Circulating Tumor Cell Clusters Are Oligoclonal Precursors

13 Pages·2014·4.25 MB·English

by  Nicola Aceto

Circulating Tumor Cell Clusters
Are Oligoclonal Precursors
of Breast Cancer Metastasis
NicolaAceto,1,2AdityaBardia,1,2DavidT.Miyamoto,1,5MariaC.Donaldson,1,2BenS.Wittner,1,2JoelA.Spencer,3,4
MinYu,1,2AdamPely,3,4AmandaEngstrom,1,2HuiliZhu,1,2BrianW.Brannigan,1,2RaviKapur,6ShannonL.Stott,1,2,6
ToshiShioda,1,2SridharRamaswamy,1,2DavidT.Ting,1,2CharlesP.Lin,3,4MehmetToner,6,7DanielA.Haber,1,2,8,*
andShyamalaMaheswaran1,7,*
1MassachusettsGeneralHospitalCancerCenter,HarvardMedicalSchool,Boston,MA02129,USA
2DepartmentofMedicine,HarvardMedicalSchool,Boston,MA02129,USA
3AdvancedMicroscopyProgram,WellmanCenterforPhotomedicineandCenterforSystemsBiology,MassachusettsGeneralHospital,
Boston,MA02114,USA
4HarvardStemCellInstitute,1350MassachusettsAvenue,Cambridge,MA02138,USA
5DepartmentofRadiationOncology,MassachusettsGeneralHospital,HarvardMedicalSchool,Boston,MA02129,USA
6CenterforBioengineeringinMedicine,HarvardMedicalSchool,Boston,MA02129,USA
7DepartmentofSurgery,HarvardMedicalSchool,Boston,MA02129,USA
8HowardHughesMedicalInstitute,ChevyChase,MD20815,USA
*Correspondence:[email protected](D.A.H.),[email protected](S.M.)
http://dx.doi.org/10.1016/j.cell.2014.07.013
SUMMARY cer metastasis is derived primarily from mouse models and it
isthoughtto involveaseriesofsequentialsteps:epithelial-to-
Circulating tumor cell clusters (CTC clusters) are mesenchymaltransition(EMT)ofindividualcellswithinthepri-
present in the blood of patients with cancer but marytumorleadingtotheirintravasationintothebloodstream,
their contribution to metastasis is not well defined. survivalofsuchcirculatingtumorcells(CTCs)withintheblood-
Usingmousemodelswithtaggedmammarytumors, stream, and finally their extravasation at distant sites, where
we demonstrate that CTC clusters arise from oligo- mesenchymal-to-epithelial transition (MET) culminates in their
proliferation as epithelial metastatic deposits (Hanahan and
clonal tumor cell groupings and not from intravas-
Weinberg,2011).WhileEMThasindeedbeendemonstratedin
cular aggregation events. Although rare in the
human breast cancer cells in the circulation (Yu et al., 2013),
circulationcomparedwithsingleCTCs,CTCclusters
therequirementforEMTtoinitiatemetastasishasbeendebated
have 23- to 50-fold increased metastatic potential.
(Ledford,2011;Tarinetal.,2005).Alternativemodelsproposed
In patients with breast cancer, single-cell resolution
includetumor-derivedmicroembolithatmaybreakofffrompri-
RNA sequencing of CTC clusters and single CTCs, mary tumors, lodging into distal capillaries where they initiate
matched within individual blood samples, identifies metastatic growth (Fidler, 1973; Liotta et al., 1976; Molnar
the cell junction component plakoglobin as highly et al., 2001). Using diverse technological platforms, we and
differentially expressed. In mouse models, knock- others have indeed detected clusters of CTCs, ranging from
down of plakoglobin abrogates CTC cluster forma- 2–50cancercells,withinthecirculationofpatientswithmetasta-
tionandsuppresseslungmetastases.Inbreastcan- ticepithelialcancers(Choetal.,2012;Fidler,1973;Molnaretal.,
2001;Stottetal.,2010;Yuetal.,2013).
cer patients, both abundance of CTC clusters and
Studiesofcancermetastasishaveemphasizedtheconceptof
high tumor plakoglobin levels denote adverse out-
‘‘seedversussoil’’asakeydeterminantofmetastaticpropensity
comes.Thus,CTCclustersarederivedfrommulticel-
(Fidler, 2003). This model matches the importance of mutated
lular groupings of primary tumor cells held together
genetic drivers within tumor cells conferring proliferative and
through plakoglobin-dependent intercellular adhe-
invasive properties, with that of the microenvironment of the
sion, and though rare, they greatly contribute to the distantorganor‘‘niche,’’whichmayfacilitatemetastaticgrowth.
metastaticspreadofcancer. However,thephysicalcharacteristicsofsingleCTCsandCTC
clustersmayalsocontributetometastaticpropensity,especially
astheyimpacttheabilityofepithelialtumorcellstosurvivethe
INTRODUCTION loss of cell adherence and shear forces in the blood stream,
i.e., different survival signals among the cancer cell ‘‘seeds’’
Themetastaticspreadofbreastcancer,typicallytobone,lung, maybeimportant.Forinstance,inamouseendogenouspancre-
liver,andbrain,accountsforthevastmajorityofcancer-related aticcancermodel,noncanonicalWntsignalingiselevatedwithin
deaths(Nguyenetal.,2009).Ourunderstandingofepithelialcan- CTCs, where it appears to suppress anoikis (Yu et al., 2012),
1110 Cell158,1110–1122,August28,2014ª2014ElsevierInc.
whileinasubcutaneoustumorxenograftmodel,theadmixtureof depletionoferythrocytesandleukocytesfrombloodspecimens,
tumorandstromalcellswithinmicroembolimaycontributestro- yielding untagged CTCs and small CTC clusters in solution,
mal-derivedsurvivalsignals(Dudaetal.,2010).Inarecentstudy where they can be micromanipulated for single-cell RNA
of human breast cancer, mesenchymal markers indicative of sequencing(Ozkumuretal.,2013).Here,weusebothofthese
EMT were expressed within the cancer cells comprising CTC devices,alongwithinvivoflowcytometryandnextgeneration
clusters(Yuetal.,2013).Takentogether,bothhumanandmouse RNA sequencing, to interrogate CTCs from both patients with
modelingstudiespointtothecomplexityofbloodbornemetas- metastatic breast cancer and mouse tumor models. We find,
tasisandtheneedtocaptureandcharacterizeCTCstobetter usingmousemodels,thatCTCclustersarederivedfromoligo-
understandthisprocess. clonal clumps of primary tumor cells and constitute a rare but
CTCshavebeendetectedinthemajorityofepithelialcancers, very highly metastasis-competent subset of CTCs, compared
where they represent cancer cells captured as they transit with single circulating breast cancer cells. RNA sequencing of
throughthebloodstream(Alix-Panabie`resandPantel,2013;Yu human breast CTC clusters identifies plakoglobin as a key
etal.,2011).Assuch,theyholdthekeytounderstandingcritical mediatoroftumorcellclustering,whichisexpressedinahetero-
pathwaysthatmediatethebloodbornedisseminationofcancer, geneouspatternwithintheprimarytumor.Knockdownofplako-
whichmaynotbereadilyevidentthroughanalysesofbulkpri- globinexpressioninthemousemodelsuppressesCTCcluster
mary or metastatic tumor populations. Factors leading to the formationandreducesmetastaticspread.
generationofCTCsfromaprimarytumorareunknown,including
thefractionderivedfromcancercellsthathaveactivelyintrava- RESULTS
satedintothebloodstream,versusthosethatarepassivelyshed
asaresultofcompromisedtumorvasculature.Althoughexceed- EndogenousCTCClustersHaveIncreasedMetastatic
inglyrarecomparedwithnormalbloodcells,thenumberofCTCs PotentialComparedtoSingleCTCs
inthebloodstreamfarexceedsthenumberofmetastaticlesions To define the origin and functional properties of CTC clusters,
in patients, indicating that the vast majority CTCs die in the compared with single CTCs, we made use of mouse models,
bloodstream, with only a minor fraction representing viable wheretumorcellcomposition,transitofCTCsthroughtheblood-
metastatic precursors. Epithelial cells that have lost adhesion- stream,andmetastaticdepositscanbemonitoredandquanti-
dependentsurvivalsignalsrapidlyundergoanoikis,afatelikely fied. We first established a model to test the generation of
tomeetmostCTCsinthebloodstream.Itisinthiscontextthat endogenous CTCs and metastases from a primary orthotopic
eithermesenchymaltransformation,stromal-derivedfactors,or tumor xenograft. These experiments were designed both to
persistentinterepithelialcelljunctionsmayprovidesurvivalsig- test the metastatic propensity of CTC clusters versus single
nals that attenuate this apoptotic outcome (Duda et al., 2010; CTCs, as well as to determine whether CTC clusters originate
Manietal.,2008;Robsonetal.,2006;Yuetal.,2012).Dissecting fromanoligoclonalgroupingofprimarytumorcellsorfromthe
thecontributionsofthesevariousmechanismstohumancancer clonal progeny of an individual tumor cell. MDA-MB-231-LM2
requires the ability to isolate individual CTCs from the blood- (LM2) cells, a lung-metastatic variant of MDA-MB-231 human
streamandsubjectthesetodetailedmolecularanalyses. breastcancercells(Minnetal.,2005),wereengineeredtoex-
MultipletechnologieshavebeendevelopedforCTCcapture, press either green fluorescent protein (LM2-GFP) or mCherry
taking advantage of tumor-specific epitopes absent in normal (LM2-mCherry),anda1:1mixtureofthesedifferentiallytagged
bloodcells,variationsintheirphysicalpropertiessuchassize, cellswasinjectedintothemammaryfatpadofimmunodeficient
density, and electromechanical characteristics, or by applying (NSG) mice. As expected, overt primary breast tumors were
high throughput imaging to unpurified blood cell preparations observedafter5weeksandtheseretainedanequaldistribution
(forreview,seeYuetal.,2011).ThefactthatCTCsareextremely of LM2-GFP and LM2-mCherry tagged cells, as confirmed by
rare, even in patients with advanced metastatic cancers (esti- IHCstaining(FiguresS1AandS1Bavailableonline).Wesampled
mated at one CTC/billion normal blood cells), and that they the blood of tumor-bearing animals for presence of single or
maybepoisedonthevergeofapoptosis,hasmadetheiranal- clusteredCTCsusingaterminalbleedandsimultaneouslyhar-
ysiscontingentupontechnologicalconstraints.Wehaveintro- vestedthelungsforanalysisofmetastaticdeposits.Inaddition
ducedaseriesofmicrofluidicdevicesthathavetheadvantage to enumeration of CTCs, we reasoned that clonally-derived
oflow-shear,yethighthroughput,interrogationofunprocessed CTC clusters would uniformly express either GFP or mCherry,
whole blood, providing highly enriched and unfixed CTCs that whereas aggregations of cells from the primary tumor would
are suitable for detailed molecular analysis (Nagrath et al., beheterogeneousforthetwomarkers(Figure1A).Weobserved
2007; Ozkumur et al., 2013; Stott et al., 2010). Among these, ameanof2,486CTCeventspermouse(n=5mice),ofwhicha
theherringbone(HBCTC-Chip)makesuseofgrooveswithinthe meanof65(2.6%)wereCTCclustersand2,421(97.4%)were
ceilingofthemicrofluidicchambertogenerateturbulentmicro- singleCTCs(Figures1B,1C,andS2A).Virtuallyall(91%)CTC
fluidicflow,directingcellsagainstantibody-coatedwallsofthe clusters were dual positives for GFP and mCherry. A mean of
device,whereCTCsarecaptured(Stottetal.,2010).Thisdevice, 5.6 (9%) CTC clusters per mouse (with fewer than three cells
whose highly efficient design enabled our initial detection of percluster)werecomprisedofcellsexpressingonlyoneofthe
large CTC clusters, requires on-chip cell lysis for nucleic acid twomarkers,consistentwithexpectedprobabilitiesgivena1:1
extractionandhenceprovidesanenrichedbutheterogeneous mixture ofGFP/mCherry expressing cellsin theprimary tumor
CTCpopulationforanalysis(Yuetal.,2012,2013).Incontrast, (Figures 1C and S2B). Thus, CTC clusters do not result from
our recently described negCTC-iChip achieves highly efficient theproliferationofasingletumorcellinthevasculature,instead
Cell158,1110–1122,August28,2014ª2014ElsevierInc. 1111
A B C
D
E F
G
Figure1. CTCClustersDemonstrateIncreasedMetastaticPotentialComparedtoSingleCTCs
(A)Schematicoftheexperiment.MDA-MB-231-LM2(LM2)cellsexpressingGFP(LM2-GFP)ormCherry(LM2-mCherry)cellsweremixedat1:1ratioandinjected
intherightmammaryglandofimmunodeficientmicetogenerateone-colorsingleCTCsandmulticolorCTCclusters.Accordingly,one-colormetastaticfociare
derivedfromasingleCTC,whilemulticolorfociarisepredominantlyfromaCTCcluster.
(B)RepresentativeimagesofsingleCTCs(GFP-ormCherry-positive)andCTCclusters(GFP-andmCherry-positive)capturedontheHBCTC-Chip(left).Lung
metastaticfociderivedfromasingleCTC(GFP-ormCherry-positive)oraCTCcluster(GFP-andmCherry-positive)areshown(right).GFP(brown),mCherry(red).
Bloodsamplesandlungspecimenswereisolated5weeksafterprimarytumordevelopment.n=5.
(C)Bargraphsshowingthemeanpercentageofone-colorversusmulticolorCTCeventscapturedbytheHBCTC-Chip(left),themeanpercentageofone-color
versusmulticolorCTCclusters(middle),aswellasthemeanpercentageofone-colorversusmulticolorlungfoci(right).n=5.
(D)BargraphshowingthenormalizedmetastaticpotentialofsingleCTCsandCTCclusters.ErrorbarsrepresentSEM.n=5,*p=0.031byStudent’sttest.
(E)Schematicoftheexperiment.LM2-GFPcellswereinjectedintherightmammaryglandwhileLM2-mCherrycellswereinjectedintheleftmammaryglandof
immunodeficientmicetogeneratetumorsthatgiverisetoone-colorsingleCTCsandCTCclusters,aswellasraremulticolorCTCclusters(resultingfrom
aggregationevents).Accordingly,one-colormetastaticfociarederivedfromasingleCTCoraCTCcluster,whilemulticolorfociderivefromCTCaggregates.
(F)RepresentativeimagesofsingleCTCs(GFP-ormCherry-positive)andCTCclusters(GFP-ormCherry-positive)capturedontheHBCTC-Chip(left).Lung
metastaticfociderivedfromasingleCTC(GFP-ormCherry-positive)oraCTCcluster(GFP-ormCherry-positive)areshown(right).GFP(brown),mCherry(red).
Bloodsamplesandlungspecimenswereisolated5weeksafterprimarytumordevelopment.n=5.
(G)Bargraphsshowingthemeanpercentageofone-colorversusraremulticolorCTCeventscapturedbytheHBCTC-Chip(left),themeanpercentageof
one-colorversusmulticolorCTCclusters(middle),aswellasthemeanpercentageofone-colorversusmulticolorlungfoci(right).n=5.
SeealsoFiguresS1andS2.
they appear to represent the aggregation of neighboring cells, in CTC clusters, we reasoned that metastatic tumors derived
mostlikelywithintheprimarytumormass(seebelow). fromasingleCTCwouldbepositiveforasinglemarker,while
Metastaticdepositsinthelungswereanalyzedforbothnum- thosederivedfromCTCclusterswouldstainforbothGFPand
ber and composition using anti-GFP and anti-mCherry anti- mCherry(Figures1Aand1B).Ameanof323lungfociwereiden-
bodies,simultaneouslywiththeCTCanalyses(Figures1B,1C, tifiedpermouse(n=5mice),ofwhich171(53%)weremulticolor,
and S2A). Given the distribution of GFP and mCherry staining and therefore derived from CTC clusters, versus 152 (47%)
1112 Cell158,1110–1122,August28,2014ª2014ElsevierInc.
unicolor derivatives of single CTCs (Figures 1C and S2A). lungmetastasesrelativetotheprevalenceofsingleCTCsand
Normalizing the number and distribution of lung metastases CTC clusters, we calculated a 23-fold increase in metastatic
withthatofsingleCTCsandCTCclusters,wecalculatethata competenceforCTCclustersversussingleCTCs(FigureS2D),
CTCclusteris(cid:1)50timesmorelikelytogiverisetoametastatic anestimatethatiscomparabletothe50-foldincreasederived
depositthanasingleCTC(Figure1D).Thus,whileCTCclusters from LM2 cell experiments. Taken together, these two mouse
aremuchmorerarethansingleCTCsinthisorthotopicmouse tumor models indicate that CTC clusters constitute only 2%–
modelofbreastcancer,theycontributeequallytothemetastatic 5% of all CTC events detected in the circulation, but their
burdeninthelung. dramatically elevated metastatic potential (23–50 times that of
Tofurthervalidate(1)thatoligoclonalCTCclustersarisefrom singleCTCs)contributestoapproximatelyhalfofallmetastatic
thefragmentingofprimarytumorcellclumpsintothevasculature lesionsinorthotopicbreastcancermodels.
and not from intravascular aggregation of single CTCs, and
(2) that oligoclonal lung metastases arise from CTC clusters ClusteredCancerCellsAreMoreResistantthanSingle
and not from the reseeding of a metastatic site by multiple CellstoApoptosisfollowingDisseminationtotheLung
singleCTCs,weundertookasecondseriesoforthotopicmouse Wegeneratedaninvitroassaythatallowedustoobtainasus-
xenograft experiments, injecting LM2-GFP cells into the right pensionofeithersinglecellsorclusteredcells(2–30cells)from
mammaryfatpadandLM2-mCherrycellsintheleftfatpadof cultures of GFP-Luciferase-tagged LM2 cells (see Extended
immunodeficient mice (Figure 1E). Five weeks after injection, ExperimentalProcedures).Weinjected200,000LM2cellspre-
mice harbored two independent and differentially tagged tu- pared either as single cells (LM2-SC) or as clusters (LM2-CL)
mors, and we again simultaneously harvested the blood for into the tail vein of immunodeficient mice and subjected them
analysisofCTCsandthelungsforenumerationofmetastaticde- to serial luciferase-based imaging (Figures 2A and 2B). Both
posits.Asexpected,singleCTCsinthecirculationdemonstrated LM2-SC and LM2-CL cells reached the lungs with equal effi-
equalcontributionsfromtheGFPandm-Cherry-taggedprimary ciency (day 0), as shown by both bioluminescence and GFP
tumors.However,unlikethepreviousmultitaggedsingletumor immunohistochemical(IHC)staining(Figure2B).However,over
model,inmicewithtwoindependentindividually-taggedtumors, thefollowingdays,theLM2-SClungsignalprogressivelydimin-
thevastmajorityofCTCclusters(96%)wereofasinglecolor, ishedasthecellsunderwentmassiveapoptosis,demonstrated
withequalcontributionsfromGFP-ormCherry-positiveprimary by staining for cleaved caspase 3 (Figure 2C). In contrast, the
tumors(Figures1F,1G,andS2A).Thus,thevastmajorityofCTC LM2-CLlungsignalpersistedfollowingintravascularinoculation,
clusters are derived from individual primary tumors, excluding withcellsshowingresistancetoapoptosisandtumorsexpanding
intravascularaggregationofsingleCTCsasasignificantsource more rapidly (Figures 2B–2D). Lung tumors eventually grew in
ofCTCclusters. micesubjectedtotailveininjectionwitheitherofthetwoLM2
A very small fraction of CTC clusters observed in the dual derivatives, but injection of clustered cells resulted in reduced
tumor-bearingmiceweremulticolor(4%ofCTCclusters,corre- overallsurvival,with12.7weeksforLM2-CLversus15.7weeks
spondingto0.12%oftotalCTCevents)(Figures1GandS2A). forLM2-SC(p<0.016)(Figure2E).Weconfirmedthedifferential
While extraordinarily rare, the presence of such CTC clusters rate of apoptosis and metastatic growth in the lung for single
derived from two independent tumors may originate either versusclusteredcancercellsusingtailveininjectionoftwoaddi-
from the uncommon intravascular aggregation of single CTCs tionalbreastcancercelllines,BT474and4T1(FiguresS3A–S3D).
orfromamixingofcancercellswithinthetwoprimarytumors,
duetothepreviouslyreported‘‘tumorreseeding’’phenomenon CalculationofCTCClustersandSingleCTCCirculatory
(Kimetal.,2009)(FigureS1B).Consistentwiththelatterhypoth- ClearanceRateUsingInVivoFlowCytometry
esis,wefoundthat3%–5%ofcellswithintheGFP-taggedpri- Clustersoftumorcellsmayexhibitconsiderableflexibilityasthey
marytumorwerepositiveformCherryand3%–5%ofcellswithin navigate through narrow channels, and capillary beds them-
themCherry-labeledtumorwerepositiveforGFP(FigureS1B).In selvesmayhaveunevenvesseldiametersorbypasstractsthat
addition to rare multicolor CTC clusters, we observed a small allow transit of large multicellular structures. However overall,
fraction (8%) of multicolor tumors in the lung (Figures 1G and CTCclustersaremorelikelythansingleCTCstobetrappedin
S2A).Thesemetastaticlesionscouldresulteitherfromtherare small capillaries of the lung and distal organs. Thus, the low
multicolor CTC clusters or from the reseeding of metastatic steady-statelevelofCTCclustersinthecirculationmayreflect
lesionsbymultiplesingleCTCs. a considerably higher generation rate if their clearance rate is
WeconfirmedthefindingsderivedfromthetwoLM2mouse veryhigh.TotestifCTCclustersindeedhaveafasterclearance
xenograft experiments with a second, mouse-derived breast ratefromthebloodstreamthansingleCTCs,weusedinvivoflow
cancer cell line, 4T1 (Figures S2C and S2D). Consistent with cytometry(IVFC)tomonitorLM2-SCandLM2-CLcellslabeled
theLM2results,a1:1mixtureof4T1-GFPand4T1-mCherrycells with the lipophilic carbocyanine membrane dye DiD, following
within an orthotopic mammary tumor generated CTC clusters tailveininjectioninimmunodeficientmice(Figure3A).DiDwas
that were overwhelmingly multicolored (90%), whereas two selectedtoachieveoptimaldetectionofCTCswithourIVFCset-
separateprimary4T1tumorslabeledeitherwithGFPormCherry tings. Circulating DiD-labeled cells were detected in real time
produced CTC clusters that were of a single color (87%). withintheearbloodvesselsforatotalof55minineachmouse.
These observations further support that CTC clusters arise as Injected LM2-CL cells were cleared at least three times more
oligoclonalfragmentsderivedfromasingletumor(FigureS2C). rapidlythanLM2-SC(half-life:6–10minforLM2-CLversus25–
Normalizing the number and color distribution of 4T1-derived 30 min for LM2-SC) (Figure 3B). Together, these observations
Cell158,1110–1122,August28,2014ª2014ElsevierInc. 1113
Figure2. CTCClustersAreMoreResistanttoApoptosisatDistalMetastaticSites
(A)SchematicshowingMDA-MB-231-LM2-GFP-Luciferase(LM2)cellspreparedassinglecells(LM2-SC)orasclusters(LM2-CL)priortoinjectionintothetail
veinofimmunodeficientmice.Cells(23105)wereinjectedasLM-SCorLM2-CLpermouse.
(B)Representativebioluminescenceimagesofmiceat0,6,and12daysaftertailveininjectionwithLM2-SCorLM2-CLcells(left).n=4.Representativeimagesof
GFP-stainedsectionsofmouselungsafterinjectionwithLM2-SCorLM2-CLcells(right).ThebargraphshowsthemeanpercentageofGFP-positivecellsinlungs
fromLM2-SC-orLM2-CL-injectedmice.ErrorbarsrepresentSEM.n=4;NS,notsignificant,*p=0.03byStudent’sttest.
(C)Representativeimagesofcleavedcaspase3-stainedsectionsofmouselungs24hrafterinjectionwithLM2-SCorLM2-CLcells.Thebargraphshowsthe
meanpercentageofcleavedcaspase3-positivecellsinlungsfromLM2-SC-orLM2-CL-injectedmice.ErrorbarsrepresentSEM.n=4;*p<0.02byStudent’s
ttest.
(D)LungmetastasisgrowthcurvefrommiceinjectedwithLM2-SCorLM-CL.ErrorbarsrepresentSEM.n=4;p<0.03byStudent’sttest.
(E)Kaplan-MeiersurvivalplotshowingsurvivalratesformiceinjectedwithLM2-SCorLM2-CL.n=4;p<0.016bylogranktest.
SeealsoFigureS3.
defineacirculatingtimeforCTCsinthebloodstream:theshorter HER2-positive(n=13),triplenegative(n=17)subtypesofbreast
circulationhalf-lifeofCTCclustersisconsistentwiththeirmore cancer(total:265datapoints).Fortheseexperiments,wemade
rapidentrapmentwithincapillariesofdistalorgans,wherethey useoftheHBCTC-Chip,whichishighlyefficientincapturingboth
mayinitiatemetastaticgrowth(Liottaetal.,1976). largeandsmallCTCclusters(Stottetal.,2010).Wecoatedthe
microfluidicchamberwithacombinationofantibodies,targeting
ThePresenceofCTCClustersinPatientswithBreast theepithelialcelladhesionmolecule(EpCAM),aswellastheline-
andProstateCancerCorrelateswithPoorPrognosis agemarkersepithelialgrowthfactorreceptor(EGFR)andhuman
HavingcharacterizedtheoriginandmetastaticpotentialofCTC epithelialgrowthfactorreceptor2(HER2/ErbB2),whichtogether
clustersinmousemodels,weundertooktostudytheirproperties efficientlycapturebothepithelialandmesenchymalbreastCTCs
in patients with cancer. To first test the clinical significance of (Yu et al., 2013). After processing 3 ml of whole blood from
CTCclustersinthebloodofpatientswithprogressingmetastatic patients with breast cancer, the CTCs captured on the chip
breastcancer,wemeasuredtheirpresenceinbloodspecimens werestainedwithantibodiesagainstwidespectrumcytokeratin
fromatotalof79patients,drawnatmultipletimepointsovera (CK)toidentifyCTCsandagainsttheleukocytemarkerCD45to
periodof19months.PatientswererecruitedtoanIRB-approved assesswhitebloodcell(WBC)contamination(Figure4A).CTCs
study at the Massachusetts General Hospital Cancer Center, wereidentifiedin54outof79patients(68%).Amongpatients
including women with estrogen receptor-positive (n = 49), with CTCs, 3 (5.6%) had CTC clusters evident across more
1114 Cell158,1110–1122,August28,2014ª2014ElsevierInc.
A B
Figure3. CTCClustersDemonstrateaFasterClearanceRatefromtheBloodstream
(A)SchematicshowingtheexperimentalsetupformeasuringtheclearancetimeofsingleCTCsandCTCclusters.Briefly,DiD-stainedLM2cellswerepreparedas
LM2-SCorLM2-CLandinjectedintothetailveinofimmunodeficientmice.InvivoflowcytometrywasappliedtotheearbloodvesselstodetectsingleCTCsand
CTCclustersovera55minperiodafterinjection.GraphsshowrepresentativefluorescencepeakscorrespondingtothetransitofasingleCTCorCTCcluster
throughtheearbloodvessel.
(B)GraphshowingsingleCTCsandCTCclustersclearancecurves.ErrorbarsrepresentSEM.n=5forsingleCTCsandn=4forCTCclusters,*p<0.01by
two-wayANOVA.
thanthreetimepoints,while16(29.6%)hadCTCclustersduring results point to the potential relevance of CTC clusters in the
onetothreetimepointsand35(64.8%)hadnodetectableclus- progressionofhumancancer.
ters(Figure4B).WecorrelatedthepresenceofCTCclusterswith
progression-freesurvival(PFS)forallpatientswheresuchdata Single-CellResolutionRNASequencingofMatchedCTC
wereavailable(n=30)(TableS1).Ofnote,PFSwascalculated ClustersandSingleCTCsPurifiedfromPatientswith
astimefrominitiationoftherapytodiscontinuationbythetreat- BreastCancer
ingclinician(blindedtotheCTCresults),andPFSdataanalysis TheabilitytocapturebothsingleCTCsandCTCclustersfrom
wasperformedonlywhenclinicalmeasurementsbracketedthe the same blood specimen made it possible to undertake sin-
CTCs isolation time frame. Patients with CTC clusters across gle-cell resolution RNA sequencing, searching for differences
morethanthreetimepointshadameanprogression-free sur- inexpressionprofilesmatchedtoindividualpatients.Forthese
vivaltimeof32.6days,comparedwith134.8daysforpatients experiments,weappliedthenegCTC-iChip,whichenablesisola-
whereCTCclusterswerefoundduringonetothreetimepoints tion and single-cell manipulation of untagged CTCs, together
and160.5daysforpatientswithsingleCTCsonly(p=0.0002) withanoptimizedprotocolfornextgenerationRNAsequencing
(Figure4C;TableS1).Thus,evenamongpatientswithadvanced from minute amounts of template (Ozkumur et al., 2013; Tang
metastaticbreastcancer,thecontinuouspresenceofCTCclus- etal.,2010).Bloodspecimensfromtenpatientswithmetastatic
tersisassociatedwithanadverseclinicaloutcome. breastcancerweresubjectedtomicrofluidicdepletionofRBCs
Giventherelativelyshorttimetoprogressioninpatientswith andCD45-andCD66b-positiveWBCs,leavinguntaggedsingle
advancedbreastcancer,wesoughttotestthecorrelationbe- CTCsandsmallCTCclustersinthefinalproduct(Ozkumuretal.,
tween CTC clusters and adverse prognosis in patients with a 2013).Unfixedtumorcellswerestainedforcellsurfaceexpres-
longer clinical course. We measured the number of CTCs in a sion of EpCAM, HER2, and the mesenchymal marker CDH11
totalof64patientswithprostatecancerusingbloodspecimens (Alexa488-conjugated), and counterstained with antibodies
drawnatmultipletimepointsoveraperiodof53months(total: againstCD45,CD14,andCD16toidentifycontaminatingleuko-
202datapoints).CTCsinprostatecancerpatientswerevisual- cytes (TexasRed-conjugated) (Figure 5A). Individual CTC clus-
izedbystainingwithacocktailofantibodiesagainstprostate- ters (median of three cells per cluster) were isolated using a
specificantigen(PSA)andprostate-specificmembraneantigen micromanipulator and compared with numerically matched
(PSMA); anti-CD45 staining was used to exclude white blood pools of single CTCs from the same specimen, followed by
cells (Miyamoto et al., 2012). CTCs were detected in 48/64 next generation RNA sequencing (SOLiD 5500XL) (Figure 5A).
patients (75%). CTC clusters were present in 6/48 samples Wederivednormalizedexpressionprofilesforatotalof29sam-
(12.5%) (Figures 4D and 4E). In this cohort, the presence of ples(15poolsofsingleCTCsand14CTCclusters)isolatedfrom
CTCclustersduringatleastonetimepointstronglycorrelated tenbreastcancerpatients.
with a dramatically shorter overall survival time (mean survival UnsupervisedhierarchicalclusteringofRNAsequencingdata
time was 115.8 days for patients with CTC clusters versus showed no obvious distinctions at the global gene expression
930.1daysforpatientswithsingleCTCs;p=0.00001)(Figure4F; levelbetweensingleCTCsandCTCclusters,withbothofthese
TableS1).Whilefurtherstudieswillberequiredtoascertainthe clusteringcloselybypatientoforigin(Figure5B).Consistentwith
clinicalutilityofCTCclustersversussingleCTCsasprognostic themicroscopicappearanceofCTCclustersasprimarilytumor
determinants in either breast or prostate cancer, these initial cell-derived, we did not identify RNA signatures of other cell
Cell158,1110–1122,August28,2014ª2014ElsevierInc. 1115
A B C Figure4. ThePresenceofCTCClustersin
PatientswithCancer CorrelateswithPoor
Prognosis
(A) Representative images of a CTC cluster, a
singleCTC,andawhitebloodcell(WBC)isolated
from a breast cancer patient using the HBCTC-
Chipandstainedwithwide-spectrumcytokeratin
(CK,red),CD45(green),andDAPI(nuclei,blue).
(B) A total of 79 breast cancer patients (corre-
spondingto265timepoints)wereanalyzedforthe
presence of CTCs, with 54 of the 79 patients
scoringpositiveforCTCs.Thebargraphshows
the percentage of CTC-positive patients having
CTCclustersduringmorethanthreetimepoints
(red),CTCclustersacrossonetothreetimepoints
(blue)orsingleCTCsonly(black).
(C) Kaplan-Meier progression-free survival plot
showingprogressionratesforbreastcancerpa-
D E F tientshavingCTCclustersduringmorethanthree
timepoints(red),CTCclustersacrossonetothree
timepoints(blue)orsingleCTCsonly(black).The
meanprogression-freesurvivaltimeforeachgroup
isgiveninparentheses.p=0.0002bylogranktest.
(D)RepresentativeimagesofaCTCcluster,asin-
gleCTCandawhitebloodcell(WBC)isolatedfrom
aprostatecancerpatientusingtheHBCTC-Chip
andstainedwithprostate-specificantigen(PSA,
red),prostate-specificmembraneantigen(PSMA,
yellow),CD45(green),andDAPI(nuclei,blue).
(E)Atotalof64prostatecancerpatients(corre-
spondingto202timepoints)wereanalyzedforthe
presence of CTCs, with 48 of the 64 patients
scoringpositiveforCTCs.Thebargraphshows
the percentage of CTC-positive patients having
CTCclustersduringatleastonetimepoint(red)or
singleCTCsonly(black).
(F)Kaplan-MeieroverallsurvivalplotshowingprogressionratesforprostatecancerpatientshavingCTCclustersduringatleastonetimepoint(red)orsingle
CTCsonly(black).Themeanoverallsurvivaltimeforeachgroupisgiveninparentheses.p=0.0001bylogranktest.
SeealsoTableS1.
types,includingTcells,Bcells,dendriticcells,naturalkillercells, adherence junctions (Aktary and Pasdar, 2012), which has
hematopoietic stem cells, macrophages/monocytes, granulo- beenreportedtohavebothpositiveandnegativerolesindiverse
cytes, endothelial cells, or fibroblasts (Figure S4). Markers for malignancies(Hakimelahietal.,2000;Kolligsetal.,2000;Shiina
platelets were present in both single CTCs and CTC clusters, etal.,2005).AlongwithupregulationofplakoglobinRNA,multi-
consistentwiththeirknownadherencetocancercellsinthecir- plecomponentsofbothdesmosomesandadherencejunctions
culation.Foreachpatient,wecomparedgeneexpressiondata weresignificantlyenrichedinCTCclusters(FiguresS6A–S6E).
of CTC clusters versus single-CTCs, generating a list of 31 ConsistentwiththeRNAsequencingresults,weconfirmedpla-
CTC-cluster-associatedgenessharedacrossdifferentpatients koglobinproteinexpressioninmultipleCTCclusters,butnotin
(q<0.01,log2FC>1,inmorethan70%ofallintrapatientcom- matchedsingleCTCsfromabreastcancerpatient(Figure5E).
parisons) (Figures 5C and 5D; Table S2). To identify potential WhileCTCclustersexpressepithelialcelljunctioncomponents,
drivers of metastasis among CTC-cluster-enriched genes, we includingplakoglobinandE-cadherin,wehavepreviouslyshown
tested for correlation between their overexpression in primary that some mesenchymal markers may also be upregulated in
tumor specimens and clinical outcomes in a cohort of 1,956 suchclusters,aneffectthatmaybeassociatedwithadherence
patients with ER-positive, HER2-positive, and triple-negative inthebloodstreamwithTGFb-richplatelets(Labelleetal.,2011;
breastcancers.AmongthecandidateCTCclustergenes,plako- Yu et al., 2013). Matched primary and metastatic tumors bi-
globinwasuniqueinitshighlevelofoverexpressioninCTCclus- opsieswereavailablefromthispatient:plakoglobinexpression
terscomparedwithsingleCTCs(219-fold)andthefactthatits wasremarkablyheterogeneousinboththeprimaryandmetasta-
expression in primary tumors associated with a significantly ticbreasttumors,withfociofhighexpressioninterspersedwith
reduced distant metastasis-free survival (p = 0.008) (Figures regionswithoutdetectableprotein(Figure6B).Thus,whilepla-
5D,6A,andS5).WethereforeselectedplakoglobinasaCTC- koglobinisakeycomponentofintercellularjunctions,itsvariable
cluster-enrichedtranscriptformoredetailedanalysis. expressionlevelswithinprimarytumorsraisesthepossibilitythat
Plakoglobin (JUP) is a member of the Armadillo family of it might demarcate tightly adherent groups of cells that may
proteins and an important component of desmosomes and constituteprecursorstoCTCclusters.
1116 Cell158,1110–1122,August28,2014ª2014ElsevierInc.
A B
C
D E
Figure5. RNASequencingofCTCClustersandSingleCTCsRevealsaCTC-Clusters-AssociatedGeneSet
(A)Illustrationoftheexperimentalsetup(top).Representativeimagesofalabeledwhitebloodcell(WBC,red),asingleCTCandaCTCcluster(green)(bottom)are
shown.
(B)Heatmapshowingunsupervisedhierarchicalclusteringof15singleCTCspoolsandmatched14CTCclustersisolatedfromtenbreastcancerpatients
(SC,singleCTCs;CL,CTCcluster).
(C)Heatmapshowingthetop31transcriptsupregulatedinCTCclusters.n=15forsingleCTCsandn=14forCTCclusters;q<0.01,log2foldchange(FC)>1in
morethan70%intrapatientcomparisons(SC,singleCTCs;CL,CTCcluster).
(D)Graphshowinglog2foldincreaseinplakoglobin(JUP)foreachcomparisonbetweenmatchedCTCclustersversussingleCTCs.Thethresholdlinerepresents
aq<0.01andlog2foldincrease>1.
(E)RepresentativeimagesofasingleCTCandaCTCclustercapturedontheHBCTC-Chipfromabreastcancerpatientandstainedwithwide-spectrum
cytokeratin(CK,red),plakoglobin(green),andDAPI(nuclei,blue).
SeealsoFigureS4andTableS2.
Cell158,1110–1122,August28,2014ª2014ElsevierInc. 1117
A B
Figure6. PlakoglobinExpressionCorrelateswithDecreasedDistantMetastasis-FreeSurvival
(A)Kaplan-Meierdistantmetastasis-freesurvivalplotshowingprogressionratesforpatientswhoseprimarytumorexpressedeither‘‘lowplakoglobin’’or‘‘high
plakoglobin’’transcript.n=1,956;p=0.008bylogranktest.
(B)Representativeimagesofplakoglobin(red)andCD31(bloodvessels,brown)-stainedtissuesectionsofmatchedprimarytumor(left)andbonemetastasis
(right)fromahormonereceptor-positivebreastcancerpatientwithhighCTCclustercounts.Arrowsindicatebloodvesselsin‘‘highplakoglobin’’regions.Nuclei
arestainedwithhematoxylin.Thebargraph(middle)showsplakoglobinreadspermillioninmatchedsingleCTCsandCTCclustersisolatedfromthesame
patient.ErrorbarsrepresentSEM.n=3;*p=0.031.
SeealsoFiguresS5andS6.
PlakoglobinIsRequiredforCTCClusterFormationand andS7C).Thus,plakoglobinknockdownabrogatesintercellular
ContributestoBreastCancerMetastasis interactionsrequiredtogenerateclusteredcancercells,thereby
To define the functional consequences of plakoglobin expres- reducingtheirpotentialtoproducelungfociafterdirectintravas-
sioninthecontextofCTCclusters,wefirstappliedaninvitro cularinjection.
assay (Vybrant), which utilizes a fluorogenic dye to measure Finally, we generated orthotopic xenografts, injecting LM2-
cell-to-cell adhesion under a variety of culture conditions (El GFP-Luciferase cells expressing either control or plakoglobin
Khouryetal.,1996).Wecomparedsevenbreastcancercelllines shRNAs into the mammary fat pad of immunodeficient mice
(MDA-MB-231-LM2, BT474, MCF7, T47D, BT549, BT20, and and measuring tumor growth as well as tumor-derived CTCs.
ZR-75-1) with two nontransformed human mammary epithelial Plakoglobinknockdowndidnotaltertheprimarytumorgrowth
cells(HMECandMCF10A),followingstablelentiviral-mediated rate,measuredforupto30days(Figures7CandS7D),nordid
plakoglobin knockdown. shRNA-mediated plakoglobin sup- itaffectthetotalnumberofsingleCTCsderivedfromtheprimary
pressiontriggereddisruptionofcell-cellcontactsin6/7breast tumor (Figure 7D). Remarkably, the number of tumor-derived
cancerlinesgrownasamonolayer,whileithadnodetectable CTCclusterswassignificantlyreducedinmicebearingLM2pla-
effectineitherofthetwonontransformedbreastepithelialcells koglobin shRNA-expressing tumors compared to control mice
(p<0.04)(Figures7A,S7A,andS7B).Thus,breastcancercells (Figure7D).Inparallel,bioluminescenceimagingofmouselungs
maybemoredependentonplakoglobin-mediatedcelljunctions demonstratedastriking80%reductioninlungnodulesformice
thannormalepithelialcells,whichmaybenefitfromadditionalor bearingplakoglobin-suppressedprimarytumors(Figure7E).
alternativepathwaysinformingintercellularconnections(Alford Together,ourdatasuggestamodelwherebyplakoglobin-ex-
and Taylor-Papadimitriou, 1996; Cavallaro and Christofori, pressing regions within a primary tumor produce aggregated
2004). tumor cells, i.e., CTC clusters, that are shed into the blood-
To extend these observations in vivo, we introduced either stream,wherethey demonstraterapidclearing atdistantsites
plakoglobin shRNAs or nontarget controls into GFP-Lucif- andenhancedmetastaticpotential(Figure7F).Thismechanism
erase-taggedLM2andBT474cellsandpreparedtheseassingle ofmetastaticspreadandthepossibilitythatCTCclustersmay
cells(SC)orclusters(CL)fortailveininjectionintoimmunosup- betargetedtherapeuticallythroughdisruptionofcell-celljunc-
pressedmice(FigureS7C).Consistentwithourpreviousresults, tionsprovideanopportunityforstrategiestoreducethemeta-
bothLM2andBT474cellsexpressingcontrolshRNAsshowed staticspreadofbreastcancer.
dramatically increased persistence in the lung when prepared
underCLversusSCconditions.Incontrast,despiteCLcondi- DISCUSSION
tions, plakoglobin knockdown in both LM2 and BT474 cells
dissociatedclustersintosinglecells,consistentwiththerequire- By applying microfluidic CTC isolation technologies to both
ment for plakoglobin for intercellular adhesion in these cells. patientswithbreastcancerandmousemodels,wehavecharac-
Following plakoglobin knockdown, tail vein inoculation of CL terized CTC clusters, a striking but poorly understood feature
andSCpreparationsofbothLM2andBT474werecomparable of bloodborne metastasis. CTC clusters have been observed
inproducingareducednumberoflungmetastases(Figures7B in patients with cancers of different origin and using multiple
1118 Cell158,1110–1122,August28,2014ª2014ElsevierInc.
A B
C D
E F
Figure7. PlakoglobinIsRequiredforCTCClusterFormationandLungMetastasis
(A)Bargraphshowingtherelativecell-to-celladhesioninapanelofmammaryepithelialcellsandbreastcancercelllinesgrowninthepresenceorabsenceof
plakoglobin.ErrorbarsrepresentSEM.n=5;*p<0.04.
(B)LungmetastasisgrowthcurvesfrommiceinjectedwithLM2-GFP-Luciferase(left)orBT474-GFP-Luciferase(right)cellsexpressingcontrolorplakoglobin
shRNAsandpreparedassinglecells(SC)orclusters(CL).ErrorbarsrepresentSEM.n=4;*p<0.05,**p<0.04byStudent’sttest.
(C)LM2-GFP-Luciferasetumorgrowthcurvesinthepresenceorabsenceofplakoglobin.n=4;NS,notsignificant.
(D)BargraphsshowingthenormalizednumberofCTCclusters(left)andsingleCTCs(right)permlofblood.Bloodsampleswereisolated4weeksafterprimary
tumordevelopmentandprocessedwiththeHBCTC-Chip.ErrorbarsrepresentSEM.n=4;*p<0.05byStudent’sttest.
(E)BargraphshowingnormalizedlungphotoncountsfrommicebearingaLM2-GFP-Luciferasecontrolorplakoglobinknockdownprimarytumorfor4weeks.
ErrorbarsrepresentSEM.n=4;*p<0.045byStudent’sttest.
(F)Schematicshowingthat‘‘highplakoglobin’’regionsintheprimarytumorarelikelytogenerateCTCclusterswithincreasedmetastaticpotential.
SeealsoFigureS7.
technologies(Choetal.,2012;Fidler,1973;Liottaetal.,1976; normal capillaries. Our in vivo flow cytometry studies indicate
Molnar et al., 2001; Stott et al., 2010; Yu et al., 2013). While thatclustersaremorerapidlyclearedfromthecirculationthan
mostclustersarerelativelysmall,somecomprisedozensoftu- single CTCs. Nonetheless, both the structural deformability of
mor cells, raising the question of how they navigate through theaggregatedcellswithintheseclustersandthepresenceof
Cell158,1110–1122,August28,2014ª2014ElsevierInc. 1119