Download Epistatic networks jointly influence phenotypes related to

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Dual inheritance theory wikipedia , lookup

Biology and consumer behaviour wikipedia , lookup

Neurogenomics wikipedia , lookup

Genetic algorithm wikipedia , lookup

Twin study wikipedia , lookup

Behavioural genetics wikipedia , lookup

Heritability of IQ wikipedia , lookup

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 The copyright holder for this preprint (which was not
peer-reviewed) is the author/funder.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 The copyright holder for this preprint (which was not
peer-reviewed) is the author/funder.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