復(fù)旦大學(xué)醫(yī)學(xué)遺傳學(xué)casecontrolTDT分析課件

FudanGeneticWorkshop2005

GeneticEpidemiologyinPopulations

YinYaoyyao@GeneticAssociationStudies:StudieswithUnrelatedIndividualsFudanGeneticWorkshop,Jan042005Assessingriskfordiseaseduetospecificalleles/genotypesContrastdirect/indirectgeneticassociationstudiesDiscussgeneticmodelsTestsandestimatesofassociationbetweenalleles/genotypesanddiseaseDesignstobeusedLearningObjectivesforThisLectureApplicationsforgeneticassociationanalysesDirectversusindirectassociationIndirectassociation:marker-baseddiseaseassociationExploitsLDbetweenmarkersandunobserveddiseasevariantsExamplesofdirectandindirectassociationstudiesStudydesignsfordirectandmarker-based(LD)associationGeneticAssociationStudiesinEpidemiologyCommonDesigns:Cross-sectionalCase-controlCohortClinicalTrial(drugresponsehypothesisisreallyacase-control)Case-family(trios,sibs,extendedfamilies)Case-onlyLearningObjectivesforThisLectureApplicationsforassociationanalysesDirectversusindirectassociationIndirectassociation:marker-basedassociationExploitsLDExamplesofdirectandindirectassociationstudiesStudydesignsfordirectandmarker-based(LD)associationDirectmethodTestingwhetheraparticularalleleisadisease-predisposing(causative)allele‘exposurestatus’directlymeasuredGeneticepidemiologystudies–twodifferentconcepts

..GACTAAGGCCCCCGTTCAAGGAA.. C/TAPOEgeneonc19Eg:AparticularAPOEallele(e4)changesproteinisoformGenotypethatparticularsiteforassociationstudyDirectTestsTestforassociationbetweenobservedgenotypesanddiseaseoutcomeCaseControlTotalCCabm1CTcdm2TTefm3TotalndnhN…CCCCCG……CCCCCG……CCCCCG……CCTCCG……CCTCCG……CCTCCG…DirectlymeasurepotentiallypredisposingallelesIndirectTestsTestforassociationbetweenmarkergenotypesanddiseaseoutcomeMarkerCaseControlTotalAAabm1AGcdm2GGefm3TotalndnhN…CC?CCG…CAG…CC?CCG…CAGUsemarkergenotypes

assurrogatefor

‘functional’genotype

byexploitinganycorrelationbetween

CandA…CC?CCG…CAG…CC?CCG…CGG…CC?CCG…CGG…CC?CCG…CGGDirectMethod:CandidateAlleleTestingMustknowpotentiallyfunctionalpolymorphismsSNPsmayoffersetofcandidatelociforthismethodEg,lookatnon-synonymouscSNPs(thoseincodingregionsthatarelikelytobefunctional)Note:Assumingdiseaseiscausedby(relatively)commonallelesCommondisease-commonvarianthypothesisThishypothesisiscontroversialIfgenome-wideproject:multiplecomparisons-~30,000genes.Evenifonlyonefunctionallocus/genetested,veryhighnumberoffalse+sduetochanceIndirectAssociationStudiesinEpidemiologyApplications:Localization(whatisthebestestimateofthediseasegenelocation?)FinemappingWholegenomescanCandidategeneeffectsGene–environmenteffectsDrugresponsemodificationLocalizationIndirectAssociationStudiesinEpidemiologyApplications:Localization(whatisthebestestimateofthediseasegenelocation?)FinemappingWholegenomescanCandidategeneeffectsGene–environmenteffectsDrugresponsemodificationIndirectAssociationStudiesinEpidemiologyApplications:Localization(whatisthebestestimateofthediseasegenelocation?)FinemappingWholegenomescanCandidategeneeffectsGene–environmenteffectsDrugresponsemodificationCandidategeneLDstudiesInstrumentalinidentifyinggeneticriskfactorsfordiseaseHLAType1diabetesRheumatoidarthritisMultiplesclerosisAPOEAlzheimer’sdiseaseAtherosclerosisAngiotensinConvertingEnzyme(ACE)MyocardialinfarctionAtherosclerosisCandidategeneLDstudiesPolymorphismincandidategenescanbeusedasa‘marker’forvariationinthegene.ThehighertheLDbetweenthecausativevariantandthemarkeralleles,thebetterthesurrogateinformationprovidedbythemarker.MarkerCaseControlTotalAAabm1AGcdm2GGefm3TotalndnhN…CC?CCG…CAG…CC?CCG…CAG…CC?CCG…CAG…CC?CCG…CGG…CC?CCG…CGG…CC?CCG…CGGThepowerofGtodetectdiseaseriskat?isonlyasstrongastheLDbetweentheloci!Genotype-leveltestsfordiseaseassociation

TestforassociationbetweengenotypesanddiseaseoutcomeGCaseControlTotal11abm112cdm222efm3TotalndnhNEx:Alzheimer’sdiseaseandAPOESNPs…Single-markerc2testsfordiseaseassociation:M1GCaseControlTotalCC101(.50)116(.75)217CT75(.37)32(.21)107TT26(.13)6(.04)32Total202154356ModelingissuesIfwedomeasuretheputativevariationofinterest,whatkindsofstatisticalanalysesshouldbeperformed?Mostgeneralgeneticmodelforrisk:assumetoapriorirelationshipbetweenheterozygoteandhomozygoteriskConsidereachgenotypeasaseparateexposurecategory:GenotypeRiskparameterRelativeRiskAARAARRAAA*RA*RRA***R**1Inalogisticregressionmodel:LogisticregressionH0:bi=0

Geneticmodelinterpretations:Assume“11”genotypecodingrepresentsgenotypewithlowestabsoluterisk(baseline) b1=b2=0 noassociationwiththatpolymorphism b1=0,b2>0 (completely)recessive b1=b2>0 (completely)dominant 0<b1<b2 additiveormultiplicativeNote:Ifrecessive,dominant,ormultiplicativemodelsareappropriate,couldcreateasinglevariablewith0/1codingfordominantandrecessive,and0/1/2codingfornumberofcopiesofaparticularallele.Note:ThiscanbeextendedthroughGLMtomanytypesofoutcomes(ratherthansimplyoddsofdisease/notdisease,asabove)

IndirectAssociationStudiesinEpidemiologyApplications:Localization(whatisthebestestimateofthediseasegenelocation?)FinemappingWholegenomescanCandidategeneeffectsGene–environmenteffectsDrugresponsemodificationEx:IncorporationofEnvironmentMomsmokedduringpreg.CPControlG-72127G+121984146MomdidnotsmokeCPControlG-2515G+1554020SmokeTGFaCPControlORNoG-721271NoG+12191.11YesG-25152.94YesG+1555.29MaternalsmokingisassociatedwithoralcleftandTGFaCouldbeaconfounder,orinteractEx:IncorporationofEnvironmentSmokeTGFaCPControlORNoG-721271NoG+12191.11YesG-25152.94YesG+1555.29G+effectintheabsenceofE: ORG=1.11SmokeTGFaCPControlORNoG-721271NoG+12191.11YesG-25152.94YesG+1555.29EeffectintheabsenceofG: ORE=2.94Assumingamultiplicativemodel,Ifnointeraction,ORforcombinedgroupG+,Eshouldbe: ORGE=1.11*2.94=3.26ObservedORGE=5.29>>Possibleinteraction!SmokeTGFaCPControlORNoG-721271NoG+12191.11YesG-25152.94YesG+1555.29Logisticregressionwithinteraction Testforinteraction:H0:gi=0

ModelingissuesForCleftexample,dominancewasmodeled,sologisticmodelwouldlooklike:

InterpretationsofIndirectAssociationStudiesApositiveassociationcanmean:ThetargetedalleleiscausalThetargetedalleleisinLDwithacausalalleleThereisconfoundingduetopopulationstratificationThereisconfoundingorbiasforsomeotherreasonTypeIerrorAnegativefindingcanmean:ThegeneorregionunderstudyisnotassociatedwithdiseaseriskThetargetedalleleisnotinLDwiththecausalalleleAppropriatestratificationorotheraccommodationofheterogeneitywasnotidentifiedTypeIIerror(notenoughpower)LearningObjectivesforThisLectureLinkageversusassociationUsesofassociationanalysesDirectversusindirectassociationIndirectassociation:marker-basedassociationExploitsLDExamplesofdirectandindirectassociationstudiesStudydesignsfordirectandmarker-based(LD)associationChoiceofDesignandAnalysisSamplingdesignUnrelatedindividualsFamily-basedsamplingUnitofanalysisSingle-locusHaplotypesStatisticalProceduresChi-squaretestsLikelihood-basedtestsAsymptoticpvaluesEmpiricalsignificancefromresamplingAppropriatesignificancethresholdsChoiceofDesignandAnalysisDesignoptionsforassociationstudiesSamplingunrelatedindividualsFamily-baseddesignsThesedifferonhow‘controls’aredefinedStudyDesignsforLDMappingStrategiesUnrelatedSamples(Eg,Casecontrol)UnrelatedcontrolsaresampledfromthesamepopulationasthecasesMatchedorunmatchedonotherfactorsPerformchi-squaredtestforassociationFamily-basedEx:TDT,looksforexcesstransmissionofparticularallelesfromparentstoaffectedchildrenControlsare‘untransmittedalleles’MarkeralleleCaseControlAAabAacdaaefForeachindividual,have2x2tableof0s,1s,or2sUseallsuchtablestogetamatchedchi-squaretestforexcessoccurrenceincellsbandc[McNemar’stest]A-Nottransmitteda–NottransmittedA-Transmitted02a-Transmitted00A,aA,aA,AContrastsbetweenepidemiologicandfamily-basedassociationdesignsUnrelatedindividualsDonotneedparentalgeneticinformationProvidesestimatesofallelefrequencies(ifappropriatelysampled)MaybemorepowerfulandsimplertocollectforsomephenotypesOpensthepotentialforconfoundingduetopopulationstratificationFamily-baseddesignsNeedparentalgenotypes(oratleastotherfamilymembers)DonotneedunrelatedcontrolsAvoidspopulationstratificationconfounding(analysisismatchedbyfamily)DesignsforLDstudiesofunrelatedindividualsAllunrelatedsamplingLDmethodslookforassociationbetweenmarkersandtheoutcomeofinterestacrossindividualsTheutilityofthisapproachwillbeafunctionoftheactualLDbetweenthemarkersandthetruediseaseallele(s)Cross-sectionalCase-controlCohortAnalysesaresimilar,interpretationsmaybedifferentbecausesamplingandtemporalrelationshipsaredifferentClinicaltrial?Suchdataareusedtoassesspharmacogeneticresponseoutcomes.However,thisisoftenacase-controldesign(thegenotypesarenotrandomized!)StudyDesignsusedforLDmappingCase-Controldesign(unrelatedindividuals)Advantages:Commonlyusedtoolinepidemiology--methodologyiswell-knownConvenienttocollect--opportunitytodrawverylargesamplesMoreefficientrecruitmentthanfamily-basedsamplingmethodsPopulation-baseddesignscanallowthesimultaneouscharacterizationofdiseaseallelefrequency,penetrance,andattributableriskUnrelatedcontrolscanprovideincreasedpowerinmanysituationsStudyDesignsusedforLDmappingCase-Controldesign(unrelatedindividuals)Limitations:DifficulttoestablishphasewhenfocusingonhaplotypesMaybesusceptibletoconfoundingduetostratificationDifficulttomeasureparent-of-origineffectsorotherparent-specificeffectsDifficulttoestimaterecombinationfractions(localization)usingLDinanunrelated-subjectsettingCase-ControlSusceptibilitytopopulationstratificationTDTandotherfamily-basedassociationmethodscommonlyusedtoguardagainstconfoundingduetopopulationstratificationBut,RequiresrecruitmentofadditionalfamilymembersAddedcostFamilymembersmaynotbeavailableInefficientifmanymembersrequiredormembersareuninformativeProbablynotasprevalentaproblemasoncethoughtCanbedealtwiththroughassessmentandadjustmentofpopulationstratificationwithinacase-controldatasetStudyDesignsusedforLDmappingCase-ControldesignLimitations:MaybesusceptibletoconfoundingduetostratificationSolution:Mea