* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Epistatic networks jointly influence phenotypes related to
Survey
Document related concepts
Transcript
bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Epistaticnetworksjointlyinfluencephenotypesrelatedtometabolicdiseaseandgene expressioninDiversityOutbredmice AnnaL.Tyler,BoJi,DanielM.Gatti,StevenC.Munger,GaryA.Churchill,KarenL.Svenson,GregoryW. Carter* [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] *Correspondingauthor:GregoryW.Carter TheJacksonLaboratory,600MainSt,BarHarbor,ME,USA bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 Abstract Multiplegeneticandenvironmentalfactorscontributetometabolicdisease,witheffectsthatrange acrossmolecular,organ,andwhole-organismlevels.Dissectingthismulti-scalecomplexityrequires systemsgeneticsapproachestoinferpolygenicnetworksthatinfluencegeneexpression,serum biomarkers,andphysiologicalmeasures.Inrecentyears,multi-parentmodelorganismcrosses,such astheDiversityOutbred(DO)mice,haveemergedasapowerfulplatformforsuchsystems approaches.TheDOmiceharborextensivephenotypicandgeneticdiversity,allowingfordetection ofmultiplequantitativetraitloci(QTL)andtheirinteractionsathighgenomicresolution.Inthis study,weused474DOmicetomodelgeneticinteractionsinfluencinghepatictranscriptome expressionandphysiologicaltraitsrelatedtometabolicdisease.Bodycomposition,serum biomarker,andlivertranscriptomedatafrommicefedeitherahigh-fatorstandardchowdietwere combinedandsimultaneouslymodeled.Modulesofco-expressedtranscriptswereidentifiedwith weightedgeneco-expressionnetworkanalysis,withsummarymodulephenotypesrepresenting coordinatedtranscriptionalprogramslinkedtospecificbiologicalfunctions.Wethenusedthe CombinedAnalysisofPleiotropyandEpistasis(CAPE)tosimultaneouslydetectdirectedepistatic interactionsbetweenhaplotype-specificQTLfortranscriptmodulesandphysiologicalphenotypes. Bycombininginformationacrossmultiplephenotypiclevels,weidentifiednetworksofQTLwith numerousinteractionsthatrevealhowgeneticarchitectureaffectsmetabolictraitsatmultiplescales. Specifically,thesenetworksmodelhowgeneregulatoryprogramsfromdifferentinbredfounder strainsinfluencemorecomplexphysiologicaltraits.Byconnectingthreelevelsoftheorganismal hierarchy–geneticvariation,transcriptabundance,andphysiology–werevealedadetailedpicture ofgeneticinteractionsinfluencingcomplextraitsthroughdifferentialgeneexpression. 50 Introduction 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 Traitsrelevanttometabolicdisease,suchasobesity,andbloodlipidprofiles,havecomplexgenetic architecture(Schork1997).Manygeneticfactorscontributetothesetraitsandpotentiallyinteractto influencemultipletraitssimultaneously.Identifyingthesegenesandtheirinteractionswillplaya criticalroleinpredictingindividualsusceptibilitytometabolicdiseaseandprioritizingdrugtargets fortargetedtreatments(MooreandWilliams2009).However,despiteavailabilityoflarge-scale genotypeandphenotypedatainmultiplehumanpopulations,littleisknownaboutthegenetic architectureofmetabolicdisease-relatedtraits. Thereareanumberofchallengesassociatedwithmappingthegeneticarchitectureofcomplextraits inhumanpopulations.IncontrasttoMendeliantraits,inwhichasinglegeneticvariantisresponsible forthevastmajorityofphenotypicvariation,complextraitsareinfluencedbymanyvariantswith smalleffects,whicharedifficulttodetect.Largevariationinenvironmentalexposuresbetween individualscaneasilyoverwhelmsmallgeneticeffects,compoundingtheproblem.Human populations,moreover,haveintricatepopulationstructure(Rosenbergetal.2002)whichcancause spuriousassociationsingeneticmappingexperiments(Pritchardetal.2000).Detectinggenetic interactions,orepistasis,inhumansraisesadditionalchallenges.Epistaticinteractionstendtobe weakerthanmaineffectsandcangenerateadditivegeneticvariance(HuangandMackay2016),and variationinallelefrequenciesbetweenpopulationsmakesreplicationoftrueinteractionsbetween populationsdifficult(Greeneetal.2009). Highlydiversemulti-parentpopulations,suchastheDiversityOutbred(DO)mice(Svensonetal. 2012)[@Gatti]offerapowerfulalternativetohumanpopulationsformappingthegenetic architectureofcomplextraits.Asanoutbredpopulation,theDOmicearepotentiallyabettermodel ofhumanpopulationsthaninbredmice.BecausetheDOfoundersincludedthreestrainsrecently derivedfromwildmice,thepopulationcontainsextensiveallelicvariationthatisevenlydistributed bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 acrossthegenome(Philipetal.2011;Svensonetal.2012;Loganetal.2013).Thisdensityof polymorphismsallowsmuchmoreextensivemappingthancanbedoneintypicalcrossesbetween inbredstrains,whichcansharelargeregionsofidenticalsequence(Yangetal.2011).Furthermore, thebreedingparadigmintheDOisdesignedtomaintainallelicdiversity,reducelinkage disequilibrium,andgenerateminimalpopulationstructure(Svensonetal.2012;Chesleretal.2016). Thusvariationinallelefrequencydoesnotconfounddetectionofvarianteffectsorepistasisasit doesinhumanpopulations,andeffectscanbemappedtorelativelynarrowgenomicloci,whichwill enhancethediscoveryofgeneticinfluencesonphenotype. Alargenumberoftraits,includingmanyclinicallyrelevanttraits,havebeenmeasuredinDOmice (Svensonetal.2012;Bogueetal.2015)[and@Gatti].Whileheritable,fewofthesetraitshavea singleQTLofexceptionaleffect[@Gatti].TheDOmicethusprovideanidealplatformfor investigatingthegeneticarchitectureofcomplextraits.Theirphenotypicdiversitycombinedwith extensivegeneticvariationthatisevenlydistributedandhighlyrecombinedfacilitatesdetectionof bothgeneticmaineffectsandinteractionsinfluencingmanyclinicallyrelevanttraits. Inthisstudyweusecombinedanalysisofpleiotropyandepistasis(CAPE){tyler2013cape}to investigatethegeneticarchitectureofmultiplecomplextraitsrelatedtometabolicdiseasein474 maleandfemaleDOmicefedeitherahigh-fatorstandardchowdiet.Specifically,weanalyzed epistasisinfluencingfatmass,leanmass,andcirculatinglevelsofcholesterol,triglycerides,and leptin,aswellasthreegeneexpressionphenotypes.CAPEisanapproachthatcombinesinformation acrossmultiplephenotypestoinferdirectedgeneticinteractions.Itinfersasinglemodelformultiple quantitativetraits,andleveragesstatisticalpowerfrommultiplephenotypestoenhancethe detectionofQTLandtheirinteractions.Withthisapproach,werecentlyanalyzedthegenetic architectureofbodycompositionandbonedensityinawell-poweredF2mouseintercross(Tyleret al.2016)thatrevealedalargenetworkofweakinteractionsthatgenerallyreducedphenotypic variationacrossthepopulation.Hereweapplytheprinciplesofthisanalysistoinvestigatethe contributionsofwithin-strainandbetween-strainepistaticinteractionsintheDO,augmentedby interactiverolesofsexandhigh-fatdietinthenetwork. 105 Results 106 Transcriptswithtransgeneticeffectsclusterintofunctionallyenrichedmodules 107 108 109 110 111 112 113 114 115 116 117 118 119 Becausewewereinterestedingeneticinteractionsthatinfluenceexpressiontraits,whichmust includeatleastonetranseffect,wefirstfilteredthelivertranscriptometo3635transcriptsthatwere influencedby transgeneticloci(Methods).Weperformedweightedgenecorrelationnetwork analysis(WGCNA)(LangfelderandHorvath2008)onthesetranscriptsandobtained11distinct modules.UsingtheDatabaseforAnnotation,VisualizationandIntegratedDiscovery(DAVID)(Huang etal.2009a;b)wefoundthatthreeofthesemoduleshadsignificantlyenrichedfunctional annotations(Benjamini-adjustedp≤0.05):(1)cellularmetabolicprocess(MetabolismModule)(p =6.3x10-17),(2)oxidationreductionprocess(RedoxModule)(p=7.7x10-7),and(3)immune response(ImmuneModule)(p=5.2x10-15)(Table1).Weusedthemoduleeigengenesfromthese modulesasphenotypesforCAPEanalysis(Methods)(Ghazalpouretal.2006;Philipetal.2014).We refertothemhereafterbytheirfunctionalannotations. 120 PleiotropicQTLinfluencephysiologicalandexpressiontraits 121 122 123 Wecombinedthemoduleeigengenesdescribedabovewithfivephysiologicaltraits:leantissuemass, fattissuemass,aswellascholesterol,leptin,andtriglyceridelevels.Fatmasswaslog-transformedto reproduceamorelinearrelationshipwithleanmass(Forbes1987).Thesetraitsweremodestly bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 correlated(Figure1),implyingthatsomegeneticfactorsmaybesharedamongthetraits,while othersmaybedistinct.Todeterminewhetherthetraitsherewereinfluencedbothbypleiotropicloci andlocispecifictoindividualtraits,weperformedlinearregressiontoassociatethehaplotypeat eachlocuswitheachofoureightphenotypes(Methods).Acrossalltraits,onlyoneQTLfor cholesterolondistalChr1reachedgenome-widesignificance(permutation-basedp<0.05). However,thereweremultiplelociwhereindividualhaplotypeshadsubstantialeffectsthat potentiallycontributetopolygenicetiology(Figure2).Insomecases,asinglehaplotypehadan apparenteffectonasinglephenotype.Forexample,apositiveeffectoftheNZOhaplotypeon cholesterolcanbeseenondistalChr11(Figure2).Likewise,theA/Jhaplotypeatanearbylocushad apositiveeffectonleptinlevels(Figure2).Otherlociwerepleiotropic.TheCASThaplotypeatathird locusonchromosome11hadnegativeeffectsonfatmass,cholesterol,leptin,triglycerides,andthe ImmuneModule(Figure2).ThiseffectwassharedtoalesserextentbythePWKhaplotypeinfat mass,leptin,andtriglyceridelevels.Thiscomplexpatternofeffectssuggestsacomplexunderlying geneticarchitecture.Thehaplotypeeffectsthatarecommonacrossmultiplephenotypesmay representacommongeneticfactorinfluencingmultipletraits.Wecombinedthesecommonsignalsto gaininformationaboutindividualloci.Haplotypesthatinfluenceasinglephenotype,forexamplethe NZOhaplotypeeffectoncholesterol,providenon-redundantinformationthatcanbeusedtoidentify geneticfactorswithspecificphenotypiceffects. 143 Singularvaluedecompositionconcentratesfunctionalgeneticeffects 144 145 146 147 148 149 150 Wedecomposedthetraitmatrixusingsingularvaluedecompositiontoobtaineigentraits(ETs) (Figure3A).InouranalysisweusedthefirstthreeETs,whichcaptured88.3%oftheoverall variance.ETsrecombinecovaryingelementsofthemeasuredtraits,andpotentiallyconcentrate functionallyrelatedeffects.Forexample,leptin,cholesterol,andfatmass,alongwiththeRedoxand ImmuneModules,wereaveragedinET2.ThisETappearstocapturetheCAST/PWKeffectonChr11 notedearlytoinfluencemultipletraits.(Figure3B). 151 Anepistaticnetworkinvolvingallhaplotypesinfluencesphysiologicalandexpressiontraits 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 Becausethereweremoremarkersgenotypedthancouldbetestedexhaustivelyinpairs,weuseda subsetofhaplotypeswiththegreatesteffect-sizesinallthreeETs(seeMethods).Thehaplotypewith greateststandardizedeffectfromeachpotentialQTLpeakwasretainedandthepeakwasfurther sampledtokeep10%ofmarkerswithinit.Thisprocessyieldedatotalof515markersrepresenting allsevenhaplotypeson17chromosomes(TheC57BL/6Jhaplotypewasexcludedbecauseweusedit asthereferencestrain).Becausemarkerselectionwasbasedoneffectsize,thehaplotypeswere unevenlyrepresented(Figure5A).A/Jwasthemosthighlyrepresentedhaplotypewith100markers oneightchromosomes,andNODwastheleastrepresentedwith32.WSBalleleswerethemost widelydistributed,beingselectedfrom12differentchromosomes.WeranCAPEonthesemarkers andthefirstthreeETstofindanepistaticnetworkbetweenloci(Methods). Theresultingnetworkconsistedof89interactionsamong49lociandtwocovariates(Figure4).All haplotypesparticipatedinatleastoneinteraction(Figure5A).WSBhaplotypeswereinvolvedinthe largestnumberofinteractions(32),whileNZOparticipatedinthefewest(8).Thenumberoftotal interactionseachhaplotypeparticipatedindidnotcorrelatewithitsrepresentationinthe515 markersselectedfortheCAPEpipeline(Figure5A)(p=0.1).Thefinalepistaticnetworkwas directed,meaningthatinteractionsmodelasourcemarkerthatactsonatargetmarker,andwecan thusmeasurethenumberoftimeseachhaplotypewasthesourceofaninteractionorthetargetofan interaction.Themajorityofhaplotypeswereroughlyevenlyrepresentedasbothsourcesandtargets. However,the129haplotypewasatargetofinteractionsaboutfourtimesmorefrequentlythanit wasasource,whiletheNZOhaplotypewasasourceabouttwiceasmanytimesasitwasatarget bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 173 174 175 176 177 178 179 180 181 182 183 184 (Figure5A).Thecovariates,sexanddietwerebothmuchmorefrequentlysourcesofinteractions thantheyweretargets(Figure5A). Theinteractionsbetweenhaplotypesweremostoftenbetweenstrainsratherthanwithin-strain (Figure5B).Inter-straininteractionswereconcentratedamongthe129,WSB,NZOandA/J haplotypes,whichareallintheMusmusculusdomesticussubspecies.CAST,M.musculuscastaneus, interactedwitheachoftheotherstrainsrelativelyevenly,whilePWK,M.musculusmusculus,wasthe mostisolatedstrain,anddidnotinteractatallwiththeNZOorNODhaplotypes.Theonlyhaplotype withmultipleintra-straininteractionswasWSB.Thismaybeduetothewidesamplingofthe selectedWSBallelesfrom12differentchromosomesresultinginmoreuniquelociwithpotentialfor interactingwitheachother. 185 Sexinteractedwithallfounderhaplotypes 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 Sexsignificantlyaffectedallphysiologicaltraitsexceptleptinlevels.Thiseffectwaspositiveforall phenotypesmeaningthatmaleshadhigherlogfatmass(males1.9g,females:1.7g,p=5.7x10-2), leanmass(males:25.1g,females:18.3g,p<2x10-16),cholesterol(males110.4mg/dl,females:93.8 mg/dl,p=4.3x10-10),andtriglycerides(males:156.0mg/dl,females:115.0mg/dl,p=7.6x10-14).All expressionmodulesweresignificantlylowerinmales(allp<2x10-16).Sexalsoparticipatedin interactionswithgeneticloci.Themajorityofgeneticinteractionswithsex(12of15)involveda suppressionofalleleeffectsbysex,indicatingthatthealleleshadlargereffectsinfemalesthanin males.Allelesfromallfounderstrainswereaffected.Onelocus,theCASTalleleonChr11,enhanced theeffectsofsex.Thealleleoverallhadnegativeeffectsonleptin,cholesterol,andleanmass,butin males,thesemeasureswerehigherinthepresenceofthisallelethanexpectedfromtheadditive model.Therewasalsoasinglelocus,theWSBalleleonChr17,thatsuppressedtheeffectsofsex, indicatingthatmalescarryingthisallelehadlowerthanexpectedleanmass,fatmass,etc.For example,boththisalleleandsexhadpositiveeffectsoncholesterol,butcholesterollevelsinmale micecarryingthisWSBallelewerelowerthanexpectedfromtheadditivemodel.Finally,phenotypic effectsofthemalesexwereenhancedbythehigh-fatdiet,suggestingthatmalesweremore susceptibletotheeffectsofthehigh-fatdiet. 203 Dietinteractedwithasubsetofparentalhaplotypes 204 205 206 207 208 209 210 211 212 213 214 Dietsignificantlyincreasedlogfatmass(chow:1.6g,HF:2.1g,p<2x10-16),cholesterol(chow:85.8 mg/dl,HF:119.1mg/dl,p<2x10-16),andleptin(chow:7.7mg/dl,HF:19.7mg/dl,p<2x10-16)and significantlydecreasedtriglyceridelevels(chow:146.7mg/dl,HF:124.3mg/dl,p=1x10-4).Italso significantlydecreasedallexpressionmodules(allp<0.001).Similartosex,themajorityofgenetic interactionswithdiet(fiveofseven)werethoseinwhichhigh-fatdietsuppressedgeneticeffects. Thatis,thealleleshadgreaterphenotypiceffectinchow-fedmicethanmiceonthehigh-fatdiet. Therewasonelocus,theCASTalleleonChr2,thatenhancedtheeffectsofdiet,indicatingthat animalscarryingthisalleleweremoresusceptibletotheeffectsofthehigh-fatdiet.Theeffectsofdiet werealsoenhancedbysex,asmentionedabove,indicatingthatmalesinthispopulationweremore susceptibletotheeffectsofthehigh-fatdietthanfemales. 215 Networkmotifshadbothredundantandsynergisticeffectsonphenotypes 216 217 218 219 220 Tobetterunderstandtheoverallinfluenceofgeneticinteractionsontraits,weperformedanetwork motifanalysisasdescribedin(Tyleretal.2016).Networkmotifsarecomposedofoneinteraction betweentwoloci,eachofwhichhasamaineffectononephenotype(Figure6A).Theinteractioncan eitherbesuppressingorenhancing,andthetwomaineffectscandrivethephenotypeeitherinthe samedirection(coherent)orinopposingdirections(incoherent).Hereweinvestigatedtheeffectsof bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 221 222 223 224 225 226 227 228 229 230 231 232 networkmotifsontraitsintheDOandcompareourresultstoourpreviousresultsfromresultsfrom anF2intercrossbetweeninbredstrainsinTyleretal.(2016)(Tyleretal.2016). Onlyenhancing-incoherentandsuppressing-coherentmotifswerepresentintheDOepistatic network(Figure6B).Theyinvolvedallparentalhaplotypesandwerepredominantlyinteractions betweenhaplotypesfromdifferentparents(enhancing-incoherent:72%differentparental haplotypes,suppressing-coherent:96%differentparentalhaplotypes).Incontrasttotheintercross, theenhancing-incoherentmotifswerenotpredominantlybalancing,buttendedtodrivetraitsaway fromthepopulationmean.Thevastmajorityofthesemotifs(92%)hadadestabilizingeffecton phenotype,and80%drovethephenotypepastanyadditiveprediction(Figure7).Asubstantial fractionofthesuppressing-coherentmotifs(25%)werenon-redundant,meaningtheypushed phenotypesfartherfromthepopulationmeanthanpredictedbytheadditivemodel(Figure7). 233 Discussion 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 Traitsassociatedwithmetabolicdisease,suchascholesterollevels,bodyfatmass,andtriglyceride levelshavecomplexgeneticarchitecture.Mappinggenesinfluencingthesetraitswillhelpidentify mechanisticfactorsinfluencingthemand,togetherwithmolecularbiomarkers,mayultimately providetargetsfortherapies.Mappingcomplexgeneticeffects,however,ischallenging,especiallyin humanpopulationsinwhichenvironmentalfactorsandpopulationstructurecanoverwhelmweak geneticeffects.Miceofferanexcellentalternativeaspre-clinicalmodelorganismsinwhichtodissect complextraitsmechanistically.However,themajorityofinbredstrainsusedinmedicalresearchare closelyrelatedtoeachother.Thehavelimitedphenotypicdiversityandlargegeneticblindspotsdue toalackofgeneticvariantsbetweenthem.TheDOmiceprovideapowerfulalternativeplatformfor fine-mappingcomplextraits.Theyharborimmensegeneticandphenotypicdiversity,andhave minimalpopulationstructure,therebyallowingmuchmoredetailedassessmentsofcomplexgenetic architectureinfluencingcomplextraits.ThegeneticdiversityintheDOdoescreateitsownissues, however,inthatlargegeneticeffectscanbedifficulttofind.Usingstandardmappingmethods,we andothershaveshownthatmosttraitsareinfluencedbymanyQTLwithsmalleffects(@gatti),and fewQTLrisetogenome-widesignificance.HereweusedCombinedAnalysisofPleiotropyand Epistasis(CAPE)tocombinemulti-dimensionalphenotypeinformationandtestforgenetic interactionsinfluencingasuiteofrelatedtraits.Wefoundnumerousindividualeffectsandan epistaticinteractionnetworkinfluencingbothphysiologicalandexpressiontraits.Theinteraction effects,whichtendedtobeweak,identifiedgeneticelementsthatpotentiallyinfluencethetraitsand informedonthegeneralgeneticarchitectureofthesetraits. Thetwofactorswiththelargestinfluenceonmostphenotypesweresexanddiet.Sexinfluencedall traitsexceptserumleptinlevels.Inournetwork,sexalsointeractedwith14geneticloci.Although multiplesex-specificQTLhavebeenmappedinhumans(Weissetal.2006;Oberetal.2008),the studiesareoftenoflowpowerandfewindividualresultshavebeenreplicated(Oberetal.2008).The DOmiceofferapowerfulplatformtoinvestigatetheroleofsexincomplextraitsinmammalian systems.Inourstudy,themajorityofthegeneticinteractionswithsexwereasuppressionofallele effectsinmales.Theseallelesmayhelpidentifyimportantriskandprotectiveallelesformetabolic diseaseinfemales.Forexample,the129alleleatalocusonChr19hadpositiveeffectsonthe MetabolismExpressionModuleandtriglyceridelevels,suggestingthatthislocuscontainsagenethat increasestriglyceridesthroughgeneexpressiondifferencesinmetabolicpathways.Theeffectsofthe 129alleleweresuppressedbysex,indicatingthatithadalargereffectinfemalesthanmales. CombiningthealleleandinteractioninformationfromtheCAPEnetwork,wecangeneratea hypothesisaboutthecausalgeneinthislocus.Therearesixgenesknowntoinfluencetriglyceridesin theChr19.4locusandoneofthese,Sorbs1,hasacis129-specificeffectincreasingSorbs1expression bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 (Figure8A).Sorbs1isfurthermoreexpressedmorehighlyinfemales(p=0.002)(Figure8B),andis significantlycorrelatedwithtriglyceridelevelsintheDOmice(r2=0.17,p<2x10-16).Previouswork hasshownthatmicewithhomozygousdeletionsofthisgenehavereducedtriglyceridelevels (Lesniewskietal.2007).Increasedexpressionduetothegain-of-function129alleleisconsistent withincreasedtriglyceridesincarriers,andthereforethe129alleleofthisgenemayincreaseriskfor elevatedtriglyceridelevelsinfemalemice. Inadditiontosex,dietisanimportantfactorindeterminingriskofmetabolicdiseaseandtheir relatedphenotypes.Thehigh-fatdietinourstudyhadasubstantialimpactonalltraitsexceptthe MetabolismModule.High-fatdietenhancedtheeffectsofsexindicatingthatmalesintheDO populationweremoresusceptibletotheeffectsofthedietthanfemales.Ithasbeenshownthat inbredmaleB6micegainmoreweightandhavehigherbloodlipidprofileswhengivenahigh-fatdiet (Hwangetal.2010).AndalthoughnotrepresentedintheDO,maleBALB/cAmicehavealsobeen showntobemoresusceptiblethanfemalestoweightgainandhepaticlipidaccumulation(Nishikawa etal.2007).Dietinteractedwithanumberofgeneticloci,andlikesex,mostlysuppressedtheeffects oftheseloci,indicatingthatthealleleshadalargereffectinanimalsfedstandardchow.Multiple studieshaveshowninteractionsbetweengenesanddietininfluencingfactorsrelatedtotraits associatedwithmetabolicdisease(forreviewsee(Ordovas2006)).TheresolutionintheDOgenome combinedwithinformationaboutgeneticinteractionswillhelpspeedidentificationofgenes interactingwithdietandhelpelucidatehowhigh-fat,high-sucrosedietsleadtoobesityand metabolicdisease,aswellashowhealthydietshelppreventtheseconditions. Inadditiontotheinteractionswithsexanddiet,geneticlocialsointeractedwitheachotherto influencephenotypesinnetworkmotifs.InapreviousstudyofanF2intercross(Tyleretal.2016),we foundthatsuppressing-coherentandenhancing-incoherentmotifsweresignificantlyenrichedinthe epistaticnetwork.InthisF2population,bothtypesofmotifstendedtohavemoderatingeffectson phenotypes. Thesuppressing-coherentnetworkmotifstendedtoreflectredundancy,whiletheenhancingincoherentinteractionshadabalancingphenotypiceffectdrivingphenotypestowardinbredstrain means(Tyleretal.2016).Animalshomozygousforoneparentalalleleatbothinteractinglocihad lessextremephenotypesthanthosewithamixofparentalallelesatthetwoloci(Tyleretal.2016). Similartoourpreviousfindings,networkmotifsintheDOwerepredominantlyenhancingincoherentorsuppressing-coherent(Figure6B).However,incontrasttotheintercross,the enhancing-incoherentmotifsfrequentlydrovetraitsfartherfromthepopulationmeanthan predictedbytheadditivemodel.Themajorityofthesuppressing-coherentmotifshadredundant effects,i.e.thetwolocihadlessthanadditiveeffects,butasubstantialfraction(36%)also destabilizedphenotypes,drivingthemawayfromthepopulationmean. Thisphenotypicdestabilizationislikelyduetothedifferenceinalleliccombinationsbetweenthe multi-parentDOmiceandaclassicintercrossdesign.Inanintercrossallinteractionsbydefinition arebetweenallelesfromasinglenon-referenceparent,whereasinteractionsintheDOweremost frequentlybetweenallelesfromdifferentparentalancestries.Inbothdesigns,eachoftheparental strainshasdevelopeditsownuniquesetofallelestomaintainquantitativetraitsatstrain homeostasis.Inanintercross,accumulationofallelesfromasingleparentalstrainmay combinatoriallyachievehomeostaticphenotypesforthatparent.Bycontrast,intheDOthemixingof parentalallelesmayinsteaddestabilizephenotypesbydrivingthemtoextremesandcreatingthe immensephenotypicdiversityseeninthispopulation.Furthermore,thatweseemoredestabilizing interactionsamongtheenhancing-incoherentmotifsmayimplysomethingaboutmolecularpathway structure.Wehypothesizethatsuppressing-coherentmotifsrepresentinteractionsbetweengenes withinasinglepathway,whileenhancing-incoherentmotifsrepresentinteractionsbetweengenesin different,butfunctionallyrelatedpathways.Thisisconsistentwithearlierworkonperturbationsof bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 fruitflysignalingpathways(Hornetal.2011;Carter2013).Thepatternsofstabilizingand destabilizingmotifsinourstudysuggeststhatrecombiningparentalalleleswithinpathwaysiswell toleratedandoftenredundant,whilerecombinationbetweenrelatedpathwaysmorefrequently destabilizesphenotypes. AlthoughthegeneticdiversityintheDOallowsrelativelyfinemapping,wecannotdefinitively identifywhichgenesintheselociareresponsibleforthephenotypiceffects.Wecan,however, combinetheinformationinepistaticinteractionswithestimatedfunctionalinteractionstogenerate hypothesesaboutcausalgenes.Forexample,wefoundaninteractionbetweentheA/Jhaplotypeon Chr9locus2(Chr9.2:5Mbto36Mb)andtheCASThaplotypeonChr2locus2(Chr2.2:123Mbto 133Mb)thatinfluencedtheImmuneModule.EachlocushadanegativemaineffectontheImmune Module,andtheircombinedeffectwasredundantwiththeeffectoftheChr2.2locus(Figure9A). Thispatternofeffectsindicatesaredundantinteractionandthepossibilitythatthecausalgeneson thetwolocioperateinthesamepathway.Tofurtherinvestigatethispremise,weidentifiedallthe genesinthetworegionsthathadstrain-specificpolymorphisms(A/JonChr9.2andCASTonChr 2.2),andfilteredthesetoincludegenesthathadbeenpreviouslyannotatedtothemammalian phenotype(MP)term“immunephenotype”(seeMethods).WethenusedIntegrativeMulti-species Prediction(IMP)(Wongetal.2015)toidentifythemostlikelyamongthesegenestointeract functionally.ThisfilteringprocessidentifiedCasp4onChr9.2andIl1bonChr2.2asthemostlikely genesinthesetwolocitointeract.IntheIMPnetwork,thetwogenesinteracteddirectlyinanetwork functionallyenrichedforcytokineproductionandsecretion(p=4.3x10-12)(Motenkoetal.2015) (Figure9B).InsupportofthehypothesisthatCasp4andIl1binteract,bothtranscriptsarecorrelated withtheImmuneModule(Figure9C,Casp4:r2=0.48,p=2.6x10-28,Il1b:r2=0.49,p=1x10-30),and witheachother(r2=0.32,p=7.4x10-13)(Figure8C).Casp4,alsoknownasCasp-11,isamemberof thecysteine-asparticacidproteasefamilyandisessentialforIL1Bsecretion.Micewithhomozygous mutationsofCasp4havedecreasedlevelsofcirculatingIL1B(Wangetal.1998).ThatCasp4is directlyinvolvedinIL1Bsecretionisconsistentwiththeredundantgeneticinteractionweobserved betweenChr9.2andChr2.2intheCAPEnetwork.Redundantinteractionsarehypothesizedtooccur betweenvariantsencodinggeneswithinasinglepathway(AveryandWasserman1992;Lehner 2011).Eachvarianthasasimilareffectonthepathway,butbecausethepathwaycanonlybe disruptedonce,thecombinationofthetwovariantsdidnothaveafurthereffectdespitebeingfrom differentparentalstrains.Suchcombinatorial,polygeniccandidategeneswererevealedbyour geneticinteractionanalysisthatidentifiedredundantgeneticeffects. Elsewhereinthenetwork,wehypothesizethatgenesinteractinginenhancing-incoherentnetwork motifsfunctionindistinctpathwaysthatneverthelessinfluenceeachother.Inadditiontothe redundantinteractionabove,weprioritizedinteractinggenesinasecondinteractionbetweenthe sameA/JhaplotypeonChr9.2anotherQTLonChr2.Thissecondlocus,Chr2locus4(Chr2.4:165 Mbto171Mb)representedaneffectoftheNODhaplotypeanddidnotoverlaptheCASTQTL(123 Mbto133Mb)thatalsointeractedwiththeChr9.2A/JQTL.ThisQTLthusrepresentsadistinct interaction.TheA/JChr9.2andtheNODChr2.4lociinfluencedtheImmuneExpressionModulein oppositedirections,andtogether,theydrovethetraittobeslightlymorenegativethanpredictedby theadditivemodel(Figure10A).Followingthesamegeneselectionpipelinedescribedabove,we identifiedCasp4againfortheChr9.2A/Jlocus,andSrcasalikelyinteractingpartnerintheChr2.4 NODlocus(Figure10B).Transcriptsforbothgenesaresignificantlycorrelatedwiththeimmune expressionmodule(Casp4:r=0.47,p=6.3x10-28;Src:r=0.47,p=3.7x10-27)andwitheachother(r= 0.21,p=3.2x10-6)(Figure10C).IntheIMPnetworkCasp4andSrcoccupytwolobesofaconnected graph,indicatingthattheyarelessdirectlyfunctionallyrelatedthanCasp4andIl1b.TheCasp4sideof thenetworkisenrichedforgenesinvolvedininflammasomepathways(p=2.9x10-6)(Motenkoetal. 2015),whiletheSrcsideofthenetworkisenrichedforEGFRsignaling(p=2.7x10-4)(Motenkoetal. 2015).TheIL-1andEGFfamiliesofproteinsareupregulatedinhumankeratinocytesduringwound bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 healingandinpsoriasis,andtheyhavebeenshowntointeractsynergisticallyinupregulating transcriptsinvolvedinantimicrobialdefenses(Johnstonetal.2011).Conversely,inhibitingEGFR signalinginkeratinocytesreducestheirIL-1secretioninresponsetoStaphylococcusaureusinfection (Simanskietal.2016).Insum,theseobservationssuggestthattheA/JalleleofCasp4andtheNOD alleleofSrcmayinteracttoinfluenceimmune-relatedexpressioninmice. OuranalysisofgeneticinteractionsinDOmicehasrevealedanumberofinterestingfeaturesofthe geneticarchitectureofcomplextraitsrelatedtometabolicdisease.First,wedetectednumerous significantgeneticinteractionsinfluencingbothphysiologicalandexpressiontraitsinanoutbred population.Althoughtheseeffectsweresmallrelativetothemaineffectsweidentified,wewereable todetectthembycombininginformationacrossmultiplephenotypes.Theinteractionsprimarily involvedallelesfromdifferentparentalhaplotypes.Thispatternindicatesthatmulti-parent populationsmaybemorepowerfulplatformsthanstandardintercrossesfordetectingepistasisdue totheincreasedgeneticdiversity.Interactionsinanintercrossarebydefinitionbetweenallelesfrom thesameparentalstrain,butintheDOinteractionswithinstrainhaplotypesarerelativelyrare.With theadditionalallelicvariationintheDO,moregeneticcombinationswithdiversephenotypiceffects arepresent.Second,wefoundthatnetworkstructureofgeneticinteractionsinoutbredmiceis distinctfromthenetworkstructurewefoundpreviouslyinamouseintercross.Intheintercross interactionsdescribedbynetworkmotifspredominantlyreducevariationintraits,drivingthem towardtheparentalstrainmean.Incontrast,theenhancing-incoherentmotifsintheoutbredmice tendedtodrivetraitsawayfromthepopulationmean.Theextremetraitsweremostfrequently causedbyinteractionsbetweenallelefromdifferentparentalhaplotypes.Extremephenotypesupon recombinationofallelesintheDOmayhavethebenefitofmakingepistasisinoutbredpopulations easiertodetectthanepistasisinintercrossesbetweentwoinbredstrains.Finally,weshowedthatwe canusegeneticinteractionsasinformationtoprioritizecandidategenesingenomicregions. Interactionsbetweentwolociimplyafunctionalrelationshipbetweenelementsencodedinthetwo loci.Bycombininginformationabouthaplotype-specificgeneticinteractionswithgenomicfunctional data,liketheIMPnetwork,wecangenerateplausiblehypothesesregardingcausalgenes.The hypothesesgeneratedbythismethodinthisstudyweresupportedbyexpressiondatanotusedin thehypothesisgeneration.Togethertheseresultsspeaktothevalueofmulti-parentoutbred populationsinthedissectionofthegeneticarchitectureofclinicallyrelevantcomplextraits. 402 Methods 403 Mice 404 405 406 407 408 409 410 411 412 MicewereobtainedfromTheJacksonLaboratory(BarHarbor,ME)asdescribedin(Svensonetal. 2012)and@Gatti.Theanimalswerenon-siblingDOmicerangingfromgeneration4to11,andmales andfemaleswererepresentedequally.AllanimalprocedureswereapprovedbytheAnimalCareand UseCommitteeatTheJacksonLaboratory(AnimalUseSummary#06006).Micewerehousein same-sexcageswithfiveanimalspercageasdescribedin(Svensonetal.2012)[email protected] hadfreeaccesstoeitherstandardrodentchow(6%fatbyweight,LabDiet5K52,LabDiet,Scott Distributing,Hudson,NH)orahigh-fat,high-sucrosediet(HFD)(EnvigoTekladTD.08811,Envigo, Madison,WI)forthedurationofthestudyprotocol(26weeks).CaloriccontentoftheHFDwas45% fat,40%carbohydratesand15%protein.Dietswereassignedrandomly. 413 Phenotype Measurements 414 415 416 417 418 Phenotypesweremeasuredasdescribedin(Svensonetal.2012)[email protected] weeksofage,bloodwascollectedretro-orbitallyafteradministrationoflocalanesthetic.Cholesterol andtriglyceridesweremeasuredusingtheBeckmanSynchronDXC600ProClinicalchemistry analyzer.Leptinwasmeasuredinnon-fastedplasmapreparedaspreviouslydescribed(Svensonet al.2012).LevelswereanalyzedusingtheMesoScaleDiscoveryelectrochemiluminescentsystem bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 419 420 421 422 423 accordingtothemanufacturer'srecommendedprotocol(MesoScaleDiagnostics,Rockville,MD). Bodycomposition(leanmassandtotalmass)weremeasuredatage12weeksusingdualX-ray absorptiometry(DEXA)usingaLunarPIXImusdensitometer(GEMedicalSystems).Fatmasswas calculatedaslog(totalmass-leanmass).Measurementswereperformedattwotimepoints.All measurementsinthisstudyweretakenfromthefirsttimepoint. 424 Genetic analysis 425 426 427 428 429 Genotypingwasperformedontailbiopsiesasdescribedin(Svensonetal.2012)usingtheMouse UniversalGenotypingArray(MUGA).Asubsetoftheanimals(293)weregenotypedonthe Megamuga(GeneSeek,Lincoln,NE).Theintensitiesfromthearrayswereusedtoinferthehaplotype blocksineachDOgenomeusingahiddenMarkovmodel(HMM)(Gattietal.2014b). 430 Merging Haplotype Reconstructions from Different Methods 431 432 433 434 435 436 437 438 439 440 441 442 443 444 Genotypes were measured with the MUGA (7,854 markers), Megamuga (77,642 markers) and by GBRS, which is a set of software tools that uses RNA-Seq data to reconstruct individual sample genomes in multiparental population (MPP) (@Gatti, and http://https://github.com/churchilllab/gbrs).Tomergediplotypeprobabilitiesfromallsources,weinterpolatedmarkersonanevenly spaced64,000markergrid(0.0238cMbetweenmarkers). Transcriptomeprofiling Transcriptome-wideexpressionlevelsweremeasuredasdescribedin(Chicketal.2016),(Mungeret al.2014)and@Gatti.TotalliverRNAwasisolatedfromeachmouseandsequencedusingsingle-end RNA-Seq(Mungeretal.2014).Transcriptswerealignedtostrain-specificgenomesfromtheDO founders(Chicketal.2016).Weusedanexpectationmaximizationalgorithm(EMASE, https://github.com/churchill-lab/emase)toestimatereadcounts.Readcountsineachsamplewere normalizedusingupper-quantilenormalizationandarankZtransformationwasappliedacross samples. 445 Filteringtranscriptsfortranseffects 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 Wewereinterestedinmappingeffectstotranscriptsthatwereinfluencedbydistant(trans)genetic loci.Todeterminewhichtranscriptshadtransloci,wefirstusedDOQTL(Gattietal.2014a)tomap QTLforalltranscriptsexpressedinatleast50samples(26,875transcripts).DOQTLeffectsusing founderallelehaplotypeprobabilitiescalculatedasdescribedin(Gattietal.2014b).Inaddition,we usedsex,dietandbatchasadditivecovariatesandusedhierarchicallinearmodelstocorrectfor geneticrelatedness(Kangetal.2008). Fromthismappingweidentifiedcis-eQTLsfortranscripts,whichwedefinedasasuggestiveeQTL (LOD>=7.4)within2Mbpoftheencodinggene’stranscriptionstartsite.Foreachtranscript,we regressedouttheeffectsofthecis-eQTL(Pierceetal.2014)andre-mappedQTLusingDOQTL.We identified3635trans-eQTLsdefinedasaQTL(LOD>=7.4)onachromosomeotherthanthe transcriptsencodinggeneoratleast10Mbawayonthesamechromosome.Additionally,forthe followingclusteringanalysis,weusedtheresidualexpressionbyremovingtheeffectsofcishaplotypeandbatcheffectvialinearregression.TheprocedureisoutlinedinSupplementaryFigure 1. WeightedGeneCo-expressionNetworkAnalysis Co-expressiongenemoduleswereobtainedbyclusteringtrans-actingtranscriptsusingtheWGCNA packageinR(LangfelderandHorvath2008;undefinedauthor2016).WGCNAcomputestheabsolute valueofPearsoncorrelationforallgenepairsandgeneratesanadjacencymatrixbyraisingthe bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 correlationmatrixtoauser-definedpower.Wesetthepowertosixtoachieveanetworkwithscalefreedegreedistribution.Toconstructthemodulenetwork,WGCNAuseshierarchicalclusteringto produceadendrogramofgenes.Individualbranchesofthedendrogramrepresentmodules,which areclustersofhighlyco-expressedgenes.Themoduleswithsimilarexpressionprofilescanbe mergedbasedontheircorrelation.Wesettheminimummodulesizeto30andtheminimumheight formergingto0.25(correspondingtoaPearsoncorrelationof0.75)toobtainrelativelylargeand distinctmodules.Thefirstprincipalcomponentforeachmodule(termedeigengenesinWGCNA)is usedtorepresentthesummaryco-expressionpatternforeachmodule.Theseeigengenesare hereafterreferredtoasmodulephenotypesforCAPEanalysis.Eachmodulewasassessedfor functionalenrichmentusingtheDAVIDdatabase(Huangetal.2009a;b).TheGOenrichment significancethresholdforallgeneontologyenrichmentanalyseswasp≤0.05,withBenjamini correctionformultiplecomparisons. Combinedanalysisofpleiotropyandepistasis Combinedanalysisofpleiotropyandepistasis(CAPE)isamethodforderivinggeneticinteraction networksofgeneticvariantsthatinfluencemultiplephenotypes{tyler2013cape}.Theopen-sourceR packageofcapewasadapted(below)touseforDOmicewithextensiontomultipleallelesinour analysis. WebeganouranalysisbyregressingoutbreedinggenerationoutofeachtraitandapplyingarankZ transformationtoeachphysiologicaltrait.Thesewerecombinedwiththethreemoduleeigengenes representingsignificantlyenrichedmodulesfromWGCNA(seeabove).Wethenperformedsingular valuedecomposition(SVD)onthetraitmatrixtoobtaineightorthogonaleigentraits(ET’s).TheET’s combinecommonsignalsacrossalltraits.Inthisanalysis,weusedthefirstthreeETs,whichcaptured 88.3%ofthevariationinthetraits.Wethenperformedlinearregressiontoassociateeachmarker witheachET. Foreachmarkerweusedaseven-statemodeltoestimatetheeffectofthefounderhaplotypeson eachtrait.WeusetheB6alleleasthereference,andthusB6allelesarenotexplicitlyincludedinour finalresults.Wealsoincludedtwocovariates,sex(female:0,male:1)anddiet(chow:0,high-fat:1). 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 Theindex𝑖isfrom1tonumberofsamplesand𝑗 isfrom1tonumberofET’s. Pi,aistheprobabilityofeachalleleaatthelocus,andxc,iisthepresenceorabsenceofeachcovariate. Weusedtheresultsofthesingle-locusregressiontoselectmarkersforthelocus-pairregressions. Variantselectionforpairwiseregression Becausethereweremoremarkersgenotypedthancouldbetestedinapairwiseregression,we selectedasubsetofvariantsbasedonstandardizedeffectsize.Weselectedindividualhaplotypes(for exampletheA/Jhaplotypeatmarker1)suchthathaplotypesfrommultiplefounderstrainsand multiplechromosomeswouldberepresentedinthelocus-pairregression.Todothis,wepickedan arbitrarythresholdandidentifiedhaplotypepeaksineffectsizethatroseabovethisthreshold.We pickedthemarkerwiththelargesteffectsizeinthispeakandsampled10%oftheremaining markersinthepeakuniformlyatrandom.Weprogressivelyloweredthethresholduntilwehad sampledapproximately500individualvariants.(SupplementaryTable2)Thefinalnumberof variantsselectedwas515,representingallhaplotypesacross17chromosomes. Pairwiseregression bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 513 Weexpressthefullmodelfortwovariantslabeled1and2as: 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 Theindex𝑖isfrom1tonumberofsamplesandjisfrom1tonumberofET’s. Pi,aistheprobabilityofeachalleleaatthelocus,andxc,iisthepresenceorabsenceofeachcovariate. EijistheETforsample𝑖.P1,iandP2,iaretheprobabilitiesofthealleleateachoftwovariantsfor samplei.P1,iP2,iistheinteractionoftwovariants,𝛽! and 𝛽! aretheeffectsoftwovariantsontheET𝑗, and𝛽!" istheinteractioncoefficient. Foreachmarkerpair,theregressioncoefficientsacrossallET’swerereparametrizedtoobtaintwo newparameters(𝛿! and𝛿! ).The𝛿termsareindependentofphenotypeandcanbedefinedasthe degreetowhichonevariantinfluencestheeffectoftheotheronthephenotypes.𝛿! representsthe inferredgeneticactivityofthefirstvariantwhenthesecondvariantispresent.Anegative𝛿 coefficientindicatesonevariantsuppressinganother.Forexample,anegative𝛿! indicatesthat variant1suppressestheeffectofvariant2onthatphenotype.The𝛿termsarecomputedintermsof coefficientsfrompairwiseregressionasfollows: 530 531 532 533 534 535 Next,the 𝛿 termsaretranslatedintodirectedvariables𝑚!" and𝑚!" ,whichdescribevariant-tovariantinfluencesthatfitallphenotypesviaindirectassociations.Theterm𝑚!" and𝑚!" aredirect influencesofonevariantontheother,withnegativeinfluencesindicatingsuppressionandpositive influencesindicatingenhancement.Theterms𝑚!" and𝑚!" aredefinedintermsof𝛿! and𝛿! : ! ! 𝑚!" = !!!! ,𝑚!" = !!!! 536 537 538 539 540 541 542 543 ! ! Errorsareestimatedthroughstandardleast-squaresregressionandasecond-orderTaylor expansionontheregressionparameters(Carteretal.2012).Wedefinedtheabsolutevalueofthe ratioofanestimatedcoefficientanditsstandarderror(|β/se|)asthestandardizedeffecttoevaluate themaineffectsofthevariantsonthephenotypesandtheinteractiveeffectsofthevariants.The significancethresholdofthestandardizedeffectisdeterminedbasedongenotypepermutationtest andadjustedformultipletesting.Toavoidfalsepositivesduetolinkagedisequilibrium(LD),we excludedvariantpairswithPearson’scorrelationcoefficientabove0.5inthepairwiseregression. 544 Permutationtesting 545 546 547 548 549 550 551 Permutationtestingwasconductedtogeneratenulldistributionsofmparameters.Foreach permutation,weshuffledtheETsrelativetogenotypes.Wethenperformedasinglelocusscanand selectedthetop~500markersforapairwisemarkerscanasdescribedabove.Werepeatedthis processuntil500,000markerpairsweretested.Wecombinedpermutationsacrossmarkerpairsto generateasinglenulldistribution(Tyleretal.2014).Empiricalpvaluesforeachmodelparameter werecalculatedandcorrectedusingfalsediscoveryrate(FDR){benjamini1995controlling}. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 552 Grouping linked markers 553 554 555 556 557 558 Finalresultsarereportedforlinkageblocksratherthanindividualmarkers.Theblockswere determinedasdescribedin(Tyleretal.2016).Briefly,foreachhaplotype,weusedthecorrelation matrixbetweenvariantsasanadjacencymatrixtoconstructaweightednetwork,andusedthefast greedycommunitydetectionalgorithminR/igraphtoestimateboundariesbetweenblocksofsimilar markers(CsardiandNepusz2006). 559 Phenotypic Effects of Motifs 560 561 562 563 564 565 566 567 Foreachmotifintheepistaticnetwork,weexaminedthephenotypiceffectsofeachoftheindividual lociaswellastheinteractioneffect.Foreachindividuallocus,wedividedtheanimalsintotwobins: thosecarryingthealternateallele(e.g.atleastheterozygousfortheA/Jalleleatlocus1),andall others.Wecalculatedthemeantraitvalueacrossalltraitsforbothgroups,anddefinedthemain effectofthealleleasthedifferencebetweenthegroups.Thepredictedadditiveeffectwasthesumof thetwomaineffects.Tocalculatetheactualeffectoftheinteraction,webinnedtheanimalsintotwo groups:thosecarryingthealternatealleleatbothloci(e.g.atleastheterozygousfortheA/Jalleleat locus1andatleastheterozygousfortheNODalleleatlocus2),andallothers. 568 Prioritization of genes in interacting loci 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 Weusedafunction-orientedmethodtogeneratehypothesesaboutwhichgenesininteracting regionsmightbecontributingtotheepistaticeffectsinferredbyCAPE.Wefocusedontwo interactionsthatinfluencedtheImmuneModule,themoduleeigengenefromthegenemodule enrichedforimmunefunction.BothinteractionsinvolvedtheA/JhaplotypefromaregiononChr9. ThisregioninteractedwiththeNODhaplotypeonChr2andtheCASThaplotypeonChr2to influencetheImmuneModule.WefirstusedbiomaRtfoundallproteincodinggenesintheregionby findingallgenesintheeffectsizepeakcreatedbythehaplotype(Durincketal.2005;2009).Weused theRpackageSNPTools(Gatti)toquerytheSangerSNPdatabase(Keaneetal.2011;Yalcinetal. 2011)tofindgenesharboringvariantsprivatetothestrainofinterest.Thus,wefoundallprivateA/J variantsintheregiondefinedbytheA/JeffectonChr9,andallvariantsprivatetoNODandCASTon theChr2regionsdefinedbythesehaplotypeeffectsrespectively. Becausethemaineffectsoftheseregionswererelatedtotheimmunemodule,wefurtherfilteredthe genesineachregiontogenesannotatedtotheMousePhenotype(MP)Ontology(Smithetal.2005) term“immunephenotype.”Wethenlookedforthemostprobablefunctionalinteractionsbetween thegroupsofgenesfromeachchromosomalregionusingIntegrativeMulti-speciesPrediction(IMP) (Wongetal.2015).IMPisaBayesiannetworkbuiltthroughintegrationofgeneexpressiondata, protein-proteininteractiondata,geneontologyannotationsandotherdata.Itpredictsthelikelihood thatpairsofgenesinteractfunctionallyinmultiplemodelorganismsandhumans.WeusedIMPto findthehighestlikelihoodconnectedcomponentthatcontainedatleastonegenefromeach chromosomalregionparticipatingintheepistaticinteraction.Weselectedthegenepairwiththe highestlikelihoodofinteractingfunctionallyasourtopcandidategenepairfortheinteraction. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 591 Figures 592 593 594 595 Table1.FunctionalenrichmentforthreegeneexpressionmodulesfoundbyWGCNA. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 596 597 598 599 600 Figure1. Correlationplotsforallphenotypesusedinthisstudy. Traits tend to be modestly correlated with each other. Physiological traits and expression traits are positively correlated within their groups, but negatively correlated between groups. Malesareshown asgreentrianglesandfemalesarebluesquares.Darkershadeindicateshigh-fatdiet(HF). 601 602 603 604 605 606 607 Figure2.EffectsizesofeachstrainhaplotypeonChr11onfivetraits:leanmass,logfatmass,cholesterol,triglycerides,andthe metabolismexpressionmodule.Individualhaplotypeshavedistincteffectsontraits.TheCASThaplotypeondistalChr11has pleiotropiceffectsonalltraits(greenboxes).TheNZOandA/Jhaplotypeshaveindividualeffectsoncholesterol(bluebox)andleptin (yellowbox)respectively. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 608 609 610 611 612 613 614 615 616 Figure3.Eigentrait(ET)selectionfromdecompositionoftraits.A)Traitsweredecomposedbysingularvaluedecomposition(SVD)to orthogonalETs.ThegraybarsshowtheproportionofthetotalvariancecapturedbyeachET,andtheheatmapshowsrelative contributionsofeachtraittoeachET.B)HaplotypeeffectsforChr11onthefirstthreeETs. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 617 618 619 620 621 622 Figure4.Thefinallocusinteractionnetwork.Maineffectsareshowningrayconcentriccircles.Significantmaineffectsarecoloredfor thehaplotypethathadthesignificanteffects.Positive(brown)andnegative(blue)effectsareonlyshownforSexandDiet.Interactions areshownasarrowsbetweenchromosomalregionsandarecoloredtoindicateanenhancingeffect(brown)orasuppressingeffect (blue). 623 624 625 626 627 628 629 630 Figure5.Tabulationofalleleparticipationinepistaticinteractions.A)Thenumberoftimeseachhaplotypewasthesourceofan interactionorthetargetofaninteraction,andthetotalnumberofinteractionseachhaplotypeparticipatedin.Rowsaresortedbytotal numberofinteractions.Thefinaltwocolumnsindicatehowmanymarkersweretestedinthepairwisemarkertestsforeach haplotype,andhowmanychromosomesthesemarkerswerefoundon.Darkerbluehighlightingindicateshighercounts.B)Adetailed countoftheinteractionseachhaplotypeparticipatedinwitheachotherhaplotypeandeachcovariate(SexandDiet).Darkerblue squaresrepresenthighercounts.Countsof0arerepresentedbydashesforvisualizationpurposes. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 631 632 633 634 635 636 637 Figure6.NetworkMotifsA)Cartoonsdepictingfourtypesofnetworkmotif.Eachmotifconsistsoftwomarkersinteractingtoinfluence onephenotype.Themarkerscaneitherhavethesame(coherent)ordifferent(incoherent)maineffect.Theinteractionbetweenthem canbeeitherenhancingorsuppressing.B)Countsofeachdifferentmotiftypeforeachphenotype.Darkershadesofblueindicate highercounts. 638 639 640 641 642 643 644 645 646 647 Figure7.Phenotypiceffectsofenhancing-incoherent(left)andsuppressing-coherent(right)networkmotifs.“Main1”and“Main2” showtheaveragedeviationfrompopulationmeaninnormalizedphenotypeforanimalscarryingthealternatealleleatmarker1and marker2inthemotifrespectively.Marker1andmarker2aresortedsuchthatmarker1alwayshasthesmaller(morenegative)effect. “Additive”showsthepredictedadditiveeffectgiventheMain1andMain2effects.“Actual”showstheactualdeviationfromthe populationmeanofanimalscarryingthealternatealleleatbothmarker1andmarker2inthemotifs.Linesaredrawntoconnectdots fromindividualmotifs.Bluelinesindicatemotifsthatbringphenotypesclosertothepopulationmeanthanpredictedbytheadditive model.Brownlinesindicatemotifsthatdrivethephenotypefartherfromthepopulationmeanthanpredictedbytheadditivemodel. Redlinesindicateasubsetofmotifsthatcreatemoreextremephenotypesthanpredictedbyanyadditivemodel. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 648 649 650 651 652 653 Figure8.Evidencesupportingaroleofthe129alleleofSorbs1increasingtriglyceridelevelsthroughincreasedtranscription.A)LOD score(top)andhaplotypecoefficients(bottom)forexpressionofSorbs1.TheverticalblacklinemarksthepositionofSorbs1inthe genomeonChr19.B)ExpressionofSorbs1inmaleandfemaleDOmice(a.u.=arbitraryunits)..C)Correlationbetweentriglyceride levelsandSorbs1expression(r=1.7,p<2x10-16).Femalemiceareshowninblue,andmalesareshowningreen. 654 655 656 657 658 659 660 661 662 663 Figure9.Geneprioritizationininteractingloci.A)EffectsofaninteractionbetweenChr9locus2(Chr9.2)andChr2locus2(Chr2.2). TheA/JhaplotypeonChr9.2andtheCASThaplotypeonChr2.2haveindividualnegativeeffectsontheImmuneModule.Together, theyhavethesameeffectastheCASTalleleonChr2.2,indicatingaredundantinteraction.Errorbarsshowstandarderror.B)The transcriptsofCasp4,onChr9,andIl1b,onChr2,arebothcorrelatedwiththeImmuneModule.Thetranscriptsarealsocorrelatedwith eachother.C)ThefunctionalconnectionsbetweenCasp4andIl1bfromtheIMPnetwork.Thetwogenesarepredictedtointeract functionallywithhighconfidence. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 664 665 666 667 668 669 670 671 672 673 674 Figure10.Geneprioritizationininteractingloci.A)EffectsofaninteractionbetweenChr9locus2(Chr9.2)andChr2locus2(Chr2.4). TheA/JhaplotypeonChr9.2hasanegativeeffectontheImmuneModuleandtheNODhaplotypeonChr2.4hasapositiveeffectonthe ImmuneModule.Together,theyhaveaneffectsimilartothatoftheA/JalleleonChr9.2.Errorbarsshowstandarderror.B)The transcriptsofCasp4,onChr9,andSrc,onChr2,arebothcorrelatedwiththeImmuneModule.Thetranscriptsarealsocorrelatedwith eachother.C)FunctionalconnectionsbetweenSrcandCasp4fromtheIMPnetwork.Thetwogenesarepredictedtointeract functionallybyoperatinginrelated,butdistinctpathways. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 675 676 677 678 679 680 SupplementaryFigure1.Analysispipelinetogenerateco-expressionmodules 681 682 BogueM.A.,ChurchillG.A.,CheslerE.J.,2015CollaborativeCrossandDiversityOutbreddata resourcesintheMousePhenomeDatabase.MammGenome26:511–520. 683 684 CarterG.W.,2013InferringgenefunctionandnetworkorganizationinDrosophilasignalingby combinedanalysisofpleiotropyandepistasis.G3(Bethesda)3:807–814. 685 686 CarterG.W.,HaysM.,ShermanA.,GalitskiT.,2012Useofpleiotropytomodelgeneticinteractionsin apopulation.PLoSGenet.8:e1003010. 687 688 689 690 CheslerE.J.,GattiD.M.,MorganA.P.,StrobelM.,TrepanierL.,OberbeckD.,McWeeneyS.,Hitzemann R.,FerrisM.,McMullanR.,ClayshultleA.,BellT.A.,deVillenaF.P.-M.,ChurchillG.A.,2016 DiversityOutbredMiceat21:MaintainingAllelicVariationintheFaceofSelection.G3(Bethesda) 6:3893–3902. 691 692 693 ChickJ.M.,MungerS.C.,SimecekP.,HuttlinE.L.,ChoiK.,GattiD.M.,RaghupathyN.,SvensonK.L., ChurchillG.A.,GygiS.P.,2016Definingtheconsequencesofgeneticvariationonaproteomewidescale.Nature534:500–505. 694 695 CsardiG.,NepuszT.,2006Theigraphsoftwarepackageforcomplexnetworkresearch.InterJournal ComplexSystems:1695. 696 697 DurinckS.,MoreauY.,KasprzykA.,DavisS.,DeMoorB.,BrazmaA.,HuberW.,2005BioMartand Bioconductor:apowerfullinkbetweenbiologicaldatabasesandmicroarraydataanalysis. AveryL.,WassermanS.,1992Orderinggenefunction:theinterpretationofepistasisinregulatory hierarchies.TrendsGenet.8:312–316. bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 698 Bioinformatics21:3439–3440. 699 700 DurinckS.,SpellmanP.T.,BirneyE.,HuberW.,2009Mappingidentifiersfortheintegrationof genomicdatasetswiththeR/BioconductorpackagebiomaRt.NatProtoc4:1184–1191. 701 702 ForbesG.B.,1987LeanBodyMass-BodyFatInterrelationshipsinHumans.NutritionReviews45: 225–231. 703 GattiD.M.,SNPtools.2013,https://CRAN.R-project.org/package=SNPtools. 704 705 Gatti,DanielM.,etal."Quantitativetraitlocusmappingmethodsfordiversityoutbredmice."G3: Genes|Genomes|Genetics4.9(2014):1623-1633. 706 707 708 GattiD.M.,SvensonK.L.,ShabalinA.,WuL.-Y.,ValdarW.,SimecekP.,GoodwinN.,ChengR.,PompD., PalmerA.,CheslerE.J.,BromanK.W.,ChurchillG.A.,2014bQuantitativetraitlocusmapping methodsfordiversityoutbredmice.G3(Bethesda)4:1623–1633. 709 710 711 GhazalpourA.,DossS.,BinZhang,WangS.,PlaisierC.,CastellanosR.,BrozellA.,SchadtE.E.,DrakeT. A.,LusisA.J.,HorvathS.,2006IntegratingGeneticandNetworkAnalysistoCharacterizeGenes RelatedtoMouseWeight.PLoSGenet.2:e130. 712 713 GreeneC.S.,PenrodN.M.,WilliamsS.M.,MooreJ.H.,2009Failuretoreplicateageneticassociation mayprovideimportantcluesaboutgeneticarchitecture.(TIASorensen,Ed.).PLoSONE4:e5639. 714 715 HornT.,SandmannT.,FischerB.,AxelssonE.,HuberW.,BoutrosM.,2011Mappingofsignaling networksthroughsyntheticgeneticinteractionanalysisbyRNAi.Nat.Methods8:341–346. 716 717 HuangD.W.,ShermanB.T.,LempickiR.A.,2009aSystematicandintegrativeanalysisoflargegene listsusingDAVIDbioinformaticsresources.NatProtoc4:44–57. 718 719 HuangD.W.,ShermanB.T.,LempickiR.A.,2009bBioinformaticsenrichmenttools:pathstowardthe comprehensivefunctionalanalysisoflargegenelists.NucleicAcidsRes.37:1–13. 720 721 HuangW.,MackayT.F.C.,2016TheGeneticArchitectureofQuantitativeTraitsCannotBeInferred fromVarianceComponentAnalysis.(XZhu,Ed.).PLoSGenet.12:e1006421. 722 723 724 HwangL.-L.,WangC.-H.,LiT.-L.,ChangS.-D.,LinL.-C.,ChenC.-P.,ChenC.-T.,LiangK.-C.,HoI.-K.,Yang W.-S.,ChiouL.-C.,2010Sexdifferencesinhigh-fatdiet-inducedobesity,metabolicalterationsand learning,andsynapticplasticitydeficitsinmice.Obesity(SilverSpring)18:463–469. 725 726 727 JohnstonA.,GudjonssonJ.E.,AphaleA.,GuzmanA.M.,StollS.W.,ElderJ.T.,2011EGFRandIL-1 SignalingSynergisticallyPromoteKeratinocyteAntimicrobialDefensesinaDifferentiationDependentManner.JournalofInvestigativeDermatology131:329–337. 728 729 730 KangH.M.,ZaitlenN.A.,WadeC.M.,KirbyA.,HeckermanD.,DalyM.J.,EskinE.,2008Efficient controlofpopulationstructureinmodelorganismassociationmapping.Genetics178:1709– 1723. 731 732 733 734 735 KeaneT.M.,GoodstadtL.,DanecekP.,WhiteM.A.,WongK.,YalcinB.,HegerA.,AgamA.,SlaterG., GoodsonM.,FurlotteN.A.,EskinE.,NellåkerC.,WhitleyH.,CleakJ.,JanowitzD.,HernandezPliegoP.,EdwardsA.,BelgardT.G.,OliverP.L.,McIntyreR.E.,BhomraA.,NicodJ.,GanX.,Yuan W.,vanderWeydenL.,StewardC.A.,BalaS.,StalkerJ.,MottR.,DurbinR.,JacksonI.J.,Czechanski A.,Guerra-AssunçãoJ.A.,DonahueL.R.,ReinholdtL.G.,PayseurB.A.,PontingC.P.,BirneyE., bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 736 737 FlintJ.,AdamsD.J.,2011Mousegenomicvariationanditseffectonphenotypesandgene regulation.Nature477:289–294. 738 739 LangfelderP.,HorvathS.,2008WGCNA:anRpackageforweightedcorrelationnetworkanalysis. BMCBioinformatics9:559. 740 741 LehnerB.,2011Molecularmechanismsofepistasiswithinandbetweengenes.TrendsGenet.27: 323–331. 742 743 744 LesniewskiL.A.,HoschS.E.,NeelsJ.G.,deLucaC.,PashmforoushM.,LumengC.N.,ChiangS.-H., ScadengM.,SaltielA.R.,OlefskyJ.M.,2007Bonemarrow|[ndash]|specificCapgenedeletion protectsagainsthigh-fatdiet|[ndash]|inducedinsulinresistance.Nat.Med.13:455–462. 745 746 747 LoganR.W.,RobledoR.F.,ReclaJ.M.,PhilipV.M.,BubierJ.A.,JayJ.J.,HarwoodC.,WilcoxT.,GattiD. M.,BultC.J.,ChurchillG.A.,CheslerE.J.,2013High-precisiongeneticmappingofbehavioraltraits inthediversityoutbredmousepopulation.Genes,BrainandBehavior12:424–437. 748 749 MooreJ.H.,WilliamsS.M.,2009EpistasisandItsImplicationsforPersonalGenetics.TheAmerican JournalofHumanGenetics85:309–320. 750 751 MotenkoH.,NeuhauserS.B.,O'KeefeM.,RichardsonJ.E.,2015MouseMine:anewdatawarehouse forMGI.MammGenome26:325–330. 752 753 MungerS.C.,RaghupathyN.,ChoiK.,SimonsA.K.,2014RNA-Seqalignmenttoindividualized genomesimprovestranscriptabundanceestimatesinmultiparentpopulations.…. 754 755 NishikawaS.,YasoshimaA.,DoiK.,NakayamaH.,UetsukaK.,2007Involvementofsex,strainandage factorsinhighfatdiet-inducedobesityinC57BL/6JandBALB/cAmice.Exp.Anim.56:263–272. 756 757 OberC.,LoiselD.A.,GiladY.,2008Sex-specificgeneticarchitectureofhumandisease.Nature PublishingGroup9:911–922. 758 759 OrdovasJ.M.,2006Geneticinteractionswithdietinfluencetheriskofcardiovasculardisease.Am.J. Clin.Nutr.83:443S–446S. 760 761 762 763 764 PhilipV.M.,SokoloffG.,Ackert-BicknellC.L.,StrizM.,BranstetterL.,BeckmannM.A.,SpenceJ.S., JacksonB.L.,GallowayL.D.,BarkerP.,WymoreA.M.,HunsickerP.R.,DurtschiD.C.,ShawG.S., ShinpockS.,ManlyK.F.,MillerD.R.,DonohueK.D.,CuliatC.T.,ChurchillG.A.,LariviereW.R., PalmerA.A.,O'HaraB.F.,VoyB.H.,CheslerE.J.,2011GeneticanalysisintheCollaborativeCross breedingpopulation.GenomeResearch21:1223–1238. 765 766 PhilipV.M.,TylerA.L.,CarterG.W.,2014Dissectionofcomplexgeneexpressionusingthecombined analysisofpleiotropyandepistasis.PacSympBiocomput:200–211. 767 768 769 770 PierceB.L.,TongL.,ChenL.S.,RahamanR.,ArgosM.,JasmineF.,RoyS.,Paul-BrutusR.,WestraH.-J., FrankeL.,EskoT.,ZamanR.,IslamT.,RahmanM.,BaronJ.A.,KibriyaM.G.,AhsanH.,2014 Mediationanalysisdemonstratesthattrans-eQTLsareoftenexplainedbycis-mediation:a genome-wideanalysisamong1,800SouthAsians.(JKPritchard,Ed.).PLoSGenet.10:e1004818. 771 772 PritchardJ.K.,StephensM.,RosenbergN.A.,DonnellyP.,2000Associationmappinginstructured populations.Am.J.Hum.Genet.67:170–181. 773 RCoreTeam(2016).R:Alanguageandenvironmentforstatisticalcomputing.RFoundationfor bioRxiv preprint first posted online Jan. 5, 2017; doi: http://dx.doi.org/10.1101/098681. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It is made available under a CC-BY-NC 4.0 International license. 774 775 776 StatisticalComputing,Vienna,Austria.URLhttps://www.R-project.org/.AveryL.,WassermanS., 1992Orderinggenefunction:theinterpretationofepistasisinregulatoryhierarchies.Trends Genet.8:312–316. 777 778 RosenbergN.A.,PritchardJ.K.,WeberJ.L.,CannH.M.,KiddK.K.,ZhivotovskyL.A.,FeldmanM.W., 2002Geneticstructureofhumanpopulations.Science298:2381–2385. 779 SchorkN.J.,1997GeneticallyComplexCardiovascularTraits.Hypertension29:145–149. 780 781 782 783 SimanskiM.,RademacherF.,SchröderL.,GläserR.,HarderJ.,2016TheInflammasomeandthe EpidermalGrowthFactorReceptor(EGFR)AreInvolvedintheStaphylococcusaureus-Mediated InductionofIL-1alphaandIL-1betainHumanKeratinocytes(ICAllen,Ed.).PLoSONE11: e0147118. 784 785 SmithC.L.,GoldsmithC.-A.W.,EppigJ.T.,2005TheMammalianPhenotypeOntologyasatoolfor annotating,analyzingandcomparingphenotypicinformation.GenomeBiol.6:R7. 786 787 788 SvensonK.L.,GattiD.M.,ValdarW.,WelshC.E.,ChengR.,CheslerE.J.,PalmerA.A.,McMillanL., ChurchillG.A.,2012High-resolutiongeneticmappingusingtheMouseDiversityoutbred population.Genetics190:437–447. 789 790 TylerA.L.,DonahueL.R.,ChurchillG.A.,CarterG.W.,2016WeakEpistasisGenerallyStabilizes PhenotypesinaMouseIntercross(JMCheverud,Ed.).PLoSGenet.12:e1005805–22. 791 792 TylerA.L.,McGarrT.C.,BeyerB.J.,FrankelW.N.,CarterG.W.,2014AGeneticInteractionNetwork ModelofaComplexNeurologicalDisease.Genes,BrainandBehavior. 793 794 WangS.,MiuraM.,JungY.-K.,ZhuH.,LiE.,YuanJ.,1998MurineCaspase-11,anICE-Interacting Protease,IsEssentialfortheActivationofICE.Cell92:501–509. 795 796 WeissL.A.,PanL.,AbneyM.,OberC.,2006Thesex-specificgeneticarchitectureofquantitativetraits inhumans.NatureGenetics38:218–222. 797 798 799 WongA.K.,KrishnanA.,YaoV.,TadychA.,TroyanskayaO.G.,2015IMP2.0:amulti-species functionalgenomicsportalforintegration,visualizationandpredictionofproteinfunctionsand networks.NucleicAcidsRes.43:W128–33. 800 801 802 803 YalcinB.,WongK.,AgamA.,GoodsonM.,KeaneT.M.,GanX.,NellåkerC.,GoodstadtL.,NicodJ., BhomraA.,Hernandez-PliegoP.,WhitleyH.,CleakJ.,DuttonR.,JanowitzD.,MottR.,AdamsD.J., FlintJ.,2011Sequence-basedcharacterizationofstructuralvariationinthemousegenome. Nature477:326–329. 804 805 806 YangH.,WangJ.R.,DidionJ.P.,BuusR.J.,BellT.A.,WelshC.E.,BonhommeF.,YuA.H.-T.,Nachman M.W.,PialekJ.,TuckerP.,BoursotP.,McMillanL.,ChurchillG.A.,deVillenaF.P.-M.,2011 Subspecificoriginandhaplotypediversityinthelaboratorymouse.Nat.Genet.43:648–655. 807