1

Themostglobalcancerincidenceinwomen

Rank Cancer Newcasesdiagnosedin2018 %ofallcancers(excl.non-melanomaskincancer)

1 Breast 2,088,849 25.4

2 Colorectal 794,958 9.7

3 Lung 725,352 8.8

4 Cervixuteri 569,847 6.9

•  Breast cancer causes the greatest number of cancer-relateddeaths among women. In 2018, it is esGmated that 627,000womendiedfrombreastcancer–thatisapproximately15%ofallcancerdeathsamongwomen.

•  While breast cancer rates are higher among women in moredeveloped regions, rates are increasing in nearly every regionglobally.

BreastCancer

CatalanoM,V.,Bertaglia,V.,Tariq,N.,Califano,R.(2014).TreatmentofAdvancedBreastCancer(ABC):TheExpandingLandscapeofTargetedTherapies.JCancerBiolRes,2(1),1036.

2

40% 20% 10-15% 15-20%

TripleNegaGve

Humanepidermalgrowthfactor2

HER2PosiGveLuminalA LuminalB

Percentageatdiagnosis

Receptorexpression

Treatmentstrategies

Estrogensandprogesterone

ChemotherapyAnG-HER2therapies

HormonaltherapiesNoveltargettherapies

DiversityofBreastCancer:Subtypes

3

1.   EarlydetecKon2.   New progression mechanism for

targetedtherapies(e.g.TNBC)3.   Heterogeneity4.   DrugResistance5.   Companiondiagnosis6.   Racialdisparity

Clinicalunmetneedsinbreastcancer

4

Stage0

Benign StageI

StageII

StageIII

StageIV

TNMStageFiveyearsurvivalrate

>95%

About90%

About25%

About70%

Progression&DiversityofBreastCancer

DCIS:Ductalcarcinomainsitu

TNM Stage T = size of primary tumor N = the extent of spread to nearby lymph nodes M = presence or absence of distant metastases

5

DiversityofBreastCancer:RacialDisparity

SurveillanceResearch,2017

Year Year

Rateper100,000

Rateper100,000

Disease Biomarkers From Liquid Biopsy to Precision Medicine

Liquid Biopsy

•  Predictive, •  Personalized, •  Preventive, •  Participatory

“Doctors have always recognized that every patient is unique, and doctors have always tried to tailor their treatments as best they can to individuals. You can match a blood transfusion to a blood type — that was an important discovery. What if matching a cancer cure to our genetic code was just as easy, just as standard? What if figuring out the right dose of medicine was as simple as taking our temperature?” - President Obama, January 30, 2015

LiquidBiopsy:•Lessinvasive,•lesscostly,•lessrisky,•containmoredynamicinformaGonthanconvenGonalGssuebiopsies.

Varietyofbiomarkersexistedinbodilyfluidssuchasblood,saliva,urine,ascites,etc

8

miR-331andmiR-21=>gastriccancer

8miRNAs=>coloncancer

MicroRNA(miRNA)1.  Non-codingRNA2.  17-25nucleoGdes3.  Post-transcripGonalregulaGon:

base-pairwithmRNAsandsilencethosemRNAs

4.  Appeartotargetabout60%ofthegenesofhuman

miRNABiomarkersinLiquidBiopsy

9

1.  IdenGfypossiblebiomarkersforearlydetecGonofbreastcancerfromliquidbiopsy(e.g.miRNAs).

2.  Generateprofilingonthevariousstagesofbreastcancer. IdenGfy more insigheul informaGonregardingthecriGcalfactorsthatprogresscancertothenextstage.

Problemstobesolved

Comprehensivedata(containingallvariaGon)+High-throughputtechniques(highsensiGvity)+Properdataanalysis(lowsamplenumberissue)+ArGficialintelligenceforopGmizaGon(predictable)

10

Currentbiomarkersforbreastcancer

TumorMarker N(Total) SensiKvity

CA15-3 35(145) 24.1%

CA27.29 37(145) 25.5%

CEA 27(145) 18.6%

BreastCancer:3tumormarkers•  canceranGgen15-3(CA15-3),•  canceranGgen27.29(CA27.29),and•  carcinoembryonicanGgen(CEA)

Hou,MF,etal.(1999).EvaluaGonofserumCA27.29,CA15-3andCEAinpaGentswithbreastcancer.KaohsiungJMedSci.,23(1),pp.88-93.

Comprehensivedata

11

Wen-HongKuo,MD;PhD

Noncancer Benign 10 DCIS 20

CancerStage

TotalI II III IV

Subtypes

LuminalA 8 8 12 5 33

LuminalB 8 8 12 5 33

HER2+ 8 8 12 5 33

TripleNegaGve 8 8 12 4 32

Total 32 32 48 19 131

Serum 300µL� 3D-Gene®

①

Addition

Vortex & cfg.

Water layer

Organic layer RNA purification

(96 well)

Extraction Concentration

Serum miRNA

4µL

RNA extraction� RNA purification�

miRNA

Exosome

12

TakahiroOchiya,PhD�

High-throughputtechniques1.  miRNAextracGonfromserum2.  miRNAanalysiswithmicroarray

miRNAoligochip(2565miRNA,miRBasever.21)

RNA 2µL

Fluorescent labeling

Hybridization

Wash

Digitizing

Scan RNA labeling �

DNA chip analysis� Analysis�3D-Gene®

13

High-throughputtechniques1.  miRNAextracGonfromserum2.  miRNAanalysiswithmicroarray

miRNAexpressiondata(Rawdata)

1.DifferenGalexpressionanalysis2.ElasGcnet(LASSO&Ridge)

1.Supportvectormachine(SVM)2.Lineardiscriminantanalysis(LDA)3.Generalizedlinearmodel(GLM)

Modeling

NormalizedData

FeaturemiRNAs

1.VariancestabilizingnormalizaGon(VSN)

2.Spike-inVSN

3.CrossNorm

Ochiyaet.al.

NormalizedData

NA-imputedData

FeaturemiRNAs

14

RawData

NA-handledData

1.Removing2.ImputaGon(trimmedmean)

ThisStudy

PredicGonmodel PredicGonmodel

PotenKalbiomarkers&PredicKonmodels

1.AccuracyFiltering2.PredicGonmodeling

FurtherModeling

ImputaGon(smallvalue)

CalibraGonwithnegaGvecontrolsandinternalcontrols

DifferenGalexpressionanalysisandexpressionlevelfiltering

FeaturesselecKon

NA-handling

NormalizaKon

DataAnalysisPipeline

Chen-AnTsaiTakahiroOchiya �

15

-log 1

0(P-value)

log2(foldchange)

-log 1

0(P-value)

log2(foldchange)

-log 1

0(P-value)

log2(foldchange)

DifferenKalExpressionAnalyses

AdjustedP-values<0.05(Benjamini–Hochbergprocedure) Significantlog2foldchange=1.5

VSN Spike-inVSN CrossNorm

103probesareselected 437probesareselected

269probesareselected

16

miRNAsSelectedbyElasKcNetRegression

hsa-miR-4783-5p

miRNAAnalysisFlowChart RawData

NormalizedData

FeaturemiRNAs

1.  DifferenGalExpressionAnalysis2.  ElasGcnet(LASSO&Ridge)

1.  SupportVectorMachine(SVM)2.  Lineardiscriminantanalysis(LDA)3.  Generalizedlinearmodel(GLM)

PredicGonmodels

NormalizaGon

FeaturesselecGon

Modeling

EvaluaGon:LeaveoneoutCrossvalidaGon

18

WholemiRNA 2565

AperremovingNA 2462

NormalizaGon VSN Spike-inVSN

CrossNorm

DifferenGalExpression

0103

307130

0269

ElasGcNet 3231

5014

022

Modelingwithatmost3selectedmiRNAoreachsinglemiRNAs->Construct17,423,153models->Evaluatewith10-foldcrossvalidaGon

ResultsofmiRNAthroughAnalysisPipeline

103 125

1221

1

19

SVM LDA GLM

ModelingwithSelectedmiRNAsSelecGngthemiRNAswithpredicGonability

sensiGvity&specificity>85%

20

hsa-miR-61xxhsa-miR-61xhsa-miR-66xxhsa-miR-42xxhsa-miR-42yyhsa-miR-7xxhsa-miR-3xxhsa-miR-6xxxhsa-miR-5xxhsa-miR-12xxhsa-miR-67xx

1

11

2

2 1

1 1

SVM

LDA GLM

OverlapandConsistencyofEachModelingMethod

SelectedmiRNAswithPredicKonAbility

⇒ 18miRNAs(overlapfrom4methods)

21

NA-omiMngdatawithVSN

PCAandheatmapwithSelected18miRNAs

22

NA-omiMngdatawithVSN

PCAplotsshowselected18miRNAsfit-inearlydetecKon

Further modeling

23

Aper previous predicGon modeling, we used the union of selecGon from eachpipeline tobuildmorepredicGonmodels. Fisher’s lineardiscriminantanalysis (LDA)was performedwith each of thesemiRNAmarker or a combinaGon of atmost sixmiRNAmarkers.To evaluaGon the predicGon performance, 10-fold cross validaGonwere applied toeachmodel.Weseparatedthedatainto10groups,builtthemodelwith9groupsandused the residual group as tesGng cohort to calculate the predicGon accuracy,sensiGvity and specificity. Aper repeaGng the esGmaGon process with differenttesGng group 10 Gmes, the average values of each test result were calculated formodelevaluaGon.TheresulGngvaluesofthediscriminantfuncGonswereusedtopreparethediagnosGcindex.Indexscore≥0:breastcancerIndexscore<0:non-breastcancerorotherclinicalcondiGons

Modelingwith1miRNA(hsa-miR-614)

FurtherModeling(with1miRNA)

24

Threshold:0.9850Specificity:0.8656±0.0964SensiGvity:0.8700±0.0881Accuracy:0.8683±0.0552AreaUnderCurve:0.9320

PredicGonscore=11.3262-1.5682xhsa-miR-614

Threshold:1.0250Specificity:0.9156±0.0822SensiGvity:0.9307±0.0621Accuracy:0.9250±0.0470AreaUnderCurve:0.9660

FurtherModeling(with3miRNAs)

25

PredicGonscore=5.4049-1.7271xhsa-miR-614+0.0937xhsa-miR-42XX+1.1171xhsa-miR-61XX

Threshold:1.4590Specificity:0.9533±0.0635SensiGvity:0.9747±0.0517Accuracy:0.9667±0.0419AreaUnderCurve:0.9871

FurtherModeling(with4miRNAs)

26

PredicGonscore=9.225-0.9554xhsa-miR-614+0.8076xhsa-miR-42xx+1.4167xhsa-miR-61xx-1.9153xhsa-miR-66XX

Threshold:1.5280Specificity:0.9611±0.0533SensiGvity:0.9800±0.0396Accuracy:0.9729±0.0326AreaUnderCurve:0.9885

FurtherModeling(with5miRNAs)

27

PredicGonscore=8.763-0.665xmiR-614+0.865xmiR-42xx+1.413xmiR-61xx-1.697xmiR-66xx-0.716xmiR-1xxx

Validation • PaGentSerumCollecGon:Healthy,Benign,pre-Cancer&Cancer

• Workflow:1.  IsolateRNA2.  OpGmizeprimer3.  ReversetranscripGon4.  AmplifycDNA5.  RunqPCR/nanostring/(liquid)chip6.  Analyze

29

Conclusions

2. Several serum miRNAs are enough to idenGfy the group(cancerornoncancer)ofapaGentatahighaccuracylevel.Thus,theseselectedmiRNAscouldbeviewedaspotenGalbiomarkersforimplemenGngearlydetecGonofbreastcancer.

1. While applying different analysis pipeline would get quietdifferent outcomes, there are some overlaps, which show theconsistencyofthesemethods.

30

PerspecKves1.  The models seem to be precise enough to fit early

detecGon, more validaGons are sGll required toestablishrobustcriteria.

2.  The established useful analysis pipeline enablesapplying for other different expression data derivedfromotherdiseases.

3.  ThemechanismsoftheseselectedmiRNAsrelatedareunknown. It ismuchmoremeaningful and criGcal forthe understanding of these idenGfied biomarkers. Bycomprehending themolecularmechanismsunderlyingthesebiomarkers, thedevelopingeffecGvetreatmentsandtranslaGonalresearchwouldbepromoted.

Acknowledgements

LabmembersBao-HongLee,PhD

Yu-LingTai,PhD

YumiYasua,PhD

EmilyLo, PhD

Shion-ShanShen, PhD

Shang-JerKuo(PhDstudent)

Chia-YuYang(Master)

Ru-YingFang(Master)

Yi-TingChen(Master)

Dan-JungChuang(Master)

Shi-ChenWang(Master)

MitchLin(Master)AngelaWang(RA)

JeffKu(RA)

CornellDavidC.Lyden,

MD,PhD

AyukoHoshino,PhD

HéctorPeinado,PhD(CNIO,Spain)

NTUKing-JenChang,

MD,PhD

Wen-HongKuo,MD,PhD

Ming-YangWang,MD,PhD

Chen-ChihHsu,PhD

Chen-AnTsai,PhD

JapanTakahiroOchiya,PhD

(NCC)

HidetoshiTahara,PhD(UHiroshima)

KojiUeda,PhD(CancerFound.)

OthersArthurLander,

MD,PhD(UCIrvine)

Kuo-KanLiang(AS)

AliMortazabi,PhD(UCIrvine)

ZhongminZhao,PhD(UTHealth)