METHODSARTICLE published:28January2013 doi:10.3389/fphar.2013.00002 Clonal evolution and therapeutic resistance in solid tumors MichaelT.Barrett1,2*,ElizabethLenkiewicz1,LisaEvers1,TaraHolley1,ChristianRuiz3,LukasBubendorf3, AleksanderSekulic4,RameshK.Ramanathan1,5 andDanielD.VonHoff1,5 1TheTranslationalGenomicsResearchInstitute,Scottsdale,AZ,USA 2MayoClinicArizona,Scottsdale,AZ,USA 3InstituteforPathology,UniversityofBasel,Basel,Switzerland 4Dermatology,MayoClinicArizona,Scottsdale,AZ,USA 5VirginiaPiperCancerCenter,ScottsdaleHealthcare,Scottsdale,AZ,USA Editedby: Tumorsfrequentlyariseasaresultofanacquiredgenomicinstabilityandthesubsequent GeraldBatist,McGillUniversity, evolution of neoplastic populations with variable genomes. A barrier to the study of the Canada somatic genetics of human solid tumors in vivo is the presence of admixtures of non- Reviewedby: neoplasticcellswithnormalgenomesinpatientsamples.Thesecanobscurethepresence BrionW.Murray,PfizerOncology ResearchUnit,USA of somatic aberrations including mutations, homozygous deletions, and breakpoints in FernandoRodríguez-Serrano, biopsies of interest. Furthermore, clinical samples frequently contain multiple neoplastic UniversityofGranada,Spain populations that cannot be distinguished by morphology. Consequently, it is difficult to *Correspondence: determine whether mutations detected in a sample of interest are concurrent in a sin- MichaelT.Barrett,TheTranslational gle clonal population or if they occur in distinct cell populations in the same sample.The GenomicsResearchInstitute, Scottsdale,AZ85259,USA. adventoftargetedtherapiesincreasestheselectionforpreexistingpopulations.However e-mail:[email protected] theasymmetricdistributionoftherapeutictargetsinclonalpopulationsprovidesamecha- nismfortherapidevolutionofresistantdisease.Thus,thereisaneedtonotonlyisolate tumorfromnormalcells,buttoalsoenrichdistinctpopulationsofclonalneoplasticcellsin ordertoapplygenometechnologiestoidentifyclinicallyrelevantgenomicaberrationsthat drivediseaseinpatientsinvivo.Toaddressthiswehaveappliedsingleandmultiparame- terDNAcontentbasedflowassaystothestudyofsolidtumors.Ourworkhasidentified examples of clonal resistance to effective therapies.This includes androgen withdrawal in advanced prostate cancer. In addition we demonstrate examples of co-existing clonal populations with highly aberrant genomes and ploidies in a wide variety of solid tumors. We propose that clonal analysis of tumors, based on flow cytometry and high resolution genome analyses of purified neoplastic populations, provides a unique approach to the studyoftherapeuticresponsesandtheevolutionofresistance. Keywords:solidtumors,flowcytometry,clonalevolution,aCGH,nextgenerationsequencing INTRODUCTION evolutionarymodelsoftumordevelopmentandprogressionbased Nowell (1976) proposed that tumors arose as a result of an onheterogeneouspatternsofmutationsinprimaryandmetastatic acquired genomic instability and the subsequent selection and lesions(Gerlingeretal.,2012).Theseadvancesintechnologyand evolution of clonal populations of neoplastic cells with unique cancergenomicsprovidearichopportunitytostudythebasisof patternsof aberrations.Consequently,eachpatient’scancermay therapeuticresponsesandtheevolutionofresistance. evolve and become dependent on distinct sets of selected aber- Afundamentalhypothesisincancerbiologyisthatthegenes rations during its clinical history. This clonal evolution model and cellular pathways deregulated by those selected events in wasbasedonstudiesprimarilyof liquidtumorsusinglowreso- eachtumorgenomerepresentenrichedcandidatesfordeveloping lutioncytogeneticsthatshoweduniformkaryotypesofallcellsin diagnosticmarkersandtherapeutictargets.However,thecellular individualpatientsamples,biochemicalassays,andallele-specific heterogeneityofclinicalsamplesandthegeneticdiversityofcancer chromosome X expression of glucose-6-phosphate isoenzymes genomesarebarrierstothetranslationofgenomicsforimproved in all cells of a variety of neoplasms in heterozygous women, patientcare.Thepresenceofadmixturesofnon-neoplasticcellsin andimmunologystudiesofhomogeneousimmuno-globulinsin patientsamplescanobscurethedetectionofsomaticaberrations plasmacelltumors.Furthermorethismodelpredictedthatclonal including mutations,homozygous deletions,and breakpoints in behaviorwouldmediateemergenceofresistanceevenafteranini- biopsiesofinterest.Furthermore,clinicalsamplesfrequentlycon- tialclinicallysignificantresponse.Advancesingenometechnolo- tainmultipleneoplasticpopulationsthatcannotbedistinguished gies, notably oligonucleotide-based arrays and next generation by morphology based methods (Rabinovitch et al.,1999; Maley sequencing (NGS) platforms, have greatly increased the resolu- et al., 2006). Consequently, it is difficult to determine whether tion potential for studying the basis of therapeutic resistance in mutations and aberrations in a sample of interest are concur- humancancers.Anumberof recentNGSstudieshaveproposed rentinasingletumorpopulationoriftheyoccurindistinctcell www.frontiersin.org January2013|Volume4|Article2|1 Barrettetal. Clonalevolutionandsolidtumors populationsinthesamesample.Thus,thereisaneedtonotonly genomic aberrations including somatic mutations that target isolatetumorfromnormalcells,buttoalsoenrichdistinctpopu- clinicallyrelevantsignalingpathwaysinpatientsinvivo. lationsofclonalneoplasticcellsinordertostudycancergenomes Inordertoaddressthecomplexityandheterogeneityofclinical inpatientsinvivo. sampleswehaveappliedhighlyquantitativeDNAcontentbased Cancer genome studies have relied primarily on two main flow cytometry assays to identify and subsequently sort distinct approachesforselectingandpreparingsamplesforanalyses.The fractions of tumor and non-tumor populations in each sample first is to pre select biopsies that exceed an arbitrary threshold of interest (Ruiz et al., 2011; Liu et al., 2012). Flow cytometry for tumor cell content and necrosis based on histological meth- cellsorterscanselectandobjectivelymeasureindividualparticles odssuchasH&Estaining(CancerGenomeAtlasResearch,2011; suchascellsornucleiusingdesiredfeaturesdefinedbyfluorescent Guichard et al., 2012; Nik-Zainal et al., 2012). However many andlightscatteringparametersinaflowstream.Onceidentified samples fail the criteria; this is especially true for tumors aris- desiredsubpopulationscanbedeflectedinthestreambyanelectric ing in solid tissues where a high degree of tissue heterogeneity, fieldandcollectedaspureobjectivelydefinedpopulations.Recent withvariedadmixturesofreactivestroma,inflammatorycells,and advancesinthistechnologycombinehighefficiencyflowrateswith necrosisinimmediatecontactwithtumorcells,isaverycommon thedetectionof relativelyrareeventsindilutesamplesenabling feature. Laser capture microdissection (LCM) can enrich tumor theapplicationof flowcytometrytoclinicalbiopsiesfromclini- cell content prior to analysis. However LCM is limited in abil- caltissues(IbrahimandvandenEngh,2003,2008).Forexample, itytoobjectivelydistinguishmultipleclonalpopulationsinsingle therearewellestablishedDNAstainingbasedmethodsforisolat- biopsiesandinprocessingthroughputforheavilyadmixedsam- ingnucleiofaneuploidanddiploidneoplasticpopulationsfrom ples. The second approach is to passage biopsies either in tissue solid tumor samples (Rabinovitch,1994; Glogovac et al.,1996). cultureorinmousexenografts(Jonesetal.,2008).Thesemeth- Tumorpopulationscanbeobjectivelyandquantitativelypurified ods apply a selective pressure on the complex mixtures of cells togreaterthan95%purityformolecularanalyseseveninheav- and clones present in a patient sample. Furthermore they can ily admixed clinical samples. In addition these methods can be be time consuming, labor intensive, and cannot be efficiently combinedwitheitherproliferationmarkerssuchasKi67ortissue applied in most clinical settings. Consequently the number of specific markers such as cytokeratins to purify distinct popula- samples that can be successfully passaged varies from site to tions of diploid neoplastic cells from samples of interest (Prevo site, and the biological complexity and clinical context of the et al., 1999; Loo et al., 2004; Galipeau et al., 2007). Flow sort- patientsamplemaynotbereflectedinthefinalprocessedsample. inghasbeensuccessfullyusedforavarietyof humanneoplasias Morerecently,studieshaveusedmixturesof tumorandnormal includingbreast,esophageal,lung,andcoloncarcinomas(Lofberg cells to estimate thresholds for the detection of somatic muta- etal.,1992;Tanakaetal.,1995;Takanishietal.,1996;Barlettaetal., tions to compensate for the complex composition and cellular 1998;Looetal.,2004;Laietal.,2007).Recentlywehavedeveloped admixtures of clinical samples (Biankin et al., 2012). However methodologiesthatenablethestudyofflowsortedclinicalsamples thesemethodsdonotfullyaccountforclonalheterogeneity,are with high definition whole genome assays, including oligonu- dependentondepthofreadsacrosseachgenomeofinterest,and cleotidearraybasedcomparativegenomichybridization(aCGH) requireadditionalprocessingandfilteringofincreasinglycomplex andNGS(Ruizetal.,2011;Holleyetal.,2012).Theapplication datasets. of clonalanalysis,basedonwholeexomeandgenomemeasure- Amajorobjectiveforcancergenomestudiesistodistinguish ments of purified populations from clinical samples,provides a “driver”aberrationsthattargetclinicallysignificantsignalingpath- highlyfavorableapproachtostudyingtherapeuticresponsesand ways from “passenger” aberrations that arise as a result of the advancingmoreeffectivetargetedtherapies. biological background in genomically unstable tumors. A com- mon approach for identifying clinically relevant cancer genome MATERIALSANDMETHODS aberrations is to characterize lesions, including any loss or any Pancreaticductaladenocarcinoma(PDA)sampleswereobtained gain for each chromosome, occurring at rates that are statisti- underaWIRBprotocol(20040832)foranNIHfundedbiospeci- callysignificantinsamplesof interest(Aguirreetal.,2004;Weir menrepository(NCIP01GrantCA109552)andtwoAACR/Stand et al., 2007; Chen et al., 2008; Kimmelman et al., 2008; Bredel uptoCancer(SU2C)sponsoredclinicaltrials20206-001and2026- et al.,2009;Liu et al.,2009). Single copy losses and gains occur 003.Additional PDA samples as well as prostate carcinoma and as part of the random events associated with genomic instabil- ovariancarcinomasampleswereobtainedwithapprovedconsent itypresentintumorgenomes.Thusmanycancergenomestudies oftheEthicsCommitteeofBasel(252/08,302/09).Themelanoma determine a statistical threshold based on the background rates tissue samples analyzed in this study were obtained under the of losses and gains for detecting selected copy number changes. Institutional Review Board approved protocols at Mayo Clinic. However this requires relatively large numbers of patient sam- In all cases the tissues were acquired and the study was con- ples to account for the genomic instability and the patient spe- ducted according to Good Clinical Practice and the Declaration cificvariationstypicallypresentincancergenomes.Furthermore ofHelsinkiPrincipleswithinformedconsentfromeachpatient. recentNGSbasedstudieshaveshownthatthegenomiclandscapes of most solid tumors contain a wide variety of low prevalent FRESHTISSUESAMPLEPREPARATIONANDFLOWSORTING mutations (McKenna et al., 2010; Biankin et al., 2012). Thus Whereverpossibleclinicalsamplesarecollectedinliquidnitrogen a fundamental challenge in the translation of cancer genomes andstoredat−80˚C.Theseincludeneedlebiopsies(e.g.,18g),sur- for improved patient outcomes is the identification of selected gicallyresectedtissues,pleuraleffusions,andcoreneedlebiopsies FrontiersinPharmacology|PharmacologyofAnti-CancerDrugs January2013|Volume4|Article2|2 Barrettetal. Clonalevolutionandsolidtumors fromvariousorgansites.Alternativelysamplesmaybecollected fluidpressureistypically17–18psiwitha100µmnozzle.Forsin- insmallaliquotsoftissueculturemediasupplementedwith10% gleparameterDNAcontentassaysDAPIemissioniscollectedat DMSOandheldonwetice.Thesecanthenbetransferredto−80˚C >450nm.Ineachsortingexperimentweusedoneormore50µm forlongtermstorage.TheuseofmediawithDMSOprovidesfur- FFPEscrollstoobtainsufficientnumbersof intacttumornuclei thertissuepreservationinthosecaseswheremultiplesamplingof forsubsequentmolecularassays.DNAcontentandcellcycleare thetissueinvolvescyclesof freezethawing.Priortosortingeach then analyzed using the software program MultiCycle (Phoenix biopsyisquicklythawedonicethenmincedinthepresenceofNST FlowSystems,SanDiego,CA,USA). buffer and DAPI according to published protocols (Rabinovitch GatingbasedonDNAcontentprovidesarobustquantitative etal.,2001;Maleyetal.,2006;Galipeauetal.,2007).Nucleiare measureforidentifyingandsortingtumorpopulationsfromsam- mechanicallydisaggregatedthenfilteredthrougha40-µmmesh ples of interest. The ploidy and the relative distribution of each priortoflowsortingwithanInfluxcytometer(Becton-Dickinson, population present in a biopsy can be recovered by fitting the SanJose,CA,USA)withultravioletexcitationandDAPIemission G /G andG /MpeaksasGaussiancurvesandtheSphasedistri- 0 1 2 collected at >450nm. DNA content and cell cycle are analyzed butionasaGaussianbroadeningdistribution.TheDNAcontent using the software program MultiCycle (Phoenix Flow Systems, histograms from tumor tissue are frequently suboptimal (broad SanDiego,CA,USA). c.v’s, high debris and aggregation) and often complex (multi- pleoverlappingpeaksandcellcycles)withfrequentskewingand FORMALINFIXEDPARAFFINEMBEDDEDSAMPLEPREPARATIONAND non-Gaussianpeakshapes.ThisiseventruerforFFPEspecimens FLOWSORTING thatoftencontainhigherlevelsofdamagedorfragmentednuclei Formalinfixedparaffinembedded(FFPE)samplesarefixedinfor- (debris)resultingineventsusuallymostvisibletotheleftof the malin at the time of collection then stored according to routine diploidG peakandthatfallrapidlytobaseline.Forreproducible 1 pathology methods. Prior to sorting excess paraffin is removed phasemeasurementswetypicallyacquire10,000events.However with a scalpel from either side of 40–60µm scrolls to reduce if a substantial proportion of events are from debris or aggre- accumulationof debrisduringthesortingprocess.Eachscrollis gates,thetotalnumberofeventsacquiredmustbecorrespondingly collectedintoindividualmicrocentrifugetubesthenwashedthree higherinordertoassuretherequiredminimumnumberofintact times with 1ml Xylene for 5min to remove remaining paraffin. singlenucleiforaccuratecurvefitting.Despitethesechallengesdis- Each sample is rehydrated in sequential ethanol washes (100% tinctpopulationscanbeefficientlysortedfromroutinelyprepared 5min ×2, then 95, 70, 50, and 30% ethanol) and washed two FFPEsamples. times in 1ml of 1mM EDTA pH 8.0. A 1-ml aliquot of 1mM EDTApH8.0isaddedtothesamplesandincubatedat95˚Cfor DNAEXTRACTION 80mintofacilitatetheremovalof proteincross-linkspresentin The genomic DNA from sorted nuclei is extracted using an FFPE tissue. Samples are then cooled to room temperature for amendedprotocolfromQIAamp®DNAMicroKitfromQiagen ≥5min, followed by addition of 300µl PBS pH 7.4 and gentle (Valencia,CA,USA). Briefly each sorted sample is resuspended centrifugation for 2min at 3.6×g. The supernatant is carefully in 180µl bufferATL and 20µl proteinase K then incubated for removed and the pellet washed three times with 1ml PBS pH 3h at 56˚C for complete lysis. Samples are bound and washed 7.4/0.5mMCaCl toremoveEDTA.Eachsampleisthendigested accordingtoQIAamp®DNAMicroKitinstructions,elutedinto 2 overnight(6–17h)in1mlofafreshlypreparedenzymaticcocktail 50µl of H 0,then precipitated overnight with 5µl 3M sodium 2 containing50units/mlofcollagenasetype3,80units/mlofpuri- acetateand180µl100%EtOH.Eachsampleisthencentrifuged fied collagenase, and 100units/ml of hyaluronidase in PBS pH for30minat20,000×g,washedin1mlof70%EtOHfor30min 7.4/0.5mMCaCl buffer.EachenzymeisrehydratedwithPBSpH at 20,000×g. The samples are carefully decanted and the DNA 2 7.4/0.5mMCaCl bufferthenstoredat−20˚Cimmediatelyprior pelletwasdriedbyspeedvacuumthenresuspendedinasmallvol- 2 to addition to the cocktail mixture. Following overnight diges- ume(e.g.,10–50µl)ofH Oforfinalconcentrationssuitablefor 2 tion 500µl NST is added to each sample to facilitate pelleting. accuratequantification. Samples are then centrifuged for 5min at 3000×g,after which pelletsareresuspendedin750µlofNST/10%fetalbovineserum DNAAMPLIFICATION and then passed through a 25-G needle 10–20 times. The sam- Tomakeefficientuseofsortedclinicalsampleswehaveoptimized plesarefilteredthrougha35µmmeshandcollectedintoa5ml wholegenomeamplificationmethodstogenerateDNAtemplates Polypropylene round bottom tube. The mesh is rinsed with an suitableforwholegenomeandexomeanalyses.Sortedpopulations additional750µlofNST/10%fetalbovineserumandplacedon fromfreshfrozensamplescanbeamplifiedwithhighlyprocessive icewhileprocessingremainingsamples.Thetotalvolumeinthe enzymessuchasphi29thatrequireintactgenomicDNAasstarting tubeforeachsampleisapproximately1.5ml.Anequalvolumeof material.ExtractedfreshfrozensourcedgenomicDNAisampli- 20µg/mlDAPIisthenaddedtoeachtubetoachieveafinalcon- fiedusingtheIllustraGenomiPhiV2AmplificationkitfromGE centrationof10µg/mlDAPIpriortoflowsortingwithanInflux Healthcare Bio-sciences, Corp. (Piscataway, NJ, USA) according cytometerwithultravioletexcitation(Becton-Dickinson,SanJose, to our published protocols (Ruiz et al., 2011). Typically we use CA,USA).TheoptimalsettingsforsortingFFPEsampleswiththe >50ng genomic DNA from each sorted population as input to Influxsorterareasfollows:dropformationisachievedwithpiezo ensurelinearamplificationacrossthegenome. amplitude of 6–10V and a drop frequency of 30kHz. The sort ForFFPEsamplestheDNAtemplatesarenotsuitableforampli- modeissettopurityyieldwithadropdelayof 31.5–32.Sheath fication protocols with highly processive polymerases. Thus we www.frontiersin.org January2013|Volume4|Article2|3 Barrettetal. Clonalevolutionandsolidtumors validated an alternative method based on single primer based kit,andIlluminaindexedadaptersarenextligatedontoA-tailed amplification of size selected double strand DNA input (Hol- products. Samples are next PCR amplified using Herculase II ley et al.,2012). Genomic DNAs from sorted FFPE samples are polymerase and purified. Samples are then run on an Agilent amplifiedusingOvation®WGAFFPESystemfromNuGEN®Tech- Bioanalyzertoverifyamplificationandtoquantifysamples.Sam- nologies(SanCarlos,CA,USA).DNAisprocessedinaccordance ples are adjusted to 147ng/µL for 24h hybridization to exonic withOvation®WGAFFPEstandardSPIAprotocolwithanalter- RNA probes using Agilent’s SureSelect All Exon 50Mb Plus kit, nateT7endonucleasefragmentationstep.Theseprotocolsrequire whichcontains561,823probestargeting202,124exons.Captured input from 50,000 sorted nuclei as starting material to ensure productsarenextselectedfor,purified,andPCRamplified.Final a linear amplification. Resulting amplified products are either librariesareverifiedandquantifiedusinganAgilentBioanalyzer. usedastemplateforaCGHanalysisorprocessedwiththeNugen Encore ds-DNA module according to the supplier’s instructions PAIREDENDNEXTGENERATIONSEQUENCING in order to generate double-stranded (ds) end-repaired DNA as Libraries are denatured using 2N NaOH and diluted with HT2 inputforlibrarysuitableforNGS.InallcasesA50–100ngaliquot buffer (Illumina). One percent of denatured and diluted phiX ofpooled46,XXDNA(Promega,Madison,WI,USA)isamplified is spiked into each lane to allow for error rate reporting on the with the matching amplification protocol to generate a suitable HiSeq.ClustergenerationisperformedusingIllumina’scBotand referenceforeachaCGHexperimentusingamplifiedDNAtem- HiSeq Paired End Cluster Generation Kit. Flow cells are paired plates.Thequalityofeachamplificationproductisassessedbygel endsequencedonIllumina’sHiSeq2000usingIllumina’sHiSeq electrophoresispriortodownstreamanalysis. Sequencing Kit. Raw sequencing data are converted to standard FASTQ format using CASAVA pipeline with in-house custom aCGHANALYSIS scripts1,2. FASTQC program is used for quality control and all Freshfrozenphi29amplifiedandFFPEnon-amplifiedDNAsare readsaretrimmedto90high-qualitybasepairs.Inordertogen- treatedwithDNAse1priortoKlenow-basedlabeling.Highmol- erateatleast100millionpassfilterreadsforeachexomelibrary, ecular weight phi29 templates are digested for 30min while the twolanesofaHiSeq2000flowcellaresequencedforeachofthe smallerfragmentedFFPEsamplesaredigestedforonly1min.In FFPEandfreshfrozenexomes.Dataisalignedtohg18assemblyof eachcase1µlof10×DNase1reactionbufferand2µlofDNase humangenomeusingBWAsequencealignmentsoftware(version 1dilutionbufferareaddedto7µlofDNAsampleandincubated 0.5.9)andrawalignmentBAMfilesarefurtherprocessedforqual- atroomtemperaturethentransferredto70˚Cfor30mintodeac- ityrecalibration,duplicateremoval,andlocalrealignmentusing tivateDNase1.IncontrasttheamplifiedFFPEsourcedDNAsdo acustomin-housepipelinebasedonPicardandGATKtools3(Li not require DNase 1 treatment prior to Klenow-based labeling. andDurbin,2009;McKennaetal.,2010;DePristoetal.,2011).For InallcasessampleandreferencetemplatesarelabeledwithCy-5 eachsample,variantsarecalledfromBAMfilesusingsamtoolsand dUTPandCy-3dUTPrespectivelyusingaBioPrimelabelingkit varscanusingaminimumcoveragecut-offof10,andonlythose (Invitrogen,Carlsbad,CA,USA)accordingtoourpublishedpro- variantsthatarecalledbybothalgorithmsareretained(Koboldt tocols(Ruizetal.,2011).Alllabelingreactionsareassessedusing etal.,2009;Lietal.,2009). a Nanodrop assay (Nanodrop, Wilmington, DE, USA) prior to mixingandhybridizationto400kCGHarrays(AgilentTechnolo- RESULTS gies,SantaClara,CA,USA)for40hinarotating65˚Coven.All CLONALANALYSIS:SINGLEPARAMETERDNACONTENTBASEDFLOW microarrayslidesarescannedusinganAgilent2565CDNAscan- SORTING nerandtheimagesareanalyzedwithAgilentFeatureExtraction Flow cytometry methods can discriminate distinct populations version 10.7 using default settings. The aCGH data are assessed in each biopsy of interest based on one or more features. The with a series of QC metrics then analyzed using an aberration basic principle for flow sorting is to interrogate single parti- detectionalgorithm(ADM2;Lipsonetal.,2006).Thelatteriden- cles in suspension for desired parameters. These include ploidy tifiesallaberrantintervalsinagivensamplewithconsistentlyhigh as defined by the mean DNA content, as well as differentiation or low log ratios based on the statistical score derived from the or cell lineage markers. Clonal populations in tumors can then averagenormalizedlogratiosofallprobesinthegenomicinterval be distinguished by differences in ploidy, copy number aberra- multipliedbythesquarerootofthenumberoftheseprobes.This tions,mutations,ordifferentiation.Forsolidtumorsamplestissue scorerepresentsthedeviationoftheaverageofthenormalizedlog are mechanically disaggregated and the nuclei suspended in the ratiosfromitsexpectedvalueof zeroandisproportionaltothe presence of DAPI. The isolation of nuclei provides an efficient heighth(absoluteaveragelogratio)ofthegenomicinterval,and mechanism to prepare single particle suspensions that can be tothesquarerootofthenumberofprobesintheinterval. interrogated in a flow stream.A hallmark of many human can- cersisthedevelopmentof genomicinstabilityandtheevolution EXOMELIBRARYPREPARATION of aneuploid tumor cell lineages. Thus for many solid tumors Atotalof3µgofhigh-qualityDNAtemplatewitha260/280ratio DNAcontentassayscanbeveryeffectiveinidentifyingandsubse- between 1.8 and 2.1 is fragmented to a target size of 150–200 quentlysortingneoplasticpopulationsforgenomicanalysis.For base pairs on the Covaris E210 system. Fragmentation is veri- fied on a 2% TAE gel and fragmented samples are end-repaired 1http://www.bioinformatics.babraham.ac.uk/projects/fastqc/ using New England Biolab’s NEB Next kit (Ipswich,MA,USA). 2http://www.illumina.com/support/sequencing/sequencing_software/casava.ilmn Repairedsamplesareadenylatedatthe3(cid:48)endusingtheNEBNext 3http://picard.sourceforge.net FrontiersinPharmacology|PharmacologyofAnti-CancerDrugs January2013|Volume4|Article2|4 Barrettetal. Clonalevolutionandsolidtumors exampleweidentifiedfourdistinctpopulationsinabiopsyfroma tumor.Theabilitytoprofilepatientsamplesregardlessoftumor PDAsurgicalresection(Figure1).Eachofthesepopulationswas cellcontentprovidesanunbiasedapproachtothestudyof PDA collected in our sorting assay then processed for whole genome genomes. analysis. These included a genomically normal diploid popula- A striking observation in our ongoing studies of PDA is the tion as well as a small tumor aneuploid population. The ability unique nature of the genomes profiled from each patient. For tocollectfoursimultaneousstreamsintheseassaysoptimizesthe example a 3.7N aneuploid population representing <8.0% of useofeachclinicalsampleandprovidesisogenicpopulationsfor a PDA sample had a focal amplicon on 6p21 as the most sig- analysis. nificant copy number gain in its genome (Figure 2). This high levelfocalampliconincludestheVEGFAgeneandthenucleoside TUMORMARKERSANDTHERAPEUTICTARGETS transportergeneSLC29A(ENT1)thatregulatesintracellulargem- Pancreatic ductal adenocarcinoma is a highly lethal tumor type citabinelevels.Thepresenceoftheseco-amplifiedgenessuggests that is difficult to molecularly characterize at the biopsy level a clinical hypothesis that this tumor has a context of vulnera- due to complex genomes and heterogeneous cellularity, as can- bility to therapy based on anti-angiogenic agents and would be cer cells represent on average only 25% of the cells within the responsive to gemcitabine. In addition this same genome had a tumor(Seymouretal.,1994).PreviousstudiesofPDAhaverelied focalhomozygousdeletiontargetingJARID2,aregulatorof his- ontumorderivedcelllinesandxenograftstoobtainsamplesfor tone methyltransferase complexes during normal development. genomestudies(Jonesetal.,2008).Howeverthesemethodscre- A current hypothesis in our PDA clinical studies is that loss of ateselectivepressuresontumorcells,aretypicallylaborintensive, normal epigenetic regulation may make tumors responsive to and require fresh tissues limiting their utility for patient based combinationtreatmentsthatincludehypomethylationagentssuch studies.Morerecently,studieshaveusedmixturesof tumorand as5-azacytidine. normal cells to estimate thresholds for the detection of somatic Singlecopygainsandlossescanariseathighratesasaresultof mutations to compensate for the complex composition and cel- thegenomicinstabilitythatisahallmarkofmanytumors.Incon- lular admixtures of clinical PDA samples (Biankin et al.,2012). trast, events such as homozygous deletions and focal high level Incontrastweapplyourflowsortingmethodstoclinicalsamples amplifications typically require multiple independent genomic andprofilethegenomesofeachsortedpopulation.Thesehighly eventsandthuscanrepresentbiologicallyselectedeventsincan- purifiedclonalpreparationsareusedforhighdefinitiongenomic cer genomes that target known and putative tumor suppressor analysestodeterminethelandscapeofaberrationspresentineach genesandactivatedoncogenes.Theidentificationofgenestargeted FIGURE1|Clonalprofiling:flowsortingneoplasticcellsfromsolid thensubsequentlysortpopulationsofinterest.Tabledisplaysquantitative tissuebiopsies.(A)BiopsiesaremincedinthepresenceofDAPIthen analysisofthefourpeaks(P1–P4)detectedinapancreaticadenocarcinoma mechanicallydisaggregatedtocreatesinglenucleisuspensions.Theseare sample.(C)Eachpeakwassortedandcollectedintoindividualtubes.(D) thenflowsortedinsingleormultiparameterassays.(B)Scatterplot, GenomicDNAfromeachpurifiedpopulationisextractedthenprocessedfor displayingeventsduringsorting,areusedtogeneratehistogramstoidentify assaysofinterestincludingwholegenomeandwholeexomeanalyses. www.frontiersin.org January2013|Volume4|Article2|5 Barrettetal. Clonalevolutionandsolidtumors FIGURE2|Clonalanalysesofpancreaticductaladenocarcinoma focalamplicons.(B)Chromosome6and(C)locus-specificviewsofthe (PDA)biopsyP1026Z.(A)DAPI-basedDNAcontentanalysisdetected JARID2AhomozygousdeletionandtheVEGFASLC29A1ampliconin a3.7Npopulationrepresenting7.6%ofthecellularcontentofthe the3.7NPDAgenome.BlueshadeareasdenoteADM2aberrant biopsy.Onlythe3.7Npopulationshowedhomozygousdeletionsand intervals. by these selected events provides insights into signaling path- The detection of the complete loss of genomic sequence is ways that may drive the clinical behavior of individual tumors. highlysensitivetoaslittleas5%admixturesofnon-tumorcellsin Forexamplepreviousreportshaveidentifiedsomaticmutations clinicalsamples(Zhaoetal.,2004).Thisisconsistentwithrecur- of NUMB in breast carcinoma and of PARK2 in GBM, colon, ringobservationsthatfeweraberrationscanbedetectedinpatient andlungcancers(Colalucaetal.,2008;Veeriahetal.,2010).We samplesinvivothaninpassagedmodelsystems(Jonesetal.,2008; haveidentifiedhomozygousdeletionsinbothofthesegenespro- Leary et al.,2008;Parsons et al.,2008). However a limitation in vidingevidencethattheirtumorsuppressorfunctionextendsto many cancer genome studies is the inability to objectively dis- PDA.Thepotentialclinicalsignificanceofhomozygousdeletions criminate single copy losses from homozygous deletions and to is highlighted by the prediction that complete loss of NUMB accuratelymapanddeterminehighlevelamplificationincancer results in increased Notch signaling which can be targeted by genomesarisinginindividualpatientsamples.Thuscopynumber gamma secretase inhibition or by blocking the Notch receptor data is typically reduced to reporting frequencies of any loss or directly(Figure3).Thesesamegenomicanalyseshavealsoiden- anygainforeachchromosomeinrelativelylargecohortsofsam- tified homozygous deletions in a series of other known (e.g., ples.Thisfurtherlimitsthetranslationalpotentialofthegenomic SMAD2,SMAD3,JARID2)andputative(SMYD3,USP25)tumor studyofcancersincludingthosewithvariablecomplexgenomes, suppressorgenesthatareselectivelydisruptedinPDAgenomes. rare cancers, and biopsies with high admixtures of genomically Given that many tumor suppressor genes targeted by homozy- normal cells and insufficient tumor cell content (Aguirre et al., gous deletion have frequent sequence mutations the latter pro- 2004).Furthermore,histology-basedmethodscannotreadilydis- videhighlyfavorablecandidatesforresequencingandIHCbased tinguish whether aberrations in a tumor are present in a sin- assays. gle cancer genome or if they are distributed in multiple clonal FrontiersinPharmacology|PharmacologyofAnti-CancerDrugs January2013|Volume4|Article2|6 Barrettetal. Clonalevolutionandsolidtumors FIGURE3|ClinicalsignificanceofNUMBhomozygousdeletionin fromaPDAsample.Shadedareasrepresentcopynumberaberrantregions pancreaticductaladenocarcinoma(PDA)genome.(A)Homozygous calledbyADM2algorithm.(B)ConceptmapofNumbanditsinteractions deletionoftheNumbtumorsuppressorgeneinananeuploidcloneisolated withNotchandTP53pathways. populations.Consequentlycurrentapproachesfortheanalysesof wedetectedaseriesof somaticgenomicaberrationsinthe3.1N cancergenomesarelimitedintheirabilitytodeterminetheclinical population.Theseincludedafocalgainat18p11.32thatincluded contextofeachpatient’stumor. thethymidylatesynthase(TYMS)locuswhoseproteinproductis Inordertofurthervalidateourclonalmethodswehaveapplied targetedby5-FUbasedtherapies.IHCrevealedanincreasedlevel thosetoourongoingStandUptoCancer(SU2C)PDAclinicaltri- ofTYMS.Thissuggeststhatpriortreatmentresultedinamplifica- als.Thefirstofthesetrials,2026-001,involvesmolecularprofiling tionofthisregionandincreasedproteinlevelsandgaverisetothe ofpatientswithpreviouslytreatedmetastaticdisease.Foreachof therapeuticresistancethataroseinthispatientduringtreatment. the35patientsinthistrialonetothree18gneedlebiopsiesfrom a liver metastasis was available for analysis. These samples were CLONALHETEROGENEITYINSINGLEBIOPSIES subsequently analyzed for a panel of IHC markers under CLIA Our ongoing clinical studies have provided a unique view of conditionstoidentifyaclinicallyactionabletarget.Asacorrelative clonal heterogeneity and show that clonal evolution can be dri- samplesofthesesametissueswerealsoprocessedforclonalanaly- venbydifferentbiologicaleventsinpatientsamples.Forexample sisusingsingleparameterDNAcontentsortingandaCGHforeach thepresenceof distinctaneuploidpopulationswithoverlapping sortedpopulation.Anexampleofthesestudiesisapatientwhohad copy number aberration profiles suggests a mitotic event con- previouslybeentreatedwith5-FUbasedtherapy(Figure4).Three tributed to the heterogeneity of the tumor. In some of these distinctpopulationswereidentifiedandsubsequentlysortedfrom cases simultaneous ploidies acquire relatively subtle copy num- thelivermetastasisbiopsy.Thegenomeofeachsortedpopulation berdifferencesintheirgenomes.SingleparameterDNAcontent wasthenprofiledbyaCGH.Thediploidandtetraploidfractions sortingidentifiedtwodistinctploidies(3.2Nand2.7N)inaliver were genomically normal by copy number analysis. In contrast metastasisthatsharedaseriesof genomicaberrationsincluding www.frontiersin.org January2013|Volume4|Article2|7 Barrettetal. Clonalevolutionandsolidtumors FIGURE4|Clonalanalysisofpancreaticductaladenocarcinoma whilethe3.1Npopulationhadmultiplecopynumberaberrations (PDA)livermetastasisfromSU2Cclinicaltrial.(A)DNAcontent throughoutitsgenome.(C)Focalampliconon18p11.32thatincluded flowsortingidentifiedandpurifiedthreedistinctpeaksinthesingle thethymidylatesynthase(TYMS)locus.Blueshadeareasdenote biopsy.(B)The2.0Nand4.0Npopulationsweregenomicallynormal ADM2aberrantintervals. focal amplification of FZD3 (8p21) and homozygous deletions advancing personalized therapies for patients with solid tumors of MAP2K4 (17p12) and CDKN2A (9p21.3;Figure5). In addi- will be to objectively measure genomic lesions in one or more tioneachoftheclonalpopulationshadadeletionwithmatching genomespresentinclinicalsamplesofinterest. boundarieson21q.Inthe3.2Ngenomethedeletionwashomozy- gousasmeasuredbyalog ratiovalueof<−3.0forCGHprobes CLONALHETEROGENEITYINMETACHRONOUSSAMPLES 2 in the interval. In contrast the same interval was only partially Astrikingexampleofclonalheterogeneityisshowninanovarian deleted in the 2.7N population. Given that the MAP2K4 and carcinoma with biopsies from the primary tissue and a pleural CDKN2A deletionswerehomozygousinbothpopulationsthese effusionsampleacquired1yearlater.DNAcontentsortingiden- datasuggeststhatthedistinctaneuploidpopulationsevolvedafter tified two aneuploid populations in each of the clinical samples metastasizingtotheliverandthatthe3.2Ncloneacquiredcom- (Figure6).Strikinglytheploidyof eachof thefourpopulations pletelossofthisregionasaresultofongoinggenomicinstability. was distinct. Two hypertetraploid populations were identified Thepresenceofthesetwodistinctbuthighlyrelatedclonalpop- in the primary tumor while the pleural effusion sample had a ulationsinasinglelivermetastasishighlightstheongoingnature hypodiploid and a hyperdiploid population. Each of these four ofevolutioninPDA. aneuploidpopulationswassortedandprofiledforcopynumber Wehavemadesimilarobservationsofongoinginstabilityand aberrations.Thegenomesofthesefourdistinctpopulationshad evolutioninawidevarietyofsolidtumorsincludingmelanoma, sharedaswellasdistinctaberrationsthatfurtherdefinetheclonal adrenalcorticalcarcinoma,breastcancer,hepatocellularcarcino- heterogeneityof thistumor.Forexamplethesharedaberrations mas,andcoloncarcinoma.Thesignificanceofclonallydiverging includedhighlevelfocalampliconsonchromosomes3pand10q. eventsandtheirpossibleroleinresistancetotherapyanddisease Wealsodetectedaberrationsthatwerepresentinoneofthepri- progression is being followed in ongoing clinical trials. A fun- marypopulationsandinbothofthepopulationsdetectedinthe damental hypothesis is that distinct clones will have differential pleuraleffusion.ThisincludedanamplicononchromosomeXp11 responsestoselectivetherapies.Achallengeforclinicaltrialsand thatincludestheBMP15locus(Figure7A).Theproteinencoded FrontiersinPharmacology|PharmacologyofAnti-CancerDrugs January2013|Volume4|Article2|8 Barrettetal. Clonalevolutionandsolidtumors FIGURE5|Clonalanalysisofpancreaticductaladenocarcinoma genomecopynumberprofilesofthe2.7Nand3.2Npopulations.(C) (PDA)livermetastasisfromSU2Cclinicaltrial.(A)DNAcontent Aberrationdetectiononchromosome21and(D)21q12-q21.2in flowsortingidentifiedandpurifiedadiploidandtwodistinct bothaneuploidpopulations.BlueshadedareasdenoteADM2 aneuploid(2.7N,3.2N)peaksinthesinglebiopsy.(B)Whole aberrantregions. bythisgeneisamemberofthebonemorphogeneticproteinfamily amplicon in a non-linear lineage. This relatively simple exam- which is part of the transforming growth factor-beta superfam- pleusingsingleparametersorting,aCGH,andtwochronological ily. This protein is believed to be involved in oocyte maturation samplesfromasinglepatienthighlightthepotentialcomplexity andfolliculardevelopmentasahomodimerandbyforminghet- ofclonalevolutionduringtumorprogression. erodimerswitharelatedprotein;Gdf9.Inadditionwedetecteda homozygousdeletionat21qthatsimultaneouslydeletedthemet- CLONALHETEROGENEITY,EVOLUTION,ANDACQUIREDRESISTANCE allopeptidasesADAMTS1andADAMTS5thatwaspresentonlyin TOTARGETEDTHERAPY the pleural effusion populations (Figure 7B). The presence and Thedetectionandsortingofmorethanonetumorpopulationin patterns of these amplicons and homozygous deletions in these abiopsyandtheavailabilityofmultiplebiopsiesfromindividual samplessuggestthatBMP15 amplificationcontributedtoclonal patientsextendsthestudyofclinicalphenotypesandthebehaviors selectionduringprogression,andthathomozygousdeletionofthe of clonal populations in vivo. For example the aneuploid pop- metallopeptidasescontributedtotheestablishmentofmetastatic ulations with adverse histological features and multiple selected lesionsinthispatient’stumor. genomicaberrationsincludinga5.7Npopulationwithhighlevel Howevertherewerealsoaberrationsthatwerepresentinthe ARamplificationthataroseduringtheclinicalhistoryofapatient primary populations but absent in the pleural effusion samples. whodevelopedadvancedPC,wereuniquelysensitivetotherapeu- Specifically another focal amplicon on chromosome 6p target- ticregimenandwereerasedafterhormonewithdrawal(Figure8). ingtheprogestinandadipoQreceptorfamilymemberVIIIgene The patterns of acquired clonal aberrations,including losses on locus (PAQR8; Figure 7C). The latter encodes a steroid mem- chromosomes 1p, 1q, 12p, and 18q that persisted throughout branereceptorthatbindsprogesteroneandhasbeenimplicated progression, suggest that the aneuploid populations arose from in normal oocyte maturation. The absence of this amplicon in diploidprogenitorsduringtheevolutionofdisease.Strikingly,the the pleural effusion samples suggests that the evolution of this co-occurring diploid cells acquired a low-level focal AR ampli- tumor may be driven by progenitor populations lacking the 6p conafterbilateralorchiectomyleadingtoincreasedsensitivityto www.frontiersin.org January2013|Volume4|Article2|9 Barrettetal. Clonalevolutionandsolidtumors FIGURE6|Clonalanalysisofmatchingmetachronousprimary pleuraleffusiontumorsamplesfrom2010.Wholegenomecopy ovariantumorandpleuraleffusion.(A)DNAcontentdetectionof numberplotsforeachsortedpopulation.Bluearrowsshared twodistinctaneuploidpopulationsinprimarytumorsamplesfrom aberrations;redarrowspleuraleffusionspecificaberrations;green 2009.Wholegenomecopynumberplotsforeachsortedpopulation. arrowsprimaryspecificaberrations;blackarrowssingleprimaryand (B)DNAcontentdetectionoftwodistinctaneuploidpopulationsin sharedpleuraleffusionaberrations. FIGURE7|Genomicaberrationsandclonalanalysisofmatching homozygousdeletiontargetingADAMTS1-ADAMTS5.(C)Chromosome metachronousprimaryovariantumorandpleuraleffusion.(A) 6p12.1-p12.3ampliconandPAQR8.BlueshadedareasdenoteADM2defined ChromosomeX11pfocalampliconandBMP15.(B)Chromosome21q21.3 aberrantintervals. FrontiersinPharmacology|PharmacologyofAnti-CancerDrugs January2013|Volume4|Article2|10