Chapter-12-Gene-Gene-Interaction-and-Epistasis_201

CHAPTER12Gene-GeneInteractionandEpistasisDavidM.EvansMRCCentreforCausalAnalysesinTranslationalEpidemiology,DepartmentofSocialMedicine,UniversityofBristol,UKOUTLINEIntroduction198Whatis“Epistasis”?198“Biological”Epistasis199StatisticalEpistasis199Two-LocusQuantitativeTraitModelsIncorporatingEpistaticInteractions201TestforAssociationIncorporatingInteractions203Two-LocusBinaryModelsIncorporatingInteraction206WhyModelEpistasis?206StrategiesforDetectingEpistasisinGenome-WideAssociationStudies207Two-StageStrategiestoDetectEpistasis209OtherSimpleTestsforGene-GeneInteraction209Higher-OrderInteractions210MoreSophisticatedApproachestoModelingandDetectingInteractions211Conclusions211Reference211197AnalysisofComplexDiseaseAssociationStudies
2011ElsevierInc.INTRODUCTIONThereismountingevidencesuggestingthatgene-geneor“epistatic”interactionsplayanimportantroleintheetiologyofcomplextraitsanddiseasesinhumans.StudiesinexperimentalorganismssuchasSaccha-romycescerevisaeandDrosophillamelanogastersuggestthatgene-geneinteractionsoccurcommonly,mayinvolveseveralloci,andmayberesponsibleforappreciableportionsofthetotalphenotypicvariance[1e5].Withtheadventofgenome-wideassociation(GWA)studies,thefieldhasprogressedquicklyfromahandfulofdiscoveriesattheturnof2006,toidentifyinghundredsofcommonvariantsreliablyassociatedwithcomplextraitsanddisease[6].WhileGWAstudieshaveundoubtedlycontributedspectacularlytoidentifyinggeneticvariantsinvolvedincomplexdiseaserisk,thevastmajorityofgenomescanshaveonlyemployedsingle-locustestsofassociationthatdonotexplicitlyconsiderthepossibilityofgene-geneinteractions.Attentionisnowturningtowhetherfittingstatisticalmodelswhichincludeinteractioncomponentsmayproveprofitableingenefinding.Indeed,theexistenceofundetectedepistasishasbeenofferedasonereasonwhy,formanycommondiseases,eventhoughmanyriskdisposinglocihavebeenidentified,themajorityoftheheritabilitystillremainstobeexplained[7,8](foracontraryviewseeHilletal.[9]).Inthischapter,wedefinethedifferingandsometimesconfusingmeaningsofthewordsgene-geneinteractionand“epistasis”.Weshowhowthegeneticvarianceinanysystemcanbedecomposedintosingle-locusandepistaticcomponentsofvariance.Weexplainwhyincludinginteractioneffectsinmodelsofassociationmightbeimportant.Finally,wedescribesomeofthesimplermethods,chal-lengesandissuesinfittingmodelsthatcontaininteractionstoGWAdata.WHATIS“EPISTASIS”?Therearefewtermsinthescientificliteraturethathavegeneratedasmuchwidespreadconfusionastheword“epistasis”.Asweshallsee,thisispartlyduetothehistoryoftheword,buthascertainlynotbeenhelpedbythemanyconflictingandoftenunstateddefinitionsofthetermfoundintheliteraturesinceitfirstappearedin1909.Epistasiscanbelooselydefinedastheinteractionbetweentwoormoregenes.However,whatismeantbytheword“interaction”isnotalwaysclear,andcanvarybetweenbiologists,statisticians,statisticalgeneticistsandepidemiologists[10,11].12.GENE-GENEINTERACTIONANDEPISTASIS198“BIOLOGICAL”EPISTASISTheterm“epistatic”wasfirstcoinedinthegeneticsliteraturebyWilliamBateson[12]todescribeamaskingeffect,wherebyavariantoralleleatonelocusmaskstheexpressionofaphenotypeatanotherlocus.Thissortofdefinitionisanalogoustothetraditionalmeaningoftheword“dominance”fromMendeliangenetics,whichreferstoasituationwhereanallelemaskstheexpressionofotherallelesatthesamelocus.AclassicexampleofabiologicalepistaticinteractionisthecoatcolorinaLabradorretrieverdog[13].ThecoatcolorofLabradorscanbeblack,brownorgolden.Coatcolorisprimarilycontrolledbytwodifferentlocieablack/brownlocus(B/b)andagoldlocus(E/e).Theblackalleleisdominanttothebrownallele,sodogsthatarehetero-zygousatthislocuswillpreferentiallyhaveablackcoatcolor.However,expressionatthislocusalsodependsuponthedog’sgenotypeatthegoldlocus.Dogshomozygousfortherecessive“e”alleleatthegoldlocushaveagoldencoatcolorregardlessoftheirgenotypeattheblack/brownlocus.Inotherwords,thegoldlocus“masks”orisepistatictotheeffectattheblack/brownlocus.Infact,theexpectedratioofamatingbetweentwodogsthatareheterozygousatbothlociis9:4:3(black:golden:brown).STATISTICALEPISTASISInthecaseofquantitativetraits,epistasisreferstoadeviationfromadditivityintheeffectofallelesatdifferentlociwithrespecttotheircontributiontothequantitativephenotype.Inotherwords,itdescribesthesituationwherethecombinedeffectoftwoormorelocicannotbepredictedfromthesumoftheirindividualsingle-locuseffects.Thisdefinitionofepistasis,whichisoftenemployedbystatisticalgeneticists,wasfirstusedbyRonaldFisherinhisclassic1918paper“Thecorre-lationbetweenrelativesonthesuppositionofMendelianinheritance”,whichunifiedthedisparateGaltonianandMendelianschoolsofgenetics[14].Thetotalgeneticvarianceofaquantitativetraitcanbepartitionedintocomponentsofvarianceduetosingle-locuseffectsandgene-gene(“epistatic”)interactions.Considerasinglebi-alleliclocuswithallelesdenotedXandx.Letthemeantraitvalueofindividualswithgeno-typesXX,XxandxxbemXX,mXxandmxx,respectively.Withoutlossofgeneralityweassumethatthesegenotypicmeansarescaledasdevia-tionsfromthetraitgrandmean(i.e.,theweightedaverageofmXX,mXxandmxx).ThedegreetowhichmXxdeviatesf