Genomic analysis reveals few genetic alterationsin pediatric acute myeloid leukemiaIna Radtkea, Charles G. Mullighana, Masami Ishiia, Xiaoping Sua, Jinjun Chenga, Jing Mab, Ramapriya Gantia,Zhongling Caia, Salil Goorhaa, Stanley B. Poundsc, Xueyuan Caoc, Caroline Obertb, Jianling Armstrongb, Jinghui Zhangd,Guangchun Songa, Raul C. Ribeiroe, Jeffrey E. Rubnitze, Susana C. Raimondia, Sheila A. Shurtleffa,and James R. Downinga,1

Departments of aPathology, cBiostatistics, and eOncology, and the bHartwell Center for Bioinformatics and Biotechnology, St. Jude Children’s ResearchHospital, 262 Danny Thomas Place, Memphis, TN 38105; and dCenter for Biomedical Informatics and Information Technology, National Cancer Institute,National Institutes of Health, 2115 E. Jefferson Street, Rockville, MD 20892

Edited by Janet D. Rowley, University of Chicago Medical Center, Chicago, IL, and approved June 11, 2009 (received for review March 20, 2009)

Pediatric de novo acute myeloid leukemia (AML) is an aggressivemalignancy with current therapy resulting in cure rates of only 60%.To better understand the cause of the marked heterogeneity intherapeutic response and to identify new prognostic markers andtherapeutic targets a comprehensive list of the genetic mutations thatunderlie the pathogenesis of AML is needed. To approach this goal,we examined diagnostic leukemic samples from a cohort of 111children with de novo AML using single-nucleotide-polymorphismmicroarrays and candidate gene resequencing. Our data demonstratethat, in contrast to pediatric acute lymphoblastic leukemia (ALL), denovo AML is characterized by a very low burden of genomic alter-ations, with a mean of only 2.38 somatic copy-number alterations perleukemia, and less than 1 nonsynonymous point mutation per leu-kemia in the 25 genes analyzed. Even more surprising was theobservation that 34% of the leukemias lacked any identifiable copy-number alterations, and 28% of the leukemias with recurrent trans-locations lacked any identifiable sequence or numerical abnormali-ties. The only exception to the presence of few mutations was acutemegakaryocytic leukemias, with the majority of these leukemiasbeing characterized by a high number of copy-number alterations butrare point mutations. Despite the low overall number of lesions acrossthe patient cohort, novel recurring regions of genetic alteration wereidentified that harbor known, and potential new cancer genes. Thesedata reflect a remarkably low burden of genomic alterations withinpediatric de novo AML, which is in stark contrast to most other humanmalignancies.

copy number alterations � single-nucleotide-polymorphism (SNP) �microarray � candidate gene resequencing � loss-of-heterozygosity (LOH)

Leukemia results from multiple genetic and epigenetic alterationswithin hematopoietic stem cells (HSCs) or progenitors that alter

their normal self-renewal, proliferation, differentiation, and apop-totic pathways (1–3). These alterations include point mutations,gene rearrangements, deletions, amplifications, and a diverse arrayof epigenetic changes that influence gene expression. For mostleukemias the full complement of oncogenic lesions remains to bedefined.

To define the lesions in acute leukemia, we recently usedsingle-nucleotide-polymorphism (SNP) microarrays to performgenome-wide DNA copy-number and loss-of-heterozygosity(LOH) analyses on primary leukemic blasts from pediatric patientswith acute lymphoblastic leukemia (ALL) (4, 5). These studiesidentified a high frequency of genetic alterations of key regulatorsof B lymphoid development and cell cycle in B-progenitor ALL.More recently, similar approaches have been used to explore thetype of copy-number alterations (CNAs) in adult myeloid malig-nancies (6–9), although these studies have used relatively lowresolution platforms.

We have now extended these analyses to pediatric de novo acutemyeloid leukemia (AML). AML comprises 15–20% of the acuteleukemias diagnosed in this age group and remains a challenging

disease with an inferior treatment outcome compared to ALL.Despite the introduction of new drugs and allogeneic bone marrowtransplantation, overall cure rates in most contemporary treatmentprotocols remain below 60% (10–12).

Like pediatric ALL, de novo AML is a heterogeneous diseasecomposed of different genetic subtypes with distinct clinical fea-tures and responses to contemporary therapies. The best charac-terized subtypes include the core-binding factor leukemias(t(8;21)[RUNX1(AML1)-RUNX1T1(ETO)] and inv(16)/t(16;16)[CBF�-MYH11]), cases with rearrangements of the MLLgene on chromosome 11q23, cases with distinct morphology in-cluding acute promyeloctic leukemia with t(15;17)[PML-RARA]and acute megakaryoblastic leukemia (FAB-M7), and cases withnormal cytogenetics. Although some cooperating lesions have beenidentified in AMLs, including point mutations or CNAs of NRAS,KRAS, FLT3, KIT, PTPN11, RUNX1, MLL, NPM1, CEBPA, andTP53 (13–17), the full complement of cooperating lesions remainsto be defined. The identification of the complete complement ofgenetic lesions within AML will not only improve our understand-ing of the molecular pathology of acute leukemia, but should alsodirectly impact diagnosis and risk stratification, and may lead to theidentification of new targets against which novel therapies can bedeveloped.

We report the results of a study of genome-wide DNA CNAs,LOH, and targeted gene resequencing analyses on primary leuke-mic blasts from 111 pediatric AML patients. Our data demonstratethat, in contrast to pediatric ALL, de novo AML is characterizedby a very low burden of genomic alterations. Despite the lownumber of lesions, however, unique recurring regions of geneticalteration were identified that harbor known, and potential newcancer genes. Moreover, the spectrum of CNAs and sequencemutations was found to vary significantly across the differentgenetic subtypes of AML.

ResultsAML Leukemic Cells Contain Few Copy-Number Alterations. As an initialapproach to define the total complement of genetic lesions inpediatric de novo AML, we performed high resolution genome-wide analysis on leukemic blasts from diagnostic bone marrowaspirates from 111 patients using both Affymetrix 100K and 500KSNP microarrays (combined resolution of 615K). The leukemias

Author contributions: I.R., C.G.M., S.A.S., and J.R.D. designed research; I.R., C.G.M., M.I.,X.S., J.C., R.G., Z.C., S.G., J.A., and S.C.R. performed research; X.S., S.B.P., and J.Z. contributednew reagents/analytic tools; I.R., C.G.M., M.I., X.S., J.C., J.M., S.G., S.B.P., X.C., C.O., J.Z., G.S.,R.C.R., J.E.R., S.C.R., and S.A.S. analyzed data; and I.R. and J.R.D. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

1To whom correspondence should be addressed. E-mail: james.downing@stjude.org.

This article contains supporting information online at www.pnas.org/cgi/content/full/0903142106/DCSupplemental.

12944–12949 � PNAS � August 4, 2009 � vol. 106 � no. 31 www.pnas.org�cgi�doi�10.1073�pnas.0903142106

Dow

nloa

ded

by g

uest

on

July

22,

202

1

included a representation of the different genetic subtypes of thepediatric de novo AML (in SI Appendix, Tables S1 and S2). Germline DNA was available for 65 of the patients allowing a definitiveidentification of somatically acquired CNAs. Two-hundred sevenCNAs were detected across the cohort with a mean number ofCNAs/patient of 2.38 (range 0–45), with no significant differencein the average number of gains (1.32, range 0–41) and losses (1.06,range 0–12) (Fig. 1 and SI Appendix, Table S2). The frequency ofCNAs was similar across the various AML genetic subtypes with theexception of FAB-M7, which had an average of 9.33 CNAs/patient,with the majority consisting of gains (SI Appendix, Table S2, P �0.013). Excluding FAB-M7 leukemias the average number ofCNAs/de novo AML patient was 1.76. Notably, 34% of the patientslacked any identifiable CNAs (SI Appendix, Table S2). Importantly,no association was detected between clinical outcome and thenumber of gains, losses, total CNAs, or the amount of the genomealtered in either univariate or multivariate analysis.

Gene Targets of Recurrent Copy-Number Alterations. Recurrent CNAswithin a patient cohort can be used to identify alterations ofpotential biological significance (driver versus passenger muta-tions). Surprisingly, when large regions (whole chromosomes orchromosome arms) of gains or losses were excluded the majority ofthe remaining lesions were nonrecurrent, being identified in only asingle patient (SI Appendix, Table S2). Using the genomic identi-fication of significant targets in cancer (GISTIC) algorithm (18),only 5 significant regions of gains and 13 regions of deletion werealtered more often than would be predicted by chance (Fig. 1 B andC). These included 1 broad lesion (gain of chromosome 22), 4 focallesions that were contained within broader lesions [8q24.21 (0.091Mb), 7p21.3 (3.523 Mb), 7q36.1 (1.372 Mb), and 9q22.3 (0.943Mb)], and 13 focal lesions containing �25 genes (3 loci), 6–25 genes(3 loci), 2–5 genes (3 loci), a single gene (3 loci), or lacking anannotated gene (1 locus) (Fig. 1, Fig. 2, and SI Appendix, Table S3).In addition, we identified 24 other recurrent lesions where the

Deletions

Amplifications

10-0.8510-1.1 10-1.6 10-2.6 10-4.7 10-8.810-0.810-110-1.4 10-2.2 10-3.8 10-7

CCDC26

ABCC4

88 genes

81 genes (*ERG, *TMPRSS2)

410 genes

Deletions

4 genes

12 genes (*CDKN2A)

3 genes

22 genes

RUNX1T1

TUSC1

13 genes (*XPA)

7 genes

4 genes (*MLL)

31 genes

No gene

17 genes (*CBFB)

5 genes (*MYH11)

0.25 0.25False Discovery Rate False Discovery Rate

2

3

4

5

6

7

8

910

11

121314

15-16

17-19

20-22X

log2ratio-2 0 +2patients

1

t(8;21) inv(16) MLL t(15;17) M7 Miscellaneous Normal

SN

Ps

by

chro

mo

som

e

B C

A

Fig. 1. DNA copy-number abnormalities in pediatric de novo AML. (A) Summary of CNA (log2 ratio) from a combined 100K and 500K Affymetrix SNP array analysisof diagnostic leukemia cells from 111 pediatric de novo AML patients. Each column represents a case and the 615K SNPs are arranged in rows according to chromosomallocation. Cases are arranged by subgroup. Diploid regions are white. Blue represents deletion, red amplification (see color scale). Gross changes can be observed forexample in chromosome 8 (10 cases with trisomy 8). (B) GISTIC (18) analysis of copy-number gains. (C) GISTIC analysis of copy-number losses. False discovery rate q valuesare plotted along the x axis with chromosomal position along the y axis. Altered regions with significance levels exceeding 0.25 (marked by vertical green line) aredeemed significant. Five significant regions of amplification and 13 significant regions of deletion were identified. Chromosomal position and relevant genes are shown foreach significant region on the right side of the plots. Genes indicated in blue are associated with known translocations, genes marked with * are cancer census genes (19).

Radtke et al. PNAS � August 4, 2009 � vol. 106 � no. 31 � 12945

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

July

22,

202

1

minimal altered regions (MAR) was less than 20 Mb, but fell belowstatistical significance by the GISTIC algorithm (SI Appendix, TableS4). Together, these 41 recurrent focal lesions contained a total of1,158 genes (290 from GISTIC peaks, 868 from other MARs), ofwhich only 30 (2.6%) were contained within the Cancer GeneCensus listing (http://www.sanger.ac.uk/genetics/CGP/Census/)(19), suggesting that new AML cancer genes are likely to existwithin the identified MAR. Twenty-one of the 30 cancer consensusgenes have been previously implicated in AML as targets oftranslocations or sequence mutation including CBFB, CDKN2A,ELL, ERG, ETV6, FLI1, GMPS, JAK3, LYL1, MLF1, MLL,MLLT1, MLLT4, MYH11, MYST4, NPM1, NSD1, PBX1, RUNX1,SH3GL1, and WT1. A number of other cancer consensus geneswere contained within nonrecurrent CNAs suggesting that a subsetof the nonrecurrent lesions may provide a selective advantage to theleukemic cell and thus constitute driver mutations (SI Appendix,Tables S5 and S6).

When broad and focal CNAs were combined the most commonaffected region was on chromosome 8, band q24, which showed acopy-number gain in 14% of patients (Fig. 2 and SI Appendix, TableS6). Focal gains were identified in 4 patients, and in an additional11 patients an increase in copy number was observed either as aresult of a larger region of chromosomal gains (1 case), or as theresult of trisomy 8. No significant difference in the frequency of thisalteration was detected across the different AML genetic subtypes.Interestingly, the MAR targets a locus that was originally identifiedin a forward genetic screen to identify genes that are required forretinoic acid (RA)-induced myeloid cell differentiation (20, 21).Retroviral integration into this locus disrupted RA-induced differ-entiation. The region contains a putative gene referred to asCCDC26, and a highly conserved large intervening noncodingRNA of unknown function (22). Which of these transcripts nor-mally functions in myeloid differentiation, and whether the iden-tified copy-number changes alter their expression and functionremain to be determined.

The MARs in 8 other recurrent CNAs were each limited to asingle gene in at least 1 patient, thus identifying the gene as a targetof the lesion (SI Appendix, Table S6). These included monoallelicdeletions involving RUNX1T1 (ETO), a known target of the AML-associated t(8;21) translocations, MLL involved in 11q23 translo-cations, FAM20C, which is expressed in hematopoietic cells and isthe target of mutations in Raine’s syndrome (23) a lethal osteo-sclerotic bone disease, and the putative tumor suppressors TUSC1(24) and BCOR (25), and amplifications of ABCC4, encoding amultidrug resistance membrane protein (26), MLLT4, a target of

the AML-associated t(6;11) translocations (27), and PRDM5 (28),a putative tumor suppressor. The CNAs of RUNX1T1 were ob-served in three t(8;21) leukemias, whereas the MLLT4 alterationwas detected in 2 patients, 1 with normal cytogenetics and the otherwith a complex karyotype that did not include a detectable struc-tural alteration of 6q (SI Appendix, Fig. S1).

In addition to RUNX1T1 and MLLT4, 7 other genes that are thetargets of AML-associated chromosomal translocations were af-fected by recurrent CNAs, including MYH11, CBF�, MLL, NSD1,MLF1, ERG, and MYST4 (SI Appendix, Tables S3 and S4). Priorstudies have demonstrated that focal micro deletions and amplifi-cations can occur near the breakpoints of chromosomal transloca-tions (4, 29). Consistent with these observations, the CNAs ofMYH11 and CBF� were observed in cases with inv(16)/t(16;16), andthe CNAs of MLL were seen in a subset of cases with MLLtranslocations (SI Appendix, Fig. S1 and S2). On the basis of theseobservations, we examined whether any other recurrent CNAs ofgenes were the result of cryptic translocations. Our analysis iden-tified 4 leukemias that expressed the t(5;11)-encoded NUP98-NSD1chimeric transcript, with 2 having CNAs adjacent to 1 or both genes(SI Appendix, Fig. S2) and 2 cases that expressed the t(6;11)-encoded MLL-MLLT4(AF6) chimeric transcript, with associatedCNAs involving both genes (SI Appendix, Fig. S2). These dataindicated that at least 14% of the cytogenetically normal ormiscellaneous karyotype subgroups within this cohort contained acryptic translocation. Whether the CNAs involving MLF1, ERG,and MYST4 are also the result of cryptic translocations remains tobe determined.

Copy Neutral Loss of Heterozygosity. The genotypes generated by theAffymetrix SNP array platform enable the detection of regions ofsomatic copy neutral-loss of heterozygosity (CN-LOH), which mayidentify reduplication of mutated or aberrantly methylated genesthat contribute to tumorigenesis. Prior publications have identifiedCN-LOH of 11p (involving WT1), 13q (FLT3), and 19q (CEPBA)in AML (30, 31), and have suggested that CN-LOH is the mostfrequent lesion in myeloid malignancies, occurring in up to 48% ofcases (6, 7). However, these studies have often failed to compare theSNP genotypes of the tumor to the patient’s own constitutionalDNA and consequently have been unable to definitively distinguishsomatic CN-LOH from inherited homozygosity. We initially per-formed paired CN-LOH analysis for 60 patients with matchedconstitutional and leukemia cell DNA and identified only 6 leuke-mias that contained somatic CN-LOH that were defined by 3 ormore contiguous SNPs (SI Appendix, Fig. S3 and Table S7). These

centtel

8q24.21

e1 e2 e3 e4e1a

50 kb

ORF

CCDC26

AML-AE-#20

AML-misc-#5AML-AE-#8

AML-N-#15

n=11NB-4

HL60

lincRNA

Fig. 2. Amplification of CCDC26 in pediatric AML. The genomic organization of CCDC26 is illustrated relative to the telomere (tel) and centromere (cent) ofchromosome 8, with exons labeled by lowercase e, and an alternative transcript initiating from exon e1a. The vertical blue lines show the location of the SNPs oncombined 100K and 500K Affymetrix arrays. The putative ORF encoded by exons e3 and e4 is shown by a dotted line with the arrowhead illustrating the direction fortranscription. The vertical red arrow marks the integration site found in a retroviral integration screen of retinoic acid resistant myeloid cell lines (20). The green boxrepresents the 10-kb lincRNA, identified in ref. 22. The extent of amplification across this genomic locus for each case is illustrated by a horizontal red line with the caseIDnumbernext tothe lineor thenumberofcases thatcontained largeamplifications that includedtheentire locus.TwoAMLcell lines,HL-60andNB-4,werealso foundto have focal amplification of this locus.

12946 � www.pnas.org�cgi�doi�10.1073�pnas.0903142106 Radtke et al.

Dow

nloa

ded

by g

uest

on

July

22,

202

1

leukemias included 5 cases with large regions of CN-LOH (chro-mosome 21, n � 1; 11p, n � 2; 9p, n � 1; and 15q, n � 1) along witha single case with a focal region of CN-LOH on chromosome 8q11that contained no genes.

We next performed unpaired CN-LOH analysis for the 51patients that lacked constitutional DNA by using the 60 constitu-tional DNA samples from our AML cohort as a reference pool toeliminate common regions of inherited homozygosity (32). Becausethis approach can fail to exclude private regions of inherited LOH,we decided to limit our calls to regions of CN-LOH that were �20Mb, thus significantly improving the accuracy of identifying truelesions (SI Appendix, Fig. S3 and S7). When the unpaired and pairedanalyses were combined, we identified somatic CN-LOH in only13% of the entire cohort, including chromosome 13 (n � 4), 11p(n � 3), 9p (n � 2), and 6p, 8q, 15q, 17q, and chromosome 21 (1patient each). The 4 leukemias with chromosome 13 CN-LOH hadhomozygous FLT3-ITD mutations (see below), and the two leuke-mias with 9p CN-LOH had homozygous deletions of CDKN2A/Bthat were included in the regions of LOH. Thus, somaticallyacquired CN-LOH occurs in 13% of de novo AML, with many ofthe identified lesions targeting genes previously shown to beinvolved in the pathogenesis of AML.

Integration of Sequence Mutation and Copy-Number Alteration Data. Togain further insights into the complement of oncogenic lesions inAML, we sequenced genes previously found to be mutated in AML(AML-associated cancer genes, including NRAS, KRAS, PTPN11,FLT3, KIT, RUNX1, CEBPA, ETV6, NPM1, TP53, CDKN2A/B,GATA1, and MLL), along with a subset of the genes targeted byCNAs in this cohort (CCDC26, FAM20C, TUSC1, ERG, IKZF1,PAXIP1, PTEN, FBXW7, BTG1, XRCC2, BRAF, and LYL1).Surprisingly, nonsynonymous somatic sequence mutations wereonly identified in the AML-associated cancer genes (Fig. 3 and seeSI Appendix, SI Text).

Forty-seven leukemias contained somatic activating mutations

of genes within the RAS-signaling pathway or upstream receptortyrosine kinases (NRAS, n � 26; KRAS, n � 2; PTPN11, n � 3;FLT3-ITD, n � 15; and KIT, n � 5). Consistent with previouslypublished data, mutations of NRAS were frequent in the corebinding factor (44%) and MLL rearranged AMLs (24%) (33), butwere uncommon in other AML subtypes (12%) (P � 0.015) (Fig.3 and SI Appendix, Table S2). By contrast, mutations in FLT3 wereonly found in acute promyelocytic leukemia and in patients with anormal or miscellaneous karyotype (Fig. 3, P � 0.0015). Themutations in these genes were heterozygous with the exception of4 leukemias that contained homozygous FLT3 mutations associ-ated with chromosome 13 CN-LOH. Although mutations in thesegenes are usually mutually exclusive (33), two t(8;21) containingAMLs had mutation in both NRAS and KIT (Fig. 3).

Mutations in CEBPA, RUNX1, ETV6, and MLL partial tandemduplications (MLL-PTD) were also common (Fig. 3 and SI Appen-dix, Table S2) and were more frequent in AMLs with either anormal or miscellaneous karyotype. For patients with matchedconstitutional DNA, we were able to show that each mutation wassomatic except for CEBPA, which was germ line in 3 of 6 patients(SI Appendix, Table S2). Germ line mutations in CEBPA have beenpreviously identified in rare pedigrees of familial AML (34). Pointmutations in the other analyzed AML-associated cancer genes wererare in this patient cohort (Fig. 3 and SI Appendix, Table S2).

Importantly, 42% of patients had no point mutations in the 25genes analyzed, 38% had only 1 gene mutated, 13% had 2 genesmutated, and only 7% had 3 genes with point mutations. Moreover,the frequency of mutations varied across the AML genetic subtypeswith a higher number of lesions observed in the cases with normalor miscellaneous karyotypes as compared to the other subtypes(1.36 sequence mutations/patients in normal or miscellaneouskaryotypes versus 0.54/patient in other subtypes, P � 0.0001).

Integration of the CNA/CN-LOH analysis and candidate generesequencing revealed several important findings (Fig. 3 and SIAppendix, Table S2). First, 28% of leukemias with recurrent trans-

Fig. 3. Integrated analysis of copy-number alterations, CN-LOH, and point mutations. Each column illustrates the results of the integrated mutational analysis fora single patient, with the patients grouped into the 7 genetic AML subtypes, which are color coded as illustrated. The rows depict the presence of mutations fromsequenceanalysis (Top13rows)orcopy-numberalterations (Bottom2rows).Thepresenceofamutation inagene is shownasabox,withthetypeofmutation indicatedby the color of the box as shown. The presence of copy-number alterations is shown in the Bottom 2 rows with amplifications in red, and deletions in green, and theintensity of the color corresponding to the number of copy-number alterations according to the scale shown at bottom right.

Radtke et al. PNAS � August 4, 2009 � vol. 106 � no. 31 � 12947

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

July

22,

202

1

locations lacked any identifiable sequence or numerical abnormal-ities. Second, although the FAB-M7 leukemias had the highestnumber of CNAs per case, they rarely contained point mutations,with only a single patient having a GATA1 mutation. Third, theleukemic cells from AML patients with normal or miscellaneouskaryotypes each contained 1 or more alterations, with over 40% ofthese leukemias containing both CNAs and sequence mutations.These data demonstrate that genomewide copy-number alterationand target gene resequencing complement routine cytogeneticanalysis and identify new genetic lesions in more than half ofpediatric de novo AMLs (Fig. 3). However, although new lesionsare identified, the total burden of genomic alterations is low, withonly 21% of the AML having �5 lesions (Fig. 4 and SI Appendix, TableS2). This is in stark contrast to results from pediatric ALL where 77%of cases have �5 lesions excluding point mutations (4, 5).

DiscussionOur analyses of CNAs, CN-LOH, and sequence mutations inpediatric de novo AML identified remarkably few somaticgenetic alterations within the leukemia cells. We identified amean of only 2.38 somatic CNAs per leukemia and less than 1nonsynonymous point mutation per leukemia in the 25 genesanalyzed. Moreover, somatic CN-LOH was observed in only13% of patients. Even more surprising was the observation that34% of the leukemias lacked any identifiable CNAs, and 28% ofthe leukemias with recurrent translocation lacked any identifi-able sequence or numerical abnormalities. The only exception tothe presence of few mutations was acute megakaryocytic leu-kemias, with the majority of these leukemias being characterizedby a high number of CNAs but rare point mutations.

These data reflect a very low burden of genomic alterations inpediatric de novo AML, which is in stark contrast to most othercancers. Recent studies using similar approaches in pediatric ALL,and adult cancers [lung (35), pancreatic (36), glioblastoma multi-forme (37, 38), breast and colon (39, 40)], have demonstrated amuch higher number of CNAs and point mutations, with themajority of these cancers containing a very large number ofmutations. Although the possibility exists that AMLs may containsmall regions of CNAs that are below the resolution of detectionusing the combined 100K and 500K SNP platforms, this appearsunlikely on the basis of the absence of smaller lesions in pediatricand adult AMLs analyzed using the higher resolution Affymetrix6.0 microarrays.

It is generally believed that the large number of mutations withincancer cells arise either from inherent genomic instability or as theresult of a single mutational crisis. Although we have performedgenomewide analysis for only a single type of mutation (CNAs), andhave sequenced only a limited number of potential cancer genes,our data suggest that AML may arise in the absence of an increasedmutational rate. Moreover, our data raise the possibility that thedevelopment of AML may require fewer genetic alterations thanother cancers and that a very limited number of biological processesmay need to be altered in hematopoietic stem cells, multipotentialprogenitors, or committed myeloid progenitors to convert themfrom a normal cell into an acute myeloid leukemic cell. Analternative but not mutually exclusive possibility is that epigeneticchanges may play a more predominant role in AML, working inconcert with genetic alterations to alter a wider range of biologicalprocesses to induce overt leukemia. Detailed genomewide epige-netic analysis and whole genome resequencing will ultimately berequired to determine the range of mutations and epigeneticchanges required for the development of AML. However, therecent sequence of all coding exons in the DNA of an AMLpatient’s leukemia cells revealed only 10 somatically acquirednonsynonymous mutations in the coding region of annotated genes(41), suggesting that even with single base pair resolution fewmutations may be the rule in AML.

Although the majority of the focal CNAs occurred in only a singlecase, the recurrent lesions targeted 30 genes known to be involvedin cancer, including 21 that have been previously implicated inAML. In addition, many of the recurrent lesions lacked any cancerconsensus genes, suggesting that new AML cancer genes existwithin the identified regions. A few of the latter lesions target singlegenes, thereby directly implicating the target genes in AML patho-genesis. The top-ranked recurrent lesion in this category targeteda MAR that contained a putative gene CCDC26, along with arecently described �10-kb highly conserved noncoding RNA (lin-cRNA-CCDC26) that resides within an intron of CCDC26 (22).Interestingly, the lincRNA-CCDC26 resides within 1 kb of theretroviral integration site that disrupted myeloid differentiation(20). Moreover, multiple transcripts are encoded within this locusand some are normally downregulated during myeloid cell differ-entiation (20, 21). Exactly how the AML-associated CNAs affectexpression of these transcriptional units and whether their alter-ations contribute to the development of AML remains to bedetermined. Nevertheless, the locus is a good candidate to contain

A

B

Acute Myeloid Leukemia

Acute Lymphoblastic LeukemiaCytogenetics+ 615K SNP analysis

Cytogenetics+ 615K SNP analysis

Cytogenetics+ 615K SNP array+ sequence mutation analysis

Cytogenetics

Cytogenetics

Fig. 4. Spectrum of number of lesions per case in pediatric AML and ALL. Percentage of cases containing zero (normal), 1, 2, 3, 4, 5, or �5 alterations basedon cytogenetics, cytogenetics plus CNAs using 615K SNP arrays, or cytogenetics, 615K SNP arrays, and targeted gene resequencing of 25 candidate genes in AAML (n � 111 cases) and B ALL (n � 212, cases from refs. 4, 5).

12948 � www.pnas.org�cgi�doi�10.1073�pnas.0903142106 Radtke et al.

Dow

nloa

ded

by g

uest

on

July

22,

202

1

gene(s) whose alterations may contribute to the development ofAML. Other genes implicated include the known or putative tumorsuppressors TUSC1, BCOR, and PRDM5, ABCC4, which encodesa multidrug resistance membrane protein, and FAM20C, a geneexpressed in hematopoietic cells that is involved in the pathogenesisof osteosclerotic bone disease. Determining how genetic alterationsof these genes contribute to the AML pathogenesis will requiredirect functional studies on the biological role of the encoded geneproducts in normal and leukemic hematopoiesis.

The limited number of recurrent lesions and the low frequencyof individual recurrent lesions across the cohort are notable andlikely reflect the marked heterogeneity among the AML cases thatwere analyzed. Within the patient cohort, 7 distinct genetic subtypesof de novo AML were included. Each subtype might arise from adifferent combination of genetic lesions, with the complement oflesions being heavily influenced by the initiating event (a chromo-somal translocation in many pediatric de novo AMLs). Our inte-grated analysis of cytogenetics, CNAs, and sequence mutations isconsistent with this interpretation, with marked difference in thespectrum of changes observed across the AML genetic subtypes.Extending the detailed genomic characterization of AML to a largenumber of leukemias for each individual known genetic subtype ofde novo AML should yield valuable information on the spectrumof mutations in this disease. It will also be of interest to see whetherthe limited number of CNAs and point mutations found in de novopediatric AML, will also be observed in other subtypes of AMLincluding myelodysplasia-related AML, and secondary AML thatresults from prior chemotherapy and/or radiation therapy.

MethodsPatients and Samples. Leukemic samples from 111 pediatric AML patients treatedat St. Jude Children’s Research Hospital (SJCRH) were studied. Written informedconsent and institutional-review-board approval was obtained for each patient.No commercial entity was involved in the conduct of the study, the analysis orstorage of the data, or the preparation of the manuscript. The authors vouch forthe completeness and accuracy of the data and analysis.

Genomic Analyses. DNA extracted from leukemic cells obtained at diagnosis andfromsamplesobtainedduringremissionwasgenotypedwith theuseof50KHind240,50KXba240,250KSty,and250KNspSNParrays (Affymetrix) foreachsample.Data from all 4 arrays were combined before analysis for the presence of CNAs,resulting in an average intermarker distance of less than 5 kb. SNP array analysisof copy-number alterations and LOH was performed as previously reported (4)and is described in SI. A subset of the identified copy-number alterations wasvalidated using fluorescence in situ hybridization (FISH) or quantitative genomicPCR. The primary SNP data are available upon request. Identifiers for each caseare listed in SI Appendix, Table S1.

Genomic Resequencing of Candidate AML Cancer Genes. Genomic sequencing ofexons and adjacent splice sites was performed on the genes listed in the text. Adetailed description of the sequencing methods are in SI.

ACKNOWLEDGMENTS. This study was supported by a Cancer Center Core Grant21765 from the National Cancer Institute, a Leukemia and Lymphoma SocietySpecialized Center of Research Grant (LLS7015, to J.R.D.), a grant from theNational Health and Medical Research Council of Australia (C.G.M.), and theAmerican Lebanese Syrian Associated Charities of St. Jude Children’s ResearchHospital. We thank Claire Boltz, Letha Phillips, and James Dalton for technicalassistance.

1. Kelly LM, Gilliland DG (2002) Genetics of myeloid leukemias. Annu Rev Genomics HumGenet 3:179–198.

2. Reya T, Morrison SJ, Clarke MF, Weissman IL (2001) Stem cells, cancer, and cancer stemcells. Nature 414:105–111.

3. Schimmer AD (2006) Induction of apoptosis in lymphoid and myeloid leukemia. CurrOncol Rep 8:430–436.

4. Mullighan CG, et al. (2007) Genome-wide analysis of genetic alterations in acutelymphoblastic leukaemia. Nature 446:758–764.

5. Mullighan CG, et al. (2008) BCR-ABL1 lymphoblastic leukaemia is characterized by thedeletion of Ikaros. Nature 453:110–114.

6. Dunbar AJ, et al. (2008) 250K single nucleotide polymorphism array karyotypingidentifies acquired uniparental disomy and homozygous mutations, including novelmissense substitutions of c-Cbl, in myeloid malignancies. Cancer Res 68:10349–10357.

7. Gondek LP, et al. (2008) Chromosomal lesions and uniparental disomy detected by SNParrays in MDS, MDS/MPD, and MDS-derived AML. Blood 111:1534–1542.

8. Raghavan M, et al. (2005) Genome-wide single nucleotide polymorphism analysisreveals frequent partial uniparental disomy due to somatic recombination in acutemyeloid leukemias. Cancer Res 65:375–378.

9. Rucker FG, et al. (2006) Disclosure of candidate genes in acute myeloid leukemia withcomplex karyotypes using microarray-based molecular characterization. J Clin Oncol24:3887–3894.

10. Ravindranath Y, et al. (2005) Pediatric Oncology Group (POG) studies of acute myeloidleukemia (AML): A review of four consecutive childhood AML trials conducted be-tween 1981 and 2000. Leukemia 19:2101–2116.

11. Ribeiro RC, et al. (2005) Successive clinical trials for childhood acute myeloid leukemiaat St Jude Children’s Research Hospital, from 1980 to 2000. Leukemia 19:2125–2129.

12. Rubnitz JE, et al. (2007) Prognostic factors and outcome of recurrence in childhoodacute myeloid leukemia. Cancer 109:157–163.

13. Bacher U, Haferlach T, Schoch C, Kern W, Schnittger S (2006) Implications of NRASmutations in AML: A study of 2502 patients. Blood 107:3847–3853.

14. Frohling S, et al. (2004) CEBPA mutations in younger adults with acute myeloidleukemia and normal cytogenetics: Prognostic relevance and analysis of cooperatingmutations. J Clin Oncol 22:624–633.

15. Loh ML, et al. (2004) PTPN11 mutations in pediatric patients with acute myeloidleukemia: Results from the Children’s Cancer Group. Leukemia 18:1831–1834.

16. Renneville A, et al. (2008) Cooperating gene mutations in acute myeloid leukemia: Areview of the literature. Leukemia 22:915–931.

17. Zwaan CM, et al. (2003) FLT3 internal tandem duplication in 234 children with acutemyeloid leukemia: Prognostic significance and relation to cellular drug resistance.Blood 102:2387–2394.

18. BeroukhimR,etal. (2007)Assessingthesignificanceofchromosomalaberrations incancer:Methodology and application to glioma. Proc Natl Acad Sci USA 104:20007–20012.

19. Futreal PA, et al. (2004) A census of human cancer genes. Nat Rev Cancer 4:177–183.20. Yin W, Rossin A, Clifford JL, Gronemeyer H (2006) Co-resistance to retinoic acid and

TRAIL by insertion mutagenesis into RAM. Oncogene 25:3735–3744.21. Hirano T, et al. (2008) Genes encoded within 8q24 on the amplicon of a large

extrachromosomal element are selectively repressed during the terminal differentia-tion of HL-60 cells. Mutat Res 640:97–106.

22. Guttman M, et al. (2009) Chromatin signature reveals over a thousand highly conservedlarge non-coding RNAs in mammals. Nature 458:223–227.

23. Simpson MA, et al. (2007) Mutations in FAM20C are associated with lethal osteoscle-rotic bone dysplasia (Raine syndrome), highlighting a crucial molecule in bone devel-opment. Am J Hum Genet 81:906–912.

24. Shan Z, Parker T, Wiest JS (2004) Identifying novel homozygous deletions by micro-satellite analysis and characterization of tumor suppressor candidate 1 gene, TUSC1,on chromosome 9p in human lung cancer. Oncogene 23:6612–6620.

25. Huynh KD, Fischle W, Verdin E, Bardwell VJ (2000) BCoR, a novel corepressor involvedin BCL-6 repression. Genes Dev 14:1810–1823.

26. Rius M, Hummel-Eisenbeiss J, Keppler D (2008) ATP-dependent transport of leukotri-enes B4 and C4 by the multidrug resistance protein ABCC4 (MRP4). J Pharmacol ExpTher 324:86–94.

27. Taki T, et al. (1996) Fusion of the MLL gene with two different genes, AF-6 andAF-5alpha, by a complex translocation involving chromosomes 5, 6, 8 and 11 in infantleukemia. Oncogene 13:2121–2130.

28. Deng Q, Huang S (2004) PRDM5 is silenced in human cancers and has growth suppres-sive activities. Oncogene 23:4903–4910.

29. Zhang Y, Rowley JD (2006) Chromatin structural elements and chromosomal translo-cations in leukemia. DNA Repair (Amst) 5:1282–1297.

30. Gupta M, et al. (2008) Novel regions of acquired uniparental disomy discovered inacute myeloid leukemia. Genes Chromosomes Cancer 47:729–739.

31. Serrano E, et al. (2008) Uniparental disomy may be associated with microsatelliteinstability in acute myeloid leukemia (AML) with a normal karyotype. Leuk Lymphoma49:1178–1183.

32. Beroukhim R, et al. (2006) Inferring loss-of-heterozygosity from unpaired tumors usinghigh-density oligonucleotide SNP arrays. PLoS Comput Biol 2:e41.

33. Nanri T, et al. (2005) Mutations in the receptor tyrosine kinase pathway are associatedwith clinical outcome in patients with acute myeloblastic leukemia harboringt(8;21)(q22;q22). Leukemia 19:1361–1366.

34. Owen C, Barnett M, Fitzgibbon J (2008) Familial myelodysplasia and acute myeloidleukaemia: A review. Br J Haematol 140:123–132.

35. Weir BA, et al. (2007) Characterizing the cancer genome in lung adenocarcinoma.Nature 450:893–898.

36. Jones S, et al. (2008) Core signaling pathways in human pancreatic cancers revealed byglobal genomic analyses. Science 321:1801–1806.

37. Anonymous (2008) Comprehensive genomic characterization defines human glioblas-toma genes and core pathways. Nature 455:1061–1068.

38. Parsons DW, et al. (2008) An integrated genomic analysis of human glioblastomamultiforme. Science 321:1807–1812.

39. Leary RJ, et al. (2008) Integrated analysis of homozygous deletions, focal amplifica-tions, and sequence alterations in breast and colorectal cancers. Proc Natl Acad Sci USA105:16224–16229.

40. Sjoblom T, et al. (2006) The consensus coding sequences of human breast and colorectalcancers. Science 314:268–274.

41. Ley TJ, et al. (2008) DNA sequencing of a cytogenetically normal acute myeloidleukaemia genome. Nature 456:66–72.

Radtke et al. PNAS � August 4, 2009 � vol. 106 � no. 31 � 12949

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

July

22,

202

1