [173] Whole transcriptome analysis of hematopoieticprogenitor cells of U2af1-mutated mice revealed alteredhematopoiesis and changes in premRNA splicing Compar-ing the results of the analysis with human RNA-seq data fromTCGAAML cohort displayed enrichment ofU2AF1-inducedsplicing alterations in processing genes ribosomal genesand recurrently mutated MDS and AML-associated genes(eg NPM1 BCOR and KMT2D) The authors concludedsequence-specific AS pattern triggered bymutant U2AF1wassimilar in mouse and human cells

Li et al (2018) integrated AS events derived from RNA-seq with H3K79me2 ChIP-seq data across 34 human normaland cancer cell types [174] Clustering based on skippingexon-associated sites divided all cell types to 6 clusters Twoof them consisted predominately of cell lines derived fromhematological malignancies Moreover four AML cell linesmainly with KMT2A rearrangements were found in onecluster together with one CML cell line Deregulated genesassociatedwith this particular clusterwere involved inmRNAsplicing via spliceosome The obtained results corroboratedcontribution of epigenetic-mediated splicing events to pro-gression of KMT2A-AML and associated alternative splicingmediated by K79me methyltransferase encoded by DOT1Lgene with leukemogenesis

22 Chimeric RNAs Newly DiscoveredContribution to Transcriptome Complexity

Apart from well-established fusions and alternative tran-scripts another class of transcripts has been recently discov-ered in tumor as well as in normal cells by RNA-seq [175 176]The new class of functional and potentially oncogenic RNAscalled chimeric RNAs (chRNAs) not only are transcribedfrom genome regions modified by translocation inversionor more complex chromosomal rearrangement but can begenerated as a result of posttranscriptional RNA processingeg cis- or trans-splicing By combining two or more geneloci chRNAs also differ from conventional splicing variantsThe existence of chRNAs in AML was proved by Ruffleet al in 2017 [177] In RNA-seq data from three AMLpatients 17 chRNAs were identified including new PML-RARA transcripts with exon junctions not described earlierin t(1517) and expression changes with time and treatmentOther chRNAs originated from two adjacent genes (egVAMP8-VAMP5) nonadjacent exons separated by thousandsof base pairs (eg SLC16A3-METRNL andUBR5-AZIN1) two

distant genes (eg NONE-CTDP1) or gene fragments corre-sponding to opposite chromosome strands Extending theirresearch to a larger AML patient cohort and normal CD34+transcriptome the authors identified four new types oftumor-specific chRNAs recurrently expressed in AML sam-ples (TRIM28-TRIM28 DHRS7B-TMEM11 PLXNB-BLRD1and SLC16A3-METRNL) On the basis of PML-RAR fusionsit was shown in one patient several isoforms of the samechimeric transcript can coexist and present various sensitivityto therapeutic agents The resistance to treatment can beindicated by the appearance or increase of fusion transcriptsHowever the oncogenic potential of chRNAs needs to beverified by further research

23 Small RNAs from the NGS Perspective

Compared to microarrays NGS enabled more comprehen-sive and quantitative analysis of miRNome Ramsingh et al[178] analyzed miRNome of one CN-AML patient femalediagnosed as FAB M1 to assess miRNA expression andmutations in miRNA or miRNA binding sites Small RNAsequencing of leukemic myeloblasts and CD34+ cells pooledfrom 5 healthy donors revealed expression of 472 miRNAsincluding 7 novel miRNAs The most highly expressedmiRNA in both AML and CD34+ cells wasmiR-233 In AMLit represented almost 50 of all miRNA reads miRNAswhich displayed differential expression between AML andcontrol CD34+ pool includedmiR-362-3p andmiR-25 over-expressed and underexpressed in AML respectively Com-parison of NGS-based miRNA profiling with the array- andRT-PCR-based approaches showedmicroarray and real-timeanalyses underestimated the expression of some miRNAalthough the general correlation between platforms wassignificant The authors did not find acquired mutations inmiRNA genes but revealed several novel germline polymor-phisms Comparing the results of miRNA expression withthe sequence of this particular AML patient genome whichwas known earlier [21] they identified a single mutation inthe putative tumor suppressor gene TNFAIP2 and provedthis mutation generated a new miRNA binding site Asthis mutation resulted in a Dicer-dependent translationalrepression of a reporter gene the consequence could be atranslational repression of TNFAIP2 previously describedas a target of PML-RARA or PZLF-RARA fusion genes andhighly expressed in hematopoietic cells TNFAIP2 mutationwas predicted to generate imperfect binding sites formiR-223and miR-181b but the experiments conducted on AML celllines did not confirm contribution of miR-223 and miR-181band any other known miRNA to the translational repressionof mutant TNFAIP2 31015840-UTR The authors did not excludepossibility of regulation by a new unknown yet miRNANevertheless the TNFAIP2 31015840-UTRmutation must be rare asit was not found in any other AML samples from 187 patientsscreened

Integrating SNP and mRNA arrays with microRNAprofiling of 16 myeloid cell lines Garcıa-Ortı et al [179]associated expression levels of 19 miRNAs with CNVsaffecting their loci One of these miRNAs miR-370 often

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 23

upregulated in AML was proven to target tumor suppressorNF1 downregulated in more than 30 AMLs BecauseNF1 suppression activates RAS similarly as RAS-activatingmutation AML patients with miR-370 overexpression maypotentially benefit from additional treatment with either RASor mTOR inhibitors

Starczynowski et al [180] who globally analyzed miRNAlocalization and expression in human genome using cell linemodels discovered that 77 miRNAs mapped to leukemia-associated copy-number alterations and the expression ofonly 18of themwas detectable Furthermore they found theloss of two selected miRNAmiR-145 andmiR-146a localizedin a commonly deleted in AML region 5q initiated leukemiainmice Using small RNA sequencing they identified 28 novelmiRNAs 18 of which mapped to leukemia-associated copy-number alterations and may play a role in leukemogenesis

Using t(821) AML mRNA- and miRNA-sequencing datafrom TCGA project Junge et al [181] constructed a net-work of 605 transcripts potential competitors of RUNX1T1in miRNA binding The so called competing endogenousRNAs (ceRNAs) cross-regulate each other by competing forbinding to shared miRNAs The predicted set of ceRNAscontained multiple oncogenes and members of the integrincadherin and Wnt signaling pathways One-third of thosegenes were differentially expressed between t(821) AML andnormal granulocyte-macrophage progenitor cells Takinginto account experimentally validated miRNA binding sitesthe authors selected 21 top RUNX1T1 ceRNAs including13 which shared miRNA binding sites with RUNX1T1 egPLAG1 TCF4 NFIB and YWHAZ Therefore the authorssupported the hypothetical miRNA sponge function ofRUNX1T1 gene particularly its 3rsquoUTR present in a leukemicRUNX1-RUNX1T1 fusion transcript and overexpressed up to1000 times in t(821) versus other and control samples

24 piRNAs Extracellular Vesicles andTransposable Elements

From small RNAs miRNAs are definitely best characterizedand their association with normal and malignant develop-ment is well established Recently a little longer (25-33 nt)P-element-induced wimpy testis (PIWI)-interacting RNAs(piRNAs) responsible for epigenetic silencing of transpos-able elements (TEs) in germline tissue have been corre-lated with brain functioning and tumor transformation [104182] Aberrant piRNA expression was detected in multiplemyeloma and various solid tumors [104] In cancer cellspiRNAs and PIWI proteins may contribute to tumorigenesisthrough aberrant DNA methylation leading to genomicsilencing and promotion of a ldquostem-likerdquo state or oppo-sitely through gene expression activation via regulation ofhistone acetylation and euchromatin formation Comparativeanalysis of malignant and normal tissues from 11 organsshowed that out of approximately 20000 piRNA presentin human genome less than 300 are expressed in somatictissues and more than 500 in corresponding tumors [183]Although most piRNAs were commonly upregulated acrosstumors some piRNAs were expressed in tumor-specific

manner A fraction of small RNAs abundant in miRNAsand piRNAs was detected in extracellular vesicles (EVs)secreted by bone marrow mesenchymal stem cells (BM-MSC) which are a component of hematopoietic microen-vironment [184] EVs treatment of hematopoietic stem cellsextracted from umbilical cord blood (UCB-CD34+ cells)induced cell survival suppressed apoptosis and decreased celldifferentiation However piRNA role in AML remains to beelucidated Up to date the only evidence of direct relationbetween piRNAs and AML pathogenesis was presented byShiva Bamezai in her PhD thesis devoted to the role ofArgonaute protein PIWIL4 in hematopoiesis and AML [185]The author showed upregulation of PIWIL4 gene encodingone of PIWI proteins in AML samples the highest in AMLwith KMT2A-AF9 translocation conferring poor prognosisOf note prognostically favorable AML with PML-RARA andinv16 showed the lowest levels of PIWIL4 Overexpressionof PIWIL4 was correlated with high expression of genesinvolved in cell proliferation such as FLT3 CBL and NRASContrary depletion of PIWIL4 in AML with KMT2A rear-rangements drastically reduced leukemic cell growth in vitroand in vivo but did not affect normal cord blood CD34+cell growth Out of 10 thousand unique piRNAs detectedin wt THP-1 cells over 1000 revealed changed expressionfollowing PIWIL4 knockdown including 80mapped to genesindicated by RNA-seq as deregulated by PIWIL4 depletionInterestingly most of these genes belonged to the actincytoskeleton regulation pathway

The role of TEs in AML and MDS was highlighted byColombo et al [186] While highly expressed in embryo-genesis TEs are usually methylated and silenced by hete-rochromatin in the somatic cells Activation of TEs is beingobserved in ageing tissues and cancers In LSCs whichare the most therapy-resistant fraction of AML cells lowexpression of TEs was noted along with the suppressionof genes involved in interferon pathway inflammation andimmune response Significant suppression of TE expressionwas also identified in high-risk MDS compared to low-riskMDS Considering the suppression of TEs in in AML andMDS as a mechanism for immune escape indicates thepotential targets to activate cancer immunogenicity in thesemyeloproliferative malignancies

25 snoRNA

Small nucleolar RNAs (snoRNAs) are basically involved inthe posttranscriptional modification of ribosomal RNAsin cooperation with protein partners [102 187] In recentyears new functions of snoRNAs which can be submittedto an extensive processing have been discovered namelyin alternative splicing regulation of chromatin structuremetabolism and neoplastic transformation [187] NGS-basedanalysis of snoRNAs is more tricky as their size rangingfrom 60 to 250 nucleotides [187] overlaps with a gapbetween conventional small RNA sequencing and RNA-seqdevoted to mRNAs and other RNA molecules longer than200 nt Developing their own sequencing approach Warneret al [188] showed that snoRNAs eg orphan snoRNAs

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

24 Journal of Oncology

contained in the imprinted DLK-DIO3 and SNURFSNRPNloci are expressed in a lineage- and developmentallyrestricted manner in human hematopoiesis Moreover 120snoRNAs including SNORA21 SNORA36C and SCARNA15displayed consistent differential expression in AML andwhat is even more interesting all of them were decreasedin AML samples compared to normal CD34+ cells Of notealthough the majority of snoRNAs were embedded in theintrons of host genes expression levels of snoRNAs did notcorrelated with expression or alternative splicing of hostgenes which suggested cellular levels of mature snoRNAswere determined by other factors No somatic mutationswere detected in the snoRNA genes either By the waythe authors found a few novel snoRNAs and proved stan-dard transcriptome sequencing cannot reliably distinguishunspliced primary host gene RNA from correctly processedsnoRNA [188]

26 Long Noncoding RNAs

A novel class of functional RNAs which do not encodeproteins and are longer than 200 nt are referred to as longnoncoding RNAs (lncRNAs) [189] Their role in imprintingand regulation of cell cycle cell differentiation and apoptosishas been postulated [190] In 2014 Garzon et al [191] firstreported significance of lncRNAs for AML pathogenesisand prognosis Using a custom microarray platform theyevaluated lncRNA expression in 148 older CN-AML patientsand validated the results in an independent cohort of 71patients Distinct lncRNApatterns were determined for AMLwith FLT3-ITD and mutations in such genes as NPM1CEBPA IDH1 IDH2 ASXL1 and RUNX1 For examplepatients with mutated NPM1 revealed upregulation of severalantisense transcripts ofHOX genes (HOXB-AS3MEIS1-AS2)plasmacytoma variant translocation 1 (PVT1) and the coiled-coil domain containing 26 (CCD26) lncRNAs Wilms tumor1 antisense RNA (WT1-AS) lncRNA was found as typicalof FLT3-ITD signature whereas downregulation of HOXB-AS3 lncRNA was noted in CEBPA-mutated AML RUNX1mutationwas associatedwith the increase of lncRNAs locatedin the proximity of lymphoid marker genes (eg BLNK) theimmunoglobulin heavy locus (IGH) complex and vault RNA1-1 (VTRNA1-1) which was linked with multidrug resistanceOf note no specific lncRNAprofileswere found forDNMT3Aand TET2 mutations which frequently occur in older CN-AML patients Instead prognostic signature composed of 48lncRNAs was identified indicating correlation of lncRNAexpression with AML treatment response and survival

In 2017 Schwarzer et al [192] presented the com-prehensive transcriptome landscape of the normal humanhematopoietic stem cells and their differentiated progeniesShort and long noncoding RNAs (ncRNAs) together withmRNAs were characterized with the use of three microar-ray platforms in 12 distinct cell populations purified withmulticolor flow cytometry from blood of healthy donorsThe observed cell-type-specific ncRNA expression indicatedthe tight regulation and coordinated function of this RNAclass in human hematopoietic system Functional analysis

of the identified ncRNA fingerprints in the studied cellsand in two independent datasets of more than 600 AMLsamples revealed 80 overlap of associated gene setsFor example HOTAIRM1 granulocyte-specific lncRNA wasassociated with inflammatory and innate immune responsepathways andwas strongly correlatedwith genes upregulatedin AML with NPM1 mutation In addition novel ncRNAregulators of granulopoiesis were predicted eg LINC00173expressed specifically in mature granulocytes and negativelyassociated with the expression of genes related to stemnesscell cycle progression and cancer The role of LINC00173in granulocyte proliferation and differentiation was thenconfirmed by its transcriptional repression with CRISPR-interference (CRISPRi) in NB4 leukemia cell line Similarlygain- and loss-of-function experiments validated the func-tion ofmiRNAs and lncRNAsof the humanDLK1-DIO3 locusin the differentiation and maintenance of megakaryocytesTranscriptome analysis of 46 pediatric AML samples allowedidentification of prognostically relevant ncRNA signaturesshared by normal HSCs and AML blasts of distinct cytoge-netic and morphologic subgroups

In the same year a systematic analysis of lncRNAsin hematopoiesis and hematological malignancies was alsoconducted by Delas et al [193] with a murine model andAML cell lines First a catalog of lncRNAs was madein the 11 types of cells representing different stages ofhematopoietic differentiation and blood cancers Remark-ably similar expression patterns were observed for protein-coding genes and lncRNAs through hematopoietic systemand disease development which indicated involvement ofsimilar mechanisms of expression regulation A loss-of-function screen revealed that 20 lncRNAs were requiredfor leukemia progression in vivo Some of them seemed topromote leukemia stem-cell signatures eg Pvt1 and Lilamwhose functions were correlated with the function of MYConcogene Another lncRNA termed Pilna was requiredfor the myeloid lineage during bone marrow reconstitu-tion

Transcriptome analysis of an individual primary AMLsample fromTCGA dataset enabled discovering 194 unanno-tated small RNAs in the 17-35 nt size range 258 unannotatedsmall RNAs in the range of 36-100 nt and 977 previouslyunannotated multiexon lncRNA transcripts [194] Amajorityof them were also found in the sequencing data of otherAML patients from TCGA collection Integration of thecollected data with RNA-seq data from 179 other AMLcases led to identification of a subset of lncRNAs withenriched expression in AMLM3 (egMEG3) and other FABsubtypes comparing to normal CD34+ cells Reanalysis of200 transcriptomes from TCGA AML dataset was also usedto construct prognosis-related lncRNAmodule pathway net-work [195] First lncRNA coexpression network composedof 42 functionalmodules was generated thanks to integrationof data from small and mRNA sequencing Then survivalanalysis was performed for each of the identified lncRNAmodules and 8 of them significantly enriched in 70 pathways(including AML pathway chemokine signaling pathway andpathway in cancer) appeared to be correlated with patientoutcome

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 25

Relation between lncRNAexpression andAMLprognosiswas also studied by Mer et al [196] who sequenced tran-scriptomes of 274 intensively treated AML patients from aSwedish cohort and found 33 individual lncRNAs signifi-cantly associated with OS Based on lncRNA expression theauthors classified all AML patients to four distinct molecularsubtypes The reproducibility and prognostic significance ofthe identified lncRNA-based signatures were validated inan independent AML patient cohort (142 TCGA samples)Remarkably neither of the lncRNA-driven AML subtypeswas found to be highly concordant with any of the con-ventional clinical or genetic factors although enrichmentin CEBPA NPM1 FLT3-ITD and TP53 mutation was notedfor particular subtypes Despite some similarities betweenlncRNA and mRNA expression both types of transcriptsstratified AML patients in different ways This suggested alimited overlap in information retrieved from various levelsof transcriptome analysis and underpinned the rationale forfurther lncRNA studies

In 2019 Bill et al [197] while analyzing a large set of tran-scriptome data collected from 450 CN-AML patients iden-tified a 111-lncRNA-based signature specific for LSCs Inter-estingly only four out of 111 lncRNAs were downregulated insamples with high stem cell like gene expression profile ThelncRNA signature was mainly composed of long intergenicnoncoding RNAs (lincRNAs 30) antisense RNAs (19)and sense intronic RNAs (13) One of the upregulatedlncRNAs in LSCs was DANCR (Differentiation AntagonizingNonproteinCoding RNA) lincRNAhighly conserved betweenmice and humans overexpressed in hepatocellular carci-noma DANCR described as a regulator of the Wnt pathwaycrucial for the biology of LSCs was proved to play a rolein LSC self-renewal and quiescence Decreased expression ofMYC and other genes from Wnt pathway in AML cell lineafter DANCR knockdown confirmed association of DANCRwith Wnt pathway

27 Circular RNA

Thoughmost human and mammalian premRNAs are splicedinto linear molecules the existence of circular RNAs (cir-cRNAs) generated by noncanonical splicing (also termedbacksplicing) was sporadically reported within the last 30years [198] In leukemia a spectrum of abnormal KMT2Atranscripts including circular isoforms resulted from exonscramblingwas shown byCaldas et al in 1998 [199] Howeveruntil deep transcriptome sequencing was developed thisphenomenon could not be studied on a large scale At presentit is known that circRNAs arise from hundreds of humangenes in normal and malignant cells [198] The abundancediversity and enhanced stability of circular transcripts inhuman cells suggest that they not only are a result of splicingside effect but can play a role in the regulation of importantphysiological and pathological processes

You and Conrad who elaborated the acfs algorithm foridentification and quantification of circRNAs from single-and paired-ended RNA-seq data [200] demonstrated its effi-ciency on published AML datasets Comparing 5 APL and 5

CN-AML cases the authors found 80 circRNAswith oppositepattern of expression generated from the host genes crucialto the differentiation and proliferation of myeloid cells Forexample circ EMB generated from EMB gene encodingembigin transmembrane protein from the immunoglobulinsuperfamily considered as a cancer biomarker was highlyabundant in APL Upregulation in CN-AML was noted forcirc SMARCA5 originated from SMARCA5 gene encoding acore component of chromatin remodeling and spacing factorRSF which promotes cell proliferation In compliance withthe theory of circRNA functioning as miRNA sponges theauthors found miR-10b (a member of the miR-99 family)binding site on circ SMARCA5 As SMARCA5 expressionwas shown to be affected by miR-99 binding of miR-10bby circ SMARCA5 could relieve linear SMARCA5 transcriptsfrom repression and contribute to the accumulation ofundifferentiated myeloid cells Analysis of data from a largerAML cohort revealed that the gene loci frequently mutated inAML produced significantly more circRNAs

In 2017 circRNAs generated from NPM1 gene have beenextensively studied by Hirsch et al [201] In total in six AMLcell lines (including OCI-AML3 line with NPM1 mutation)one CML cell line and 3 samples derived from the PBMCfraction of healthy volunteers 15 circular NPM1 transcriptswere identified including those previously deposited in cir-cBase and novel ones Oxford Nanopore technology of longread sequencingwas harnessed to verify the internal structureof identified transcripts One of them (hsa circ 0075001)which exhibited highly differential expression in theAML celllines was quantified in a cohort of 46 AMLpatients Based onhsa circ 0075001 expression patients were divided into twogroups that differed in the expression ofmore than 2000 othergenes AML patients with high hsa circ 0075001 expressionpresented upregulation of ribosomal protein genes increaseof total NPM1 expression and downregulation of genesinvolved in the Toll-like receptor (TLR) signaling pathwayand genes targeted by miR-181 (eg CARD8 CASP1 MSR1SLC11A1 TLR4) which is deregulated in CN-AML Theexpression of hsa circ 0075001 correlated with total NPM1expression but was not affected by the NPM1 mutationalstatus Global analysis of circRNA expression in 10AML sam-ples and 10 sorted cell fractions form healthy hematopoieticcontrols evidenced the existence of circRNA transcripts foralmost half of all highly expressed genes [201] Despite ageneral tendency towards higher circRNA expression fromgenes with higher parental gene expression there were someexceptions eg circFLT3 expression was not correlated withFLT3 gene expression in AML samples (though it was inhealthy samples) AML patients and healthy controls differedin the expression levels of circRNAs arisen from 27 genesincluding ANGPT1 UGCG and FLT3 In addition AMLsubgroup-specific circRNA signatures were identified egNPM1-mutated patients could be distinguished from NPM1-wt patients based on their global circRNA expression (but notbased on circNPM1 expression)

LrsquoAbbate et al [202] who analyzed the architecture andexpression pattern of chromosome 8 region with MYCamplification in 23 cases of AML detected a significantoverexpression of circPVT1 a circular transcript of PVT1

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

26 Journal of Oncology

gene in the studied AML cases compared to NK-AML Theexpression of circPVT1 was correlated with PVT1 gene copy-number increase and high PVT1 expression in AML patientswithMYC amplification

Contribution of circRNA to drug resistance was recentlyshown by Shang et al with the use of doxorubicin-resistantTHP-1 AML cell line (THP-1ADM) [203] The authorsidentified 49 circRNAsdifferentially expressed betweenTHP-1ADM and THP-1 cells One of them circPAN3 wasoverexpressed not only in THP-1ADM cells but also inrefractory and recurrent AML samples Silencing of thiscircRNA restored sensitivity to doxorubicin in THP-1ADMcells which indicated its significant role in chemoresistancemediation

28 Focus on Stem Cells

Leukemic blasts which accumulate in BM and PB of AMLpatients represent cells blocked at a particular stage ofdifferentiation However in the majority of GEP studiesperformed within the first years of microarray technologydevelopment bulk AML cells were compared to the wholepool of mononuclear cells from healthy control samplesThen the cell sorting techniques were introduced to selectparticular cell fractions The example of a study where AMLcells were compared not only to unselected healthy BM andPB samples but also to normal hematopoietic CD34+ cellsextracted from BM and PB was that of Stirewalt et al [204]They found 13 genes deregulated in AML compared to allsubpopulations of normal hematopoietic cells including 7upregulated (BIK CCNA1 FUT4 IL3RA HOMER3 JAG1WT1) and 6 downregulated (ALDHA1A PELO PLXNC1PRUNE SERPINB9 TRIB2) Moreover the expression lev-els of WT1 FUT4 CCNA1 HOMER3 JAG1 TRIB2 andSERPINB9 were strongly associated with FAB classificationwhereas WT1 JAG1 ALDH1A1 TRIB2 and PLXNC1 withcytogenetics For example WT1 was the most overexpressedin AML with inv(16) or t(1517) CCNA1 in t(1517) whileBIK expression was absent or extremely low in t(821) 7upregulated genes were also measured in pediatric AMLwhere BIK FUT4 and WT1 showed the most significantincrease in expression

After the discovery of the self-renewing leukemic stemcells (LSCs) the main focus was shifted to these earlyprogenitor cells capable of initiating leukemogenesis Howthey are different from normal human hematopoietic stemcells (HSCs) was first shown in 2009 by Majeti et al [205]who identified over 3000 DEGs between normal HSCsand LSCs extracted from AML patients The selected genesencoded mainly proteins with kinase activity associated withnucleoplasm Golgi apparatus chromosomes vacuoles andactin cytoskeleton KEGG pathway analysis revealed the topdysregulated pathways in LSCswere those related to adherensjunction ribosome regulation of actin cytoskeleton tightjunction focal adhesion apoptosis MAPK signaling T-cellreceptor signaling pathway JAK-STAT signaling and Wntsignaling Some of these pathways were already associatedwith stem-cell biology and cancer development other were

not studied in cancer stem cells yet The obtained resultsemphasized the importance of the LSCsrsquo interaction withtheir niche in leukemia initiation and progression

Gentles et al showed the correlation of high expressionof leukemic stem-cell genes with adverse outcomes in AML[206] Comparing subpopulations of cells extracted from 163NK-AML samples the authors identified 52 genes discrim-inating the LSC-enriched subpopulations (CD34+CD38-)from the leukemic progenitor cell (LPC)-enriched subpop-ulations (CD34+CD38+) among others genes involved inearly hematopoiesis egVNN1 RBPMS SETBP1 GUCY1A3MEF2C and HOPX Genes associated with proliferation cellcycle and differentiation were systematically repressed in theLSCs The identified LSC gene expression signature (LSCscore) reflecting self-renewal ability was validated on fourindependent datasets of 1047 patients in total leading tosimilar conclusions OS EFS and RFSwereworse for patientswith high LSC score [100]

Ng et al [207] compared gene expression profiles between138 LSC+ and 89 LSC- cell fractions from 78 AML patientsFrom the list of 104 DEGs 17 most related to stemnesswere selected to generate a LSC score (LSC17) whichcould be calculated for each patient as the weighted sumof expression of the 17 genes Strong association betweenhigh LSC17 scores and poor overall and event-free survivalwas observed as in the case of higher bone marrow blastpercentage at diagnosis higher incidence of the FLT3-ITDand adverse cytogenetics The LSC score was validated byxenotransplantation assays and by reanalysis of microarrayand RNA-seq data from five independent cohorts of morethan 900 AML patients with different subtypes Comparingwith other prognostic determinants such as age WBC countcytogenetic risk group and mutational status LSC17 scorewas the strongest and independent prognostic factor In theend the authors designed a customNanoString assay to easilyanalyze expression of 17 genes from the LSC signature Theassay should allow for rapid risk assessment at diagnosisapplication of more intensified investigational therapies inthe case of high-score patients and protection of low-scorepatients against unnecessary toxicity To test its efficiency inchildhood leukemia Duployez et al applied the LSC17 scoreto 228 de novopediatricAMLpatients [208] Indeed childrenwith low LSC17 score had significantly better outcome (OSand EFS) compared to children with high score Then thestemness signature was validated in 257 children from anindependent AML cohort However prognosis of pediatricpatients with low LSC17 was not significantly better thanthosewith intermediate LSC17 score The differences betweenadult and pediatric AML patients might result from thedifferent proportion of CBF AML in both groups (twice ashigh in children) Nevertheless the authors extended theLSC17 prognostic value to pediatric AML patients at leastwith non-CBF AML [208] Among all negative prognosticfactors including the high LSC17 score high WBC countcytogenetic group ldquoother aberrationsrdquo and presence of WT1and RUNX1 mutations high LSC17 score gained the beststatistics

Recently GEP of LSCs HSCs and leukemic progenitorsfrom the same AML bone marrow enabled identifying three

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 27

genes overlapping in the results of two pairwise comparisons(LSCs vs HSCs and LSCs vs leukemic progenitors) S100A8SOD2 and IGFBP7 [209] Significantly lower expression ofIGFBP7 encoding insulin-like growth factor-binding pro-tein 7 in LSCs was correlated with reduced sensitivity tochemotherapy Restoration of IGFBP7 expression by a recom-binant human gene resulted in differentiation inhibition ofLSC survival and improved response to therapy withoutaffecting normal hematopoiesis and HSC survival ThereforeIGFBP7 gene was postulated to be a factor responsible for thepersistence of LSCs

29 Taking Advantage of Model Organisms

Deep insight into the biology of the disease is not possibleby the analysis of cells extracted from patient BM or PB Invitro experiments with cell line models are convenient buthave also some limitations In fact animal models are thebest option to study particular gene function consequencesof early pathogenic events or treatment with potential ther-apeutics Using retroviral insertion mutagenesis in miceErkeland et al (2004) [210] identified a number of genes thatcould be involved in the pathogenesis of AML In one oftheir later works the authors studied the virus integrationsites (VISs) and virus common integration sites (CISs) inthe human GEP datasets of 285 adult AML samples [61]and 130 pediatric AML samples [68] First they notedVIS flanking genes were significantly more differentiallyexpressed between AML clusters than random genes in bothdatasets Then the authors identified five regulatory networksinvolving 110 VISCIS genes most differentially expressed inthe adult dataset Network associated with cell growth con-tained only these genes Many of them eg interleukin andSTAT genes were implicated in cytokine signaling Anothernetwork consisted of gene expression regulators was ableto discriminate between AML patients with favorable andunfavorable prognosis In the unfavorable group HOXA9MEIS1 and CCND3 genes were increased whereas BCOR andGFI1 genes were decreased

Glass et al [211] used NGS platform to identify MECOM(previously EVI1) target genes by comparison in MECOM-overexpressing murine myeloid leukemia cell lines (DA-1NFS-60) before and after shRNA-mediated MECOM knock-down MECOM oncogenic transcription factor associatedwith human myeloid malignancies of poor prognosis isoverexpressed in 8ndash10 of adult AML and up to 27 ofpediatric leukemias with KMT2A rearrangementsMECOM-induced leukemic cells present impaired myeloid differen-tiation resistance to apoptosis and aberrant cell cycle reg-ulation which results in excessive proliferation CombiningRNA-seq expression profiling with ChIP-seq the authorsfound MECOM directly bound to and downregulated Cebpegene encoding a master myeloid differentiation regulatorSerpinb2 gene encoding serine protease inhibitor involvedin cell cycle regulation and numerous genes from the Jak-Stat signaling pathway that drive cellular differentiationIn addition several P2X purinoceptors responsible forATP mediated apoptosis in neutrophils and macrophages

appeared to be significantly downregulated in MECOMleukemic cells

An interesting approach was applied by Wilhelm et al(2011) [212] who generated two related murine leukemicclones through the retroviral overexpression of Meis1 andHoxa9 genes in the purified fetal liver (FL) cells Bothclones differed in the hematopoietic stem-cell frequency andgene expression profile Considering microRNAs only afew miRNAs were really highly expressed more than 95of all miRNA transcripts came from the top 15 miRNAsFunctional annotation analysis showed the most differenti-ating genes betweenMeis1- andHoxa9-overexpressing cloneswere related to immune system development hemopoietic orlymphoid organ development hemopoiesis and myeloid celldifferentiation Both differential expression andor differen-tial splicing were observed in the studied transcriptomesMoreover hundreds of single nucleotide variants (SNVs)shared by both cell clones or unique for one of them wereidentified with respect to the public reference sequence Thestudy revealed also a number of unannotated transcribedelements

30 Gene Expression RegulationNew Directions

McKeown et al (2017) [213] matched the epigenomic cir-cuitry with the transcriptional state of leukemic cells toidentify potential new treatment strategies The authorsfocused on the large (gt20 kb) highly active chromatinregions called ldquosuper-enhancersrdquo (SEs) described previouslyas key oncogenic drivers in tumor cells [214] AlthoughSEs constitute only about 5 of all enhancers they areinvolved in the regulation of the crucial genes defining cellidentity and phenotype 66 AML patients were characterizedin terms of enhancers super-enhancers gene expressionand mutational status of blasts [213] On average 807 SEsper sample with a median length of 22 kb were identifiedMost of them were linked to multiple genes In additionSE maps from normal HSPCs and monocytes were used todefine the signature of differentiation state The two mostpervasive enhancer signatures were found in the genome andnamed Myeloid Differentiation and HOX Factor Activationthe last correlated with enhancers associated with homeobox(HOX) genes PBX3 and MEIS1 Based on overall SE distri-bution patients were classified to 6 subgroups including 4enriched in NPM1 and FLT3 mutation and one containingall AML samples with KMT2A translocation The authorsconcluded KMT2A translocations might induce a uniqueepigenetic state affecting overall enhancer landscape whichis consistent with the association of KMT2A fusion pro-teins with DOT1L deregulation and consequently aberranthistone H3K79 methylation Moreover patients classified todifferent SE-clusters showed differences in OS Discoveryof RARA-associated SE which differentiated patient samplesand correlated with significantly higher RARA expressionin 25 patients prompted the authors to test the sensi-tivity of AML cells presenting a strong SE at the RARAlocus to RAR120572 targeted compounds The results of tests

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

28 Journal of Oncology

conducted on AML cell lines AML patient derived mousexenograft models and AML cells ex vivo demonstrated thatRAR120572 agonist SY-1425 (tamibarotene) selectively inhibitedthe proliferation of AML cells with high (but not low) RARAexpression stimulating the expression of genes linked togranulocytic differentiation (eg CD38 ITGAX ITGAMand CD66) and retinoic acid response (eg DHRS3) Inthe context of efficiency comparison of SY-1425 to ATRAused clinically for APL therapy showed the prevalence ofthe tested compound Transcriptional response of APL toretinoids was similar to the response of AML cell lines withhigh RARA expression to SY-1425The authors presented thefollowing interpretation of the observed results while APL-specific PML-RARA fusion protein represses transcriptionof differentiation-related genes in non-APL AML RARA-associated SE induces overexpression of unliganded RAR120572which acts as a transcriptional repressor of genes regulatingby RAR120572 The practical consequence of clinical significanceis that some non-APL AML patient may benefit from theATRA-like therapy and increased levels of RARA mRNAcould be used as a prerequisite for the treatment by SY-1425which has a power to reset RAR120572 transcriptional activity

Not only genetic and epigenetic factors contribute toAML pathogenesis initiation progression and maturation ofAML are also affected by posttranslational modification ofproteins The important role of SUMOylation of sPRDM16in AML progression was demonstrated by Dong amp Chen onleukemic cell lines [215] PRDM16 previously termed MEL1encodes transcription factor acting as a H3K9me1 methyl-transferase responsible for maintenance of heterochromatinintegrity However only protein isoform deprived of the his-tonemethyltransferase domain encoded by a short transcriptvariant sPRDM16 was associated with AML pathogenesisThe authors showed that SUMOylation of sPRDM16 changedthe expression of genes implicated in wound responsecell proliferation chemotaxis differentiation and cell cycleprogression including 13 genes (eg KLF10 BCL3 HDAC9CCL5 IL6R LIF and NUMB) involved in proliferation anddifferentiation of hematopoietic and leukemic cells

31 Single-Cell Sequencing The Future

ConsideringAMLas amulticlonal cancer we should be awareof the fact that bulk RNA sequencing reflects what is goingon in a dominant clone that is not necessarily a clone oforigin Even application of cell sorting technique may notbe sufficient to catch the whole AML heterogeneity as far aswe sequence a pool of cells A promising approach whichcan overcome this problem is single-cell gene expressionprofiling A pilot single-cell AML study was performed ontwenty CD45-positive cells collected from an individual AMLpatient [216] GEP showed only 11 out of 20 cells selectedfor the analysis were putative blasts ie CD34-positive orHLA-DRA- and CD117-positive Moreover two of them wereoutliers in PCA Complementary targeted DNA sequencingrevealed the presence of mutations in DNMT3A and NPM1and FLT3-ITD in the analyzed AML sample HoweverRNA-seq identified only DNMT3A mutation in only one

single cell Possible explanations are not sufficient coverage(althoughNPM1 and FLT3 genes are usually highly expressedin AML) imperfect algorithms for mutation detection inRNA-seq data and mutational heterogeneity of blast cellsAn argument for the last explanation can be found in thereport of Shlush et al [217] who identified in highly purifiedpreleukemic stem cells DNMT3A mutation at high allelefrequency but did not detected concomitant NPM1mutationTherefore the authors concluded DNMT3A arose early inAML evolution The preliminary experiences with single-cellsequencing show this technology demands optimization butseems to be a strategy of future

32 Commercial Solutions forAML Research and Clinics

As a result of years of transcriptome-scale studies a fewcommercial tools dedicated to AML and other hemato-logic malignancies were developed Affymetrix technologywhich conquered the microarray market was used by Sky-lineDx (httpswwwskylinedxcom) high-tech commercial-stage biotech company headquartered in Rotterdam theNetherlands to design AMLprofiler a qualitative in vitrodiagnostic and prognostic microarray supporting the choiceof optimal therapy strategy [218] AMLprofiler combinesseven separate assays used for the purpose of cytogeneticsmutation detection and gene expression analysis reducingthe time between sampling and diagnosis from 4 weeks to3 days The time of diagnostic report generation does notexceed 15 min

Illumina (httpswwwilluminacom) the companyheadquartered in SanDiego CaliforniaUSAwhopracticallymonopolized the NGSmarket released theMiSeqDx systemthe first FDA-regulated CE-IVD-marked NGS platformfor in vitro diagnostic testing Although AML-dedicatedkits do not exist for MiSeqDx the system is customizableand a number of partners collaborates with the companyto develop new clinical assays At present kits for targetsequencing are available for other Illumina systems egTruSight Myeloid Sequencing Panel targeting 54 genesincluding CEBPA NPM1 and FLT3-ITD or AmpliSeq forIllumina Myeloid Panel targeting 40 DNA genes 29 RNAfusion driver genes and 5 gene expression levels associatedwith myeloid cancers including AML

Other integrated diagnostic platforms are being devel-oped and validated eg rapid and sensitive NGS-based assaycombining karyotyping and mutational screening of AML[219] Here three NGS libraries are generated two DNA-based librariesmdashfor whole genome sequencing and selectedvariant identificationmdashand one RNA-based library for fusiontranscript detection The whole workflow can be completedwithin 5 days

33 Conclusions

Within the last two decades an explosion of AML studiesdriven by technological progress could be observed AML

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

Journal of Oncology 29

picture emerging from transcriptome research is very com-plex and dynamic AML transcriptome affected by cytoge-netic and genetic variability resembles a mosaic composedof many elements interacting with each other AML sub-types present unique patterns of protein-coding gene andmicroRNA expressionMoreover a spectrumof fusion genesalternative transcripts and newly discovered chimeric RNAsas well as a lot of noncoding RNAs including long linearand circular forms contribute to the complexity of leukemictranscriptome

Transcriptomedata cannot be interpreted separately fromgenetic and genomic data Even a sole point mutation canaffect expression of numerous genes eg when it occurs in atranscription factor epigenetic regulator or a crucialmemberof a signaling pathway Therefore genome exome methy-lome sequencing or targeted resequencing is often combinedwith RNA-seq In some cases transcriptome analyses helpedto define which mutation was the driver one It must beremembered that a disease is not limited to leukemic blastsRecent studies on BMmicroenvironment underlined its rolein AML initiation progression and relapse Recognition ofmolecular interplay between LSCs and BM niche not onlyis necessary to understand the AML biology but also opensnovel AML treatment directions

Many reports proved the power of global transcriptomeprofiling and proposed application of GEP for AML diag-nosis and prognosis A single microarray-based test usuallyclassified AML subgroups properly without any a prioriknowledge The following studies utilized more advancedgenerations of microarrays composed of an increasing num-ber of probes and included more numerous cohorts ofpatients Of note the results obtained by different groupswere largely overlapping As gene expression-based predic-tion of the major cytogenetic subgroups was efficient in bothpediatric and adult AMLs GEP appears as an attractivealternative to classical cytogenetics which is laborious andtime-consuming On the other hand there is no need toharness high-throughput technology when simple eg PCR-based diagnostic tests can easily confirm the presence of thefusion genes

Because some genes are significantly overexpressed inAML not only in leukemic blasts but also in BM microen-vironment and their expression affects treatment responsetranscript measurement at the time of diagnosis shouldbe obligatory in AML diagnostics In my opinion reliablequantitative PCR techniques eg ddPCR have currentlythe highest potential to be routinely applied in clinicalpractice to analyze selected transcript levels To analyzehigher number of transcripts gene expression arrays or RNA-seq may be applied instead Although both high-throughputGEP approaches demand specialized expensive equipmentand well-trained staff RNA-seq seems to be a superiortechnology offering more information at a comparable costWithin the last few years small personal relatively cheapand portable sequencers started to appear which makesNGS-based tests more available Moreover alternative third-generation sequencing technologies are becoming more andmore popular I believe that future clinical diagnostic lab-oratories will offer NGS services not limited to mutation

detection as a standard Even if accepting GEP as the onlydiagnostic test would be difficult it could at least serve as afirst screening or a complementary tool

Summarizing although gene expression studies werenot directly translated into clinical practice up to datethey helped us to understand the biology of tumors andundoubtedly contributed to the improvement of classificationof hematological malignancies and risk estimation which iscrucial for optimal treatment decision and directs us towardspersonalized medicine

Conflicts of Interest

The author declares that she has no conflicts of interest

Funding

This study was supported by the Polish Ministry of Sci-ence and Higher Education under the SPUB (5536E-63SPUB20171) program

Acknowledgments

I would like to thank Professor John M Bennett Universityof Rochester Medical Center who kindly agreed to sharethe microscopic images of M1-M7 FAB AML types from hisprivate collection

Supplementary Materials

Supplementary Table 1 presents the results of PubMed searchdemonstrating a great effort of the scientific communityput into acute myeloid leukemia research All terms usedfor searching are included along with the correspondingnumbers of papers found with two additional filters appliedPart of the information presented in Supplementary Table 1was used to generate Figure 1 plots Supplementary Table 2lists all AML transcriptome papers cited in the manuscriptThe following information is included title first and lastauthor laboratory name of the journal year of publica-tion AML sample size sample description and technologyapplied All papers are divided according to the technology(microarrayPCR or NGS) and then ordered chronologically(Supplementary Materials)

References

[1] A Kuykendall N Duployez N Boissel J E Lancet and J SWelch ldquoAcute myeloid leukemia the good the bad and theuglyrdquo ASCO Educational Book no 38 pp 555ndash573 2018

[2] C D Bloomfield E Estey L Pleyer et al ldquoTime to repeal andreplace response criteria for acute myeloid leukemiardquo BloodReviews vol 32 no 5 pp 416ndash425 2018

[3] L Hartmann and K H Metzeler ldquoClonal hematopoiesisand preleukemiamdashGenetics biology and clinical implicationsrdquoGenes Chromosomes and Cancer 2019

[4] D Duarte E D Hawkins and C Lo Celso ldquoThe interplay ofleukemia cells and the bonemarrowmicroenvironmentrdquoBloodvol 131 no 14 pp 1507ndash1511 2018

30 Journal of Oncology

[5] S Li C E Mason and A Melnick ldquoGenetic and epigeneticheterogeneity in acute myeloid leukemiardquo Current Opinion inGenetics amp Development vol 36 pp 100ndash106 2016

[6] D Bonnet ldquoCancer stem cells lessons from leukaemiardquo CellProliferation vol 38 no 6 pp 357ndash361 2005

[7] J J Yang T S Park and T S Wan ldquoRecurrent cytogeneticabnormalities in acute myeloid leukemiardquo in Cancer Cytoge-netics vol 1541 of Methods in Molecular Biology pp 223ndash245Springer New York NY USA 2017

[8] S C Meyer and R L Levine ldquoTranslational implications ofsomatic genomics in acute myeloid leukaemiardquo The LancetOncology vol 15 no 9 pp e382ndashe394 2014

[9] J M Bennett D Catovsky M T Daniel et al ldquoProposals forthe classification of the acute leukaemiasrdquo British Journal ofHaematology vol 33 no 4 pp 451ndash458 1976

[10] JM Bennett D CatovskyM TDaniel et al ldquoProposed revisedcriteria for the classification of acutemyeloid leukemia A reportof the French-American-British Cooperative Grouprdquo Annals ofInternal Medicine vol 103 no 4 pp 620ndash625 1985

[11] E S Jaffe International Agency for Research on Cancer andOrganitzacio Mundial de la Salut Pathology and Genetics ofTumours of Haematopoietic and Lymphoid Tissues IARC PressWHO OMS Lyon USA 2001

[12] World Health OrganizationWHO Classification of Tumours ofHaematopoietic and Lymphoid Tissues IARC Press Lyon USA2008

[13] D A Arber A Orazi and R Hasserjian ldquoThe 2016 revisionto the World Health Organization classification of myeloidneoplasms and acute leukemiardquoBlood vol 127 no 20 pp 2391ndash2405 2016

[14] T Lapidot C Sirard J Vormoor et al ldquoA cell initiating humanacutemyeloid leukaemia after transplantation into SCID micerdquoNature vol 367 no 6464 pp 645ndash648 1994

[15] H J Sutherland A Blair and R W Zapf ldquoCharacterization ofa hierarchy in human acute myeloid leukemia progenitor cellsrdquoBlood vol 87 no 11 pp 4754ndash4761 1996

[16] D Bonnet and J E Dick ldquoHuman acute myeloid leukemiais organized as a hierarchy that originates from a primitivehematopoietic cellrdquo Nature Medicine vol 3 no 7 pp 730ndash7371997

[17] D Thomas and R Majeti ldquoBiology and relevance of humanacute myeloid leukemia stem cellsrdquo Blood vol 129 no 12 pp1577ndash1585 2017

[18] B S White and J F DiPersio ldquoGenomic tools in acute myeloidleukemia From the bench to the bedsiderdquo Cancer vol 120 no8 pp 1134ndash1144 2014

[19] H J M de Jonge G Huls and E S J M de Bont ldquoGeneexpression profiling in acute myeloid leukaemiardquo The Nether-lands Journal of Medicine vol 69 no 4 pp 167ndash176 2011

[20] T R Golub D K Slonim P Tamayo et al ldquoMolecularclassification of cancer class discovery and class prediction bygene expressionmonitoringrdquo Science vol 286 no 5439 pp 531-527 1999

[21] T J Ley E R Mardis L Ding B Fulton M D McLellan KChen et al ldquoDNA sequencing of a cytogenetically normal acutemyeloid leukaemia genomerdquoNature vol 456 no 7218 pp 66ndash72 2008

[22] E R Mardis L Ding D J Dooling D E Larson M DMcLellan K Chen et al ldquoRecurring mutations found bysequencing an acute myeloid leukemia genomerdquo The NewEngland Journal of Medicine vol 361 no 11 pp 1058ndash10662009

[23] The Cancer Genome Atlas Research Network ldquoGenomicand epigenomic landscapes of adult de novo acute myeloidleukemiardquo The New England Journal of Medicine vol 368 pp2059ndash2074 2013

[24] E Papaemmanuil M Gerstung L Bullinger et al ldquoGenomicclassification and prognosis in acute myeloid leukemiardquo TheNew England Journal of Medicine vol 374 no 23 pp 2209ndash2221 2016

[25] JW Tyner C E Tognon D Bottomly BWilmot S E Kurtz SL Savage et al ldquoFunctional genomic landscape of acutemyeloidleukaemiardquo Nature vol 562 no 7728 pp 526ndash531 2018

[26] K Mrozek M D Radmacher C D Bloomfield and GMarcucci ldquoMolecular signatures in acute myeloid leukemiardquoCurrent Opinion in Hematology vol 16 no 2 pp 64ndash69 2009

[27] M Medinger C Lengerke and J Passweg ldquoNovel prognosticand therapeutic mutations in acute myeloid leukemiardquo CancerGenomics amp Proteomics vol 13 no 5 pp 317ndash330 2016

[28] MM Showel andM Levis ldquoAdvances in treating acutemyeloidleukemiardquo F1000Prime Reports vol 6 2014

[29] N J Short M E Rytting and J E Cortes ldquoAcute myeloidleukaemiardquoThe Lancet vol 392 no 10147 pp 593ndash606 2018

[30] M Stahl B Y Lu T K Kim and A M Zeidan ldquoNoveltherapies for acutemyeloid leukemia are we finally breaking thedeadlockrdquo Targeted Oncology vol 12 no 4 pp 413ndash447 2017

[31] S Ferrari U Torelli L Selleri et al ldquoStudy of the levels ofexpression of two oncogenes c-myc and c-myb in acute andchronic leukemias of both lymphoid and myeloid lineagerdquoLeukemia Research vol 9 no 7 pp 833ndash842 1985

[32] F Mavilio N M Sposi M Petrini et al ldquoExpression of cellularoncogenes in primary cells from human acute leukemiasrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 83 no 12 pp 4394ndash4398 1986

[33] C Wang J E Curtis E N Geissler E A McCulloch and MD Minden ldquoThe expression of the proto-oncogene c-kit in theblast cells of acute myeloblastic leukemiardquo Leukemia vol 3 no10 pp 699ndash702 1989

[34] R Amson F Sigaux S Przedborski G Flandrin D Givol andA Telerman ldquoThe human protooncogene product p33pim isexpressed during fetal hematopoiesis and in diverse leukemiasrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 86 no 22 pp 8857ndash8861 1989

[35] M Saito K Helin M B Valentine et al ldquoAmplification of theE2F1 transcription factor gene in the HEL erythroleukemia celllinerdquoGenomics vol 25 no 1 pp 130ndash138 1995

[36] T Tanaka K Tanaka S Ogawa et al ldquoAn acute myeloidleukemia gene AML1 regulates hemopoietic myeloid celldifferentiation and transcriptional activation antagonisticallyby two alternative spliced formsrdquo EMBO Journal vol 14 no 2pp 341ndash350 1995

[37] C E Carow M Levenstein S H Kaufmann et al ldquoExpressionof the hematopoietic growth factor receptor FLT3 (STK-1Flk2)in human leukemiasrdquo Blood vol 87 no 3 pp 1089ndash1096 1996

[38] A Porwit-MacDonald G Janossy K Ivory et al ldquoLeukemia-associated changes identified by quantitative flow cytometry IVCD34 overexpression in acute myelogenous leukemia M2 witht(8 21)rdquo Blood vol 87 no 3 pp 1162ndash1169 1996

[39] M Sawa K Yamamoto T Yokozawa et al ldquoBMI-1 is highlyexpressed in m0-subtype acute myeloid leukemiardquo Interna-tional Journal of Hematology vol 82 no 1 pp 42ndash47 2005

[40] K Inoue H Sugiyama H Ogawa et al ldquoWT1 as a newprognostic factor andanewmarker for the detection ofminimal

Journal of Oncology 31

residual disease in acute leukemiardquo Blood vol 84 no 9 pp3071ndash3079 1994

[41] M Heuser G Beutel J Krauter et al ldquoHigh meningioma 1(MN1) expression as a predictor for poor outcome in acutemyeloid leukemiawith normal cytogeneticsrdquoBlood vol 108 no12 pp 3898ndash3905 2006

[42] C D Baldus ldquoBAALC expression predicts clinical outcomeof de novo acute myeloid leukemia patients with normalcytogenetics a Cancer and Leukemia Group B Studyrdquo Bloodvol 102 no 5 pp 1613ndash1618 2003

[43] G Marcucci C D Baldus A S Ruppert et al ldquoOverexpressionof the ETS-related gene ERG predicts a worse outcome inacute myeloid leukemia with normal karyotype a Cancer andLeukemia Group B studyrdquo Journal of Clinical Oncology vol 23no 36 pp 9234ndash9242 2005

[44] S Barjesteh vanWaalwijk vanDoorn-Khosrovani C ErpelinckW L J van Putten et al ldquoHigh EVI1 expression predicts poorsurvival in acutemyeloid leukemia a study of 319 de novo AMLpatientsrdquo Blood vol 101 no 3 pp 837ndash845 2003

[45] R Bumgarner ldquoOverview of DNA microarrays types applica-tions and their futurerdquo Current Protocols in Molecular Biology2013 Chapter 22Unit 221

[46] M Schena D Shalon R W Davis and P O Brown ldquoQuantita-tive monitoring of gene expression patterns with a complemen-tary DNA microarrayrdquo Science vol 270 no 5235 pp 467ndash4701995

[47] M Schena D Shalon R Heller A Chai P O Brown andR W Davis ldquoParallel human genome analysis microarray-based expression monitoring of 1000 genesrdquo Proceedings of theNational Acadamy of Sciences of the United States of Americavol 93 no 20 pp 10614ndash10619 1996

[48] A Alizadeh M Eisen R Davis et al ldquoThe lymphochip aspecialized cdna microarray for the genomic-scale analysis ofgene expression in normal and malignant lymphocytesrdquo ColdSpring Harbor Symposium on Quantitative Biology vol 64 pp71ndash78 1999

[49] K Virtaneva F A Wright S M Tanner et al ldquoExpressionprofiling reveals fundamental biological differences in acutemyeloid leukemia with isolated trisomy 8 and normal cytoge-neticsrdquo Proceedings of the National Acadamy of Sciences of theUnited States of America vol 98 no 3 pp 1124ndash1129 2001

[50] C Schoch A Kohlmann M Dugas et al ldquoGenomic gainsand losses influence expression levels of genes located withinthe affected regions a study on acute myeloid leukemias withtrisomy 8 11 or 13 monosomy 7 or deletion 5qrdquo Leukemia vol19 no 7 pp 1224ndash1228 2005

[51] J Lu G Getz E AMiska et al ldquoMicroRNA expression profilesclassify human cancersrdquoNature vol 435 no 7043 pp 834ndash8382005

[52] B Falini C Mecucci E Tiacci et al ldquoCytoplasmic nucleophos-min in acute myelogenous leukemia with a normal karyotyperdquoThe New England Journal of Medicine vol 352 no 3 pp 254ndash266 2005

[53] T Haferlach ldquoGlobal approach to the diagnosis of leukemiausing gene expression profilingrdquo Blood vol 106 no 4 pp 1189ndash1198 2005

[54] S Mi J Lu M Sun et al ldquoMicroRNA expression signaturesaccurately discriminate acute lymphoblastic leukemia fromacute myeloid leukemiardquo Proceedings of the National Acadamyof Sciences of the United States of America vol 104 no 50 pp19971ndash19976 2007

[55] P W Janes N Saha W A Barton et al ldquoAdam meets Eph anADAMsubstrate recognitionmodule acts as amolecular switchfor ephrin cleavage in transrdquo Cell vol 123 no 2 pp 291ndash3042005

[56] C Schoch A Kohlmann S Schnittger et al ldquoAcute myeloidleukemias with reciprocal rearrangements can be distinguishedby specific gene expression profilesrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 99 no15 pp 10008ndash10013 2002

[57] S Debernardi D M Lillington T Chaplin et al ldquoGenome-wide analysis of acutemyeloid leukemia with normal karyotypereveals a unique pattern of homeobox gene expression distinctfrom those with translocation-mediated fusion eventsrdquo GenesChromosomes and Cancer vol 37 no 2 pp 149ndash158 2003

[58] R G Verhaak B J Wouters C A Erpelinck et al ldquoPredictionof molecular subtypes in acutemyeloid leukemia based on geneexpression profilingrdquo Haematologica vol 94 no 1 pp 131ndash1342009

[59] RW Georgantas III V Tanadve M Malehorn et al ldquoMicroar-ray and serial analysis of gene expression analyses identifyknown and novel transcripts overexpressed in hematopoieticstem cellsrdquo Cancer Research vol 64 no 13 pp 4434ndash44412004

[60] L Bullinger K Dohner E Bair et al ldquoUse of gene-expressionprofiling to identify prognostic subclasses in adult acutemyeloid leukemiardquo The New England Journal of Medicine vol350 no 16 pp 1605ndash1616 2004

[61] P J Valk R G Verhaak M A Beijen et al ldquoPrognosticallyuseful gene-expression profiles in acute myeloid leukemiardquoTheNew England Journal of Medicine vol 350 no 16 pp 1617ndash16282004

[62] N C Gutierrez R Lopez-Perez J M Hernandez et al ldquoGeneexpression profile reveals deregulation of genes with relevantfunctions in the different subclasses of acutemyeloid leukemiardquoLeukemia vol 19 no 3 pp 402ndash409 2005

[63] J E Payton N R Grieselhuber L Chang et al ldquoHigh through-put digital quantification of mRNA abundance in primaryhuman acutemyeloid leukemia samplesrdquoThe Journal of ClinicalInvestigation vol 119 no 6 pp 1714ndash1726 2009

[64] T Haferlach A Kohlmann S Schnittger et al ldquoAML M3 andAMLM3 variant each have a distinct gene expression signaturebut also share patterns different from other genetically definedAML subtypesrdquoGenes Chromosomes and Cancer vol 43 no 2pp 113ndash127 2005

[65] T Haferlach A Kohlmann L Wieczorek et al ldquoClinical utilityof microarray-based gene expression profiling in the diagnosisand subclassification of leukemia report from the internationalmicroarray innovations in leukemia study grouprdquo Journal ofClinical Oncology vol 28 no 15 pp 2529ndash2537 2010

[66] D R de la Bletiere O Blanchet P Cornillet-Lefebvre et alldquoRoutine use of microarray-based gene expression profilingto identify patients with low cytogenetic risk acute myeloidleukemia accurate results can be obtained even with subopti-mal samplesrdquo BMC Medical Genomics vol 5 no 1 2012

[67] S A Armstrong J E Staunton L B Silverman et al ldquoMLLtranslocations specify a distinct gene expression profile thatdistinguishes a unique leukemiardquo Nature Genetics vol 30 no1 pp 41ndash47 2002

[68] M E Ross ldquoGene expression profiling of pediatric acutemyelogenous leukemiardquo Blood vol 104 no 12 pp 3679ndash36872004

32 Journal of Oncology

[69] A Andersson T Olofsson D Lindgren et al ldquoMolecularsignatures in childhood acute leukemia and their correlations toexpression patterns in normal hematopoietic subpopulationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 102 no 52 pp 19069ndash19074 2005

[70] A Kohlmann C Schoch M Dugas et al ldquoNew insights intoMLL gene rearranged acute leukemias using gene expressionprofiling shared pathways lineage commitment and partnergenesrdquo Leukemia vol 19 no 6 pp 953ndash964 2005

[71] M Andreeff V Ruvolo S Gadgil et al ldquoHOX expressionpatterns identify a common signature for favorable AMLrdquoLeukemia vol 22 no 11 pp 2041ndash2047 2008

[72] M D Radmacher ldquoIndependent confirmation of a prognosticgene-expression signature in adult acutemyeloid leukemia witha normal karyotype a Cancer and Leukemia Group B studyrdquoBlood vol 108 no 5 pp 1677ndash1683 2006

[73] M Heuser L U Wingen D Steinemann et al ldquoGene-expression profiles and their association with drug resistancein adult acutemyeloid leukemiardquoHaematologica vol 90 no 11pp 1484ndash1492 2005

[74] E Tagliafico E Tenedini R Manfredini et al ldquoIdentification ofa molecular signature predictive of sensitivity to differentiationinduction in acutemyeloid leukemiardquo Leukemia vol 20 no 10pp 1751ndash1758 2006

[75] C S Wilson ldquoGene expression profiling of adult acute myeloidleukemia identifies novel biologic clusters for risk classificationand outcome predictionrdquo Blood vol 108 no 2 pp 685ndash6962006

[76] K H Metzeler M Hummel C D Bloomfield et al ldquoAn86-probe-set gene-expression signature predicts survival incytogenetically normal acutemyeloid leukemiardquo Blood vol 112no 10 pp 4193ndash4201 2008

[77] M R Luskin and D J DeAngelo ldquoMidostaurinPKC412 forthe treatment of newly diagnosed FLT3mutation-positive acutemyeloid leukemiardquo Expert Review of Hematology vol 10 no 12pp 1033ndash1045 2017

[78] F Stolzel C Steudel U Oelschlagel et al ldquoMechanisms of resis-tance against PKC412 in resistant FLT3-ITD positive humanacutemyeloid leukemia cellsrdquoAnnals of Hematology vol 89 no7 pp 653ndash662 2010

[79] S Tavor M Eisenbach J Jacob-Hirsch et al ldquoThe CXCR4antagonist AMD3100 impairs survival of human AML cells andinduces their differentiationrdquo Leukemia vol 22 no 12 pp 2151ndash2158 2008

[80] EM Stein G Garcia-ManeroDA Rizzieri et al ldquoTheDOT1Linhibitor pinometostat reduces H3K79 methylation and hasmodest clinical activity in adult acute leukemiardquo Blood vol 131no 24 pp 2661ndash2669 2018

[81] J E Cortes F H Heidel A Hellmann et al ldquoRandomizedcomparison of low dose cytarabine with or without glasdegibin patients with newly diagnosed acute myeloid leukemia orhigh-risk myelodysplastic syndromerdquo Leukemia vol 33 no 2pp 379ndash389 2019

[82] R Masetti S N Bertuccio A Astolfi et al ldquoHhGli antagonistin acute myeloid leukemia with CBFA2T3-GLIS2 fusion generdquoJournal of Hematology amp Oncology vol 10 no 1 2017

[83] K C S Queiroz R R Ruela-De-Sousa G M Fuhler et alldquoHedgehog signaling maintains chemoresistance in myeloidleukemic cellsrdquo Oncogene vol 29 no 48 pp 6314ndash6322 2010

[84] S M Hoy ldquoGlasdegib first global approvalrdquoDrugs vol 79 no2 pp 207ndash213 2019

[85] B V Balgobind M M Van den Heuvel-Eibrink R X DeMenezes et al ldquoEvaluation of gene expression signatures pre-dictive of cytogenetic andmolecular subtypes of pediatric acutemyeloid leukemiardquo Haematologica vol 96 no 2 pp 221ndash2302011

[86] T Yagi ldquoIdentification of a gene expression signature associatedwith pediatric AML prognosisrdquo Blood vol 102 no 5 pp 1849ndash1856 2003

[87] A V Rao P J Valk K H Metzeler et al ldquoAge-specificdifferences in oncogenic pathway dysregulation in patients withacute myeloid leukemiardquo Journal of Clinical Oncology vol 27no 33 pp 5580ndash5586 2009

[88] H J de Jonge E S de Bont P J Valk et al ldquoAML at older ageage-related gene expression profiles reveal a paradoxical down-regulation of p16INK4A mRNA with prognostic significancerdquoBlood vol 114 no 14 pp 2869ndash2877 2009

[89] H J de Jonge C M Woolthuis E S de Bont and G HulsldquoParadoxical down-regulation of p16mRNAwith advancing agein acute myeloid leukemiardquo Aging vol 1 no 11 pp 949ndash9532009

[90] T N Tanaka and R Bejar ldquoMDS overlap disorders anddiagnostic boundariesrdquo Blood vol 133 no 10 pp 1086ndash10952019

[91] A Miyazato ldquoIdentification of myelodysplastic syndrome-specific genes by DNA microarray analysis with purifiedhematopoietic stem cell fractionrdquo Blood vol 98 no 2 pp 422ndash427 2001

[92] C Tsutsumi M Ueda Y Miyazaki et al ldquoDNA microar-ray analysis of dysplastic morphology associated with acutemyeloid leukemiardquo Experimental Hematology vol 32 no 9 pp828ndash835 2004

[93] M S Shafat B Gnaneswaran K M Bowles and S ARushworth ldquoThe bone marrow microenvironment ndash Home ofthe leukemic blastsrdquo Blood Reviews vol 31 no 5 pp 277ndash2862017

[94] A Kode J S Manavalan I Mosialou et al ldquoLeukaemogenesisinduced by an activating 120573-catenin mutation in osteoblastsrdquoNature vol 506 no 7487 pp 240ndash244 2014

[95] V Vas K Senger K Dorr A Niebel H Geiger and WWagner ldquoAging of themicroenvironment influences clonality inhematopoiesisrdquo Plos One vol 7 no 8 Article ID e42080 2012

[96] B Kumar M Garcia L Weng et al ldquoAcute myeloid leukemiatransforms the bone marrow niche into a leukemia-permissivemicroenvironment through exosome secretionrdquo Leukemia vol32 no 3 pp 575ndash587 2018

[97] S Yehudai-Resheff S Attias-Turgeman R Sabbah et alldquoAbnormal morphological and functional nature of bone mar-row stromal cells provides preferential support for survival ofacute myeloid leukemia cellsrdquo International Journal of Cancervol 144 no 9 pp 2279ndash2289 2019

[98] Y Chang D Hernandez S Alonso et al ldquoRole of CYP3A4in bone marrow microenvironmentndashmediated protectionof FLT3ITD AML from tyrosine kinase inhibitorsrdquo BloodAdvances vol 3 no 6 pp 908ndash916 2019

[99] D Passaro A Di Tullio A Abarrategi et al ldquoIncreasedvascular permeability in the bone marrow microenvironmentcontributes to disease progression and drug response in acutemyeloid leukemiardquo Cancer Cell vol 32 no 3 pp 324ndash341e62017

[100] R Le Dieu D C Taussig A G Ramsay et al ldquoPeripheral bloodT cells in acute myeloid leukemia (AML) patients at diagnosis

Journal of Oncology 33

have abnormal phenotype and genotype and form defectiveimmune synapses with AML blastsrdquo Blood vol 114 no 18 pp3909ndash3916 2009

[101] R C Lee R L Feinbaum and V Ambros ldquoThe C elegansheterochronic gene lin-4 encodes small RNAs with antisensecomplementarity to lin-14rdquoCell vol 75 no 5 pp 843ndash854 1993

[102] A G Matera R M Terns andM P Terns ldquoNon-coding RNAslessons from the small nuclear and small nucleolar RNAsrdquoNature Reviews Molecular Cell Biology vol 8 no 3 pp 209ndash220 2007

[103] K V Morris and J S Mattick ldquoThe rise of regulatory RNArdquoNature Reviews Genetics vol 15 no 6 pp 423ndash437 2014

[104] K W Ng C Anderson E A Marshall et al ldquoPiwi-interactingRNAs in cancer emerging functions and clinical utilityrdquoMolec-ular Cancer vol 15 no 1 2016

[105] S M Hammond ldquoAn overview of microRNAsrdquoAdvancedDrugDelivery Reviews vol 87 pp 3ndash14 2015

[106] E A Miska ldquoHow microRNAs control cell division differenti-ation and deathrdquo Current Opinion in Genetics amp Developmentvol 15 no 5 pp 563ndash568 2005

[107] B D Harfe ldquoMicroRNAs in vertebrate developmentrdquo CurrentOpinion in Genetics amp Development vol 15 no 4 pp 410ndash4152005

[108] C Chen ldquoMicroRNAs as oncogenes and tumor suppressorsrdquoThe New England Journal of Medicine vol 353 no 17 pp 1768ndash1771 2005

[109] A Esquela-Kerscher and F J Slack ldquoOncomirsmdashmicroRNAswith a role in cancerrdquo Nature Reviews Cancer vol 6 no 4 pp259ndash269 2006

[110] H Seitz ldquoIssues in current microRNA target identificationmethodsrdquo RNA Biology vol 14 no 7 pp 831ndash834 2017

[111] G A Calin C D Dumitru M Shimizu et al ldquoFrequentdeletions and down-regulation of micro-RNA genes miR15 andmiR16 at 13q14 in chronic lymphocytic leukemiardquoProceedings ofthe National Acadamyof Sciences of the United States of Americavol 99 no 24 pp 15524ndash15529 2002

[112] S A Melo and M Esteller ldquoDysregulation of microRNAs incancer playingwith firerdquo FEBS Letters vol 585 no 13 pp 2087ndash2099 2011

[113] C Chen L Li H F Lodish and D P Bartel ldquoMicroRNAsmodulate hematopoietic lineage differentiationrdquo Science vol303 no 5654 pp 83ndash86 2004

[114] C-Z Chen and H F Lodish ldquoMicroRNAs as regulators ofmammalian hematopoiesisrdquo Seminars in Immunology vol 17no 2 pp 155ndash165 2005

[115] R W Georgantas III R Hildreth S Morisot et al ldquoCD34+hematopoietic stem-progenitor cell microRNA expression andfunction a circuit diagram of differentiation controlrdquo Proceed-ings of the National Acadamy of Sciences of the United States ofAmerica vol 104 no 8 pp 2750ndash2755 2007

[116] N Felli L Fontana E Pelosi et al ldquoMicroRNAs 221 and 222inhibit normal erythropoiesis and erythroleukemic cell growthvia kit receptor down-modulationrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 102 no50 pp 18081ndash18086 2005

[117] F Fazi A Rosa A Fatica et al ldquoA minicircuitry com-prised of microRNA-223 and transcription factors NFI-A andCEBPalpha regulates human granulopoiesisrdquo Cell vol 123 no5 pp 819ndash831 2005

[118] U Bissels SWild S Tomiuk et al ldquoCombined characterizationof microRNA and mRNA profiles delineates early differentia-tion pathways of CD133 + and CD34 + hematopoietic stem and

progenitor cells miRNAs in CD133 + cellsrdquo Stem Cells vol 29no 5 pp 847ndash857 2011

[119] Y Wang Z Li C He et al ldquoMicroRNAs expression signaturesare associated with lineage and survival in acute leukemiasrdquoBlood Cells Molecules and Diseases vol 44 no 3 pp 191ndash1972010

[120] R Garzon F Pichiorri T Palumbo et al ldquoMicroRNA geneexpression during retinoic acid-induced differentiation ofhuman acute promyelocytic leukemiardquo Oncogene vol 26 no28 pp 4148ndash4157 2007

[121] R Garzon S Volinia C G Liu et al ldquoMicroRNA signaturesassociated with cytogenetics and prognosis in acute myeloidleukemiardquo Blood vol 111 no 6 pp 3183ndash3189 2008

[122] S Debernardi S Skoulakis G Molloy T Chaplin A Dixon-McIver and B D Young ldquoMicroRNAmiR-181a correlates withmorphological sub-class of acute myeloid leukaemia and theexpression of its target genes in global genome-wide analysisrdquoLeukemia vol 21 no 5 pp 912ndash916 2007

[123] VHavelange N Stauffer C CHeaphy et al ldquoFunctional impli-cations of microRNAs in acutemyeloid leukemia by integratingmicroRNA and messenger RNA expression profilingrdquo Cancervol 117 no 20 pp 4696ndash4706 2011

[124] A Dixon-McIver P East C AMein et al ldquoDistinctive patternsof microRNA expression associated with karyotype in acutemyeloid leukaemiardquo Plos One vol 3 no 5 Article ID e21412008

[125] M Jongen-Lavrencic S M Sun M K Dijkstra P J M Valkand B Lowenberg ldquoMicroRNA expression profiling in relationto the genetic heterogeneity of acute myeloid leukemiardquo Bloodvol 111 no 10 pp 5078ndash5085 2008

[126] Z Li J Lu and M Sun ldquoDistinct microRNA expression pro-files in acute myeloid leukemia with common translocationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 105 no 40 pp 15535ndash15540 2008

[127] F Wang X Wang G Yang et al ldquomiR-29a and miR-142-3pdownregulation and diagnostic implication in human acutemyeloid leukemiardquoMolecular Biology Reports vol 39 no 3 pp2713ndash2722 2012

[128] G Marcucci M D Radmacher K Maharry et al ldquoMicroRNAexpression in cytogenetically normal acute myeloid leukemiardquoThe New England Journal of Medicine vol 358 no 18 pp 1919ndash1928 2008

[129] B J Wouters M A Jorda K Keeshan et al ldquoDistinct geneexpression profiles of acutemyeloidT-lymphoid leukemia withsilencedCEBPA andmutations inNOTCH1rdquoBlood vol 110 no10 pp 3706ndash3714 2007

[130] M E Figueroa B JWouters L Skrabanek et al ldquoGenome-wideepigenetic analysis delineates a biologically distinct immatureacute leukemia with myeloidT-lymphoid featuresrdquo Blood vol113 no 12 pp 2795ndash2804 2009

[131] S Grisendi R BernardiM Rossi et al ldquoRole of nucleophosminin embryonic development and tumorigenesisrdquoNature vol 437no 7055 pp 147ndash153 2005

[132] B Falini I Nicoletti M F Martelli and C Mecucci ldquoAcutemyeloid leukemia carrying cytoplasmicmutated nucleophos-min (NPMc+ AML) biologic and clinical featuresrdquo Blood vol109 no 3 pp 874ndash885 2006

[133] B Falini ldquoAcute myeloid leukemia with mutated nucleophos-min (NPM1) Molecular pathological and clinical featuresrdquoCancer Treatment and Research vol 145 pp 149ndash168 2010

34 Journal of Oncology

[134] M Alcalay ldquoAcute myeloid leukemia bearing cytoplasmicnucleophosmin (NPMc+ AML) shows a distinct gene expres-sion profile characterized by up-regulation of genes involvedin stem-cell maintenancerdquo Blood vol 106 no 3 pp 899ndash9022005

[135] L Handschuh P Wojciechowski M Kazmierczak et alldquoNPM1 alternative transcripts are upregulated in acute myeloidand lymphoblastic leukemia and their expression level affectspatient outcomerdquo Journal of Translational Medicine vol 16 no1 2018

[136] R Garzon M Garofalo M P Martelli et al ldquoDistinctivemicroRNA signature of acute myeloid leukemia bearing cyto-plasmic mutated nucleophosminrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 105 no10 pp 3945ndash3950 2008

[137] R GW Verhaak C S Goudswaard andW van Putten ldquoMuta-tions in nucleophosmin (NPM1) in acute myeloid leukemia(AML) association with other gene abnormalities and previ-ously established gene expression signatures and their favorableprognostic significancerdquo Blood vol 106 no 12 pp 3747ndash37542005

[138] O M Dovey J L Cooper A Mupo et al ldquo Molecular synergyunderlies the co-occurrence patterns and phenotype ofNPM1 -mutant acutemyeloid leukemia rdquoBlood vol 130 no 17 pp 1911ndash1922 2017

[139] M W Kuhn E Song Z Feng et al ldquoTargeting chromatinregulators inhibits leukemogenic gene expression in NPM1mutant leukemiardquoCancer Discovery vol 6 no 10 pp 1166ndash11812016

[140] A V Krivtsov and S A Armstrong ldquoMLL translocationshistone modifications and leukaemia stem-cell developmentrdquoNature Reviews Cancer vol 7 no 11 pp 823ndash833 2007

[141] L Huang K Zhou Y Yang et al ldquoFLT3-ITD-associated gene-expression signatures inNPM1-mutated cytogenetically normalacute myeloid leukemiardquo International Journal of Hematologyvol 96 no 2 pp 234ndash240 2012

[142] G Zhu Y Yang Y Zhang et al ldquoHigh expression of AHSPEPB42 GYPC and HEMGN predicts favorable prognosis inFLT3-ITD-negative acute myeloid leukemiardquo Cellular Physiol-ogy and Biochemistry vol 42 no 5 pp 1973ndash1984 2017

[143] J Wellbrock E Latuske J Kohler et al ldquoExpression of hedge-hog pathway mediator gli represents a negative prognosticmarker in human acute myeloid leukemia and its inhibitionexerts antileukemic effectsrdquoClinical Cancer Research vol 21 no10 pp 2388ndash2398 2015

[144] G Marcucci K Maharry Y-Z Wu et al ldquoIDH1 and IDH2genemutations identify novelmolecular subsets within de novocytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 28no 14 pp 2348ndash2355 2010

[145] V I Gaidzik L Bullinger R F Schlenk et al ldquoRUNX1 muta-tions in acute myeloid leukemia results from a comprehensivegenetic and clinical analysis from theAMLstudy grouprdquo Journalof Clinical Oncology vol 29 no 10 pp 1364ndash1372 2011

[146] S Schnittger F Dicker W Kern et al ldquoRUNX1 mutations arefrequent in de novo AML with noncomplex karyotype andconfer an unfavorable prognosis rdquo Blood vol 117 no 8 pp2348ndash2357 2011

[147] J H Mendler K Maharry M D Radmacher et al ldquoRUNX1mutations are associated with poor outcome in younger andolder patients with cytogenetically normal acute myeloidleukemia and with distinct gene and microRNA expression

signaturesrdquo Journal of Clinical Oncology vol 30 no 25 pp3109ndash3118 2012

[148] C Langer M D Radmacher A S Ruppert et al ldquoHigh BAALCexpression associates with other molecular prognostic markerspoor outcome and a distinct gene-expression signature incytogenetically normal patients younger than 60 years withacute myeloid leukemia a Cancer and Leukemia Group B(CALGB) studyrdquo Blood vol 111 no 11 pp 5371ndash5379 2008

[149] C Langer G Marcucci K B Holland et al ldquoPrognosticimportance ofMN1 transcript levels and biologic insights fromMN1-associated gene and microRNA expression signatures incytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 27no 19 pp 3198ndash3204 2009

[150] K H Metzeler A Dufour T Benthaus et al ldquoERG expressionis an independent prognostic factor and allows refined riskstratification in cytogenetically normal acutemyeloid leukemiaa comprehensive analysis of ERG MN1 and BAALC transcriptlevels using oligonucleotide microarraysrdquo Journal of ClinicalOncology vol 27 no 30 pp 5031ndash5038 2009

[151] B G Miller J A Stamatoyannopoulos and C CreightonldquoIntegrative meta-analysis of differential gene expression inacute myeloid leukemiardquo Plos One vol 5 no 3 Article IDe9466 2010

[152] F J Staal G Cario G Cazzaniga et al ldquoConsensus guidelinesfor microarray gene expression analyses in leukemia from threeEuropean leukemia networksrdquo Leukemia vol 20 no 8 pp1385ndash1392 2006

[153] M Park S Cho K Yoo et al ldquoGene expression profile related toprognosis of acute myeloid leukemiardquo Oncology Reports 2007

[154] L Handschuh M Kazmierczak M Milewski et al ldquoGeneexpression profiling of acute myeloid leukemia samples fromadult patients with AML-M1 and -M2 through boutiquemicroarrays real-time PCR and droplet digital PCRrdquo Interna-tional Journal of Oncology 2017

[155] S Lee J Chen G Zhou et al ldquoGene expression profiles in acutemyeloid leukemia with common translocations using SAGErdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 103 no 4 pp 1030ndash1035 2006

[156] J Lee J Hwang H Kim et al ldquoA comparison of gene expressionprofiles between primary human AML cells and AML cell linerdquoGenes amp Genetic Systems vol 83 no 4 pp 339ndash345 2008

[157] H Wang H Hu Q Zhang et al ldquoDynamic transcriptomes ofhuman myeloid leukemia cellsrdquo Genomics vol 102 no 4 pp250ndash256 2013

[158] G Gosse M Celton V Lamontagne A Forest and B TWilhelm ldquoWhole genome and transcriptome analysis of a novelAML cell linewith a normal karyotyperdquo Leukemia Research vol39 no 7 pp 709ndash718 2015

[159] C Hsu C Nguyen C Yan et al ldquoTranscriptome profiling ofpediatric core binding factor AMLrdquo Plos One vol 10 no 9Article ID e0138782 2015

[160] A A Singh A Mandoli K H Prange M Laakso and J HMartens ldquoAML associated oncofusion proteins PML-RARAAML1-ETO and CBFB-MYH11 target RUNXETS-factor bind-ing sites to modulate H3ac levels and drive leukemogenesisrdquoOncotarget vol 8 no 8 2017

[161] A Eisfeld J Kohlschmidt K Mrozek et al ldquoMutationallandscape and gene expression patterns in adult acute myeloidleukemias with monosomy 7 as a sole abnormalityrdquo CancerResearch vol 77 no 1 pp 207ndash218 2017

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

Journal of Oncology 31

residual disease in acute leukemiardquo Blood vol 84 no 9 pp3071ndash3079 1994

[41] M Heuser G Beutel J Krauter et al ldquoHigh meningioma 1(MN1) expression as a predictor for poor outcome in acutemyeloid leukemiawith normal cytogeneticsrdquoBlood vol 108 no12 pp 3898ndash3905 2006

[42] C D Baldus ldquoBAALC expression predicts clinical outcomeof de novo acute myeloid leukemia patients with normalcytogenetics a Cancer and Leukemia Group B Studyrdquo Bloodvol 102 no 5 pp 1613ndash1618 2003

[43] G Marcucci C D Baldus A S Ruppert et al ldquoOverexpressionof the ETS-related gene ERG predicts a worse outcome inacute myeloid leukemia with normal karyotype a Cancer andLeukemia Group B studyrdquo Journal of Clinical Oncology vol 23no 36 pp 9234ndash9242 2005

[44] S Barjesteh vanWaalwijk vanDoorn-Khosrovani C ErpelinckW L J van Putten et al ldquoHigh EVI1 expression predicts poorsurvival in acutemyeloid leukemia a study of 319 de novo AMLpatientsrdquo Blood vol 101 no 3 pp 837ndash845 2003

[45] R Bumgarner ldquoOverview of DNA microarrays types applica-tions and their futurerdquo Current Protocols in Molecular Biology2013 Chapter 22Unit 221

[46] M Schena D Shalon R W Davis and P O Brown ldquoQuantita-tive monitoring of gene expression patterns with a complemen-tary DNA microarrayrdquo Science vol 270 no 5235 pp 467ndash4701995

[47] M Schena D Shalon R Heller A Chai P O Brown andR W Davis ldquoParallel human genome analysis microarray-based expression monitoring of 1000 genesrdquo Proceedings of theNational Acadamy of Sciences of the United States of Americavol 93 no 20 pp 10614ndash10619 1996

[48] A Alizadeh M Eisen R Davis et al ldquoThe lymphochip aspecialized cdna microarray for the genomic-scale analysis ofgene expression in normal and malignant lymphocytesrdquo ColdSpring Harbor Symposium on Quantitative Biology vol 64 pp71ndash78 1999

[49] K Virtaneva F A Wright S M Tanner et al ldquoExpressionprofiling reveals fundamental biological differences in acutemyeloid leukemia with isolated trisomy 8 and normal cytoge-neticsrdquo Proceedings of the National Acadamy of Sciences of theUnited States of America vol 98 no 3 pp 1124ndash1129 2001

[50] C Schoch A Kohlmann M Dugas et al ldquoGenomic gainsand losses influence expression levels of genes located withinthe affected regions a study on acute myeloid leukemias withtrisomy 8 11 or 13 monosomy 7 or deletion 5qrdquo Leukemia vol19 no 7 pp 1224ndash1228 2005

[51] J Lu G Getz E AMiska et al ldquoMicroRNA expression profilesclassify human cancersrdquoNature vol 435 no 7043 pp 834ndash8382005

[52] B Falini C Mecucci E Tiacci et al ldquoCytoplasmic nucleophos-min in acute myelogenous leukemia with a normal karyotyperdquoThe New England Journal of Medicine vol 352 no 3 pp 254ndash266 2005

[53] T Haferlach ldquoGlobal approach to the diagnosis of leukemiausing gene expression profilingrdquo Blood vol 106 no 4 pp 1189ndash1198 2005

[54] S Mi J Lu M Sun et al ldquoMicroRNA expression signaturesaccurately discriminate acute lymphoblastic leukemia fromacute myeloid leukemiardquo Proceedings of the National Acadamyof Sciences of the United States of America vol 104 no 50 pp19971ndash19976 2007

[55] P W Janes N Saha W A Barton et al ldquoAdam meets Eph anADAMsubstrate recognitionmodule acts as amolecular switchfor ephrin cleavage in transrdquo Cell vol 123 no 2 pp 291ndash3042005

[56] C Schoch A Kohlmann S Schnittger et al ldquoAcute myeloidleukemias with reciprocal rearrangements can be distinguishedby specific gene expression profilesrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 99 no15 pp 10008ndash10013 2002

[57] S Debernardi D M Lillington T Chaplin et al ldquoGenome-wide analysis of acutemyeloid leukemia with normal karyotypereveals a unique pattern of homeobox gene expression distinctfrom those with translocation-mediated fusion eventsrdquo GenesChromosomes and Cancer vol 37 no 2 pp 149ndash158 2003

[58] R G Verhaak B J Wouters C A Erpelinck et al ldquoPredictionof molecular subtypes in acutemyeloid leukemia based on geneexpression profilingrdquo Haematologica vol 94 no 1 pp 131ndash1342009

[59] RW Georgantas III V Tanadve M Malehorn et al ldquoMicroar-ray and serial analysis of gene expression analyses identifyknown and novel transcripts overexpressed in hematopoieticstem cellsrdquo Cancer Research vol 64 no 13 pp 4434ndash44412004

[60] L Bullinger K Dohner E Bair et al ldquoUse of gene-expressionprofiling to identify prognostic subclasses in adult acutemyeloid leukemiardquo The New England Journal of Medicine vol350 no 16 pp 1605ndash1616 2004

[61] P J Valk R G Verhaak M A Beijen et al ldquoPrognosticallyuseful gene-expression profiles in acute myeloid leukemiardquoTheNew England Journal of Medicine vol 350 no 16 pp 1617ndash16282004

[62] N C Gutierrez R Lopez-Perez J M Hernandez et al ldquoGeneexpression profile reveals deregulation of genes with relevantfunctions in the different subclasses of acutemyeloid leukemiardquoLeukemia vol 19 no 3 pp 402ndash409 2005

[63] J E Payton N R Grieselhuber L Chang et al ldquoHigh through-put digital quantification of mRNA abundance in primaryhuman acutemyeloid leukemia samplesrdquoThe Journal of ClinicalInvestigation vol 119 no 6 pp 1714ndash1726 2009

[64] T Haferlach A Kohlmann S Schnittger et al ldquoAML M3 andAMLM3 variant each have a distinct gene expression signaturebut also share patterns different from other genetically definedAML subtypesrdquoGenes Chromosomes and Cancer vol 43 no 2pp 113ndash127 2005

[65] T Haferlach A Kohlmann L Wieczorek et al ldquoClinical utilityof microarray-based gene expression profiling in the diagnosisand subclassification of leukemia report from the internationalmicroarray innovations in leukemia study grouprdquo Journal ofClinical Oncology vol 28 no 15 pp 2529ndash2537 2010

[66] D R de la Bletiere O Blanchet P Cornillet-Lefebvre et alldquoRoutine use of microarray-based gene expression profilingto identify patients with low cytogenetic risk acute myeloidleukemia accurate results can be obtained even with subopti-mal samplesrdquo BMC Medical Genomics vol 5 no 1 2012

[67] S A Armstrong J E Staunton L B Silverman et al ldquoMLLtranslocations specify a distinct gene expression profile thatdistinguishes a unique leukemiardquo Nature Genetics vol 30 no1 pp 41ndash47 2002

[68] M E Ross ldquoGene expression profiling of pediatric acutemyelogenous leukemiardquo Blood vol 104 no 12 pp 3679ndash36872004

32 Journal of Oncology

[69] A Andersson T Olofsson D Lindgren et al ldquoMolecularsignatures in childhood acute leukemia and their correlations toexpression patterns in normal hematopoietic subpopulationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 102 no 52 pp 19069ndash19074 2005

[70] A Kohlmann C Schoch M Dugas et al ldquoNew insights intoMLL gene rearranged acute leukemias using gene expressionprofiling shared pathways lineage commitment and partnergenesrdquo Leukemia vol 19 no 6 pp 953ndash964 2005

[71] M Andreeff V Ruvolo S Gadgil et al ldquoHOX expressionpatterns identify a common signature for favorable AMLrdquoLeukemia vol 22 no 11 pp 2041ndash2047 2008

[72] M D Radmacher ldquoIndependent confirmation of a prognosticgene-expression signature in adult acutemyeloid leukemia witha normal karyotype a Cancer and Leukemia Group B studyrdquoBlood vol 108 no 5 pp 1677ndash1683 2006

[73] M Heuser L U Wingen D Steinemann et al ldquoGene-expression profiles and their association with drug resistancein adult acutemyeloid leukemiardquoHaematologica vol 90 no 11pp 1484ndash1492 2005

[74] E Tagliafico E Tenedini R Manfredini et al ldquoIdentification ofa molecular signature predictive of sensitivity to differentiationinduction in acutemyeloid leukemiardquo Leukemia vol 20 no 10pp 1751ndash1758 2006

[75] C S Wilson ldquoGene expression profiling of adult acute myeloidleukemia identifies novel biologic clusters for risk classificationand outcome predictionrdquo Blood vol 108 no 2 pp 685ndash6962006

[76] K H Metzeler M Hummel C D Bloomfield et al ldquoAn86-probe-set gene-expression signature predicts survival incytogenetically normal acutemyeloid leukemiardquo Blood vol 112no 10 pp 4193ndash4201 2008

[77] M R Luskin and D J DeAngelo ldquoMidostaurinPKC412 forthe treatment of newly diagnosed FLT3mutation-positive acutemyeloid leukemiardquo Expert Review of Hematology vol 10 no 12pp 1033ndash1045 2017

[78] F Stolzel C Steudel U Oelschlagel et al ldquoMechanisms of resis-tance against PKC412 in resistant FLT3-ITD positive humanacutemyeloid leukemia cellsrdquoAnnals of Hematology vol 89 no7 pp 653ndash662 2010

[79] S Tavor M Eisenbach J Jacob-Hirsch et al ldquoThe CXCR4antagonist AMD3100 impairs survival of human AML cells andinduces their differentiationrdquo Leukemia vol 22 no 12 pp 2151ndash2158 2008

[80] EM Stein G Garcia-ManeroDA Rizzieri et al ldquoTheDOT1Linhibitor pinometostat reduces H3K79 methylation and hasmodest clinical activity in adult acute leukemiardquo Blood vol 131no 24 pp 2661ndash2669 2018

[81] J E Cortes F H Heidel A Hellmann et al ldquoRandomizedcomparison of low dose cytarabine with or without glasdegibin patients with newly diagnosed acute myeloid leukemia orhigh-risk myelodysplastic syndromerdquo Leukemia vol 33 no 2pp 379ndash389 2019

[82] R Masetti S N Bertuccio A Astolfi et al ldquoHhGli antagonistin acute myeloid leukemia with CBFA2T3-GLIS2 fusion generdquoJournal of Hematology amp Oncology vol 10 no 1 2017

[83] K C S Queiroz R R Ruela-De-Sousa G M Fuhler et alldquoHedgehog signaling maintains chemoresistance in myeloidleukemic cellsrdquo Oncogene vol 29 no 48 pp 6314ndash6322 2010

[84] S M Hoy ldquoGlasdegib first global approvalrdquoDrugs vol 79 no2 pp 207ndash213 2019

[85] B V Balgobind M M Van den Heuvel-Eibrink R X DeMenezes et al ldquoEvaluation of gene expression signatures pre-dictive of cytogenetic andmolecular subtypes of pediatric acutemyeloid leukemiardquo Haematologica vol 96 no 2 pp 221ndash2302011

[86] T Yagi ldquoIdentification of a gene expression signature associatedwith pediatric AML prognosisrdquo Blood vol 102 no 5 pp 1849ndash1856 2003

[87] A V Rao P J Valk K H Metzeler et al ldquoAge-specificdifferences in oncogenic pathway dysregulation in patients withacute myeloid leukemiardquo Journal of Clinical Oncology vol 27no 33 pp 5580ndash5586 2009

[88] H J de Jonge E S de Bont P J Valk et al ldquoAML at older ageage-related gene expression profiles reveal a paradoxical down-regulation of p16INK4A mRNA with prognostic significancerdquoBlood vol 114 no 14 pp 2869ndash2877 2009

[89] H J de Jonge C M Woolthuis E S de Bont and G HulsldquoParadoxical down-regulation of p16mRNAwith advancing agein acute myeloid leukemiardquo Aging vol 1 no 11 pp 949ndash9532009

[90] T N Tanaka and R Bejar ldquoMDS overlap disorders anddiagnostic boundariesrdquo Blood vol 133 no 10 pp 1086ndash10952019

[91] A Miyazato ldquoIdentification of myelodysplastic syndrome-specific genes by DNA microarray analysis with purifiedhematopoietic stem cell fractionrdquo Blood vol 98 no 2 pp 422ndash427 2001

[92] C Tsutsumi M Ueda Y Miyazaki et al ldquoDNA microar-ray analysis of dysplastic morphology associated with acutemyeloid leukemiardquo Experimental Hematology vol 32 no 9 pp828ndash835 2004

[93] M S Shafat B Gnaneswaran K M Bowles and S ARushworth ldquoThe bone marrow microenvironment ndash Home ofthe leukemic blastsrdquo Blood Reviews vol 31 no 5 pp 277ndash2862017

[94] A Kode J S Manavalan I Mosialou et al ldquoLeukaemogenesisinduced by an activating 120573-catenin mutation in osteoblastsrdquoNature vol 506 no 7487 pp 240ndash244 2014

[95] V Vas K Senger K Dorr A Niebel H Geiger and WWagner ldquoAging of themicroenvironment influences clonality inhematopoiesisrdquo Plos One vol 7 no 8 Article ID e42080 2012

[96] B Kumar M Garcia L Weng et al ldquoAcute myeloid leukemiatransforms the bone marrow niche into a leukemia-permissivemicroenvironment through exosome secretionrdquo Leukemia vol32 no 3 pp 575ndash587 2018

[97] S Yehudai-Resheff S Attias-Turgeman R Sabbah et alldquoAbnormal morphological and functional nature of bone mar-row stromal cells provides preferential support for survival ofacute myeloid leukemia cellsrdquo International Journal of Cancervol 144 no 9 pp 2279ndash2289 2019

[98] Y Chang D Hernandez S Alonso et al ldquoRole of CYP3A4in bone marrow microenvironmentndashmediated protectionof FLT3ITD AML from tyrosine kinase inhibitorsrdquo BloodAdvances vol 3 no 6 pp 908ndash916 2019

[99] D Passaro A Di Tullio A Abarrategi et al ldquoIncreasedvascular permeability in the bone marrow microenvironmentcontributes to disease progression and drug response in acutemyeloid leukemiardquo Cancer Cell vol 32 no 3 pp 324ndash341e62017

[100] R Le Dieu D C Taussig A G Ramsay et al ldquoPeripheral bloodT cells in acute myeloid leukemia (AML) patients at diagnosis

Journal of Oncology 33

have abnormal phenotype and genotype and form defectiveimmune synapses with AML blastsrdquo Blood vol 114 no 18 pp3909ndash3916 2009

[101] R C Lee R L Feinbaum and V Ambros ldquoThe C elegansheterochronic gene lin-4 encodes small RNAs with antisensecomplementarity to lin-14rdquoCell vol 75 no 5 pp 843ndash854 1993

[102] A G Matera R M Terns andM P Terns ldquoNon-coding RNAslessons from the small nuclear and small nucleolar RNAsrdquoNature Reviews Molecular Cell Biology vol 8 no 3 pp 209ndash220 2007

[103] K V Morris and J S Mattick ldquoThe rise of regulatory RNArdquoNature Reviews Genetics vol 15 no 6 pp 423ndash437 2014

[104] K W Ng C Anderson E A Marshall et al ldquoPiwi-interactingRNAs in cancer emerging functions and clinical utilityrdquoMolec-ular Cancer vol 15 no 1 2016

[105] S M Hammond ldquoAn overview of microRNAsrdquoAdvancedDrugDelivery Reviews vol 87 pp 3ndash14 2015

[106] E A Miska ldquoHow microRNAs control cell division differenti-ation and deathrdquo Current Opinion in Genetics amp Developmentvol 15 no 5 pp 563ndash568 2005

[107] B D Harfe ldquoMicroRNAs in vertebrate developmentrdquo CurrentOpinion in Genetics amp Development vol 15 no 4 pp 410ndash4152005

[108] C Chen ldquoMicroRNAs as oncogenes and tumor suppressorsrdquoThe New England Journal of Medicine vol 353 no 17 pp 1768ndash1771 2005

[109] A Esquela-Kerscher and F J Slack ldquoOncomirsmdashmicroRNAswith a role in cancerrdquo Nature Reviews Cancer vol 6 no 4 pp259ndash269 2006

[110] H Seitz ldquoIssues in current microRNA target identificationmethodsrdquo RNA Biology vol 14 no 7 pp 831ndash834 2017

[111] G A Calin C D Dumitru M Shimizu et al ldquoFrequentdeletions and down-regulation of micro-RNA genes miR15 andmiR16 at 13q14 in chronic lymphocytic leukemiardquoProceedings ofthe National Acadamyof Sciences of the United States of Americavol 99 no 24 pp 15524ndash15529 2002

[112] S A Melo and M Esteller ldquoDysregulation of microRNAs incancer playingwith firerdquo FEBS Letters vol 585 no 13 pp 2087ndash2099 2011

[113] C Chen L Li H F Lodish and D P Bartel ldquoMicroRNAsmodulate hematopoietic lineage differentiationrdquo Science vol303 no 5654 pp 83ndash86 2004

[114] C-Z Chen and H F Lodish ldquoMicroRNAs as regulators ofmammalian hematopoiesisrdquo Seminars in Immunology vol 17no 2 pp 155ndash165 2005

[115] R W Georgantas III R Hildreth S Morisot et al ldquoCD34+hematopoietic stem-progenitor cell microRNA expression andfunction a circuit diagram of differentiation controlrdquo Proceed-ings of the National Acadamy of Sciences of the United States ofAmerica vol 104 no 8 pp 2750ndash2755 2007

[116] N Felli L Fontana E Pelosi et al ldquoMicroRNAs 221 and 222inhibit normal erythropoiesis and erythroleukemic cell growthvia kit receptor down-modulationrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 102 no50 pp 18081ndash18086 2005

[117] F Fazi A Rosa A Fatica et al ldquoA minicircuitry com-prised of microRNA-223 and transcription factors NFI-A andCEBPalpha regulates human granulopoiesisrdquo Cell vol 123 no5 pp 819ndash831 2005

[118] U Bissels SWild S Tomiuk et al ldquoCombined characterizationof microRNA and mRNA profiles delineates early differentia-tion pathways of CD133 + and CD34 + hematopoietic stem and

progenitor cells miRNAs in CD133 + cellsrdquo Stem Cells vol 29no 5 pp 847ndash857 2011

[119] Y Wang Z Li C He et al ldquoMicroRNAs expression signaturesare associated with lineage and survival in acute leukemiasrdquoBlood Cells Molecules and Diseases vol 44 no 3 pp 191ndash1972010

[120] R Garzon F Pichiorri T Palumbo et al ldquoMicroRNA geneexpression during retinoic acid-induced differentiation ofhuman acute promyelocytic leukemiardquo Oncogene vol 26 no28 pp 4148ndash4157 2007

[121] R Garzon S Volinia C G Liu et al ldquoMicroRNA signaturesassociated with cytogenetics and prognosis in acute myeloidleukemiardquo Blood vol 111 no 6 pp 3183ndash3189 2008

[122] S Debernardi S Skoulakis G Molloy T Chaplin A Dixon-McIver and B D Young ldquoMicroRNAmiR-181a correlates withmorphological sub-class of acute myeloid leukaemia and theexpression of its target genes in global genome-wide analysisrdquoLeukemia vol 21 no 5 pp 912ndash916 2007

[123] VHavelange N Stauffer C CHeaphy et al ldquoFunctional impli-cations of microRNAs in acutemyeloid leukemia by integratingmicroRNA and messenger RNA expression profilingrdquo Cancervol 117 no 20 pp 4696ndash4706 2011

[124] A Dixon-McIver P East C AMein et al ldquoDistinctive patternsof microRNA expression associated with karyotype in acutemyeloid leukaemiardquo Plos One vol 3 no 5 Article ID e21412008

[125] M Jongen-Lavrencic S M Sun M K Dijkstra P J M Valkand B Lowenberg ldquoMicroRNA expression profiling in relationto the genetic heterogeneity of acute myeloid leukemiardquo Bloodvol 111 no 10 pp 5078ndash5085 2008

[126] Z Li J Lu and M Sun ldquoDistinct microRNA expression pro-files in acute myeloid leukemia with common translocationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 105 no 40 pp 15535ndash15540 2008

[127] F Wang X Wang G Yang et al ldquomiR-29a and miR-142-3pdownregulation and diagnostic implication in human acutemyeloid leukemiardquoMolecular Biology Reports vol 39 no 3 pp2713ndash2722 2012

[128] G Marcucci M D Radmacher K Maharry et al ldquoMicroRNAexpression in cytogenetically normal acute myeloid leukemiardquoThe New England Journal of Medicine vol 358 no 18 pp 1919ndash1928 2008

[129] B J Wouters M A Jorda K Keeshan et al ldquoDistinct geneexpression profiles of acutemyeloidT-lymphoid leukemia withsilencedCEBPA andmutations inNOTCH1rdquoBlood vol 110 no10 pp 3706ndash3714 2007

[130] M E Figueroa B JWouters L Skrabanek et al ldquoGenome-wideepigenetic analysis delineates a biologically distinct immatureacute leukemia with myeloidT-lymphoid featuresrdquo Blood vol113 no 12 pp 2795ndash2804 2009

[131] S Grisendi R BernardiM Rossi et al ldquoRole of nucleophosminin embryonic development and tumorigenesisrdquoNature vol 437no 7055 pp 147ndash153 2005

[132] B Falini I Nicoletti M F Martelli and C Mecucci ldquoAcutemyeloid leukemia carrying cytoplasmicmutated nucleophos-min (NPMc+ AML) biologic and clinical featuresrdquo Blood vol109 no 3 pp 874ndash885 2006

[133] B Falini ldquoAcute myeloid leukemia with mutated nucleophos-min (NPM1) Molecular pathological and clinical featuresrdquoCancer Treatment and Research vol 145 pp 149ndash168 2010

34 Journal of Oncology

[134] M Alcalay ldquoAcute myeloid leukemia bearing cytoplasmicnucleophosmin (NPMc+ AML) shows a distinct gene expres-sion profile characterized by up-regulation of genes involvedin stem-cell maintenancerdquo Blood vol 106 no 3 pp 899ndash9022005

[135] L Handschuh P Wojciechowski M Kazmierczak et alldquoNPM1 alternative transcripts are upregulated in acute myeloidand lymphoblastic leukemia and their expression level affectspatient outcomerdquo Journal of Translational Medicine vol 16 no1 2018

[136] R Garzon M Garofalo M P Martelli et al ldquoDistinctivemicroRNA signature of acute myeloid leukemia bearing cyto-plasmic mutated nucleophosminrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 105 no10 pp 3945ndash3950 2008

[137] R GW Verhaak C S Goudswaard andW van Putten ldquoMuta-tions in nucleophosmin (NPM1) in acute myeloid leukemia(AML) association with other gene abnormalities and previ-ously established gene expression signatures and their favorableprognostic significancerdquo Blood vol 106 no 12 pp 3747ndash37542005

[138] O M Dovey J L Cooper A Mupo et al ldquo Molecular synergyunderlies the co-occurrence patterns and phenotype ofNPM1 -mutant acutemyeloid leukemia rdquoBlood vol 130 no 17 pp 1911ndash1922 2017

[139] M W Kuhn E Song Z Feng et al ldquoTargeting chromatinregulators inhibits leukemogenic gene expression in NPM1mutant leukemiardquoCancer Discovery vol 6 no 10 pp 1166ndash11812016

[140] A V Krivtsov and S A Armstrong ldquoMLL translocationshistone modifications and leukaemia stem-cell developmentrdquoNature Reviews Cancer vol 7 no 11 pp 823ndash833 2007

[141] L Huang K Zhou Y Yang et al ldquoFLT3-ITD-associated gene-expression signatures inNPM1-mutated cytogenetically normalacute myeloid leukemiardquo International Journal of Hematologyvol 96 no 2 pp 234ndash240 2012

[142] G Zhu Y Yang Y Zhang et al ldquoHigh expression of AHSPEPB42 GYPC and HEMGN predicts favorable prognosis inFLT3-ITD-negative acute myeloid leukemiardquo Cellular Physiol-ogy and Biochemistry vol 42 no 5 pp 1973ndash1984 2017

[143] J Wellbrock E Latuske J Kohler et al ldquoExpression of hedge-hog pathway mediator gli represents a negative prognosticmarker in human acute myeloid leukemia and its inhibitionexerts antileukemic effectsrdquoClinical Cancer Research vol 21 no10 pp 2388ndash2398 2015

[144] G Marcucci K Maharry Y-Z Wu et al ldquoIDH1 and IDH2genemutations identify novelmolecular subsets within de novocytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 28no 14 pp 2348ndash2355 2010

[145] V I Gaidzik L Bullinger R F Schlenk et al ldquoRUNX1 muta-tions in acute myeloid leukemia results from a comprehensivegenetic and clinical analysis from theAMLstudy grouprdquo Journalof Clinical Oncology vol 29 no 10 pp 1364ndash1372 2011

[146] S Schnittger F Dicker W Kern et al ldquoRUNX1 mutations arefrequent in de novo AML with noncomplex karyotype andconfer an unfavorable prognosis rdquo Blood vol 117 no 8 pp2348ndash2357 2011

[147] J H Mendler K Maharry M D Radmacher et al ldquoRUNX1mutations are associated with poor outcome in younger andolder patients with cytogenetically normal acute myeloidleukemia and with distinct gene and microRNA expression

signaturesrdquo Journal of Clinical Oncology vol 30 no 25 pp3109ndash3118 2012

[148] C Langer M D Radmacher A S Ruppert et al ldquoHigh BAALCexpression associates with other molecular prognostic markerspoor outcome and a distinct gene-expression signature incytogenetically normal patients younger than 60 years withacute myeloid leukemia a Cancer and Leukemia Group B(CALGB) studyrdquo Blood vol 111 no 11 pp 5371ndash5379 2008

[149] C Langer G Marcucci K B Holland et al ldquoPrognosticimportance ofMN1 transcript levels and biologic insights fromMN1-associated gene and microRNA expression signatures incytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 27no 19 pp 3198ndash3204 2009

[150] K H Metzeler A Dufour T Benthaus et al ldquoERG expressionis an independent prognostic factor and allows refined riskstratification in cytogenetically normal acutemyeloid leukemiaa comprehensive analysis of ERG MN1 and BAALC transcriptlevels using oligonucleotide microarraysrdquo Journal of ClinicalOncology vol 27 no 30 pp 5031ndash5038 2009

[151] B G Miller J A Stamatoyannopoulos and C CreightonldquoIntegrative meta-analysis of differential gene expression inacute myeloid leukemiardquo Plos One vol 5 no 3 Article IDe9466 2010

[152] F J Staal G Cario G Cazzaniga et al ldquoConsensus guidelinesfor microarray gene expression analyses in leukemia from threeEuropean leukemia networksrdquo Leukemia vol 20 no 8 pp1385ndash1392 2006

[153] M Park S Cho K Yoo et al ldquoGene expression profile related toprognosis of acute myeloid leukemiardquo Oncology Reports 2007

[154] L Handschuh M Kazmierczak M Milewski et al ldquoGeneexpression profiling of acute myeloid leukemia samples fromadult patients with AML-M1 and -M2 through boutiquemicroarrays real-time PCR and droplet digital PCRrdquo Interna-tional Journal of Oncology 2017

[155] S Lee J Chen G Zhou et al ldquoGene expression profiles in acutemyeloid leukemia with common translocations using SAGErdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 103 no 4 pp 1030ndash1035 2006

[156] J Lee J Hwang H Kim et al ldquoA comparison of gene expressionprofiles between primary human AML cells and AML cell linerdquoGenes amp Genetic Systems vol 83 no 4 pp 339ndash345 2008

[157] H Wang H Hu Q Zhang et al ldquoDynamic transcriptomes ofhuman myeloid leukemia cellsrdquo Genomics vol 102 no 4 pp250ndash256 2013

[158] G Gosse M Celton V Lamontagne A Forest and B TWilhelm ldquoWhole genome and transcriptome analysis of a novelAML cell linewith a normal karyotyperdquo Leukemia Research vol39 no 7 pp 709ndash718 2015

[159] C Hsu C Nguyen C Yan et al ldquoTranscriptome profiling ofpediatric core binding factor AMLrdquo Plos One vol 10 no 9Article ID e0138782 2015

[160] A A Singh A Mandoli K H Prange M Laakso and J HMartens ldquoAML associated oncofusion proteins PML-RARAAML1-ETO and CBFB-MYH11 target RUNXETS-factor bind-ing sites to modulate H3ac levels and drive leukemogenesisrdquoOncotarget vol 8 no 8 2017

[161] A Eisfeld J Kohlschmidt K Mrozek et al ldquoMutationallandscape and gene expression patterns in adult acute myeloidleukemias with monosomy 7 as a sole abnormalityrdquo CancerResearch vol 77 no 1 pp 207ndash218 2017

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

32 Journal of Oncology

[69] A Andersson T Olofsson D Lindgren et al ldquoMolecularsignatures in childhood acute leukemia and their correlations toexpression patterns in normal hematopoietic subpopulationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 102 no 52 pp 19069ndash19074 2005

[70] A Kohlmann C Schoch M Dugas et al ldquoNew insights intoMLL gene rearranged acute leukemias using gene expressionprofiling shared pathways lineage commitment and partnergenesrdquo Leukemia vol 19 no 6 pp 953ndash964 2005

[71] M Andreeff V Ruvolo S Gadgil et al ldquoHOX expressionpatterns identify a common signature for favorable AMLrdquoLeukemia vol 22 no 11 pp 2041ndash2047 2008

[72] M D Radmacher ldquoIndependent confirmation of a prognosticgene-expression signature in adult acutemyeloid leukemia witha normal karyotype a Cancer and Leukemia Group B studyrdquoBlood vol 108 no 5 pp 1677ndash1683 2006

[73] M Heuser L U Wingen D Steinemann et al ldquoGene-expression profiles and their association with drug resistancein adult acutemyeloid leukemiardquoHaematologica vol 90 no 11pp 1484ndash1492 2005

[74] E Tagliafico E Tenedini R Manfredini et al ldquoIdentification ofa molecular signature predictive of sensitivity to differentiationinduction in acutemyeloid leukemiardquo Leukemia vol 20 no 10pp 1751ndash1758 2006

[75] C S Wilson ldquoGene expression profiling of adult acute myeloidleukemia identifies novel biologic clusters for risk classificationand outcome predictionrdquo Blood vol 108 no 2 pp 685ndash6962006

[76] K H Metzeler M Hummel C D Bloomfield et al ldquoAn86-probe-set gene-expression signature predicts survival incytogenetically normal acutemyeloid leukemiardquo Blood vol 112no 10 pp 4193ndash4201 2008

[77] M R Luskin and D J DeAngelo ldquoMidostaurinPKC412 forthe treatment of newly diagnosed FLT3mutation-positive acutemyeloid leukemiardquo Expert Review of Hematology vol 10 no 12pp 1033ndash1045 2017

[78] F Stolzel C Steudel U Oelschlagel et al ldquoMechanisms of resis-tance against PKC412 in resistant FLT3-ITD positive humanacutemyeloid leukemia cellsrdquoAnnals of Hematology vol 89 no7 pp 653ndash662 2010

[79] S Tavor M Eisenbach J Jacob-Hirsch et al ldquoThe CXCR4antagonist AMD3100 impairs survival of human AML cells andinduces their differentiationrdquo Leukemia vol 22 no 12 pp 2151ndash2158 2008

[80] EM Stein G Garcia-ManeroDA Rizzieri et al ldquoTheDOT1Linhibitor pinometostat reduces H3K79 methylation and hasmodest clinical activity in adult acute leukemiardquo Blood vol 131no 24 pp 2661ndash2669 2018

[81] J E Cortes F H Heidel A Hellmann et al ldquoRandomizedcomparison of low dose cytarabine with or without glasdegibin patients with newly diagnosed acute myeloid leukemia orhigh-risk myelodysplastic syndromerdquo Leukemia vol 33 no 2pp 379ndash389 2019

[82] R Masetti S N Bertuccio A Astolfi et al ldquoHhGli antagonistin acute myeloid leukemia with CBFA2T3-GLIS2 fusion generdquoJournal of Hematology amp Oncology vol 10 no 1 2017

[83] K C S Queiroz R R Ruela-De-Sousa G M Fuhler et alldquoHedgehog signaling maintains chemoresistance in myeloidleukemic cellsrdquo Oncogene vol 29 no 48 pp 6314ndash6322 2010

[84] S M Hoy ldquoGlasdegib first global approvalrdquoDrugs vol 79 no2 pp 207ndash213 2019

[85] B V Balgobind M M Van den Heuvel-Eibrink R X DeMenezes et al ldquoEvaluation of gene expression signatures pre-dictive of cytogenetic andmolecular subtypes of pediatric acutemyeloid leukemiardquo Haematologica vol 96 no 2 pp 221ndash2302011

[86] T Yagi ldquoIdentification of a gene expression signature associatedwith pediatric AML prognosisrdquo Blood vol 102 no 5 pp 1849ndash1856 2003

[87] A V Rao P J Valk K H Metzeler et al ldquoAge-specificdifferences in oncogenic pathway dysregulation in patients withacute myeloid leukemiardquo Journal of Clinical Oncology vol 27no 33 pp 5580ndash5586 2009

[88] H J de Jonge E S de Bont P J Valk et al ldquoAML at older ageage-related gene expression profiles reveal a paradoxical down-regulation of p16INK4A mRNA with prognostic significancerdquoBlood vol 114 no 14 pp 2869ndash2877 2009

[89] H J de Jonge C M Woolthuis E S de Bont and G HulsldquoParadoxical down-regulation of p16mRNAwith advancing agein acute myeloid leukemiardquo Aging vol 1 no 11 pp 949ndash9532009

[90] T N Tanaka and R Bejar ldquoMDS overlap disorders anddiagnostic boundariesrdquo Blood vol 133 no 10 pp 1086ndash10952019

[91] A Miyazato ldquoIdentification of myelodysplastic syndrome-specific genes by DNA microarray analysis with purifiedhematopoietic stem cell fractionrdquo Blood vol 98 no 2 pp 422ndash427 2001

[92] C Tsutsumi M Ueda Y Miyazaki et al ldquoDNA microar-ray analysis of dysplastic morphology associated with acutemyeloid leukemiardquo Experimental Hematology vol 32 no 9 pp828ndash835 2004

[93] M S Shafat B Gnaneswaran K M Bowles and S ARushworth ldquoThe bone marrow microenvironment ndash Home ofthe leukemic blastsrdquo Blood Reviews vol 31 no 5 pp 277ndash2862017

[94] A Kode J S Manavalan I Mosialou et al ldquoLeukaemogenesisinduced by an activating 120573-catenin mutation in osteoblastsrdquoNature vol 506 no 7487 pp 240ndash244 2014

[95] V Vas K Senger K Dorr A Niebel H Geiger and WWagner ldquoAging of themicroenvironment influences clonality inhematopoiesisrdquo Plos One vol 7 no 8 Article ID e42080 2012

[96] B Kumar M Garcia L Weng et al ldquoAcute myeloid leukemiatransforms the bone marrow niche into a leukemia-permissivemicroenvironment through exosome secretionrdquo Leukemia vol32 no 3 pp 575ndash587 2018

[97] S Yehudai-Resheff S Attias-Turgeman R Sabbah et alldquoAbnormal morphological and functional nature of bone mar-row stromal cells provides preferential support for survival ofacute myeloid leukemia cellsrdquo International Journal of Cancervol 144 no 9 pp 2279ndash2289 2019

[98] Y Chang D Hernandez S Alonso et al ldquoRole of CYP3A4in bone marrow microenvironmentndashmediated protectionof FLT3ITD AML from tyrosine kinase inhibitorsrdquo BloodAdvances vol 3 no 6 pp 908ndash916 2019

[99] D Passaro A Di Tullio A Abarrategi et al ldquoIncreasedvascular permeability in the bone marrow microenvironmentcontributes to disease progression and drug response in acutemyeloid leukemiardquo Cancer Cell vol 32 no 3 pp 324ndash341e62017

[100] R Le Dieu D C Taussig A G Ramsay et al ldquoPeripheral bloodT cells in acute myeloid leukemia (AML) patients at diagnosis

Journal of Oncology 33

have abnormal phenotype and genotype and form defectiveimmune synapses with AML blastsrdquo Blood vol 114 no 18 pp3909ndash3916 2009

[101] R C Lee R L Feinbaum and V Ambros ldquoThe C elegansheterochronic gene lin-4 encodes small RNAs with antisensecomplementarity to lin-14rdquoCell vol 75 no 5 pp 843ndash854 1993

[102] A G Matera R M Terns andM P Terns ldquoNon-coding RNAslessons from the small nuclear and small nucleolar RNAsrdquoNature Reviews Molecular Cell Biology vol 8 no 3 pp 209ndash220 2007

[103] K V Morris and J S Mattick ldquoThe rise of regulatory RNArdquoNature Reviews Genetics vol 15 no 6 pp 423ndash437 2014

[104] K W Ng C Anderson E A Marshall et al ldquoPiwi-interactingRNAs in cancer emerging functions and clinical utilityrdquoMolec-ular Cancer vol 15 no 1 2016

[105] S M Hammond ldquoAn overview of microRNAsrdquoAdvancedDrugDelivery Reviews vol 87 pp 3ndash14 2015

[106] E A Miska ldquoHow microRNAs control cell division differenti-ation and deathrdquo Current Opinion in Genetics amp Developmentvol 15 no 5 pp 563ndash568 2005

[107] B D Harfe ldquoMicroRNAs in vertebrate developmentrdquo CurrentOpinion in Genetics amp Development vol 15 no 4 pp 410ndash4152005

[108] C Chen ldquoMicroRNAs as oncogenes and tumor suppressorsrdquoThe New England Journal of Medicine vol 353 no 17 pp 1768ndash1771 2005

[109] A Esquela-Kerscher and F J Slack ldquoOncomirsmdashmicroRNAswith a role in cancerrdquo Nature Reviews Cancer vol 6 no 4 pp259ndash269 2006

[110] H Seitz ldquoIssues in current microRNA target identificationmethodsrdquo RNA Biology vol 14 no 7 pp 831ndash834 2017

[111] G A Calin C D Dumitru M Shimizu et al ldquoFrequentdeletions and down-regulation of micro-RNA genes miR15 andmiR16 at 13q14 in chronic lymphocytic leukemiardquoProceedings ofthe National Acadamyof Sciences of the United States of Americavol 99 no 24 pp 15524ndash15529 2002

[112] S A Melo and M Esteller ldquoDysregulation of microRNAs incancer playingwith firerdquo FEBS Letters vol 585 no 13 pp 2087ndash2099 2011

[113] C Chen L Li H F Lodish and D P Bartel ldquoMicroRNAsmodulate hematopoietic lineage differentiationrdquo Science vol303 no 5654 pp 83ndash86 2004

[114] C-Z Chen and H F Lodish ldquoMicroRNAs as regulators ofmammalian hematopoiesisrdquo Seminars in Immunology vol 17no 2 pp 155ndash165 2005

[115] R W Georgantas III R Hildreth S Morisot et al ldquoCD34+hematopoietic stem-progenitor cell microRNA expression andfunction a circuit diagram of differentiation controlrdquo Proceed-ings of the National Acadamy of Sciences of the United States ofAmerica vol 104 no 8 pp 2750ndash2755 2007

[116] N Felli L Fontana E Pelosi et al ldquoMicroRNAs 221 and 222inhibit normal erythropoiesis and erythroleukemic cell growthvia kit receptor down-modulationrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 102 no50 pp 18081ndash18086 2005

[117] F Fazi A Rosa A Fatica et al ldquoA minicircuitry com-prised of microRNA-223 and transcription factors NFI-A andCEBPalpha regulates human granulopoiesisrdquo Cell vol 123 no5 pp 819ndash831 2005

[118] U Bissels SWild S Tomiuk et al ldquoCombined characterizationof microRNA and mRNA profiles delineates early differentia-tion pathways of CD133 + and CD34 + hematopoietic stem and

progenitor cells miRNAs in CD133 + cellsrdquo Stem Cells vol 29no 5 pp 847ndash857 2011

[119] Y Wang Z Li C He et al ldquoMicroRNAs expression signaturesare associated with lineage and survival in acute leukemiasrdquoBlood Cells Molecules and Diseases vol 44 no 3 pp 191ndash1972010

[120] R Garzon F Pichiorri T Palumbo et al ldquoMicroRNA geneexpression during retinoic acid-induced differentiation ofhuman acute promyelocytic leukemiardquo Oncogene vol 26 no28 pp 4148ndash4157 2007

[121] R Garzon S Volinia C G Liu et al ldquoMicroRNA signaturesassociated with cytogenetics and prognosis in acute myeloidleukemiardquo Blood vol 111 no 6 pp 3183ndash3189 2008

[122] S Debernardi S Skoulakis G Molloy T Chaplin A Dixon-McIver and B D Young ldquoMicroRNAmiR-181a correlates withmorphological sub-class of acute myeloid leukaemia and theexpression of its target genes in global genome-wide analysisrdquoLeukemia vol 21 no 5 pp 912ndash916 2007

[123] VHavelange N Stauffer C CHeaphy et al ldquoFunctional impli-cations of microRNAs in acutemyeloid leukemia by integratingmicroRNA and messenger RNA expression profilingrdquo Cancervol 117 no 20 pp 4696ndash4706 2011

[124] A Dixon-McIver P East C AMein et al ldquoDistinctive patternsof microRNA expression associated with karyotype in acutemyeloid leukaemiardquo Plos One vol 3 no 5 Article ID e21412008

[125] M Jongen-Lavrencic S M Sun M K Dijkstra P J M Valkand B Lowenberg ldquoMicroRNA expression profiling in relationto the genetic heterogeneity of acute myeloid leukemiardquo Bloodvol 111 no 10 pp 5078ndash5085 2008

[126] Z Li J Lu and M Sun ldquoDistinct microRNA expression pro-files in acute myeloid leukemia with common translocationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 105 no 40 pp 15535ndash15540 2008

[127] F Wang X Wang G Yang et al ldquomiR-29a and miR-142-3pdownregulation and diagnostic implication in human acutemyeloid leukemiardquoMolecular Biology Reports vol 39 no 3 pp2713ndash2722 2012

[128] G Marcucci M D Radmacher K Maharry et al ldquoMicroRNAexpression in cytogenetically normal acute myeloid leukemiardquoThe New England Journal of Medicine vol 358 no 18 pp 1919ndash1928 2008

[129] B J Wouters M A Jorda K Keeshan et al ldquoDistinct geneexpression profiles of acutemyeloidT-lymphoid leukemia withsilencedCEBPA andmutations inNOTCH1rdquoBlood vol 110 no10 pp 3706ndash3714 2007

[130] M E Figueroa B JWouters L Skrabanek et al ldquoGenome-wideepigenetic analysis delineates a biologically distinct immatureacute leukemia with myeloidT-lymphoid featuresrdquo Blood vol113 no 12 pp 2795ndash2804 2009

[131] S Grisendi R BernardiM Rossi et al ldquoRole of nucleophosminin embryonic development and tumorigenesisrdquoNature vol 437no 7055 pp 147ndash153 2005

[132] B Falini I Nicoletti M F Martelli and C Mecucci ldquoAcutemyeloid leukemia carrying cytoplasmicmutated nucleophos-min (NPMc+ AML) biologic and clinical featuresrdquo Blood vol109 no 3 pp 874ndash885 2006

[133] B Falini ldquoAcute myeloid leukemia with mutated nucleophos-min (NPM1) Molecular pathological and clinical featuresrdquoCancer Treatment and Research vol 145 pp 149ndash168 2010

34 Journal of Oncology

[134] M Alcalay ldquoAcute myeloid leukemia bearing cytoplasmicnucleophosmin (NPMc+ AML) shows a distinct gene expres-sion profile characterized by up-regulation of genes involvedin stem-cell maintenancerdquo Blood vol 106 no 3 pp 899ndash9022005

[135] L Handschuh P Wojciechowski M Kazmierczak et alldquoNPM1 alternative transcripts are upregulated in acute myeloidand lymphoblastic leukemia and their expression level affectspatient outcomerdquo Journal of Translational Medicine vol 16 no1 2018

[136] R Garzon M Garofalo M P Martelli et al ldquoDistinctivemicroRNA signature of acute myeloid leukemia bearing cyto-plasmic mutated nucleophosminrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 105 no10 pp 3945ndash3950 2008

[137] R GW Verhaak C S Goudswaard andW van Putten ldquoMuta-tions in nucleophosmin (NPM1) in acute myeloid leukemia(AML) association with other gene abnormalities and previ-ously established gene expression signatures and their favorableprognostic significancerdquo Blood vol 106 no 12 pp 3747ndash37542005

[138] O M Dovey J L Cooper A Mupo et al ldquo Molecular synergyunderlies the co-occurrence patterns and phenotype ofNPM1 -mutant acutemyeloid leukemia rdquoBlood vol 130 no 17 pp 1911ndash1922 2017

[139] M W Kuhn E Song Z Feng et al ldquoTargeting chromatinregulators inhibits leukemogenic gene expression in NPM1mutant leukemiardquoCancer Discovery vol 6 no 10 pp 1166ndash11812016

[140] A V Krivtsov and S A Armstrong ldquoMLL translocationshistone modifications and leukaemia stem-cell developmentrdquoNature Reviews Cancer vol 7 no 11 pp 823ndash833 2007

[141] L Huang K Zhou Y Yang et al ldquoFLT3-ITD-associated gene-expression signatures inNPM1-mutated cytogenetically normalacute myeloid leukemiardquo International Journal of Hematologyvol 96 no 2 pp 234ndash240 2012

[142] G Zhu Y Yang Y Zhang et al ldquoHigh expression of AHSPEPB42 GYPC and HEMGN predicts favorable prognosis inFLT3-ITD-negative acute myeloid leukemiardquo Cellular Physiol-ogy and Biochemistry vol 42 no 5 pp 1973ndash1984 2017

[143] J Wellbrock E Latuske J Kohler et al ldquoExpression of hedge-hog pathway mediator gli represents a negative prognosticmarker in human acute myeloid leukemia and its inhibitionexerts antileukemic effectsrdquoClinical Cancer Research vol 21 no10 pp 2388ndash2398 2015

[144] G Marcucci K Maharry Y-Z Wu et al ldquoIDH1 and IDH2genemutations identify novelmolecular subsets within de novocytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 28no 14 pp 2348ndash2355 2010

[145] V I Gaidzik L Bullinger R F Schlenk et al ldquoRUNX1 muta-tions in acute myeloid leukemia results from a comprehensivegenetic and clinical analysis from theAMLstudy grouprdquo Journalof Clinical Oncology vol 29 no 10 pp 1364ndash1372 2011

[146] S Schnittger F Dicker W Kern et al ldquoRUNX1 mutations arefrequent in de novo AML with noncomplex karyotype andconfer an unfavorable prognosis rdquo Blood vol 117 no 8 pp2348ndash2357 2011

[147] J H Mendler K Maharry M D Radmacher et al ldquoRUNX1mutations are associated with poor outcome in younger andolder patients with cytogenetically normal acute myeloidleukemia and with distinct gene and microRNA expression

signaturesrdquo Journal of Clinical Oncology vol 30 no 25 pp3109ndash3118 2012

[148] C Langer M D Radmacher A S Ruppert et al ldquoHigh BAALCexpression associates with other molecular prognostic markerspoor outcome and a distinct gene-expression signature incytogenetically normal patients younger than 60 years withacute myeloid leukemia a Cancer and Leukemia Group B(CALGB) studyrdquo Blood vol 111 no 11 pp 5371ndash5379 2008

[149] C Langer G Marcucci K B Holland et al ldquoPrognosticimportance ofMN1 transcript levels and biologic insights fromMN1-associated gene and microRNA expression signatures incytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 27no 19 pp 3198ndash3204 2009

[150] K H Metzeler A Dufour T Benthaus et al ldquoERG expressionis an independent prognostic factor and allows refined riskstratification in cytogenetically normal acutemyeloid leukemiaa comprehensive analysis of ERG MN1 and BAALC transcriptlevels using oligonucleotide microarraysrdquo Journal of ClinicalOncology vol 27 no 30 pp 5031ndash5038 2009

[151] B G Miller J A Stamatoyannopoulos and C CreightonldquoIntegrative meta-analysis of differential gene expression inacute myeloid leukemiardquo Plos One vol 5 no 3 Article IDe9466 2010

[152] F J Staal G Cario G Cazzaniga et al ldquoConsensus guidelinesfor microarray gene expression analyses in leukemia from threeEuropean leukemia networksrdquo Leukemia vol 20 no 8 pp1385ndash1392 2006

[153] M Park S Cho K Yoo et al ldquoGene expression profile related toprognosis of acute myeloid leukemiardquo Oncology Reports 2007

[154] L Handschuh M Kazmierczak M Milewski et al ldquoGeneexpression profiling of acute myeloid leukemia samples fromadult patients with AML-M1 and -M2 through boutiquemicroarrays real-time PCR and droplet digital PCRrdquo Interna-tional Journal of Oncology 2017

[155] S Lee J Chen G Zhou et al ldquoGene expression profiles in acutemyeloid leukemia with common translocations using SAGErdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 103 no 4 pp 1030ndash1035 2006

[156] J Lee J Hwang H Kim et al ldquoA comparison of gene expressionprofiles between primary human AML cells and AML cell linerdquoGenes amp Genetic Systems vol 83 no 4 pp 339ndash345 2008

[157] H Wang H Hu Q Zhang et al ldquoDynamic transcriptomes ofhuman myeloid leukemia cellsrdquo Genomics vol 102 no 4 pp250ndash256 2013

[158] G Gosse M Celton V Lamontagne A Forest and B TWilhelm ldquoWhole genome and transcriptome analysis of a novelAML cell linewith a normal karyotyperdquo Leukemia Research vol39 no 7 pp 709ndash718 2015

[159] C Hsu C Nguyen C Yan et al ldquoTranscriptome profiling ofpediatric core binding factor AMLrdquo Plos One vol 10 no 9Article ID e0138782 2015

[160] A A Singh A Mandoli K H Prange M Laakso and J HMartens ldquoAML associated oncofusion proteins PML-RARAAML1-ETO and CBFB-MYH11 target RUNXETS-factor bind-ing sites to modulate H3ac levels and drive leukemogenesisrdquoOncotarget vol 8 no 8 2017

[161] A Eisfeld J Kohlschmidt K Mrozek et al ldquoMutationallandscape and gene expression patterns in adult acute myeloidleukemias with monosomy 7 as a sole abnormalityrdquo CancerResearch vol 77 no 1 pp 207ndash218 2017

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

Journal of Oncology 33

have abnormal phenotype and genotype and form defectiveimmune synapses with AML blastsrdquo Blood vol 114 no 18 pp3909ndash3916 2009

[101] R C Lee R L Feinbaum and V Ambros ldquoThe C elegansheterochronic gene lin-4 encodes small RNAs with antisensecomplementarity to lin-14rdquoCell vol 75 no 5 pp 843ndash854 1993

[102] A G Matera R M Terns andM P Terns ldquoNon-coding RNAslessons from the small nuclear and small nucleolar RNAsrdquoNature Reviews Molecular Cell Biology vol 8 no 3 pp 209ndash220 2007

[103] K V Morris and J S Mattick ldquoThe rise of regulatory RNArdquoNature Reviews Genetics vol 15 no 6 pp 423ndash437 2014

[104] K W Ng C Anderson E A Marshall et al ldquoPiwi-interactingRNAs in cancer emerging functions and clinical utilityrdquoMolec-ular Cancer vol 15 no 1 2016

[105] S M Hammond ldquoAn overview of microRNAsrdquoAdvancedDrugDelivery Reviews vol 87 pp 3ndash14 2015

[106] E A Miska ldquoHow microRNAs control cell division differenti-ation and deathrdquo Current Opinion in Genetics amp Developmentvol 15 no 5 pp 563ndash568 2005

[107] B D Harfe ldquoMicroRNAs in vertebrate developmentrdquo CurrentOpinion in Genetics amp Development vol 15 no 4 pp 410ndash4152005

[108] C Chen ldquoMicroRNAs as oncogenes and tumor suppressorsrdquoThe New England Journal of Medicine vol 353 no 17 pp 1768ndash1771 2005

[109] A Esquela-Kerscher and F J Slack ldquoOncomirsmdashmicroRNAswith a role in cancerrdquo Nature Reviews Cancer vol 6 no 4 pp259ndash269 2006

[110] H Seitz ldquoIssues in current microRNA target identificationmethodsrdquo RNA Biology vol 14 no 7 pp 831ndash834 2017

[111] G A Calin C D Dumitru M Shimizu et al ldquoFrequentdeletions and down-regulation of micro-RNA genes miR15 andmiR16 at 13q14 in chronic lymphocytic leukemiardquoProceedings ofthe National Acadamyof Sciences of the United States of Americavol 99 no 24 pp 15524ndash15529 2002

[112] S A Melo and M Esteller ldquoDysregulation of microRNAs incancer playingwith firerdquo FEBS Letters vol 585 no 13 pp 2087ndash2099 2011

[113] C Chen L Li H F Lodish and D P Bartel ldquoMicroRNAsmodulate hematopoietic lineage differentiationrdquo Science vol303 no 5654 pp 83ndash86 2004

[114] C-Z Chen and H F Lodish ldquoMicroRNAs as regulators ofmammalian hematopoiesisrdquo Seminars in Immunology vol 17no 2 pp 155ndash165 2005

[115] R W Georgantas III R Hildreth S Morisot et al ldquoCD34+hematopoietic stem-progenitor cell microRNA expression andfunction a circuit diagram of differentiation controlrdquo Proceed-ings of the National Acadamy of Sciences of the United States ofAmerica vol 104 no 8 pp 2750ndash2755 2007

[116] N Felli L Fontana E Pelosi et al ldquoMicroRNAs 221 and 222inhibit normal erythropoiesis and erythroleukemic cell growthvia kit receptor down-modulationrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 102 no50 pp 18081ndash18086 2005

[117] F Fazi A Rosa A Fatica et al ldquoA minicircuitry com-prised of microRNA-223 and transcription factors NFI-A andCEBPalpha regulates human granulopoiesisrdquo Cell vol 123 no5 pp 819ndash831 2005

[118] U Bissels SWild S Tomiuk et al ldquoCombined characterizationof microRNA and mRNA profiles delineates early differentia-tion pathways of CD133 + and CD34 + hematopoietic stem and

progenitor cells miRNAs in CD133 + cellsrdquo Stem Cells vol 29no 5 pp 847ndash857 2011

[119] Y Wang Z Li C He et al ldquoMicroRNAs expression signaturesare associated with lineage and survival in acute leukemiasrdquoBlood Cells Molecules and Diseases vol 44 no 3 pp 191ndash1972010

[120] R Garzon F Pichiorri T Palumbo et al ldquoMicroRNA geneexpression during retinoic acid-induced differentiation ofhuman acute promyelocytic leukemiardquo Oncogene vol 26 no28 pp 4148ndash4157 2007

[121] R Garzon S Volinia C G Liu et al ldquoMicroRNA signaturesassociated with cytogenetics and prognosis in acute myeloidleukemiardquo Blood vol 111 no 6 pp 3183ndash3189 2008

[122] S Debernardi S Skoulakis G Molloy T Chaplin A Dixon-McIver and B D Young ldquoMicroRNAmiR-181a correlates withmorphological sub-class of acute myeloid leukaemia and theexpression of its target genes in global genome-wide analysisrdquoLeukemia vol 21 no 5 pp 912ndash916 2007

[123] VHavelange N Stauffer C CHeaphy et al ldquoFunctional impli-cations of microRNAs in acutemyeloid leukemia by integratingmicroRNA and messenger RNA expression profilingrdquo Cancervol 117 no 20 pp 4696ndash4706 2011

[124] A Dixon-McIver P East C AMein et al ldquoDistinctive patternsof microRNA expression associated with karyotype in acutemyeloid leukaemiardquo Plos One vol 3 no 5 Article ID e21412008

[125] M Jongen-Lavrencic S M Sun M K Dijkstra P J M Valkand B Lowenberg ldquoMicroRNA expression profiling in relationto the genetic heterogeneity of acute myeloid leukemiardquo Bloodvol 111 no 10 pp 5078ndash5085 2008

[126] Z Li J Lu and M Sun ldquoDistinct microRNA expression pro-files in acute myeloid leukemia with common translocationsrdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 105 no 40 pp 15535ndash15540 2008

[127] F Wang X Wang G Yang et al ldquomiR-29a and miR-142-3pdownregulation and diagnostic implication in human acutemyeloid leukemiardquoMolecular Biology Reports vol 39 no 3 pp2713ndash2722 2012

[128] G Marcucci M D Radmacher K Maharry et al ldquoMicroRNAexpression in cytogenetically normal acute myeloid leukemiardquoThe New England Journal of Medicine vol 358 no 18 pp 1919ndash1928 2008

[129] B J Wouters M A Jorda K Keeshan et al ldquoDistinct geneexpression profiles of acutemyeloidT-lymphoid leukemia withsilencedCEBPA andmutations inNOTCH1rdquoBlood vol 110 no10 pp 3706ndash3714 2007

[130] M E Figueroa B JWouters L Skrabanek et al ldquoGenome-wideepigenetic analysis delineates a biologically distinct immatureacute leukemia with myeloidT-lymphoid featuresrdquo Blood vol113 no 12 pp 2795ndash2804 2009

[131] S Grisendi R BernardiM Rossi et al ldquoRole of nucleophosminin embryonic development and tumorigenesisrdquoNature vol 437no 7055 pp 147ndash153 2005

[132] B Falini I Nicoletti M F Martelli and C Mecucci ldquoAcutemyeloid leukemia carrying cytoplasmicmutated nucleophos-min (NPMc+ AML) biologic and clinical featuresrdquo Blood vol109 no 3 pp 874ndash885 2006

[133] B Falini ldquoAcute myeloid leukemia with mutated nucleophos-min (NPM1) Molecular pathological and clinical featuresrdquoCancer Treatment and Research vol 145 pp 149ndash168 2010

34 Journal of Oncology

[134] M Alcalay ldquoAcute myeloid leukemia bearing cytoplasmicnucleophosmin (NPMc+ AML) shows a distinct gene expres-sion profile characterized by up-regulation of genes involvedin stem-cell maintenancerdquo Blood vol 106 no 3 pp 899ndash9022005

[135] L Handschuh P Wojciechowski M Kazmierczak et alldquoNPM1 alternative transcripts are upregulated in acute myeloidand lymphoblastic leukemia and their expression level affectspatient outcomerdquo Journal of Translational Medicine vol 16 no1 2018

[136] R Garzon M Garofalo M P Martelli et al ldquoDistinctivemicroRNA signature of acute myeloid leukemia bearing cyto-plasmic mutated nucleophosminrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 105 no10 pp 3945ndash3950 2008

[137] R GW Verhaak C S Goudswaard andW van Putten ldquoMuta-tions in nucleophosmin (NPM1) in acute myeloid leukemia(AML) association with other gene abnormalities and previ-ously established gene expression signatures and their favorableprognostic significancerdquo Blood vol 106 no 12 pp 3747ndash37542005

[138] O M Dovey J L Cooper A Mupo et al ldquo Molecular synergyunderlies the co-occurrence patterns and phenotype ofNPM1 -mutant acutemyeloid leukemia rdquoBlood vol 130 no 17 pp 1911ndash1922 2017

[139] M W Kuhn E Song Z Feng et al ldquoTargeting chromatinregulators inhibits leukemogenic gene expression in NPM1mutant leukemiardquoCancer Discovery vol 6 no 10 pp 1166ndash11812016

[140] A V Krivtsov and S A Armstrong ldquoMLL translocationshistone modifications and leukaemia stem-cell developmentrdquoNature Reviews Cancer vol 7 no 11 pp 823ndash833 2007

[141] L Huang K Zhou Y Yang et al ldquoFLT3-ITD-associated gene-expression signatures inNPM1-mutated cytogenetically normalacute myeloid leukemiardquo International Journal of Hematologyvol 96 no 2 pp 234ndash240 2012

[142] G Zhu Y Yang Y Zhang et al ldquoHigh expression of AHSPEPB42 GYPC and HEMGN predicts favorable prognosis inFLT3-ITD-negative acute myeloid leukemiardquo Cellular Physiol-ogy and Biochemistry vol 42 no 5 pp 1973ndash1984 2017

[143] J Wellbrock E Latuske J Kohler et al ldquoExpression of hedge-hog pathway mediator gli represents a negative prognosticmarker in human acute myeloid leukemia and its inhibitionexerts antileukemic effectsrdquoClinical Cancer Research vol 21 no10 pp 2388ndash2398 2015

[144] G Marcucci K Maharry Y-Z Wu et al ldquoIDH1 and IDH2genemutations identify novelmolecular subsets within de novocytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 28no 14 pp 2348ndash2355 2010

[145] V I Gaidzik L Bullinger R F Schlenk et al ldquoRUNX1 muta-tions in acute myeloid leukemia results from a comprehensivegenetic and clinical analysis from theAMLstudy grouprdquo Journalof Clinical Oncology vol 29 no 10 pp 1364ndash1372 2011

[146] S Schnittger F Dicker W Kern et al ldquoRUNX1 mutations arefrequent in de novo AML with noncomplex karyotype andconfer an unfavorable prognosis rdquo Blood vol 117 no 8 pp2348ndash2357 2011

[147] J H Mendler K Maharry M D Radmacher et al ldquoRUNX1mutations are associated with poor outcome in younger andolder patients with cytogenetically normal acute myeloidleukemia and with distinct gene and microRNA expression

signaturesrdquo Journal of Clinical Oncology vol 30 no 25 pp3109ndash3118 2012

[148] C Langer M D Radmacher A S Ruppert et al ldquoHigh BAALCexpression associates with other molecular prognostic markerspoor outcome and a distinct gene-expression signature incytogenetically normal patients younger than 60 years withacute myeloid leukemia a Cancer and Leukemia Group B(CALGB) studyrdquo Blood vol 111 no 11 pp 5371ndash5379 2008

[149] C Langer G Marcucci K B Holland et al ldquoPrognosticimportance ofMN1 transcript levels and biologic insights fromMN1-associated gene and microRNA expression signatures incytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 27no 19 pp 3198ndash3204 2009

[150] K H Metzeler A Dufour T Benthaus et al ldquoERG expressionis an independent prognostic factor and allows refined riskstratification in cytogenetically normal acutemyeloid leukemiaa comprehensive analysis of ERG MN1 and BAALC transcriptlevels using oligonucleotide microarraysrdquo Journal of ClinicalOncology vol 27 no 30 pp 5031ndash5038 2009

[151] B G Miller J A Stamatoyannopoulos and C CreightonldquoIntegrative meta-analysis of differential gene expression inacute myeloid leukemiardquo Plos One vol 5 no 3 Article IDe9466 2010

[152] F J Staal G Cario G Cazzaniga et al ldquoConsensus guidelinesfor microarray gene expression analyses in leukemia from threeEuropean leukemia networksrdquo Leukemia vol 20 no 8 pp1385ndash1392 2006

[153] M Park S Cho K Yoo et al ldquoGene expression profile related toprognosis of acute myeloid leukemiardquo Oncology Reports 2007

[154] L Handschuh M Kazmierczak M Milewski et al ldquoGeneexpression profiling of acute myeloid leukemia samples fromadult patients with AML-M1 and -M2 through boutiquemicroarrays real-time PCR and droplet digital PCRrdquo Interna-tional Journal of Oncology 2017

[155] S Lee J Chen G Zhou et al ldquoGene expression profiles in acutemyeloid leukemia with common translocations using SAGErdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 103 no 4 pp 1030ndash1035 2006

[156] J Lee J Hwang H Kim et al ldquoA comparison of gene expressionprofiles between primary human AML cells and AML cell linerdquoGenes amp Genetic Systems vol 83 no 4 pp 339ndash345 2008

[157] H Wang H Hu Q Zhang et al ldquoDynamic transcriptomes ofhuman myeloid leukemia cellsrdquo Genomics vol 102 no 4 pp250ndash256 2013

[158] G Gosse M Celton V Lamontagne A Forest and B TWilhelm ldquoWhole genome and transcriptome analysis of a novelAML cell linewith a normal karyotyperdquo Leukemia Research vol39 no 7 pp 709ndash718 2015

[159] C Hsu C Nguyen C Yan et al ldquoTranscriptome profiling ofpediatric core binding factor AMLrdquo Plos One vol 10 no 9Article ID e0138782 2015

[160] A A Singh A Mandoli K H Prange M Laakso and J HMartens ldquoAML associated oncofusion proteins PML-RARAAML1-ETO and CBFB-MYH11 target RUNXETS-factor bind-ing sites to modulate H3ac levels and drive leukemogenesisrdquoOncotarget vol 8 no 8 2017

[161] A Eisfeld J Kohlschmidt K Mrozek et al ldquoMutationallandscape and gene expression patterns in adult acute myeloidleukemias with monosomy 7 as a sole abnormalityrdquo CancerResearch vol 77 no 1 pp 207ndash218 2017

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

34 Journal of Oncology

[134] M Alcalay ldquoAcute myeloid leukemia bearing cytoplasmicnucleophosmin (NPMc+ AML) shows a distinct gene expres-sion profile characterized by up-regulation of genes involvedin stem-cell maintenancerdquo Blood vol 106 no 3 pp 899ndash9022005

[135] L Handschuh P Wojciechowski M Kazmierczak et alldquoNPM1 alternative transcripts are upregulated in acute myeloidand lymphoblastic leukemia and their expression level affectspatient outcomerdquo Journal of Translational Medicine vol 16 no1 2018

[136] R Garzon M Garofalo M P Martelli et al ldquoDistinctivemicroRNA signature of acute myeloid leukemia bearing cyto-plasmic mutated nucleophosminrdquo Proceedings of the NationalAcadamy of Sciences of the United States of America vol 105 no10 pp 3945ndash3950 2008

[137] R GW Verhaak C S Goudswaard andW van Putten ldquoMuta-tions in nucleophosmin (NPM1) in acute myeloid leukemia(AML) association with other gene abnormalities and previ-ously established gene expression signatures and their favorableprognostic significancerdquo Blood vol 106 no 12 pp 3747ndash37542005

[138] O M Dovey J L Cooper A Mupo et al ldquo Molecular synergyunderlies the co-occurrence patterns and phenotype ofNPM1 -mutant acutemyeloid leukemia rdquoBlood vol 130 no 17 pp 1911ndash1922 2017

[139] M W Kuhn E Song Z Feng et al ldquoTargeting chromatinregulators inhibits leukemogenic gene expression in NPM1mutant leukemiardquoCancer Discovery vol 6 no 10 pp 1166ndash11812016

[140] A V Krivtsov and S A Armstrong ldquoMLL translocationshistone modifications and leukaemia stem-cell developmentrdquoNature Reviews Cancer vol 7 no 11 pp 823ndash833 2007

[141] L Huang K Zhou Y Yang et al ldquoFLT3-ITD-associated gene-expression signatures inNPM1-mutated cytogenetically normalacute myeloid leukemiardquo International Journal of Hematologyvol 96 no 2 pp 234ndash240 2012

[142] G Zhu Y Yang Y Zhang et al ldquoHigh expression of AHSPEPB42 GYPC and HEMGN predicts favorable prognosis inFLT3-ITD-negative acute myeloid leukemiardquo Cellular Physiol-ogy and Biochemistry vol 42 no 5 pp 1973ndash1984 2017

[143] J Wellbrock E Latuske J Kohler et al ldquoExpression of hedge-hog pathway mediator gli represents a negative prognosticmarker in human acute myeloid leukemia and its inhibitionexerts antileukemic effectsrdquoClinical Cancer Research vol 21 no10 pp 2388ndash2398 2015

[144] G Marcucci K Maharry Y-Z Wu et al ldquoIDH1 and IDH2genemutations identify novelmolecular subsets within de novocytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 28no 14 pp 2348ndash2355 2010

[145] V I Gaidzik L Bullinger R F Schlenk et al ldquoRUNX1 muta-tions in acute myeloid leukemia results from a comprehensivegenetic and clinical analysis from theAMLstudy grouprdquo Journalof Clinical Oncology vol 29 no 10 pp 1364ndash1372 2011

[146] S Schnittger F Dicker W Kern et al ldquoRUNX1 mutations arefrequent in de novo AML with noncomplex karyotype andconfer an unfavorable prognosis rdquo Blood vol 117 no 8 pp2348ndash2357 2011

[147] J H Mendler K Maharry M D Radmacher et al ldquoRUNX1mutations are associated with poor outcome in younger andolder patients with cytogenetically normal acute myeloidleukemia and with distinct gene and microRNA expression

signaturesrdquo Journal of Clinical Oncology vol 30 no 25 pp3109ndash3118 2012

[148] C Langer M D Radmacher A S Ruppert et al ldquoHigh BAALCexpression associates with other molecular prognostic markerspoor outcome and a distinct gene-expression signature incytogenetically normal patients younger than 60 years withacute myeloid leukemia a Cancer and Leukemia Group B(CALGB) studyrdquo Blood vol 111 no 11 pp 5371ndash5379 2008

[149] C Langer G Marcucci K B Holland et al ldquoPrognosticimportance ofMN1 transcript levels and biologic insights fromMN1-associated gene and microRNA expression signatures incytogenetically normal acute myeloid leukemia a cancer andleukemia group B studyrdquo Journal of Clinical Oncology vol 27no 19 pp 3198ndash3204 2009

[150] K H Metzeler A Dufour T Benthaus et al ldquoERG expressionis an independent prognostic factor and allows refined riskstratification in cytogenetically normal acutemyeloid leukemiaa comprehensive analysis of ERG MN1 and BAALC transcriptlevels using oligonucleotide microarraysrdquo Journal of ClinicalOncology vol 27 no 30 pp 5031ndash5038 2009

[151] B G Miller J A Stamatoyannopoulos and C CreightonldquoIntegrative meta-analysis of differential gene expression inacute myeloid leukemiardquo Plos One vol 5 no 3 Article IDe9466 2010

[152] F J Staal G Cario G Cazzaniga et al ldquoConsensus guidelinesfor microarray gene expression analyses in leukemia from threeEuropean leukemia networksrdquo Leukemia vol 20 no 8 pp1385ndash1392 2006

[153] M Park S Cho K Yoo et al ldquoGene expression profile related toprognosis of acute myeloid leukemiardquo Oncology Reports 2007

[154] L Handschuh M Kazmierczak M Milewski et al ldquoGeneexpression profiling of acute myeloid leukemia samples fromadult patients with AML-M1 and -M2 through boutiquemicroarrays real-time PCR and droplet digital PCRrdquo Interna-tional Journal of Oncology 2017

[155] S Lee J Chen G Zhou et al ldquoGene expression profiles in acutemyeloid leukemia with common translocations using SAGErdquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 103 no 4 pp 1030ndash1035 2006

[156] J Lee J Hwang H Kim et al ldquoA comparison of gene expressionprofiles between primary human AML cells and AML cell linerdquoGenes amp Genetic Systems vol 83 no 4 pp 339ndash345 2008

[157] H Wang H Hu Q Zhang et al ldquoDynamic transcriptomes ofhuman myeloid leukemia cellsrdquo Genomics vol 102 no 4 pp250ndash256 2013

[158] G Gosse M Celton V Lamontagne A Forest and B TWilhelm ldquoWhole genome and transcriptome analysis of a novelAML cell linewith a normal karyotyperdquo Leukemia Research vol39 no 7 pp 709ndash718 2015

[159] C Hsu C Nguyen C Yan et al ldquoTranscriptome profiling ofpediatric core binding factor AMLrdquo Plos One vol 10 no 9Article ID e0138782 2015

[160] A A Singh A Mandoli K H Prange M Laakso and J HMartens ldquoAML associated oncofusion proteins PML-RARAAML1-ETO and CBFB-MYH11 target RUNXETS-factor bind-ing sites to modulate H3ac levels and drive leukemogenesisrdquoOncotarget vol 8 no 8 2017

[161] A Eisfeld J Kohlschmidt K Mrozek et al ldquoMutationallandscape and gene expression patterns in adult acute myeloidleukemias with monosomy 7 as a sole abnormalityrdquo CancerResearch vol 77 no 1 pp 207ndash218 2017

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

Journal of Oncology 35

[162] T Herold K H Metzeler S Vosberg et al ldquo Acute myeloidleukemia with del(9q) is characterized by frequent mutationsof NPM1 DNMT3AWT1 and low expression of TLE4rdquo GenesChromosomes and Cancer vol 56 no 1 pp 75ndash86 2017

[163] A Pallavajjala D Kim T Li et al ldquoGenomic characterizationofchromosome translocations in patients with Tmyeloid mixed-phenotype acute leukemiardquo Leukemiaamp Lymphoma vol 59 no5 pp 1231ndash1238 2017

[164] A Abe Y Yamamoto A Katsumi et al ldquoRearrangementof VPS13B a causative gene of Cohen syndrome in a caseof RUNX1ndashRUNX1T1 leukemia with t(81221)rdquo InternationalJournal of Hematology vol 108 no 2 pp 208ndash212 2018

[165] H Wen Y Li S N Malek et al ldquoNew fusion transcriptsidentified in normal karyotype acute myeloid leukemiardquo PlosOne vol 7 no 12 Article ID e51203 2012

[166] M Togni R Masetti M Pigazzi et al ldquoIdentification of theNUP98-PHF23 fusion gene in pediatric cytogenetically normalacute myeloid leukemia by whole-transcriptome sequencingrdquoJournal of Hematology amp Oncology vol 8 no 1 2015

[167] H L Schuback T A Alonzo R B Gerbing K L MillerS Kahwash A Heerema-Mckenney et al ldquoCBFA2T3-GLIS2fusion is prevalent in younger patients with acute myeloidleukemia and associated with high-risk of relapse and pooroutcome a childrenrsquos oncology group reportrdquo Blood vol 124no 21 p 13 2014

[168] R Masetti M Pigazzi M Togni et al ldquoCBFA2T3-GLIS2 fusiontranscript is a novel common feature in pediatric cytogeneti-cally normal AML not restricted to FAB M7 subtyperdquo Bloodvol 121 no 17 pp 3469ndash3472 2013

[169] R Masetti M Togni A Astolfi et al ldquoDHH-RHEBL1 fusiontranscript a novel recurrent feature in the new landscape ofpediatric CBFA2T3-GLIS2-positive acute myeloid leukemiardquoOncotarget vol 4 no 10 2013

[170] R de Necochea-CampionG P ShouseQ Zhou S Mirshahidiand C Chen ldquoAberrant splicing and drug resistance in AMLrdquoJournal of Hematology amp Oncology vol 9 no 1 2016

[171] X Li X Yao S N Li et al ldquoRNA-Seq profiling revealsaberrant RNA splicing in patient with adult acute myeloidleukemia during treatmentrdquo European Review for Medical andPharmacological Sciences vol 18 no 9 pp 1426ndash1433 2014

[172] P Gao Z Jin Y Cheng and X Cao ldquoRNA-Seq analysisidentifies aberrant RNA splicing of TRIP12 in acute myeloidleukemia patients at remissionrdquo Tumor Biology vol 35 no 10pp 9585ndash9590 2014

[173] C Shirai J Ley BWhite et al ldquoMutantU2AF1 expression altershematopoiesis and pre-mRNA splicing in vivordquoCancer Cell vol27 no 5 pp 631ndash643 2015

[174] T Li Q Liu N Garza S Kornblau and V X Jin ldquoIntegrativeanalysis reveals functional and regulatory roles of H3K79me2inmediating alternative splicingrdquoGenomeMedicine vol 10 no1 2018

[175] F Mertens B Johansson T Fioretos and F Mitelman ldquoTheemerging complexity of gene fusions in cancerrdquoNature ReviewsCancer vol 15 no 6 pp 371ndash381 2015

[176] M Babiceanu F Qin Z Xie et al ldquoRecurrent chimeric fusionRNAs in non-cancer tissues and cellsrdquo Nucleic Acids Researchvol 44 no 6 pp 2859ndash2872 2016

[177] F Ruffle J Audoux A Boureux et al ldquoNew chimeric RNAs inacute myeloid leukemiardquo F1000Research vol 6 2017

[178] G Ramsingh D C Koboldt M Trissal et al ldquoCompletecharacterization of the microRNAome in a patient with acutemyeloid leukemiardquo Blood vol 116 no 24 pp 5316ndash5326 2010

[179] L Garcıa-Ortı I Cristobal C Cirauqui et al ldquoIntegrationof SNP and mRNA arrays with microRNA profiling revealsthat MiR-370 is upregulated and targets NF1 in acute myeloidleukemiardquo Plos One vol 7 no 10 Article ID e47717 2012

[180] D T Starczynowski R Morin A McPherson et al ldquoGenome-wide identification of human microRNAs located in leukemia-associated genomic alterationsrdquo Blood vol 117 no 2 pp 595ndash607 2011

[181] A Junge R Zandi J H Havgaard J Gorodkin and J BCowland ldquoAssessing the miRNA sponge potential of RUNX1T1in t(821) acute myeloid leukemiardquo Gene vol 615 pp 35ndash402017

[182] A Fu D I Jacobs and Y Zhu ldquoEpigenome-wide analysis ofpiRNAs in gene-specific DNA methylationrdquo RNA Biology vol11 no 10 pp 1301ndash1312 2015

[183] V D Martinez E A Vucic and K L Thu ldquoUnique somaticand malignant expression patterns implicate PIWI-interactingRNAs in cancer-type specific biologyrdquo Scientific Reports vol 5Article ID 10423 2015

[184] L De Luca S Trino I Laurenzana et al ldquoMiRNAs and piRNAsfrombonemarrowmesenchymal stem cell extracellular vesiclesinduce cell survival and inhibit cell differentiation of cord bloodhematopoietic stem cells a new insight in transplantationrdquoOncotarget vol 7 no 6 pp 6676ndash6692 2015

[185] S Bamezai Characterizing the Functional Role of ArgonauteProtein PIWIL4 in AcuteMyeloid Leukemia Institute for Exper-imental Cancer Research Ulm Germany 2015

[186] A R Colombo A Zubair D Thiagarajan S Nuzhdin TJ Triche and G Ramsingh ldquoSuppression of transposableelements in leukemic stem cellsrdquo Scientific Reports vol 7 no1 2017

[187] M S Scott and M Ono ldquoFrom snoRNA to miRNA Dualfunction regulatory non-coding RNAsrdquo Biochimie vol 93 no11 pp 1987ndash1992 2011

[188] W A Warner D H Spencer M Trissal et al ldquoExpressionprofiling of snoRNAs in normal hematopoiesis and AMLrdquoBlood Advances vol 2 no 2 pp 151ndash163 2018

[189] P Kapranov J Cheng S Dike et al ldquoRNA maps reveal newRNA classes and a possible function for pervasive transcrip-tionrdquo Science vol 316 no 5830 pp 1484ndash1488 2007

[190] J L Rinn and H Y Chang ldquoGenome regulation by longnoncoding RNAsrdquo Annual Review of Biochemistry vol 81 pp145ndash166 2012

[191] R Garzon S Volinia D Papaioannou et al ldquoExpression andprognostic impact of lncRNAs in acute myeloid leukemiardquoProceedings of the National Acadamy of Sciences of the UnitedStates of America vol 111 no 52 pp 18679ndash18684 2014

[192] A Schwarzer S Emmrich F Schmidt et al ldquoThe non-codingRNA landscape of human hematopoiesis and leukemiardquoNatureCommunications vol 8 no 1 2017

[193] M J Delas L R Sabin E Dolzhenko et al ldquolncRNArequirements for mouse acute myeloid leukemia and normaldifferentiationrdquo eLife vol 6 2017

[194] J Zhang M Griffith C A Miller et al ldquoComprehensivediscovery of noncoding RNAs in acute myeloid leukemia celltranscriptomesrdquo Experimental Hematology vol 55 pp 19ndash332017

[195] J-Q Pan Y-Q Zhang J-H Wang P Xu and W WangldquoLncRNA co-expression network model for the prognosticanalysis of acute myeloid leukemiardquo International Journal ofMolecularMedicine vol 39 no 3 pp 663ndash671 2017

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

36 Journal of Oncology

[196] A SMer J LindbergCNilsson et al ldquoExpression levels of longnon-coding RNAs are prognostic for AML outcomerdquo Journal ofHematology amp Oncology vol 11 no 1 2018

[197] M Bill D Papaioannou M Karunasiri et al ldquoExpression andfunctional relevanceof long non-codingRNAs in acutemyeloidleukemia stem cellsrdquo Leukemia 2019

[198] J Salzman C Gawad P L Wang N Lacayo and P O BrownldquoCircular RNAs are the predominant transcript isoform fromhundreds of human genes in diverse cell typesrdquo Plos One vol 7no 2 Article ID e30733 2012

[199] C Caldas C W So A MacGregor et al ldquoExon scrambling ofMLL transcripts occur commonly and mimic partial genomicduplication of the generdquoGene vol 208 no 2 pp 167ndash176 1998

[200] X You and T O Conrad ldquoAcfs accurate circRNA identificationand quantification from RNA-Seq datardquo Scientific Reports vol6 no 1 2016

[201] S Hirsch T J Blatte S Grasedieck et al ldquoCircular RNAs ofthe nucleophosmin (NPM1) gene in acute myeloid leukemiardquoHaematologica vol 102 no 12 pp 2039ndash2047 2017

[202] A L1015840Abbate D Tolomeo I Cifola et al ldquoMYC-containingamplicons in acute myeloid leukemia genomic structuresevolution and transcriptional consequencesrdquoLeukemia vol 32no 10 pp 2152ndash2166 2018

[203] J Shang W Chen Z Wang T Wei Z Chen and W WuldquoCircPAN3mediates drug resistance in acutemyeloid leukemiathrough the miR-153-5pmiR-183-5pndashXIAP axisrdquo ExperimentalHematology vol 70 pp 42ndash54e3 2019

[204] D L Stirewalt S Meshinchi K J Kopecky et al ldquoIdentificationof genes with abnormal expression changes in acute myeloidleukemiardquo Genes Chromosomes and Cancer vol 47 no 1 pp8ndash20 2008

[205] R Majeti M W Becker Q Tian et al ldquoDysregulated geneexpression networks in human acute myelogenous leukemiastem cellsrdquo Proceedings of the National Acadamy of Sciences ofthe United States of America vol 106 no 9 pp 3396ndash3401 2009

[206] A J Gentles S K Plevritis R Majeti and A A AlizadehldquoAssociation of a leukemic stem cell gene expression signaturewith clinical outcomes in acute myeloid leukemiardquoThe Journalof the American Medical Association vol 304 no 24 pp 2706ndash2715 2010

[207] S W Ng A Mitchell J A Kennedy et al ldquoA 17-gene stemnessscore for rapid determination of risk in acute leukaemiardquoNature vol 540 no 7633 pp 433ndash437 2016

[208] N Duployez A Marceau-Renaut C Villenet et al ldquoThe stemcell-associated gene expression signature allows risk stratifica-tion in pediatric acutemyeloid leukemiardquo Leukemia vol 33 no2 pp 348ndash357 2019

[209] H J VerhagenN vanGils TMartianez et al ldquoIGFBP7 inducesdifferentiation and loss of survival of human acute myeloidleukemia stem cells without affecting normal hematopoiesisrdquoCell Reports vol 25 no 11 pp 3021ndash3035e5 2018

[210] S J Erkeland M Valkhof C Heijmans-Antonissen et alldquoLarge-scale identification of disease genes involved in acutemyeloid leukemiardquo Journal of Virology vol 78 no 4 pp 1971ndash1980 2004

[211] C Glass C Wuertzer X Cui et al ldquoGlobal identification ofEVI1 target genes in acute myeloid leukemiardquo PLoS ONE vol8 no 6 p e67134 2013

[212] B T Wilhelm M Briau P Austin et al ldquoRNA-seq analysis of 2closely related leukemia clones that differ in their self-renewalcapacityrdquo Blood vol 117 no 2 pp e27ndashe38 2011

[213] M R McKeown M R Corces M L Eaton et al ldquoSuperen-hancer analysis defines novel epigenomic subtypes of non-aplaml including an rar120572 dependency targetable by sy-1425 apotent and selective rar120572 agonistrdquo Cancer Discovery vol 7 no10 pp 1136ndash1153 2017

[214] J Loven H A Hoke C Y Lin et al ldquoSelective inhibition oftumor oncogenes by disruption of super-enhancersrdquo Cell vol153 no 2 pp 320ndash334 2013

[215] S Dong and J Chen ldquoSUMOylation of sPRDM16 promotes theprogression of acute myeloid leukemiardquo BMC Cancer vol 15no 1 2015

[216] B Yan Y Hu K H Ban et al ldquoSingle-cell genomic profiling ofacutemyeloid leukemia for clinical use A pilot studyrdquoOncologyLetters vol 13 no 3 pp 1625ndash1630 2017

[217] L I Shlush S Zandi A Mitchell et al ldquoIdentification ofpre-leukaemic haematopoietic stem cells in acute leukaemiardquoNature vol 506 no 7488 pp 328ndash333 2014

[218] M Alessandrini S S Kappala and M S Pepper ldquoAMLprofilerA diagnostic and prognostic microarray for acute myeloidleukemiardquoMethods in Molecular Biology vol 1633 pp 101ndash1232017

[219] E K Mack A Marquardt D Langer et al ldquoComprehensivegenetic diagnosis of acutemyeloid leukemia by next-generationsequencingrdquoHaematologica vol 104 no 2 pp 277ndash287 2019

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom

Stem Cells International

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Disease Markers

Hindawiwwwhindawicom Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwwwhindawicom Volume 2013

Hindawiwwwhindawicom Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwwwhindawicom Volume 2018

PPAR Research

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwwwhindawicom Volume 2018

Journal of

ObesityJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwwwhindawicom Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwwwhindawicom Volume 2018

Diabetes ResearchJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Research and TreatmentAIDS

Hindawiwwwhindawicom Volume 2018

Gastroenterology Research and Practice

Hindawiwwwhindawicom Volume 2018

Parkinsonrsquos Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwwwhindawicom

Submit your manuscripts atwwwhindawicom