948150962November720132013ElsevierInc CORE Metadata citation and similar papers at coreacuk Provided by Elsevier Publisher Connector CancerdrivergeneshavebeendescribedasmountainsandhillsW ID: 959430
Download Pdf The PPT/PDF document "CumulativeHaploinsufciencyandTriplosensi..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
CumulativeHaploinsufÞciencyandTriplosensitivityDriveAneuploidyPatternsandShapetheCancerGenomeTeresaDavoli,AndrewWeiXu,KristenE.Mengwasser,LauraM.Sack,JohnC.Yoon,PeterJ.Park,andStephenJ.Elledge ,948–962,November7,20132013ElsevierInc. CORE Metadata, citation and similar papers at core.ac.uk Provided by Elsevier - Publisher Connector CancerdrivergeneshavebeendescribedasmountainsandhillsWoodetal.,2007).Mountainsaredrivergenesthatareveryfrequentlymutatedincancer,whereashillsrepresentlessfrequentlymutateddrivergenes.Ithasbecomeclearfromrecentinternationalsequencingeffortsthatmostpotentdrivers(moun-tains)havebeendiscovered.Akeyissueishowtodeterminetheidentityofthesigni cantbutlessfrequentlymutateddrivers,thehills.Arecentanalysissearchingforveryhighcon dencecancerdriversinadatabaseof400,000mutationsestimatedthattherewere71TSGsand54OGs(Vogelsteinetal.,2013).Itislikelythattherealsoexistadditionalfunctionallysigni cantcancerdriverswithweakerphenotypesandlowerprobabilitiesthatareselectedlessfrequently.Acentralquestionishowtoidentifythesegenes.Inprinciple,withmoresamplesanalyzed,greaterstatisticalsigni cancecanbeplacedontheoutliers,allowingdiscoveryoflowerpenetrancedrivers.However,itislikelythatthereismoreinformationpresentinthecurrentdatathatmayallowtheselowerfrequencyeventstobedetected.Toapproachthisquestion,wesoughttodeviseamethodtopredictTSGsandOGsincancerbasedonthepropertiesofgenemutationsignaturesofthesetwodistinctclassesofdrivergenes.Wehypothesizedthattheproportionofthedifferenttypesofmutationswithdifferentfunctionalimpactwouldbeinforma-tiveinpredictingthesetwotypesofdrivers(Figure1A).Eachgenehasabackgroundmutationratethatisdependentontran-scription,replicationtiming,andpossiblyotherunknownparam-eters,andthisratecanbeestimatedbythenumberofmutationsthatareunlikelytoaffectitsfunction(suchassilentorfunc-tionallybenignmutations),whoseobservedfrequencyisnotdependentonselectivepressureduringcancerevolution.Theproportionoffunctionallyrelevantmutationsofparticularclassescomparedtothisbackgroundmutationratewillbedependentonthedegreeofselectionandwillpredictthelikelihoodthatagenewillactasacancerdriver.TSGsandOGscanbedistinguishedamongthecancerdrivergenesbasedonthecharacteristicpatternofthedifferenttypesofmutations(i.e.,lossoffunction[LOF],missense,silent)thataretypicallyobservedforthosetwoclassesofdriversrelativetoneutralgenes,asillustratedinFigure1IdentiÞcationofParametersPredictingTSGsandOGsWesetouttodeterminethemostreliableparametersforthepredictionofTSGsandOGsinanunbiasedway(Figure1Weusedsequencedatafrom�8,200tumorsfromtheCOSMICForbesetal.,2010)andTCGA(databasesandarecentlypublisheddatabase(etal.,2013)comprising�1,000,000mutations(FigureS1TableS1availableonline).Wede nedalistof22parameterspri-marilybasedonthedifferentclassesofmutationsandusedtheclassi cationmethodLassoandthreetrainingsetsofknownTSGsandOGs(fromtheCancerGeneCensus,Futrealetal.,TableS2A)andneutralgenestoidentifythoseparametersthatbestpredictthetwoclassesofdrivergenes(seementalProcedures).WeemployedPolyPhen2topredictthe Prediction of TSGs of TSGs and OGs TumorSamples Analysis of Mutation Patternsfor each gene AnalysisFunctional SilentGene Tumor Suppressor Gene OncogeneNeutral geneACB for TSGs,OGs and Neutral genes MissensePatterns of mutations expected to bepredictive of TSGs and OGs DTUSON ExplorerParameters for theprediction of TSGsParameters for the1. LOF/Benign ratio3. Splicing/Benign ratio1. Entropy Score2. HiFi Missense/Benign ratio High level of missense mutationsand clustering of mutations LOFSilentSilentSilent Figure1.PredictionofTSGsandOGsBasedonTheirMutationalProÞle(A)Schematicrepresentationofourmethodforthedetectionofcancerdrivergenesbasedontheassessmentoftheoverallmutationalpro leofeachgene.Thesomaticmutationsineachgenefromalltumorsamplesarecombinedandclassi- edbasedontheirpredictedfunctionalimpact.Themainclassesofmutations(silent,missense,andLOF)aredepicted.(B)SchematicdepictingthemostimportantfeaturesofthedistributionsofmutationtypesexpectedforatypicalTSG,OG,andneutralgene.Comparedto‘‘neutral’’genes,TSGsareexpectedtodisplayahighernumberofin-activatingmutationsrelativetotheirbackgroundmutationrate(benignmutations),andOGsareexpectedtodisplayahighernumberofacti-vatingmissensemutationsand
acharacteristicpatternofrecurrentmissensemutationsinspe-ci cresidues.(C)A owchartdelineatingthemainstepsinourmethodforidentifyingTSGsandOGs,fromtheclassi cationofthemutationsbasedontheirfunctionalimpacttotheidenti cationofthebestparametersthroughLassoandtheiruseforthepredictionofTSGsandOGsbyTUSONExplorer(ortheLassomethod).(D)Schematicrelatedto(C)depictingthepa-rametersselectedbyLassoemployedbyTUSONExplorerforthepredictionofTSGsandOGs(HiFI,highfunctionalimpact).ForTSGs,theparametersaretheLOF/Benignratio,theHiFI/Benignratio,andtheSplicing/Benignratio,whereasforOGstheparametersaretheEntropyscoreandtheHiFI/Benignratio.AlsoseeFigureS1TableS1 ,948–962,November7,20132013ElsevierInc. functionalimpactofmissensemutationsinordertoclassifythemintothosewithpotentiallyhigh(HiFI)orlow(LoFI)functionalimpact(Adzhubeietal.,2010).LoFImutationsaretypicallyconservativeaminoacidchangesorchangesinpoorlyconservedresidues.Wede nedthecombinationofsilentandLoFImissenseas‘‘Benign’’mutationstoprovidealarger,morereliablevalueforestimatingbackgroundmutationrates.Wealsode nedtheLOFmutationsasthecombinationofnonsenseandframeshiftmutations.AsamajorityofknownOGsshowanatypicaldistributionofrecurrentmutationsinoneorafewkeyresidues,weutilizedentropy,awell-de nedconceptinphysicsandinformationtheory(ShannonandWeaver,1949),tomeasurethedegreeofreoccurringmutationswithinagene.TheEntropyscorerepresentstheweightedsumoftheprobabilities,acrossagene,thatasiteismutated(seeExperimentalProceduresThebestparametersfoundbyLassoforthepredictionofTSGsandOGsaredescribedbelowandarevisualizedinDandTumorSuppressorsversusNeutralGenesThemostpredictiveparametersforTSGsare:(1)theratioofLOFmutationstoBenign(p=2.51,Wilcoxon,one-tailedtest);(2)theratioofSplicingtoBenignmutations(p=4.6);(3)theratioofHiFImissensetoBenignmutations(p=);and(4)high-leveldeletionfrequency(p=1.46).A20-foldcross-validationshowsahighpredictionaccu-racyof93.2%onthesetrainingsets(Figure2AandTableS2OncogenesversusNeutralGenesThemostpredictiveparametersforOGsare:(1)theentropyformissensemutations(p=2.2);(2)theratioofHiFImissensemutationstoBenignmutations(p=1.2);and(3)high-levelampli cationfrequency(p=1.4).The20-foldcross-validationaccuracyis85.2%(Figure2BandTableS2TumorSuppressorGenesversusOncogenesOneimportantaimofourpredictionmethodisthediscriminationbetweenTSGsandOGs.Themostpredictiveparametersbe-tweenthesetwosetsare:(1)theratioofLOFtoBenignmutations(p=2.5);(2)high-levelampli cationfrequency(p=);(3)high-leveldeletionfrequency(p=7.6and(4)theratioofSplicingtoBenignmutations(p=9.9).The20-foldcross-validationaccuracyis91.9%.Overall,Lassoidenti edparametersthatmakeintuitivesensefortheseclassesofgenesandclearlydelineatedTSGsandOGsfromeachotherandfromneutralgenes.Insum,weidenti edindependentparametersthatstronglypredictanddistinguishbetweenTSGsandOGs(Figures2Cand2DandTableS2IdentifyingOGsandTSGsHavingidenti edthemostpredictiveparameters,wedevelopedamethodwecalluppressorandcogeneExplorer(TUSONExplorer)thatcombinedselectedparameterstoderiveanoverallsigni canceandrankingforeachgeneasapotentialTSGorOG(Figure1D).First,wederivedapvalueforeachgenefortheratiosofLOF/Benign,Splicing/Benign,HiFI/Benign,andMissenseEntropybasedonthecomparisontotheneutralgeneset(seeExperimentalProcedures).FortheLOF/Benignparameter,weappliedacorrectiontonormalizeforthenonuni-formcodonusageamonggenesfortheoccurrenceofnonsensemutations(seeExperimentalProcedures).Finally,weusedanextensionofLiptak’smethodtoprovideacombinedpvaluefortheselectedparametersforeachgene.ForTSGs,thecom-binedpvalues(andqvalues)werederivedfromindividualvaluesfromtheLOF/Benign,Splicing/Benign,andHiFI/Benignparam-eters.ForOGs,thecombinedvalueswerederivedfromtheMissenseEntropyandtheHiFI/Benignparameters(Figure1TheLOF/BenignparameterfordiscriminationbetweenTSGsandOGswassubsequentlyutilizedtode nea nallistofOGsandTSGs(seeExperimentalProcedures).TUSONExplorerdoesnottakeintoaccountSCNAinformation,andthisallowsustoperformarigorousanalysisofourcancerdrivergenesfortheirabilitiestopredictthefrequencyofdeletionandampli ca-tion(seebelow).AsasecondstrategytopredicttheprobabilityofagivengenebeingaTSGorOG,weemployedtheLassomodel,whichalsotakesintoaccountSCNAs(seeExperimentalProcedures).Therankedlistsofpredicted
TSGsandOGsbybothLassoandTUSONExplorerarecontainedinTablesS3AandS3B.Thislistprovidesafacilelook-uptablethatcanbeeasilysortedfordifferentparametersforallthosewhoareinterestedinthemuta-tionalbehaviorofagivengeneinthisdataset.Bothrankingstrategiesperformedsimilarlyandeliminatedtheproblemsofinappropriatelyincludinggiantgenesandgenesinhighlymutableregions(Deesetal.,2012),withouttheneedtoconsiderexpressionlevelorreplicationtiming(Lawrenceetal.,).Importantly,bothofourstrategiesdistinguishbetweenTSGsandOGs,whicharepredictedtohavefunctionallyoppo-siterolesinthecontrolofcellgrowthandhavedifferentimplica-tionsforpotentialcancertherapeutics.EstimatesoftheNumbersofTSGsandOGsArankingofthisnatureconsistsoftrulysigni cantgenesmixedwithfalse-positivegenesthatobtainlowpvaluesbychanceun-derthenullhypothesis.Thus,wesoughttogetanestimateoftheminimumnumberofTSGsandOGsbyanalyzingthedistributionofthecombinedpvaluesforeachclassofcancerdrivergenes.Toachievethis,weutilizedahistogram-basedmethod(etal.,2001)toestimatethenumberofrejectedhypothesesfromthedistributionsofthecombinedpvaluescalculatedforeachgene.Withourdataset,thismethodestimatedTSGsand250OGs(ExtendedExperimentalProceduresThislonglistofTSGsandOGssuggeststhattherearemanymoredriversthananticipatedandthattheyexistinacontinuumofdecreasingpotency(Figure7A).Fortheanalysesdescribedbelow,weconsideredthetop300TSGs(qvalue0.18)and250OGs(qvalue0.22)asourworkinglists.Giventhefactthatthedeviationofthemutationsignaturesfromthenormalpatternisafunctionofthedegreeofselectionandthefrequencyofmutation,increasingthenumberoftumorsampleswilldetectevenmorecancerdriversofprogressivelyweakerselectivepressure.TodeterminethepotentialnumberofTSGsuponadditionalsequencing,weappliedTUSONandestimatedthenumberofTSGs(usingMosig’smethod)onrandomsubsetsofthedatasetwithincreasingnumbersofsamplesandobservedthatthenumberofpredictedTSGscontinuestoincreasewithadditionalsamples.WeobservedthattherateofincreaseinthepredictednumberofTSGsdecreasesslightlyat ,948–962,November7,20132013ElsevierInc. thehighestnumberofsamplesexamined(FigureS2),indicatingapossibleplateauatverylargenumberofsamples.PAN-CancerMutationalAnalysisGeneOntology(GO)termandpathwayanalysisofourlistofpo-tentialTSGsshowedenrichmentforfunctionsthatarehighlyrelevanttotumorigenesis,includingcell-cyclecontrol,embry-onicdevelopment,promotionofdifferentiation,apoptosis,andbloodvesseldevelopment(TableS3CandFigure3).Inaddition,therewasastrongenrichmentfortranscriptionalregulation(q=)andchromatinmodi cation(q=5.7Furthermore,wenoticedanenrichmentforgenesinvolvedin 0.000.250.500.751.001.251.752.002.252.753.00 OGTSG 0.00.10.20.30.4 OGTSG 0.0000.0010.0020.0040.0050.0060.007 OGTSG 0.000.010.020.040.05 OGTSG 0.000.250.501.251.502.002.252.753.00 NeutTSG 0.000.250.501.251.501.752.002.252.50 NeutTSG 0.00.10.20.40.5 NeutTSG 0.0000.0050.0100.0150.020 NeutTSG 0.000.050.100.200.250.300.350.40 NeutOGMissense Entropy &
#1; 0.000.250.501.001.251.501.752.00 NeutOG 0.000.250.500.75 NeutOGMean Polyphen2 Score 0.000.010.020.030.040.050.06 NeutOG LOFBenign HiFIBenign Benign FrequencyLOFBenign Benign FrequencyFrequencyp=3e-28p=3e-14p=5e-13p=1e-8p=3e-16p=1e-8p=8e-6p=1e-9 TSG versus Neutral GenesOG versus TSGFrequencyMissenseHiFIBenignMean Missense Polyphen2Scorep=2e-14p=1e-9p=3e-10p=1e-6 BDAPCNPM1CASP8CDKN2AGATA3ARID1AMAP3K1B2MPTENRB1RPL22SMAD4CDKN1BTP53VHLAKT1BRAFCTNNB1EZH2IDH1KRASNRASPIK3CARAC1U2AF1EGFRFBXW7STK11IDH20.050.100.200.300.400.500.600.801.002.00 0.250.500.751.001.502.00 pvalue OGpvalue TSG110 OG versus Neutral Genes Figure2.BestParametersSelectedbyLassoforthePredictionofTSGsandOGs(A–C)Boxplotrepresentationsofthedistributionofthevaluesfortheindicatedparametersintheneutralgenes(gray),TSG(red),andOGtrainingset(green).Themedian, rstquartile,thirdquartile,andoutliersinthedistributionareshown.ThepvalueforthedifferencebetweenthetwoindicateddistributionsisshownasderivedbytheWilcoxontest.(A)BoxplotsshowingthedistributionofLOF/Benign,HiFImissense/Benign,andSplicing/Benignratiosandthehigh-levelfrequencyoffocaldeletionamongtheneutralgenesetandtheTSGset.(B)BoxplotsshowingthedistributionofMissenseEntropy,HiFImissense/Benign,meanofPolyPhen2score,andthehigh-levelfrequencyoffocalampli cationamongtheneutralgenesetandtheOGset.(C)BoxplotsshowingthedistributionofLOF/BenignandSplicing/Benignratios,high-levelfrequencyofdeletionandampli cationamongtheTSGandOGsets.(D)PlotoftheLOF/BenignratioandMissenseEntropyforeachgene,thebestparametersfordiscriminatingbetweenTSGsandOGs.Speci cgeneswithhighlevelsofLOF/BenignorMissenseEntropyareshownalongwiththeirpvaluesforbeingaTSGoranOG(TUSONExplorer).AlsoseeFigureS2TablesS2 ,948–962,November7,20132013ElsevierInc. PTPRK Response to Growth Factors B2MAntigen Presentationand Inflammation HLA-A HLA-B HLA-DRB1 Tumor Suppressor Genes p value 10-10010-2 EGFR ERBB2 GNAS PPP6C FZD6 CHD8 TCF7L2 WNT Signaling CTNNB1 WNT11 DVL1 AXIN1 GSK3B LEF RSPO2 MAP3K1 MAP2K4 BRAF KRAS NRAS HRAS MAPK1 MAP2K1 TGFBR2 TGFBR1 TGFB1 ACVR1B Activin SMAD4 SMAD2 TGF IDH1 IDH2 PSPH PGM5 PTEN RPL18 RPL5 Ribosome PIK3CA AKT1 MTOR CASP8 FADD BAX BCL2 FAS FASLG BID Mitochondria CytC CASP9 TNFR TRAF2 TNF RALGDSPRKCI FAT1 PVRL4 Cell-CellAdhesion CDH1 FOXO1 FUBP1 BCOR GPS2 CIC SETD2 RUNX1 SMARCA4 KDM5C EP300 ARID1B NSD1 MEN1 TBX3 KMT2B NCOR1 CDK12 KANSL1 KMT2A SIN3A TCF12 CHD3 EZH2 SPOP MYC MAX TRRAP Transcription GATA3 ATRX ARID1A ARID2 KTM2D KTM2CCohesin SMARCB1 Sonic Hedgehog PTCH1 SHH SUFU COS2 FU GLI GF ZFP36L1 ANAPC1 CDKN2A CDKN1B CDKN2C CDK4/6 CCND3 E2F CDK2 CCNE1 CDK1 CCNA/B CDKN1A FBXW7 Cell Cycle Regulation ZFP36L2 KEAP1 NFE2L2 CUL3 DNA Damage Response TP53 ATM USP28 UBR5 TRIP12 RBMX TP53BP1 MCM7 ATR BRCA2 BAP1 BRCA1 MLH1 MDM2 RNF168 STAG2 SF3B1 Avoiding Avoiding GenomeInstabilityAltered MetabolismNotch Signaling NOTCH1 NOTCH2 Hypoxia Stress Response HIF1A CUL2 EGLN1 IKK NFKB Invasion ERBB3 FGFR2 ACVR2A NF1 APC VHL CTCF RB1 RPL22 SMO RASA1 NF2 STK11 ARHGAP35 COL4A2 RHOA PIK3R1MBD6 IL32 IL32 10-10 10-5 RIT1 RNF43 RAC1 RNF111 BRE RAD21 SMC4 BRD7 NIPBL KIT1 PTPN11 PRKCI SRC COL4A2 PPP2R1A CHD4 Figure3.RepresentationofPredictedTSGsandOGswithinTheirCellularPathwaysPlacementofpredictedcancerdriverswithinspeci ccellularpathways.T
SGsandOGswerepredictedbyTUSONExplorer.ThepredictedTSGsandOGsbelongingtomanyknowncellularpathwaysorcomplexesareshown,alongwithhowtheygenerallycorrespondtothehallmarksofcancer.TSGsareshowninred,whereasOGsareshowningreen;colorintensityisproportionaltothecombinedpvalueasindicated.Forsomepathways,additionalgenesabsentfromthepredictedTSGsandOGswereaddedandmarkedingrayforclarityofthepathwayrepresentation.Althoughseveralgenesareknowntoaffectmultiplepathwaysandhallmarks,onlyonefunctionispresentedforthesakeoflimitingthecomplexityofthediagram.Anexternalblackboxoutsideofthecoloredgeneboxhighlightsgenespreviouslylesswellcharacterizedfortheirrolesintumorigenesis.SeealsoFigureS3TableS3 ,948–962,November7,20132013ElsevierInc. theimmunesystem(q=5.8),particularlyinantigenpro-cessingandpresentationrepresentedbytheMHCclassIsys-tem.TwomajorHLAgenes()wereinthetop90candidateTSGs(q0.0002),andthe2microglobulin)gene,whichisanobligatorycomplexcomponentofbothHLAproteins,ranked43rd(q=9.2)onourTSGlist,underscoringthatescapingfromimmunosurveillanceisasigni cantselectiveforceintumorigenesis(HanahanandWein-berg,2011Figures3B).Furthermore,,whichstimu-latestheimmuneresponsesofNKcellsandCD8+TcellsthatmonitorMHCstatus(Contietal.,2007),isalsointhetop50TSGs.Unexpectedly,negativeregulationofcelladhesion(q=4.32)wasenriched,indicatingthatincreaseofcelladhesionmayconferaselectiveadvantagetotumorcells.Tradi-tionallyithasbeenthoughtthatreducingadhesionpromotestumorigenesis;however,recent ndingssuggestapotentiallydifferentroleforcell-to-cell-adhesion.First,ithasbeenshownthatcirculatingtumorcellsexistinclustersintheblood(etal.,2011).Second,,whichrankedwellinourLassoOGlist,wasshowntopromotetransformationthroughcelladhe-sion,asdoseveralotheroncogeneslike,andlossofPavlovaetal.,2013).Thus,promotionofadhesionmaybeadrivingforceintumorigenesis. C TP53BP111714-1850 1354-H2AX Binding 956 1865-BRCTBRCT1490-1590TUDOR F 1374 MATH domainBTB domain F133W131K134SPOP 166190 E50K D140GD140N R121Q x18B 203207299 309-332 IgCTM A 1USP281077 97-116162-196400-432580649 PPP2R1A PP2A Subunit B Binding 399 8400Subunit C-Binding R144H S143FR144CT145P P179R R183W R183Q S256YS256FR258HR258CW257C E1D RASA1 161264 350426 287339SH3SH2 494575 698RasGAP 595-689C2 1 LOFSilentMissense 1044 Figure4.RepresentationofMutationPatternsinRepresentativePredictedTSGsandOGs(A–F)ThemutationalpatternsofselectedTSGsandOGsaredepicted.ForTSGs(A–D),theloca-tionsofLOF(red)andsilent(white)mutationswithinthecodingregionsareshown.ForOGs(EandF),thelocationofrecurrentmissensemu-tations(orange)andofLOFmutations(red)withinthecodingregionsareshown.TP53BP1arepreviouslylesswell-characterizedcandidateTSGsintheTUSONPAN-Canceranal-arepreviouslylesswell-characterizedcandidateOGs.SeealsoTableS3NewPotentialCancerDriversNewcomponentsofpathwayspreviouslylinkedtotumorigenesishavealsobeendetected(Figure4).Forexample,theDNAdamageresponsepathwayiscen-traltothemaintenanceofgenomicstability,andbothmembersofakeyDDRcomplex,theTP53BP1/USP28complex(Zhangetal.,2006),whicharesubstratesoftheATMkinase(etal.,2007),wereidenti edwithinthetop110TSGs(q0.15,Figures3A–4C).TwocomponentsthatregulateATM-dependentchromatinremodeling,Gudjonssonetal.,),arealsohighontheTSGlist(q,whichcontrolsexpression(Adamsonetal.,),ranked76thonthelist(q=1.1).ThereareseveralcandidateOGswithenzymaticfunctionsthatcouldserveasdrugtargets(Figure4EandTablesS3BandB),includingthreephosphatases(andregulatorssuchas,aswellasseveralkinasesamongothers).TherearemanyothernewpotentialTSGsandOGsontheseliststhatcannotbedis-cussedhereduetolimitationsofspace,butseveralofthesearepresentedinFigure3Consistentwiththeenrichmentofcell-cycleandapoptosisGOterms,integrationofthePAN-Canceranalysiswithfunctionalgenesetsrevealedthatessentialgenesaresigni cantlydepletedfordeleteriousmutations(seebelow).Anexceptiontothat ndingwasthepresenceof,andlargeribosomalsubunitgenesinthetop210TSGs(q0.07;Figure3).Interestingly,heterozygousmutationsinribosomalgenespromotetumorigenesisinzebra sh(Laietal.,2009Furthermore,familialmutationsinribosomalproteinshavebeenassociatedwithDiamond-Blackfananemia,whichisasso-ciatedwithanincreasedriskofleukemia(Willigetal.,2000AnalysisofIndividualTumorTypesIdenti cationofcancerdriversusingthePAN-Canceranalysisfavorsdiscoveryofgeneswhosefuncti
onscontributetomany ,948–962,November7,20132013ElsevierInc. differenttypesofcancer.Certaincancerdriversmaymissthecutoffforsigni canceinthePAN-Canceranalysisbecausetheyareprimarilyinvolvedincontrollingtissue-speci cdifferen-tiationnetworksorbecausetheyareratelimitingforaparticularfunctioninonlycertaintissues.Thus,weanticipatethatnewdriverscanbediscoveredthroughanalysisofmutationsigna-turesinindividualtumortypesdespitetheirlowernumbers.Weperformedthesameanalysisasaboveforeachof20tumortypes(TablesS1A,andS4B).ThisanalysisfoundmanyTSGsthatarespeci cforonetissuetypesuchasinbreastadenocarcinoma,inkidneyclearrenalcellcarcinoma,andinhematologicalmalignances(TableS4A).GeneswhoseFDRsforthedifferentsubtypeswerebelow0.25wereallalreadyrelativelyhighlyrankedinthePAN-Canceranalysis.Thisindicatesthatthema-jorityoftissue-speci cdriversweredetectedinthePAN-CancerWewantedtodeterminehowmanynewTSGsmightbeexpectedfromtheanalysisofanewcancersubtype.Forthis,wecalculatedthenumberofTSGsinthewholedatasetlackinganindividualtumortype(TablesS4C)andcomparedthislisttotheTSGsinthattumortype,whichaveraged14genes.Wefoundthattheaveragenewcancertypeaddedabout veTSGstothePAN-Cancerlist.Thus,onaverage,70%oftheTSGsdetectedinasingletumortypewerealreadydetectedinthePAN-Canceranalysisperformedafterexcludingthemutationsinthattypeoftumor.Thissuggeststhatmostcancergenesselectedduringtumorevolutionactincellularpathwayswhoseroleintumorigen-esisiswidespreadamongdifferenttumortypes.AnalysisofTSGsandOGsBehaviorofFunctionalGeneSetsThePAN-Cancermutationdatasetallowsustointerrogatethebehavioroffunctionalgenesetsderivedthroughexperimentalapproaches.WepreviouslyshowedthatSTOPgenesareover-representedinregionsofdeletion(Soliminietal.,2012).Examina-tionoftheirabundanceinthesetofcandidateTSGsshowedthatSTOPgenesaresigni cantlyenrichedintheTSGset(p=0.0031,Fisher’sexacttest)comprising10%ofthetop300TSGs(68%morethanexpected).TheSTOPgenesetshoweda50%higherratioLOF/Silentthantheaveragefortheneutralgeneset(p=2.0Figure5A).Furthermore,theSTOPgenesshowedasigni cantincreaseintheSplicing/BenignandHiFI/Benignratios,twoofthemostpotentparametersforthepredic-tionofTSGs(Figure5A).Thisanalysisfurtherunderscoresthefundamentalconnectionbetweencellproliferationandcancer.Wenextinvestigatedahigh-con dencesetof145genespredictedtobeessentialatthecellularlevelbasedontheirhousekeepingcellularfunctionsandtheirhighevolutionaryconservation(TableS5AandExperimentalProcedures).Thissetwasdepletedfromregionsofrecurringdeletions(etal.,2010)by43%(p=0.0198,Fisher’sexacttest),andalargersetof332essentialgeneswasdepletedby25%(p=0.014).ExaminationoftheLOF/Silentratioshowedthat,forthesetof145genes,thefrequencyofLOF/silentwas27%lowerthantheratefortheneutralgeneset(p=5.8Figures5B).Additionally,theLOF/kbandHiFI/Benignratioswerealsosigni cantlydecreasedintheessentialgeneset.Giventhatthevastmajorityofthemutationsanddeletionsinquestion 0.250.000.250.501.001.251.50 NeutEss 0.53.0 NeutEss 0.00.20.41.01.41.82.02.22.42.62.83.0 NeutEss 0.0000.0010.0030.0040.005 NeutEssLOF p=5.8e-5LOF p=7.1e-7HiFIBenign p=1.7e-4Frequencyp=8.9e-3 Essential versus Neutral Genes 0.00.20.60.81.0 NeutSTOP &
#1; 0.000.050.100.200.25 NeutSTOP 0.00.51.52.02.5 NeutSTOP 0.0000.0010.0020.0040.005 NeutSTOPLOF p=2e-18 p=2.5e-9HiFIBenignp=1.6e-8Frequencyp=1.4e-5 STOP versus Neutral Genes AB Figure5.BehaviorofFunctionalGeneSetsRelativetoTSGParameters(AandB)BoxplotrepresentationofthedistributionofthevaluesfortheindicatedparametersintheneutralgenescomparedwiththeessentialgenesandSTOPgenes.Themedian, rstquartile,thirdquartile,andoutliersinthedistributionareshown.ThepvalueforthedifferencebetweenthetwoindicateddistributionsisshownasderivedbytheWilcoxontest.(A)Boxplotsshowingthedistribution(orange)ofLOF/Silent,Splicing/Benign,andHiFImissense/Benignratios(gray)andthehigh-levelfrequencyoffocaldeletionamongtheneutralgeneandSTOPgenesets.(B)BoxplotsshowingthedistributionofLOF/Silent,LOF/Kb,HiFImissense/Benignratiosandthehighfrequencyoffocaldeletionamongtheneutralgeneset(gray)andtheessentialgeneset(lightblue).SeealsoTablesS3A,andS5B. ,948–962,November7,20132013ElsevierInc. areheterozygous,thereducedLOFmutationanddeletionfre-quencyoftheessentialgenesasagrouparguesthatbetween25%and45%arehaploinsuf cient.Interestingly,ourTSGswereenrichedinrecurringfocaldeletions(68%,p=0.000281)andweredepletedfromrecurringampli cations(28%,p=0.015),whereastheOGswereenrichedinampli cations(25%,p=0.046)anddepletedfromfocaldeletions(23%),indicatingthatampli cationsarealsolikelytobeCancerGeneIslands.GeneralPropertiesofCancerDriversHighInteractivityTosearchforuniquepropertiesofTSGsandOGs,weexaminedthedegreetowhichthesedriversparticipateinproteincom-plexesusingtheCORUMdatabaseofexperimentallyvalidatedhumanproteincomplexes(Rueppetal.,2010).WefoundthatbothTSGsandOGsweresigni cantlymorelikelytobeinproteincomplexesthanatypicalprotein.The13.4%ofallproteinsfoundinCORUMareinacomplex.However,36.7%ofthepredictedTSGswereincomplexes(p=3.1),and26.4%ofthepre-dictedOGswereincomplexes(p=3.510FigureS3HighBetweennessAsecondpropertyofcomplexesisthedegreetowhichtheyareconnectedtootherproteinsandcomplexes.Weexploredthisbyassessingapropertycalled‘‘betweenness,’’whichispropor-tionaltothenumberoftimestheproteinispartoftheshortestpathsbetweenallpairsofproteinsinanetwork.Highbetween-nessindicatesagreaterconnectivity.TheTSGandOGcandi-dategenelistsweremappedontothemostcurrentBioGRIDhumanprotein-proteininteractionnetwork(Starketal.,2006BoththepredictedTSGsandOGsshowahighdegreeofbetweenness(TSGp=6.16,OGp=1.68ureS3B),indicatingthattheyareoptimallypositionedtoimpactinformation owthroughnetworks.GreaterLengthProteinswithgreaterinteractivityoftenhavemoredomains.Thus,weexaminedgenelength.Cancerdriversaresigni cantlylongerthantheaveragegene(1,700nt),withthemeanforTSGsat3,234nt(p=2)andOGsat2,107nt(p=9.7Importantly,thisobservationisalsocharacteristicofthegenesinourtrainingsets(TSGs,4,133nt,p=6.7;OGs,2,260nt,p=0.0016).AnUnusuallyHighConcentrationofTSGsontheXWhileexaminingthedistributionofTSGsacrosschromosomes,wefoundthattheXisunusuallyenrichedforTSGs(p=0.0042,exactbinomialtest)relativetoautosomes.Examiningthetop300TSGs,we ndthat,althoughonly3.9%ofallgenesareontheX,itcontains7.3%ofallpredictedTSGs(86%morethanex-pected)andwastheonlychromosomewithasigni cantenrich-mentofTSGs(TableS5C).GiventhefactthattheXisfunctionallyhaploidinbothmalesandfemales,thisobservationhascertainimplicationsforevolutionaryselectionofcancerdriversduringtumorigenesisandhaploinsuf ciencyofTSGs(seeInterestingly,inthetop400TSGs,wefoundtwopotentialTSGsontheY,(q0.22).BothhavehomologsontheXthatescapeXinactivation,eachofwhichalsodisplaystumorsuppressorproperties:(p=0.019)and(p=3.3).Thiscouldexp
laintheobservationthatfrequentYnullisomyisobservedinprostate,renalcell,headandneck,Barret’sesophagealadenocarcinoma,bladder,pancreaticadenocarcinoma,andothercancersatfrequenciesof30%–80%(Bianchi,2009;Kowalskietal.,2007Furthermore,weanalyzedthesilentmutationratesalongentirechromosomesandfoundanenhancedmutationrateontheXchromosomerelativetoautosomesinmales(30%increase,p=1.1).Thisincreaseisevengreaterinfemales(77.5%,p=1.6TableS5D).Possibleexplana-tionsforthisphenomenonaredetailedinthediscussion.DistributionandPotencyofCancerDriversonChromosomesPredictArmandChromosomeSCNAInadditiontofocalSCNAs,alessfrequentbutsigni cantchro-mosomalalterationiswhole-armlossorgain.WehypothesizedthatthedistributionandpotencyofTSGsandOGsonchromo-somesmightexplaintheaveragefrequencyofchromosomalwhole-armSCNAsseenincancer.Tothisend,wegeneratedascore,Charm,thatprovidesanassessmentofeacharmbasedonthedensityofTSGsandOGsandtheirpotency(weightsofTSGsandOGsarebasedontheirrankontheirrespectivelistsandserveasametricfortheirpotency).TheCharmscorerepresentsameasureoftheamountofpositiveornegativegrowthandsurvivalpotentialthatwild-typeOGsorTSGsmightnormallyimparttoagivenarmandthereforehowSCNAsmightimpactcancerevolutionbyalteringthisbalanceduringtumorigenesis.Importantly,forCharmcalculations,weemployedtheparametersfromTUSONExplorer,whichdoesnotincludecopynumberinformation.Tolessenthedilutingimpactoffalsepositivesforthisanalysis,weappliedstringencycutoffsofaqvalueof0.25forTSGsand0.35forOGsandamin-imumof10missensemutationsforOGsand8LOFmutationsforTSGstogetastringentlistof264TSGsand219OGs(seeimentalProcedures).TheanalysisoftheCharmscoreversusfrequencyofchromosomalarmdeletionrevealedastrongposi-tivecorrelation(r=0.578,p=5.8,Pearsoncorrelation;Figure6AandTableS6A).Interestingly,theCharmalsoshowedastrongnegativecorrelationwitharmampli cationfrequency,andthusahighCharmscoreindicatesasigni -cantlyreducedtendencyforachromosomearmtobeampli ed(r=0.59,p=2.8Figure6B).SimpleTSGdensitieswithoutweightingbyrankalsoshowedcorrelationswitharmdeletions(FigureS4A),butthesecorrelationsareimprovedbyCharm.IncontrasttoCharm,theCharmscoreshowedanegativecorrelationwitharmdeletionfrequency(r=0.52,p=Figure6C).Moreover,thedensityofOGspositivelycorrelatedwitharmampli cationfrequency(r=0.45,p=1.8Figure6D)butwasnotimprovedbytheCharmscore(datanotshown).Wereasonedthat,likeGOgenesinfocaldeletions,thechro-mosomearmsmostfrequentlydeletedincancerwouldbedepletedofgenesthatpromotethe tnessofcancercells.Usingourinsilicolistofessentialgenes,weestimatedtheir tnesspotencybyestimatingtheiravoidanceofdamagingmutationsusingthe(LOF+HiFI)/Benignratios.BydeterminingaCharmscoreforeacharm,wefoundanegativecorrelationbetween ,948–962,November7,20132013ElsevierInc. r=0.578 p=5.881e-05 10 p 10q11 p 11q12 p 12q13q14q15q16 p 16q17 p 17q18 p 18q19 p 19q1 p 1q20 p 20q21q22q2 p 2q3 p 3q4 p 4q5 p 5q7 p 7q8 p 8q9 p 9q012345 0.050.100.150.20Arm Deletion Frequency r=-0.599 p=2.818e-05 10 p 10q11 p 11q12 p 12q13q14q15q16 p 16q17 p 17q18 p 18q19 p 19q1 p 1q20 p 20q21q2 p 2q3 p 3q4 p 4q 5p 5q6 p 6q7 p 7q8 p 8q9 p 9q0123 0.10.20.3Arm Amplification Frequency r=-0.523 p=0.0003214 10 p 11q12 p 12q13q14q16 p 16q17 p 17q18 p 18q19 p 19q1 p 1q20 p 20q21q22q2q3 p 3q4 p 5 p 5 q 6 p 6q7 p 7q8 p 9 p 9q0123 0.050.100.150.20Arm Deletion Frequency r=0.454 p=0.001837 10 p 11 p 11q12 p 12q13q14q15q16 p 16q17 p 17q18 p 18q19 p 19q1 p 1q20 p 20q21q22q2 p 2 q 3 p 3q4 p 4q5 p 6 p 6 q7 p 8 p 8q9 p 9 q0.000.010.02 0.10.20.3Arm Amplification Frequency r=0.771 p=4.774e-09 10 p 11p11q12 p 12q13q15q16q17 p 18 p 18q19q1 p 1q20 p 20q22q2 p 3 p 3q4 p 5 p 5q6 p 6q7q9q-5.0-2.50.02.5 0.050.100.150.20Arm Deletion Frequency r=-0.651 p=3.627e-06 10 p 10q11 p 11q12 p 12q13q14q15q16 p 16q17 p 17q18 p 18q19 p 19q1 p 20q21q22q2 p 2q3 p 3q4 p 4q5 p 5q6 p 6q7 p 8 p 8q9 p 9q-4-202 0.10.20.3Arm Amplification Frequency r=0.804 p=3.2e-06 123457891011121416171819202122-4-20 0.040.080.120.16Chromosome Deletion Frequency r=-0.648 p=0.0005522 1234578910111213141619202122-4-202 0.050.100.150.200.25Chromosome Amplification FrequencyACEGBDFH6pCharm and Amplification Frequency TSGCharm and Deletion Frequency OGDens and Amplification Frequency OGCharm and Deletion Frequency TSG-OG-EssCharm and Amplification Fre
quency TSG-OGChrom and Deletion Frequency TSG-OG-EssChrom and Amplification Frequency TSG-OGCharm and Deletion Frequency TSGCharm TSGCharm TSGCharm OGDens OGCharm TSG-OG-EssCharm TSG-OGChrom TSG-OG-EssChrom TSG-OG 22q 5q 10q 4q14q21q 17q 61715q2p10q4q11p7q20p8q2q16p8q19p10q9p7q1q15187p8p13615 Figure6.CharmScore,ChromScore,andCopyNumberAlterations:CorrelationAnalysis(A–F)ThePearson’scorrelationanalysisoftheCharmscoresorDensityscorefortheindicatedgenesets(A–D)and/orcombinationsofthesesets(EandF)relativetothefrequencyofarm-leveldeletionorampli cation.Ess,essentialgenes.(GandH)ThecorrelationsoftheChromscores(ChromandChrom)relativetothechromosome-leveldeletionorampli cationfrequency.TheCharmscoresrefertoaweighteddensityofTSGs,OGs,oressentialgenespresentoneachchromosomearm,whereeachTSGorOGisweightedbasedonitsrankpositionwithinthelistofpredictedTSGsandOGsrankedbyTUSONExplorerandeachessentialgeneisweightedbasedonits(LOF+1/2HiFI)/Benignratio.TheChromscoreistheequivalentoftheCharmscoreforwholechromosomes.SeealsoFigures4TableS6 ,948–962,November7,20132013ElsevierInc. scoresandthefrequencyofarm-leveldeletions(r=0.34,p=1.6FigureS4D).NocorrelationwasfoundbetweenCharmandampli cationfrequency,asexpected.BecausetheCharm,Charm,andCharmcorrelatewitharm-leveldeletion,wecombinedthembygivingapositiveweighttotheCharmscoreandanegativeweighttotheCharmandCharmscorestoderiveacumula-tiveCharmscore.TheCharmscoregaveanevenstrongerpositivecorrelationwitharmdeletionfre-quency(r=0.77,p=4.7Figure6EandTableS6Forampli cation,weusedtheCharmscoreandfoundastrongnegativecorrelationwithampli cationfrequency(r=0.65,p=3.6Figure6F).Wealsocombinedampli cationanddeletionfrequenciesintoasinglescoreforcopynumbervariationoneacharmandcomparedthattotheCharmscore.Thisalsogaveastrongsigni cantcorrelation(r=0.74,p=2.7FigureS5WeextendedouranalysisofcancerdriverscoresandSCNAstowhole-chromosomeaneuploidyusingitsCharmequivalentscorethatwecallChrom(Figures6G,6H,S5E,andS5F).Chromsigni cantlycorrelatedwithchromo-somedeletionfrequency(r=0.66,p=3.7FigureS4andanticorrelatedwithampli cationfrequency(r=0.54,p=;FigureS4F).Impressively,whenwecombinedallthreeclasses—TSGs,OGs,andessentialgenes—thewasstronglypredictiveofthefrequencyofchromosomeloss(r=0.80,p=3.2Figure6G),andwaspredictiveofchromosomegains(r=0.64,p=5.5Figure6H).VerysimilarresultswereobtainedusingjusttheTUSONrankingwithoutstringencycutoffs(C–S5FandTableS6Together,thesedatastronglyarguethataselectiveforceingeneratingchromosomalarmandwhole-chromosomeSCNAsderivesfromtheintegrationoftherelativedensitiesandpotenciesofpositivelyandnegativelyactingcancerdriversonaparticularchromosome.Thus,theSCNAsincancergenomesmaybeselectedduringtumorevolutionthroughcumulativehaploinsuf ciencyfordeletions(aspreviouslyproposedforSTOPgenesinfocaldeletions[Soliminietal.,2012])andthroughcumulativetriplosensitivityforampli cations(seeInthisstudyweanalyzedthemutationaldatafrom�8,200spo-radiccancerstopredictcancerdrivergenes.WedeterminedthemostpredictiveparametersforidentifyingTSGsandOGsandusedthemtodevelopanalgorithmcalledTUSONExplorertopredicttheprobabilitythatanindividualgenefunctionsasaTSGoranOGincancer.Thisunbiasedapproachdemonstratedthattheprobabilityofbeingacancerdrivercanbeassessedbythesigni canceofthedistortionofitsmutationalpatternfromthepatternexpectedfora‘‘neutral’’gene.Combiningdatafromouranalysesofdriversandcopynumberchanges,theaveragetumorinourdatasethasameannumberof1OGmutation,3TSGmutations(LOFanddamagingmissense),3chromo-somalarmgains,5chromosomalarmlosses,2whole-chromosomegain,2whole-chromosomelosses,12focaldeletions,and11focalampli cations(Zacketal.,2013Thus,SCNAscompriseaverylargeproportionofcancer-drivingAContinuumofCancerDriverGenesAcentralconclusionfromthisstudyisthattherearelikelytobemanymorecancerdriversthananticipated.OurestimateofthenumberofTSGsbasedeitheronthecombinedsigni canceofthedifferentparametersoronthesinglebestparameterforthepredictionofTSGs,i.e.,theLOF/Benignratio,predictedTSGswiththecurrentdatabasefrom8,200tumors.Likewise,wealsopredictmoreOGsthananticipated.Theviewofthecan-cerlandscapeemergingfromouranalysisdoesnot
containaclearcutoffforpredictingcancerdrivers.Instead,thereexistsacontinuumofdecreasingprobabilityofagivengenebeingadriver(eitherTSGorOG).Thisprobabilityisrevealedbythedegreeofselectionthatthegeneexperiencesduringtumorevo-lution,whichshouldbeproportionaltothephenotypiceffectcausedbyitslossorgain.Thiscontinuumofdecreasingpotencyofpotentialcancerdriversislikelytocorrespondtoacontinuumofincreasingnumbersofgeneswithdecreasingphenotypicseverity,asillustratedschematicallyinFigure7A.Inaddition,wehypothesizethateventsthatsimultaneouslyaffectmultipleweakdriverscancumulativelyhaveaneffectequaltoasinglepotentdriver.Ourmodelingoftheprogressivelyhighernumberofdrivergenesidenti edasincreasingnumbersoftumorsareanalyzedsuggeststhatthisnumberwillcontinuetoclimbasmoresequenceinformationbecomesavailablebutmaybebeginningtoplateau.However,thenewlyidenti eddriversarelikelytodisplayprogressivelylesspotencywithlowertherapeu-ticsigni cance.ThisisanalogoustoGWASstudiesforwhichincreasingsamplesizesallowtheidenti cationofprogressivelyweakeractingvariants.Ouranalysisprovidesaprobabilityofeachgenebeingacan-cerdriver,andassuch,therewillbefalsepositivesregardlessofthethresholdofminimumprobabilitythatweemploy.Identifyingbona dedriversfromtheregionswithsigni cantpvaluesbuthigherFDRvalues,i.e.,weakerphenotypicsignatures,canbeaidedbyconsideringotherinformationsuchastheirinvolvementinSCNAs,biochemicalconnectionstoknownOGsandTSGs,andfunctionalinformationgleanedfromtheliterature.Theseheuristicmethodscanbeusedtoincreasecon denceandrescuegenesontothelikelycancerdriverlist(TablesS7andS7B).PAN-CancerandTissue-SpeciÞcAnalysisAnalysisofindividualtumortypesidenti eddistinctsetsofdriversineachtumortype,butthemajorityofthesewerealsoidenti edinthePAN-Canceranalysisaslowercon dencecandidates(TablesS4A–S4C).Thus,althoughthereisclearlytissuespeci city,thereisstillsigni cantoverlapamongdifferenttumortypesandaPAN-Canceranalysissamplesasuf cientnumberofsimilartumorstodetectmostofthelargelytissue-speci cortissue-biasedcancerdrivers.Ouranalysissuggeststhatsigni cantlydeepersequencingofindividualtu-mortypesisunlikelytouncovermanynewpotentdriversbeyondwhatwehavealreadyidenti edandfurthersequencingislikelytosufferfromdiminishingreturns.Thisviewisconsis-tentwitharecentreviewthatarguesthatnearlyallpotent ,948–962,November7,20132013ElsevierInc. drivershavebeenidenti ed(Vogelsteinetal.,2013Sequencingofmorerareandrelativelyunexploredcancertypesmayidentifyafewnovelpotentdriversthatarespeci ctothosetumortypes,butthevastmajorityofpotentdriverswillalreadyhavebeenseeninothercancers.ThemajoreffectsofcontinuedsequencingwilllikelybetosolidifythecontinuumbybringingmuchweakerdriversintotherealmofstatisticalPropertiesofNewPotentialCancerDriverGenesAnalysisofthelistsenrichedforcancerdriversrevealedseveralgeneralpropertiesthatdistinguishthemfromnondrivergenes.ThelistsofbothTSGsandOGsarestronglyenrichedbothforresidenceinproteincomplexesandforapropertyknownasbetweenness,whichisameasureofthedegreetowhichasetofgenesisenrichedforhubswithinaninteractionnetwork.Thus,thedrivergenesaremuchmorehighlyconnectedthantheaverageproteininthehumangenenetworkandarelonger.Highlyconnectednodesarebetterpositionedtocontrolthe owofinformation,andtheirremovalorhyperactiva-tionwillhavethehighestimpactacrossanetworkduetotheirUnexpectedPropertiesoftheXChromosomeGiventhepotentiallydeleteriouseffectsofmutatingTSGs,weanticipatedthatTSGswouldbedepletedfromtheXchro-mosomebynaturalselection,astheXishaploidinmalesandisfunctionallyhaploidinfemalesduetodosagecompensation.However,ouranalysisrevealedjusttheopposite—namely,thattheXhas86%moreTSGsthanexpected.Oncogenes,ontheotherhand,arenotoverrepresentedontheX.Thelikelyexplana-tionisthatadeleteriousmutationinaTSGontheXismorepene-trantbecausethereisnotaWTcopytocompensateforitsloss.ThisfurthersuggeststhatnaturalselectionhasnotcompletelydepletedTSGsfromtheX,possiblybecausecancerislargelyapostreproductivedisease. ABCD Figure7.CumulativeHaploinsufÞciencyandTriplosensitivityShapetheCancerGenomeIllustrativeschematicsofdifferentconceptshighlightedintheDiscussion.(A)Thephenotypiccontinuumofcancerdrivers.(B)ThecancergeneislandmodelforfocalSCNAs.(C)Thecumulativegenedosagebalancemodelforpredictingthepatternsofan
euploidy.(Thepaneldepictstheconceptforarm-levelSCNAs.)(D)ComparisonofthepredictionsofKnudson’sTwo-HitHypothesisforTSGscomparedtotheHaploinsuf ciencyHypothesispresentedinthisstudy. ,948–962,November7,20132013ElsevierInc. WefoundahighermutationratefortheXthanforautosomes,andthisisfurtherexaggeratedinfemales.Infemales,theaddi-tionalincreaseinXmutabilityislikelyduetothepresenceoftheinactiveX,whichhasverylittletranscriptionand,hence,lesstranscription-coupledrepairandisenrichedinlate-repli-catingheterochromatin,whichtendstobemoremutagenicStamatoyannopoulosetal.,2009).Themechanismunderlyingthesedifferencesandtheirbiologicalsigni canceremainstobedetermined.However,thesedifferencesmightindicatethatthemutationratesofwholechromosomesaresetbyevolutionandthatthehighermutabilityoftheXisadvantageousoverevolutionarytimeifitalsooccursinthegermline.HaploinsufÞciencyandCancerTheclonalexpansiontheoryoftumorigenesisarguesthat,inorderforanindividualmutationtobeselected,itmustcauseanexpansionoftheclonederivedfromthatmutantcellbyincreasingitsrelativeproliferationandsurvival(VogelsteinandKinzler,1993).ThisisintuitiveforOGs,astheyaredominant,butitislesssoforTSGs.ForahemizygousmutationinaTSGtobeselectedincancer,wehavetoassumethateitherthemutationisdominantnegativeortheTSGishaploinsuf cient.Ourcurrentanalysisofthedegreetowhichessentialgenesareabsentfromhemizygousrecurringfocaldeletions,coupledwiththereducedfrequencywithwhichessentialgenesexperi-enceLOFmutationsintumors,conservativelysuggestshaploinsuf ciencyoverallamonghumangenes().Arecentanalysisofhaploinsuf ciencybythemouseknockoutconsortium(Whiteetal.,2013)foundthat42%ofgenesexaminedproducedaphenotypewhenheterozy-gous,similartoourestimates.EvidencesuggestingthatoursporadicTSGlistislargelyhaploinsuf cientcomesfromacom-parisonoftheenrichmentinfocaldeletionsofSTOPgenesversusoursporadicTSGs.STOPgenes,whichareTSG-like,areenrichedby20%(Soliminietal.,2012).Ifweassumethatonly30%ofthisgenesetishaploinsuf cientandthatalloftheselectiveenrichmentcomesfromhaploinsuf cientgenes,thenalistofpurelyhaploinsuf cientSTOPgeneswouldbeexpectedtobeenrichedby67%.Perhapscoincidentally,ourlistofTSGsisenriched68%inrecurringfocaldeletions,sug-gestingthatasigni cantproportion,andpossiblyall,ofsporadicTSGsarehaploinsuf cient.WeproposethattwoclassesofTSGsmightexist:thosethatarehaploinsuf cientandcontributetosporadiccancerandthosethatareanddonotsigni cantlycontributetosporadiccancerthroughmutation.Circumstancesunderwhichorganismsinheritonlyonefunctionalcopyofthosehaplo-suf cientTSGsmightresultincancerbecauselossofthesec-ondallelewouldproduceaselectablephenotype.ThissituationoccurswithfamilialTSGsandtheclassicTwo-Hitmodeloftumorigenesis.Ourhypothesisisconsistentwiththefactthat,outofalistof73familialTSGsculledfromtheliterature,only32%ofthemhadacombinedqvalueinthePAN-Canceranalysis(TableS3D).Anothercircumstancewithonlyonefunc-tionalallelepercelloccursontheX,whereweseeahigherdensityofTSGsthanontheautosomes.Ifthepredictedrateof30%haploinsuf ciencyiscorrect,thenonemightexpecta200%increaseoverautosomes,butnegativeselec-tivepressureontheXcouldhavereducedthatnumber.Thus,itispossiblethatthereareactuallysimilardensitiesofTSGsontheXandautosomes(haploinsuf cientsporadicTSGsandhaplosuf cientpotentialTSGs),butthoseontheXrealizetheirtumorigenicpotentialatahigherratethandothoseontheThePAN-CancerMutationalAnalysisPredictsAneuploidyinCancerAneuploidyisahallmarkofcancerandcanhavebothadvanta-geousanddeleteriousconsequencesforcells(TangandAmon,2013;Luoetal.,2009),butthereisnogeneraltheorythatexplainshowpatternsofaneuploidyemerge.Knowingtheidentityandpotentialpotencyofcancerdrivershasallowedustouncoveradrivingforcebehindselectionofarm-andchromo-some-levelSCNAs.OuranalysisusingCharmandChromasanintegratedassessmentofthedensityandpotencyofthedifferentclassesofcancerdrivergenesonchromosomesdisplayedarobustabilitytopredictthepatternsofwhole-armampli cationsanddeletionsandaneuploidy(Figures6,and).ThefactthattheCharmscoreimprovesthecorrelationswithSCNAscomparedtothesimplegenedensityofthedifferentclassesofgenesindicatesthattherankingofdrivergenesbyTUSONExplorerislikelytorepresentanaccurateestimateofthepotencyoftheirphenotypiceffectincancerandfurthersupportsthecontinuumtheory.andCh
armdonotpredictarmampli cationaswellasCharm.ThisreducedpredictivepotentialislikelytobebecausetheOGswereselectedonthebasisoftheabilitytobeactivatedbymutationandbecausesimplyincreasingthedosageby50%mightnotstronglyimpactthenetworkstheycontrol.Charm,however,doesshowastrongnegativecorre-lationwitharmdeletionfrequency,indicatingthat,normally,theWTOGsareactingtopromoteproliferationandsurvivalandthecumulativereductionoftheirlevelsby50%isdeleterious.Inthisrespect,theOGsarebehavingliketheessentialgenes,andtheinclusionofahigh-con dencelistof332essentialgenestogetherwithOGsandTSGsfurtherimprovesthepredictiveabil-ityforarmdeletions(Figure6E).Asexpected,theessentialgeneshavenopredictivepowerforampli cations.stronglypredictswhole-armdeletions.Unexpect-edly,italsostronglypredictsarmampli cation,providingastrongnegativecorrelation.ThissuggeststhatincreasingthegenedosageofagroupofTSGscanhavedeleteriouseffectsontumorigenesisthroughtheprocessofcumulativetriplosensi-tivity.IfTSGsaretrulyhaploinsuf cient,theirWTproteinlevelsmaybeonlymarginallysuf cienttoexecutetheirroles.Ifso,TSGsmaywellbemoresensitivetoincreasedgenedosagetofurtherenhancetheirpathwaysthantypicalgenes.Inotherwords,haploinsuf cientgenesmaybemorelikelytodisplaytriplosensitivity.ThispropertyofsporadicTSGsbeingbothhaploinsuf cientandtriplosensitive,therefore,maymaketheircumulativeCharmscoreanevenbetterparametertoexplainSCNAsofchromosomearmsandaneuploidyingeneral.Devel-opingacombinedCharmandChromcannowpredict80%ofthefrequencyofarmandchromo-somelossand65%oftheampli cationsobservedacrossall ,948–962,November7,20132013ElsevierInc. AlthoughthecorrelationbetweenCharm/ChromscoresandSCNAsisstriking,thereareseveralareasforimprovement.The rstareaconcernsourlackofknowledgeofthefullcomple-mentofessentialgenesandwhichofthesearehaploinsuf cient.Second,onlyasubsetofOGswillbedosagesensitive,andthisknowledgewouldimprovethecorrelation.Inaddition,therearetwoclassesofOGs.ClassIcontainsclassicaloncogenessuchasKRASthatareactivatedbymutationbutwhoseWTcopiesarenotnecessarilyoncogenicafteroverexpressionandwillnotbepredictiveofampli cation.ClassIIcontainsgenessuchascyclinDthatcanbeactivatedbyoverexpressionbutaredif culttoactivatebymissensemutationsandthuslackamutationalsignature.ClassIIOGscannotbeidenti edwithcon dencethroughmutationalsignaturesyetarelikelytodisplaytriplosen-sitivityandwouldpositivelycorrelatewithampli cation.Third,someTSGscanbedif culttodistinguishfromOGs.TheseareTGSsthathavelowhaploinsuf ciencybutcanproduceaselect-ablephenotypebygenerationofdominant-negativealleles.SuchgeneswilllackastrongLOFsignaturebutwillshowasig-ni cantnumberofdeleteriousmissensemutations,whicharelikelytopredominantlyoccurinoneorafewcrucialresidues,thusconferringasigni cantEntropyscore.Inaddition,earlySCNAeventsmightin uencesubsequentevents,asisthecasewhenspeci caneuploidyco-occurs(Ozery-Flatoetal.,),whichwouldconfoundouranalysistosomedegree.Finally,re ningtheselistsofcancerdriverswillonlyimprovetheirpredictivepower.Thecurrentprogramsforpredictionhavetheirstrengthsandweaknessesandarelikelytobefurtherimprovedinthefuture.Morepreciseknowledgeoftheseessentialandcancerdrivergenesshouldsigni cantlyimproveSCNApredict-abilityandourunderstandingofthecancergenome.Finally,theSCNAfrequenciesmightvaryaccordingtotumortype;thus,comparisonofdatasetswithinonetumortypemightprovidemorepredictivepower.Inaddition,wedonotknowtheback-groundfrequencyofSCNAsuponwhichselectionacts,sotheobservedSCNAfrequencycannotbenormalizedlikemutationratescan,andtherefore,theobservedSCNAfrequencyde-tectedmightre ectbothfrequencyoftheeventanditsselectivepower,whichcouldconfoundthecorrelation.ModelsofCancerEvolutionOurworksuggestsaveryimportantroleforcumulativehaploin-suf ciencyandtriplosensitivityoperatingduringcancerevolu-tiontodrivetumorigenesis.Ineachgenomicregion,thereareSTOP(TSG)andGO(OGandessential)genesthatwillexertanegativeorpositivephenotypiceffectontumorigenesis.BothforfocaldeletionsasillustratedbytheCancerGeneIslandModelFigure7B)andforchromosomesandchromosomalarmSCNAsasindicatedbytheCharmandChromanalysis(Figure7C),theintegratedcumulativebalanceofthesepositiveandnegativetumorigeniceffectsofindividualgenesaffectedineachSCNAeventprovidestheselect
ivepotencytothateventandcanpre-dictitsfrequencyacrosscancers.Forthepast40years,thetumorsuppressor eldhasbeenguidedbyKnudson’sclassicalTwo-HitHypothesisoftumori-genesisforfamilialcancers.Thoughtherearecertainlybona deexamplesoftheTwo-HitModelinsporadiccancer,thismodelcon ictswiththetheoryofclonalevolutionofsporadiccancerintheassumptionthatthe rsthitisfullyrecessiveandasecondhitisrequiredtocontributetotumorigenesis.Whileitisdif culttomeasurethefrequencywithwhichtheTwo-HitHypothesisoperatesincancersbecausetheroleandextentofmethylationinactivationisnotyetknownineachtumor,analysisofLOFmutationaleventssuggeststhatitmaybearelativelyinfrequenteventexceptinthecaseofafewgenessuchas(p53)and(p16),bothofwhichareinducibleresponderstooncogenicstress,whichcanincreaseduringtumorigenesis.Inthecasesofsporadiccancer,whereinthetwo-hithypothesisdoesoperate,itisstillpossibleandevenprobablethatthegenesinvolvedarehaploinsuf cienttobeginwith.Ourresultshaveledustoproposethatthevastmajority,ifnotall,ofsporadicTSGsarelikelytobehaploinsuf cientandthat,therefore,sporadicTSGsaremostlikelytooperatethroughtheHaploinsuf ciencyModelshowninFigure7D.Itisimportanttonotethatthesehypothesesarenotmutuallyexclusive,aslossofthesecondalleleofahaploinsuf cientTSG,thesecondhit,willundoubtedlyprovideastrongerselectivepressurethanthe rsthit.However,atumorhasmultiplepathsthroughwhichtoevolve,anditmaynotrequirelossofthatsecondalleleasitobtainsgrowth-promotingpowerthroughtheaccumulationofotherevents.In1914,TheodorBoveriproposedthatspeci c‘‘chromosomeconstitutionscanbeproducedsuchthatthecellsthatharboritaredriventounrestrainedproliferation’’(Boveri,1929).Recurringpatternsofaneuploidyexistintumors,butwhethertheyexistbecauseofthefrequencyofoccurrenceofeachindividualSCNAeventorbecausetheyareselectedduetoatumorigenicphenotypiceffectwasnotknown.Here,weproposethatthecu-mulativephenotypiceffectsofgenedosagealterationsofSTOPandGOgenesprovidetheselectivepressurethatisresponsiblefortherecurrentpatternsofcopynumbervariationobservedincancer.Our ndingssupportthehypothesisputforwardbyBoveriexactlyonecenturyagothataneuploidyisnotonlyahall-markofcancer,butisalsoadrivingforceduringtheevolutionofhumancancer.EXPERIMENTALPROCEDURESSomaticMutationDataSetThedatasetofsomaticmutationsincludeddatafromTheCancerGenomeAtlas(TCGA,http://cancergenome.nih.gov/)researchnetworkandfromtheCatalogueofSomaticMutationsinCancer(COSMIC,http://cancer.sanger.ac.uk/cancergenome/projects/cosmic/)andthedatasetpublishedbydrovetal.(2013).Thedatasetcontained1,200,000mutationsfrom8,207tumorsamplesfrom�20tumortypes(TableS1)andwillbeavailableathttp://elledgelab.med.harvard.edu/.AlldatarelatedtoSCNAswerederivedfromtheTCGAGenomeDataanalysisCenterattheBroadInstitute(etal.,2013TUSONExplorerPredictionsThePolyPhen2algorithm(Adzhubeietal.,2010)wasusedtopredictthefunctionalimpactofeachmissensemutationandtoclassifythemashighfunctionalimpact(HiFI)orlowfunctionalimpact(LoFI).Wede nedthefourfollowingclassesofmutations:(1)Benignmutations:Silent+LoFIMissense;(2)LossofFunctionmutations(LOF):NonsenseandFrameshiftmutations;(3)Splicingmutations:mutationsaffectingsplicingsites;and(4)HiFImissensemutations.AnadditionalparameterconsideredwastheEntropyscore,whichmeasuresthedegreeofrandomnessofthedistributionofmissense ,948–962,November7,20132013ElsevierInc. Among22potentialparameters,weselectedthemostpredictiveonesbyusingtheLassopredictionmodelandthreetrainingsetsofknownTSGsandOGs(fromtheCancerGeneCensus,Futrealetal.,2004)andputativeneutralgenes.LOF/Benign,Splicing/Benign,andHiFI/BenignratioswereselectedbyLassoforthepredictionofTSGs,andtheHiFI/BenignratioandtheEntropyscorewereselectedforthepredictionofOGs.TUSONpredictionsarebasedonthecalculationofacombinedpvalue(andqvalue)oftheselectedparam-etersbyusinganextendedversionoftheLiptakmethod(TablesS3AandS3B).BasedonthecombinedpvaluesderivedwiththeTUSONmethod,weesti-matedthetotalnumberofpredictedTSGsandOGsbyusingahistogram-basedmethod(Mosigetal.,2001CharmandChromScoreandCorrelationwithFrequencyofSCNAsForeacharmandchromosome,respectively,theCharmandChromscoresforacertaingeneset(TSGs,OGs,oressentialgenes)representthedensityofthegenescontainedinthatsetweightedbytheirpredictedpotency.Thepotencyofeachgenecorrespondstoitsra
nkpositionwithinitsgenesetlistrankedbytheTUSONpvalueorbythe(LOF+1/2HiFi)/Benignratiofortheessentialgenes.ForthecumulativeCharmTSG-OG-EssandChromTSG-OG-Essscore,thescoresofOGsandessentialgenesweresubtractedfromthescoresrelativetotheTSGs.Thecorrelationanalysiswasperformedusingone-sidedPearson’scorrelationtestbetweentheCharmorChromscoreandthefrequencyofdeletionandampli cationofeacharmorchromosomeacrossalltumors(TableS6AnalysisofFunctionalGeneSetsTheSTOPgenelistwasderivedfromananalysisperformedusingRNAigeneenrichmentranking(RIGER)algorithm(Cheungetal.,2011)onapreviouslydescribedfunctionalshRNA-basedproliferationscreen(Soliminietal.,2012TableS5A).Aninsilicolistof332essentialgeneswasderivedbyconsideringtheintersectionbetweenthelistsofgenespredictedtobehousekeepinggenesandhighlyconservedgenes(Marcotteetal.,2012TableS5A).WeusedtheFisher’sexacttesttoexaminethesigni canceoftheassociationbetweenthepresenceofageneinrecurrentSCNAs(Beroukhimetal.,2010anditspresenceamongacertaingeneset.Foradditionalinformation,seetheExtendedExperimentalProceduresSUPPLEMENTALINFORMATIONSupplementalInformationincludesExtendedExperimentalProcedures, ve gures,andseventablesandcanbefoundwiththisarticleonlineatdx.doi.org/10.1016/j.cell.2013.10.011ACKNOWLEDGMENTSWethankSimonForbesandMichaelStratton(WellcomeTrustSangerInsti-tute,UK)forthedatafromtheCosmicdataset(http://cancer.sanger.ac.uk/cancergenome/projects/cosmic/),EricWootenforhelpondataextraction,andChadCreightonandKimRathmellforallowingaccesstotheirunpublisheddataonKICHSCNAs.WealsothankAndrewFutreal,SeminLee,DavidLiving-ston,DavidPage,Mary-ClaireKing,JimLupski,RameenBeroukhim,MattMeyerson,AtanasKamburov,andPaulEdwardsforhelpfuladviceandBertVogelstein,JudithGlaven,andmembersoftheElledgelabforhelpfulcom-mentsonthemanuscript.ThisworkwasfundedbyaDODBreastCancerInnovatorAwardandNIHgranttoS.J.E.,U54LM008748toP.J.P.,andK08DK081612toJ.C.Y.S.J.E.isaninvestigatorwiththeHowardHughesMedicalInstitute.Weapologizetoourcolleagueswhosepaperswecouldnotciteduetospacelimitations.Received:August20,2013Revised:September26,2013Accepted:October8,2013Published:October31,2013Adamson,B.,Smogorzewska,A.,Sigoillot,F.D.,King,R.W.,andElledge,S.J.(2012).Agenome-widehomologousrecombinationscreenidenti estheRNA-bindingproteinRBMXasacomponentoftheDNA-damageresponse.Nat.CellBiol.,318–328.Adzhubei,I.A.,Schmidt,S.,Peshkin,L.,Ramensky,V.E.,Gerasimova,A.,Bork,P.,Kondrashov,A.S.,andSunyaev,S.R.(2010).Amethodandserverforpredictingdamagingmissensemutations.Nat.Methods,248–249.Alexandrov,L.B.,Nik-Zainal,S.,Wedge,D.C.,Aparicio,S.A.,Behjati,S.,Bian-kin,A.V.,Bignell,G.R.,Bolli,N.,Borg,A.,Børresen-Dale,A.L.,etal.;AustralianPancreaticCancerGenomeInitiative;ICGCBreastCancerConsortium;ICGCMMML-SeqConsortium;ICGCPedBrain.(2013).Signaturesofmutationalprocessesinhumancancer.Nature,415–421.Beroukhim,R.,Mermel,C.H.,Porter,D.,Wei,G.,Raychaudhuri,S.,Donovan,J.,Barretina,J.,Boehm,J.S.,Dobson,J.,Urashima,M.,etal.(2010).Theland-scapeofsomaticcopy-numberalterationacrosshumancancers.Nature899–905.Bianchi,N.O.(2009).Ychromosomestructuralandfunctionalchangesinhumanmalignantdiseases.Mut.Res.,21–27.Boveri,T.(1929).TheOriginofMalignantTumors(Baltimore,MD:WilliamsandCancerGenomeAtlasNetwork.(2012).Comprehensivemolecularportraitsofhumanbreasttumours.Nature,61–70.Conti,P.,Youinou,P.,andTheoharides,T.C.(2007).Modulationofautoimmu-nitybythelatestinterleukins(withspecialemphasisonIL-32).Autoimmun.,131–137.Dees,N.D.,Zhang,Q.,Kandoth,C.,Wendl,M.C.,Schierding,W.,Koboldt,D.C.,Mooney,T.B.,Callaway,M.B.,Dooling,D.,Mardis,E.R.,etal.(2012).MuSiC:identifyingmutationalsigni canceincancergenomes.GenomeRes.,1589–1598.Forbes,S.A.,Tang,G.,Bindal,N.,Bamford,S.,Dawson,E.,Cole,C.,Kok,C.Y.,Jia,M.,Ewing,R.,Menzies,A.,etal.(2010).COSMIC(theCatalogueofSomaticMutationsinCancer):aresourcetoinvestigateacquiredmutationsinhumancancer.NucleicAcidsRes.(Databaseissue),D652–D657.Futreal,P.A.,Coin,L.,Marshall,M.,Down,T.,Hubbard,T.,Wooster,R.,Rah-man,N.,andStratton,M.R.(2004).Acensusofhumancancergenes.Nat.Rev.,177–183.Gudjonsson,T.,Altmeyer,M.,Savic,V.,Toledo,L.,Dinant,C.,Grøfte,M.,Bart-kova,J.,Poulsen,M.,Oka,Y.,Bekker-Jensen,S.,etal.(2012).TRIP12andUBR5suppressspreadingofchromatinubiquitylationatdamagedchromo-somes.Cell,697&#
150;709.Hanahan,D.,andWeinberg,R.A.(2011).Hallmarksofcancer:thenextgener-ation.Cell,646–674.Hou,J.M.,Krebs,M.,Ward,T.,Sloane,R.,Priest,L.,Hughes,A.,Clack,G.,Ranson,M.,Blackhall,F.,andDive,C.(2011).Circulatingtumorcellsasawindowonmetastasisbiologyinlungcancer.Am.J.Pathol.,989–996.Kowalski,J.,Morsberger,L.A.,Blackford,A.,Hawkins,A.,Yeo,C.J.,Hruban,R.H.,andGrif n,C.A.(2007).Chromosomalabnormalitiesofadenocarcinomaofthepancreas:identifyingearlyandlatechanges.CancerGenet.Cytogenet.,26–35.Lai,K.,Amsterdam,A.,Farrington,S.,Bronson,R.T.,Hopkins,N.,andLees,J.A.(2009).Manyribosomalproteinmutationsareassociatedwithgrowthimpairmentandtumorpredispositioninzebra sh.Dev.Dyn.Lawrence,M.S.,Stojanov,P.,Polak,P.,Kryukov,G.V.,Cibulskis,K.,Sivachenko,A.,Carter,S.L.,Stewart,C.,Mermel,C.H.,Roberts,S.A.,etal.(2013).Mutationalheterogeneityincancerandthesearchfornewcancer-associatedgenes.Nature,214–218.Luo,J.,Solimini,N.L.,andElledge,S.J.(2009).Principlesofcancertherapy:oncogeneandnon-oncogeneaddiction.Cell,823–837.Matsuoka,S.,Ballif,B.A.,Smogorzewska,A.,McDonald,E.R.,3rd,Hurov,K.E.,Luo,J.,Bakalarski,C.E.,Zhao,Z.,Solimini,N.,Lerenthal,Y.,etal.(2007).ATMandATRsubstrateanalysisrevealsextensiveproteinnetworksresponsivetoDNAdamage.Science,1160–1166.Meyerson,M.,Gabriel,S.,andGetz,G.(2010).Advancesinunderstandingcancergenomesthroughsecond-generationsequencing.Nat.Rev.Genet.,685–696. ,948–962,November7,20132013ElsevierInc. Mosig,M.O.,Lipkin,E.,Khutoreskaya,G.,Tchourzyna,E.,Soller,M.,andFriedmann,A.(2001).AwholegenomescanforquantitativetraitlociaffectingmilkproteinpercentageinIsraeli-Holsteincattle,bymeansofselectivemilkDNApoolinginadaughterdesign,usinganadjustedfalsediscoveryratecriterion.Genetics,1683–1698.Ozery-Flato,M.,Linhart,C.,Trakhtenbrot,L.,Israeli,S.,andShamir,R.(2011).Large-scaleanalysisofchromosomalaberrationsincancerkaryotypesrevealstwodistinctpathstoaneuploidy.GenomeBiol.,R61.Pavlova,N.N.,Pallasch,C.,Elia,A.E.,Braun,C.J.,Westbrook,T.F.,Hemann,M.,andElledge,S.J.(2013).AroleforPVRL4-drivencell-cellinteractionsintumorigenesis.Elife,e00358.Ruepp,A.,Waegele,B.,Lechner,M.,Brauner,B.,Dunger-Kaltenbach,I.,Fobo,G.,Frishman,G.,Montrone,C.,andMewes,H.W.(2010).CORUM:thecomprehensiveresourceofmammalianproteincomplexes—2009.NucleicAcidsRes.(Databaseissue),D497–D501.Shannon,C.E.,andWeaver,W.(1949).TheMathematicalTheoryofCommu-nication(Urbana,IL:UniversityofIllinoisPress).Solimini,N.L.,Xu,Q.,Mermel,C.H.,Liang,A.C.,Schlabach,M.R.,Luo,J.,Bur-rows,A.E.,Anselmo,A.N.,Bredemeyer,A.L.,Li,M.Z.,etal.(2012).Recurrenthemizygousdeletionsincancersmayoptimizeproliferativepotential.Science,104–109.Stamatoyannopoulos,J.A.,Adzhubei,I.,Thurman,R.E.,Kryukov,G.V.,Mirkin,S.M.,andSunyaev,S.R.(2009).HumanmutationrateassociatedwithDNAreplicationtiming.Nat.Genet.,393–395.Stark,C.,Breitkreutz,B.J.,Reguly,T.,Boucher,L.,Breitkreutz,A.,andTyers,M.(2006).BioGRID:ageneralrepositoryforinteractiondatasets.NucleicAcidsRes.(Databaseissue),D535–D539.Stratton,M.R.,Campbell,P.J.,andFutreal,P.A.(2009).Thecancergenome.,719–724.Tang,Y.C.,andAmon,A.(2013).Genecopy-numberalterations:acost-bene tanalysis.Cell,394–405.Vogelstein,B.,andKinzler,K.W.(1993).Themultistepnatureofcancer.Trends,138–141.Vogelstein,B.,Papadopoulos,N.,Velculescu,V.E.,Zhou,S.,Diaz,L.A.,Jr.,andKinzler,K.W.(2013).Cancergenomelandscapes.Science,1546–White,J.K.,Gerdin,A.K.,Karp,N.A.,Ryder,E.,Buljan,M.,Bussell,J.N.,Salis-bury,J.,Clare,S.,Ingham,N.J.,Podrini,C.,etal.;SangerInstituteMouseGeneticsProject.(2013).Genome-widegenerationandsystematicphenotyp-ingofknockoutmicerevealsnewrolesformanygenes.Cell,452–464.Willig,T.N.,Gazda,H.,andSieff,C.A.(2000).Diamond-Blackfananemia.Curr.Opin.Hematol.,85–94.Wood,L.D.,Parsons,D.W.,Jones,S.,Lin,J.,Sjoblom,T.,Leary,R.J.,Shen,D.,Boca,S.M.,Barber,T.,Ptak,J.,etal.(2007).Thegenomiclandscapesofhumanbreastandcolorectalcancers.Science,1108–1113.Zack,T.I.,Schumacher,S.E.,Carter,S.L.,Cherniack,A.D.,Saksena,G.,Tabak,B.,Lawrence,M.S.,Zhang,C.-Z.,etal.(2013).Pan-cancerpatternsofsomaticcopynumberalteration.Nat.Genet.,1134–1140.Zhang,D.,Zaugg,K.,Mak,T.W.,andElledge,S.J.(2006).Aroleforthedeu-biquitinatingenzymeUSP28incontroloftheDNA-damageresponse.Cell,529–542. ,948–962,November7,20132013