CoactivationofReceptorTyrosineKinasesAffectstheResponseofTumorCellstoTargetedTherapies
JayneM.Stommel,1AlecC.Kimmelman,1,2HaoqiangYing,1RoustemNABIoullin,3ADItyaH.Ponugoti,3RuprechtWiedemeyer,1AlexanderH.Stegh,1JamesE.Bradner,4KeithL.Ligon,1,5CameronBrennan,6LyndaChin,1,3,7RonaldA.DePinho1,3,8* Targetedtherapiesthatinhibitreceptortyrosinekinases(RTKs)andthedownstreamphosphatidylinositol3-kinase(PI3K)signalingpathwayhaveshownpromisinganticanceractivity,buttheirefficacyinthebraintumorglioblastomamultiforme(GBM)andothersolidtumorshasbeenmodest.WehypothesizedthatmultipleRTKsarecoactivatedinthesetumorsandthatredundantinputsdriveandmaintaindownstreamsignaling,therebylimitingtheefficacyoftherapiestargetingsingleRTKs.Tumorcelllines,xenotransplants,andprimarytumorsindeedshowmultipleconcomitantlyactivatedRTKs.CombinationsofRTKinhibitorsand/orRNAinterference,butnotsingleagents,decreasedsignaling,cellsurvival,andanchorage-independentgrowtheveningliomacellsdeficientinPTEN,afrequentlyinactivatedinhibitorofPI3K.Thus,effectiveGBMtherapymayrequirecombined
regimenstargetingmultipleRTKs.
BM,themostprevalenttumorinthecentralnervoussystemofhumanadults,isamongthemostlethalcancers,withamediansurvivalof~12months(1).AberrantactivationofPI3Kpathwaycomponentsappearstobeuniversalinhumancancer,includingGBM.PI3KisactivateduponbindingphosphorylatedRTKsand/oradaptorproteinsattheplasmamembraneandsignalstomultipledownstreameffectors,suchasAkt(2).Over80%ofGBMsshowrobustAktactivation,and40to50%havelostormutatedPTEN(3),underscoringtheim-portanceofthePI3Kpathwayingliomagenesis(4).ActivationoftheRTKepidermalgrowthfactorreceptor(EGFR)isacriticalpathogeneticevent,withamplification,mutation,andrear-rangementobservedinmorethan40%ofcases,
1DepartmentofMedicalOncology,Dana-FarberCancerInstituteandHarvardMedicalSchool,Boston,MA02115,USA.2HarvardRadiationOncologyProgram,HarvardMedicalSchool,Boston,MA02115,USA.3CenterforAppliedCancerScience,BelferInstituteforInnovativeCancerScience,Dana-FarberCancerInstitute,Boston,MA02115,USA.4DivisionofHematologicNeoplasia,Dana-FarberCancerInstituteandHarvardMedicalSchool,Boston,MA02115,USA.5DepartmentofPathology,DivisionofNeuropathology,BrighamandWomen’sHospital,Boston,MA02115,USA.6DepartmentsofNeurosurgery,MemorialSloanKetteringCancerCenterandNeurologicalSurgery,WeillMedicalCollegeofCornellUniversity,NewYork,NY10021,USA.7DepartmentofDermatology,BrighamandWomen’sHospitalandHarvardMedicalSchool,Boston,MA02115,USA,8DepartmentsofMedicineandGenetics,HarvardMedicalSchool,Boston,MA02115,USA.
*Towhomcorrespondenceshouldbeaddressed.E-mail:ron_depinho@dfci.harvard.edu
makingitacompellingtargetfortherapeuticin-hibition(5,6).OtherRTKs,suchastheplatelet-derivedgrowthfactorreceptorsaandb(PDGFRaandPDGFRb)andMET(1,7),havebeenreportedtobealteredinGBM,albeitatlowerfrequencies.Notably,anti-PDGFRtherapyhasfailedinGBMpatients(8),andonly10to20%ofpatientsbenefitfromEGFRinhibition(9),point-ingtoconfoundingfactorsthatattenuatetheresponsetoRTKinhibition.
Coexpressionofwild-type(WT)PTENandaconstitutivelyactiveEGFR(vIIImutant)inGBMcorrelateswithclinicalresponsetoEGFRinhibitors,indicatingthatPTENisaresponsebioMarkerforanti-EGFRtherapyandthatitslossrenderstheseagentsineffectivebydissociatingEGFRinhibitionfromtheabrogationofPI3Kpathwayactivity(10).Alternatively,theactiva-tionstateofcriticalsurvivalpathways,suchasPI3Kandmitogen-activatedproteinkinase(MAPK),maybedeterminedbythesumofmul-tipleinputs,andmultipleRTKsbesidesEGFRmaybesimultaneouslyorsequentiallyusedbyGBMcellstomaintainsignalfluxthroughsuchpathways.Insuchamultiple-inputsystem,single-agentanti-EGFRinhibitionmightbeincapableofsufficientlysuppressingPI3KsignalinginthecontextofunopposedactivationbyPTENloss,resultinginalackofclinicalefficacy.Thatis,thetotalsignalfluxthroughthePI3KpathwaymaydictatetheresponsetoupstreamRTKinhibition,andmultipleinputstoPI3Ksignalingwouldthusconferinsensitivitytotheinhibitionofanysingleagent.
Toevaluatethispossibility,wesoughtevidencethatmultiplePI3Kactivatorscoexistingliomacells().BecausePI3Kisactivatedbybindingphosphorylatedproteinstoitsregulatorysubunit,p85a,weperformedanti-p85aimmu-noprecipitationstoidentifyPI3K-activatingpro-teins.Multipletyrosine-phosphorylatedproteinswerefoundtobeinthePI3Kcomplex(fig.S1A).GuidedbyScansite[http://scansite.mit.edu(12)],whichidentifiespotentialp85a-bindingproteins,weconfirmedtheendogenoUSPI3KinteractionwithspecificRTKsbycoimmunoprecipitationassays.Forexample,5outof14celllinesshowedactivatedERBB3(fig.S1B),whichme-diatesthebindingofEGFRandERBB2toPI3K(13),and9outof20celllineshadactivatedgrowthfactorreceptor–boundprotein2(Grb2)–associatedbinder1(GAB1),adockingproteinthatbindsactivatedRTKsdirectlyorthroughGrb2(14).Insevenofthese20celllines,highlyphosphorylatedGAB1coimmunoprecipitatedactivatedMET(fig.S2B)(15).
TodefinethecompendiumofcoactivatedRTKsinGBM,weusedanantibodyarraythatallowssimultaneousassessmentofthephospho-rylationstatusof45RTKs.Consistentwithfigs.S1andS2,threeormoreactivatedRTKs,includ-ingEGFR,ERBB3,PDGFRa,andMET,weredetectedineachof19outof20celllines(Fig.1AandtableS1).MostactivatedRTKsremainedphosphorylatedunderserumdeprivationandintumorcellxenografts(Fig.1,BandC),indi-catingthattheRTKactivationinthetransformedcellsisnotduetoligandsinserum-containingculturemedia.Finally,RTKcoactivationisnotadistinctivefeatureofgliomacells,becausesimi-larpatternsweredetectedinothersolidtumortypessuchaslungandpancreaticadenocarcino-macelllines(fig.S3).
ToexplorethetherapeuticimplicationsofRTKcoactivation,weusedU87MGgliomacellsthatconstitutivelyexpressWTEGFR,EGFRvIII(EGFR*),orakinase-deadEGFRvIII(EGFR*-KD)atlevelscomparabletothoseob-servedinprimaryGBMs(16).AlthoughMETisphosphorylatedandboundtoGAB1inU87MGcells(Fig.2Aandfig.S2B),activatedMETwassubstantiallydisplacedbyEGFRintheGAB1/PI3KcomplexwhenWTEGFRandEGFR*areexpressed.Thisoutcomerequiredthecata-lyticactivityofEGFR,becauseEGFR*-KDonlymodestlydisplacedMET(lane4inFig.2A).BecauseEGFR*-KDwasexpressedatlev-elssimilartothoseofWTEGFRandEGFR*,itisunlikelythatthedisplacementwassimplyaconsequenceofectopicoverexpression.Thisap-parent“swapping”ofRTKswithinthePI3K
Downloadedfromwww.sciencemag.orgonDecember23,2007
www.sciencemag.orgSCIENCEVOL31812OCTOBER2007287
REPORTS
Fig.1.MultipleRTKsareactivatedsimultaneouslyingliomacelllines.(A)Whole-cellextracts(WCEs)fromthegliomacelllinesLN382,SF763,LN18,andHS683wereincubatedonRTKantibodyarrays,andphosphorylationstatuswasdeterminedbysub-sequentincubationwithanti-phosphotyrosinehorse-radishperoxidase.EachRTKisspottedinduplicate:Thepairsofdotsineachcornerarepositivecon-trols.EachpairofpositiveRTKdotsisdenotedbyarednumeral,withtheidentityofthecorrespond-ingRTKslistedbelowthearrays.ThesearraysarerepresentativeofvariousRTKcoexpressionpatternsinthe20gliomacelllinesexaminedintableS1.(B)RTKantibodyarrayswereusedasin(A)withWCEsfromanimmortalizedhumanastrocytecellline[E6/E7/hTERTNHA(19)]orthegliomalineLNZ308grownin10%serum(log)orgrownfor48hoursin0.05%serum(serumstarved).(C)RTKantibodyarrayswereusedasin(A)tocompareRTKactivationinWCEsfromxenografttumorsderivedfromthegliomacelllinesSF767orLN340orfromthecorrespondinginvitroculturedcells.
Fig.2.InhibitionofmultipleRTKsisnecessarytoabrogatePI3Kandreducecellsurvival.(A)ThePI3K/GAB1adaptorcomplexcanreadilyswitchbetweenMETandEGFRbindingwithlittlediscernableeffectondownstreamsignaling.U87MGparentalcellsorcellsconstitutivelyexpressingWTEGFR,theactivat-ingvIIIdeletionmutant(EGFR*),orthevIIImutantwithaninactivatingmutationinitskinasedomain(EGFR*-KD)wereimmunoprecipitated(IP)withanantibodytotheRTK/PI3Kadaptorprotein,GAB1(left),andthenimmunoblotswereprobedwiththeindicatedantibodies.Heavychain(hc)isshowntodemonstrateequalimmunoprecipitationefficiency.WCEsfromthesamecellswereimmunoblottedwiththeindicatedantibodies.(B)(Top)U87MG-EGFR*cellsweretreatedwitheachoftheRTKinhibitors[10mMerlotinib(E),10mMSU11274(S),and/or10mMimatinib(I)],andthenWCEswereimmunoprecipi-tatedwithanantibodytoGAB1,eluted,andimmunoblottedwithantibodiestop85aorGAB1.NotethefastermigrationofGAB1inlysatesfromRTKinhibitor–treatedcells,whichisconsistentwithadecreaseinphosphorylation.(Bottom)WCEswereimmunoblottedwiththeindicatedantibodies.(C)TreatmentwithmultipleRTKinhibitorsdecreasesU87MG-EGFR*cellviability.Cellsweretreatedfor72hourswithcombinationsof10mME,10mMS,and10mMI,orwith100nMactinomycinD(actD)in0.05%serum-containingmedium,andthencellviabilitywasassayedbyadenosinetriphosphatequantitation.ErrorbarsindicateSEM;n=fourexperiments.unt,untreated.(D)ImpactofsingleandcombinationRTK-inhibitortreatmentsonsoft-agarcolonyformationofU87MG-
complexdidnotalterdownstreamsignaling(rightpanelinFig.2A),indicatingthatMETandEGFRactasredundantbutindependentinputstotheirsignalingnetworks.Consequently,coactivatedMET(orotherRTKs)wouldbeexpectedtoren-deranti-EGFRinhibitionineffectiveinextin-guishingdownstreamsignalingbyreplacingactivatedEGFRinthePI3Kcomplex.
ToaddresswhethercoactivatedRTKsconferresistancetosingleanti-RTKinhibition,weex-aminedtheconsequencesofsingleandcombinedinhibitionofEGFRandMETinU87MG-EGFR*cellswiththesetwocoactivatedRTKs,usingP-AktandP-S6ribosomalproteinasmolec-ularsurrogatesofdownstreamRTKsignaling.WefirstconfirmedthattreatmentwitheitheranEGFRinhibitor(erlotinib)oraMETinhibitor(SU11274)effectivelyblockedphosphorylationoftheirintendedtargetsinU87MG-EGFR*cells(fig.S4A).Althoughtreatmentwitheitherinhibitoralonehadnodiscern
IBLeeffectonPI3KassociationwithGAB1,combinedinhibi-tionwithbotherlotinibandSU11274resultedinthereleaseofp85afromtheRTK/GAB1com-plexandfastermigrationofGAB1,whichisconsistentwithareductioninGAB1phospho-rylation.Accordingly,downstreamsignalingasmeasuredbyP-AktandP-S6wasinhibitedonlywhentwoinhibitorswerecombined(Fig.2B).Moreover,combinedRTKinhibitionsignificant-lydecreasedviabilityofculturedU87-EGFR*cells(Fig.2C)andreducedthenumberandsizeofsoft-agarcoloniesformedinastringent
Downloadedfromwww.sciencemag.orgonDecember23,2007
28812OCTOBER2007VOL318SCIENCEwww.sciencemag.org
REPORTS
Fig.3.TargetingmultipleactivatedRTKsabro-gatescellsignaling,anchorage-independentgrowth,andviability.(A)ThegliomacelllineLN382wastreatedwiththeindicatedRTKinhibitorsin0.05%serum-containingmediumandimmunoblottedasinFig.2B.TheactivatedRTKsdetectedinthesecellsonantibodyarraysasinFig.1Aareindicatedbeneaththeblots.(B)CombinedRTK-inhibitortreatmentsinhibitsoft-agarcolonyformationinLN382cellswhenplatedandtreatedasinFig.2D.(C)TheinhibitionofmultipleactivatedRTKsdecreasescellviability.LN382cellsweretreatedwith10mME,10mMS,and/or10mMI,orwith100nMactD,andthencelldeathwasassayedasinFig.2C.ErrorbarsindicateSEM;n=fourexperiments.(D)Soft-agarcolonyinhibitionbyMETsiRNAscombinedwithRTKinhibitors.LN382cellsweretransfectedwithMETsiRNA(siM)orascramblednegativecontrol(–)andplatedandtreatedwiththeindicatedRTKinhibitorsasinFig.2D.
Fig.4.MultipleRTKsareconcomitantlyactivatedinprimaryGBMs.(A)AntibodyarrayswereperformedasinFig.1onproteinlysatesextractedfromsnap-frozenprimaryhumangliomasornormalbrainautopsymaterial.(B)Coexpressionofphospho-RTKsincellsdissociatedfromprimaryGBMMSK199.Individualtumor-derivedcellswereimmunofluorescentlystainedwiththephospho-RTKantibodiesP-EGFR,P-PDGFRa,P-InsR,orP-CSF1R.Eachrowdepictsonefieldofcellsfromaslidesimultaneouslystainedwiththeindicatedantibodies.DNAislabeledwithHoechst33342.NestinisexpressedinneuralProgenitorcells,tu-morendothelialcells,anddiffusegliomas,includ-ingastrocytomasandGBMs(20),andolig2isexpressedinneuralprogenitors,normaloligoden-droglia,anddiffusegliomas(21).Thebottomrowdepictscellsstainedonlywithsecondaryanti-bodies.
www.sciencemag.orgSCIENCEVOL31812OCTOBER2007289
REPORTS
anchorage-independentgrowthassay(Fig.2Dandfig.S4B),whereassingleinhibitorswerelesseffective.Inlinewiththelackofdetectablephospho-PDGFRorc-KITinU87MG-EGFR*cells(fig.S4A),treatmentwiththePDGFR/c-KIT/ablkinaseinhibitorimatinibonlyslightyaffectedPI3Kactivation(Fig.2B);however,whencombinedwitherlotinibandSU11274,tripleinhibitortreatmenteliminatedresidualP-AktandP-S6activity(comparelanes4and8inFig.2B)andconferredthemostdramaticin-hibitiononviabilityandanchorage-independentgrowth(Fig.2,CandD),possiblyreflectingtheinhibitoryactivityofimatinibonotherkinasesinthesecells(17).
AlthoughithasbeenreportedthatPTENlossabrogatestheresponseofGBMtoEGFRinhib-itors(10),ourfindingthatcombinationtreat-mentssignificantlyreducedP-AktandP-S6inPTEN-mutantU87MG-EGFR*cells(Fig.2B)suggeststhat,evenwithPTENdeficiency,com-binedsignalingfrommultipleRTKsisrequiredtomaintaindownstreampathwayactivation,suchasthatofPI3K.ToexaminewhetherinhibitionofcoactivatedRTKsviacombinationtherapymit-igatesPI3KactivityinotherPTEN-deficientgliomacells,wesubjectedmultiplecelllines,eitherWTormutantforPTEN,tosingleandcombinationtreatmentwitherlotinib,SU11274,andimatinib.Consistently,PI3KsignalingwasreducedorcompletelyabrogatedwithcombinedinhibitionofcoactivatedRTKsirrespectiveofPTENstatus(Fig.3Aandfig.S5A).Addition-ally,inhibitionofPI3KactivityviaRTKinhi-bitionabrogatedanchorage-independentgrowthandcellviability(Fig.3,BandC,andfig.S5,BandC).ThedecreaseincellviabilitywasmediatedinpartbyinhibitionofPI3Ksignaling,becausetransienttransfectionofeithermyristoylatedAktorp110a-CAAX(C,cysteine;A,aliphaticaminoacid;X,anyaminoacid)increasedcellviabilityindrug-treatedcells(fig.S6,P<0.001).Atthesametime,theinabilitytocompletelyrescuevia-bilityindicatesthatPI3KisnotthesolemediatoroffunctionalRTKsignaling.Inthesetreatmentstudies,weareawarethat(i)manykinasein-hibitorsexhibitactivitiesagainstmultiplekinasesinadditiontotheirprimarytargets(17)[e.g.,SU11274diminishedP-PDGFRbinLN382cells(fig.S5D)]and(ii)aparticularRTKmaynotberepresentedorbedetectedasactivatedonanantibodyarray.Nevertheless,inthreedifferentgliomacelllines,colonyformationandcellvia-bilitywerepartiallyinhibitedbysingleanddualtreatmentwithRTKinhibitorsbutweremostaffectedbycombinedtreatmentwithallthreeinhibitors(Fig.3,BandC,andfig.S5,BandC).Giventhepotentialnonspecificactionsoftheseagents,particularlySU11274[theU.S.FoodandDrugAdmiNISTration(FDA)–unapprovedMETinhibitor],weusedRNAinterference(RNAi)againstMETtoverifythATGeneticinhibitionofMETcansimilarlyconferenhancedanti-oncogenicactivityoferlotinibandimatinibinLN382.Indeed,transfectionofLN382cellswithMETsmallinterferingRNAs(siRNAs),incombinationwitherlotinibandimatinib,resultedinanearlysimilarlevelofinhibitioninsoft-agarcolonyformationinananchorage-independentgrowthassay(Fig.3D).Similartrendswereob-servedwithRNAiagainstPDGFRandEGFR(fig.S7).Theseresultssupporttheviewthat,eveninPTENmutantcells,morerobustanti-oncogeniceffectscanbeachievedthroughcombinedinhibi-tionofrelevantupstreamsignalinginputs. Finally,weassayeduntreatedprimaryhumanGBMtumorsfromnewlydiagnosedpatientsforevidenceofRTKcoactivation.IncontrastwithanormalbrainspecimenthathadnodetectableRTKactivation,eachofthe14GBMsamplesexaminedbyantibodyarrayprofilingharboredmultiplephosphorylatedRTKs(Fig.4AandtableS2).TheseincludedknownGBMRTKs,suchasEGFR,PDGFRa,andMET,aswellasRTKsnotpreviouslylinkedtoGBM,suchasRET,MST1R,andCSF1R.Giventhewell-knownintratumoralheterogeneityinGBM,weperformedimmunofluorescencestainingwithphosphospecificantibodiesagainstmultipleRTKsandobservedcoexpressionofactivatedRTKsinindividualdissociatedcellsfromapri-maryGBM(Fig.4B).Togetherwiththeinvitrodataabove,thisevidenceofinvivoRTKcoac-tivationsupportsourhypothesisthatconcomitantactivationofmultipleRTKsservestoreducede-pendenceonanysingleRTKforthemaintenanceofcriticaldownstreamsignalinginacomplextumormicroenvironment,thusrenderingsuchtumorsrefractorytosingle-agentRTKinhibition.
ThefindingsofthisstudyprovidearationalexplanationforthefeebleclinicalresponsestoRTK-inhibitormonotherapyformanysolidtumortypesandanticipatemorefavorableout-comesbycombinationsofdrugsagainstdifferentactivatedRTKsorsingledrugswithinhibitoryactivitiesagainstmultipleactivatedRTKs.More-over,bydemonstratingthecapabilitytorapidlyprofiletheactivationstatusofmostmembersoftheRTKsinresectedGBMspecimensandtheuseofsuchprofilestotailorrationalcombinationtherapies,thisstudyprovidesproof-of-conceptfortheeventualimplementationofa“personal-ized”therapeuticparadigminhumancancer(18).BecauseFDA-approvedRTKinhibitorsareavailableandadditionaldrugsareunderdevel-opment,thistreatmentparadigmcouldbereadilyimplementedforcancersthatarecurrentlyhighlyrefractorytoexistingtherapies.
ReferencesandNotes
1 E.A.Maheretal.,GenesDev.15,1311(2001).
2 L.C.Cantley,Science296,1655(2002).
3 J.Lietal.,Science275,1943(1997).
4 H.Wangetal.,Lab.Invest.84,941(2004).
5 A.El-Obeidetal.,CancerRes.57,5598(1997).
6 M.Naganeetal.,CancerRes.56,5079(1996).
7 B.Wullich,H.P.Sattler,U.Fischer,E.Meese,AnticancerRes.14,577(1994).
8 P.Y.Wenetal.,Clin.CancerRes.12,4899(2006).
9 J.N.Richetal.,J.Clin.Oncol.22,133(2004).
10 I.K.Mellinghoffetal.,N.Engl.J.Med.353,2012(2005).
11 MaterialsandmethodsandsupplementaryfiguresareavailableassupportingmaterialonScienceOnline.
12 J.C.Obenauer,L.C.Cantley,M.B.Yaffe,NucleicAcidsRes.31,3635(2003).
13 N.E.Hynes,H.A.Lane,Nat.Rev.Cancer5,341(2005).
14 H.Gu,B.G.Neel,TrendsCellBiol.13,122(2003).
15 K.M.Weidneretal.,Nature384,173(1996).
16.R.Nishikawaetal.,Proc.Natl.Acad.Sci.U.S.A.91,7727(1994).
1 M.A.Fabianetal.,Nat.Biotechnol.23,329(2005).
2 H.Varmus,Science312,1162(2006).
3 Y.Sonodaetal.,CancerRes.61,4956(2001).
4 K.Sugawaraetal.,Lab.Invest.82,345(2002).
1 K.L.Ligonetal.,J.Neuropathol.Exp.Neurol.63,499(2004).
2 WethankW.CaveneeandF.Furnariforgliomacelllines;
R.PieperfortheE6/E7/hTERTnormalhumanfibroblasts;
A.Chanforthemyr-Aktconstruct;membersoftheDePinhoandChinlaboratories,J.Engelman,andL.C.Cantleyforhelpfuldiscussions;andY.-H.Xiao,B.Feng,andJ.Zhangforbioinformatichelp.J.M.S.isfundedbytheAmericanBrainTumorAssociationandaRuthL.KirschsteinNationalResearchServiceAward.H.Y.isfundedbytheAmericanBrainTumorAssociation.A.C.K.isarecipientoftheLeonardB.HolmanResearchPathwayFellowship.R.W.issupportedbyaMildredScheelFellowship(DeutscheKrebshilfe).GrantsupportcomesfromTheClaudiaAdamBarrFoundation(A.H.S.),theGoldhirshFoundation(L.C.),andNIHgrantsRO1CA99041(L.C.)and5P01CA95616(R.A.D.,L.C.,C.B.,andK.L.L.).R.A.D.isanAmericanCancerSocietyResearchProfessorandanEllisonMedicalFoundationScholarandissupportedbytheRobertA.andReneeE.BelferFoundationInstituteforInnovativeCancerScience.ThisworkwasconductedsolelyattheDana-FarberCancerInstituteandhasnoindustrialconnectionorinfluence.R.A.D.isamemberoftheSeniorAdvisoryBoardofAbbottPharmaceuticalsandisacofounder,scientificadvisor,anddirectorofAveoPharmaceuticals.L.C.isacofounderandscientificadvisorofAveoPharmaceuticals.
SupportingOnlineMaterial
www.sciencemag.org/cgi/content/full/1142946/DC1
MaterialsandMethods
Figs.S1toS7
TablesS1andS2
23March2007;accepted31August2007Publishedonline13September2007;10.1126/science.1142946Includethisinformationwhencitingthispaper.
Downloadedfromwww.sciencemag.orgonDecember23,2007
29012OCTOBER2007VOL318SCIENCEwww.sciencemag.org
