m01Fig3Bmcrit023AlthoughthesegeneralpatternsareconsistentwiththepredictionsofYeamanandWhitlock28theeffectofthestrengthofselectionontherateofgenomeevolutionisnotaccountedforbyEqs2or4likelyinpartbecaus ID: 891326
Download Pdf The PPT/PDF document "evolutiontoclusteredarchitecturesissomew..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1 evolutiontoclusteredarchitecturesissomew
evolutiontoclusteredarchitecturesissomewhatreduced(e.g., m =0.1,Fig.3 B ; m crit =0.23).Althoughthesegeneralpatterns areconsistentwiththepredictionsofYeamanandWhitlock (28),theeffectofthestrengthofselectionontherateofgenome evolutionisnotaccountedforbyEqs. 2 or 4 ,likelyinpartbe- causetheseapproximationsignorerearrangementsthatyield clustersoflociwith r 0( Methods ).When r 0,theamountof linkagedisequilibriumbetweenloci,andthereforethe tness bene tsoflinkage,shouldscalewiththestrengthofselection (19,53),suchthatunderstrongerselectionthereiseffectively alargertargetregion( L )overwhichrearrangementscanyield clustersoflociwith tnessbene tsowingtolinkage. Therateofevolutionofclusteringisalsostronglydeterminedby boththepopulationsize( N )andtherateofoccurrenceofrear- rangements( ).Clusteringevolvesmuchmorerapidlywhen N = 10 4 thanwhen N =10 3 ,butif isincreasedbyanorderofmag- nitudefor N =10 3 ,therateofevolutionofclusteringisquite similartowhen N =10 4 (Fig.3 C ).Thisshowsthattherateof architectureevolutiondependsheavilyonthepopulation-level rateofoccurrenceofrearrangements( N ),aspredictedbyEq. 2 . Clusteredgenomicarchitecturesalsoevolvewhenenviron- mentsarebothspatiallyandtemporallyheterogeneous(Fig.4 A ) andwhenpopulationsgothroughalternatingperiodsofspatial heterogeneityandhomogeneity(Fig.4 C ).Fig.4 A showsthe evolutionofgenomicarchitecturewhenpopulationsinhabit aspatiallyheterogeneousenvironment(localoptimaare+1/ 1) thatalso uctuatesovertime(asshowninFig.4 B ).Whenthe periodoftemporal uctuationsisshort( T =1,000,Fig.4 A ), clusteringevolvesmoreslowlyandstabilizesatanintermediate level.Theamountofclusteringattheendofthesimulations variesnonmonotonicallyasafunctionof T ,withthemostclus- teringseenwhen T =5,000(Fig.4 A ).Whentheperiodisvery long( T =10 5 ),genomicarchitectureevolvestowardhighly clusteredarchitecturesbut uctuatesduetotransientincreases inthenumberofunclusteredselectedloci( SIText and Fig.S2 ). Whenpopulationsinhabitanenvironmentthatperiodically alternatesbetweenspatiallyhomogeneousandspatiallyhetero- geneous(asshowninFig.4 D ),clusteredgenomicarchitectures stillevolve,althoughtherateofevolutionisslowerforsmaller T (Fig.4 C ).Inthesesimulations,geneticdivergenceoftencollap- sescompletelyduringtheintervalsofhomogeneityandany islandsofallelicdivergencearelost,butclusteredgenomic architecturespersistandcanrapidlygiverisetonewislandsof allelicdivergencewhenheterogeneousenvironmentsarerees- tablished.Clusteredarchitecturesstillevolvewhentheenviron- mentishomogeneousformoretimethanitisheterogeneous, butataslowerrate( SIText and Fig.S3 ). Fromtheaboveresults,itisclearthattherateofrearrangement isanimportantdriverofarchitectureevolution,butistherealsoan effectofthetypeofrearrangement?Therearenumerousmech- anismsthatcangiverisetotherearrangementofsmallfragments ofthegenome:( a )nonhomologousorectopicrecombinationin- volvingtwocloselyspacedcrossovers(54,55),( b )repairfollowing multipledouble-strandedbreaks(56 58),( c )intrachromosomal recombinationformingcircularDNAthatreinsertselsewherein thegenome(59),( d )excisionandreinsertionmediatedby anking transposons(60,61),( e )transposon-mediatedtransduplication [e.g.,byhelitrons(62)o rPack-MULEs(63)],and( f )reverse transcriptionandretrotranspos itionofmRNA,whichresultsin thelossofintronsinthedaughtercopy(64).Ofthese,someresult inthemovementofagenefromoneplacetoanother(e.g., a d ), whereasothersresultinthecreationofaduplicatecopyinanew location(e.g., e and f ).Unfortunately,thereisstilllittlein- formationontherelativefreque ncyofthesedifferentmecha- nismsindifferentspecies.Alth oughalloftheabovesimulations wererunusingacut-and-pastemodelofrearrangementsimilar tomechanisms a d ( Methods ),therateofevolutiontowardclus- teredarchitectureswasnotsubstantiallychangedwhenadupli- cationmodelsimilartomechanisms e and f wasusedinstead( Fig. S4 ).However,unlikethecut-and-pastemodel,alowrateofgene deletion( )wasnecessaryforsigni cantclusteringtoevolvein theduplicationmodel,becauseotherwisetherewasnomecha- nismthroughwhichthelessclusteredparentalcopyofalocus couldbelost( Fig.S4 ). Inthesesimulations,geneduplication,genedeletion,andre- combinationbetweenrearrangedandancestralhaplotypescanall resultinchangesinthenumberofcopiesofagivenlocusthatare passedfromparenttooffspring.Empiricalstudieshavefoundthat changesingenecopynumberofteninvolvephenotypicchangesor tnesscosts,likelybecausechangesinthedosageofsomegenes upsettheoverallbalanceofexpression(65 67).Alloftheabove simulationswererunundertheassumptionofrelativelyweakse- lectionagainstcopynumbervariants(CNV; Methods ).Therateof evolutiontowardclusteredarchitecturestendedtodecreasewith increasingstrengthofselectionagainstCNV( Fig.S5 ),suggesting thatclusteredarchitecturesaremorelikelytoinvolvegenesthat haveonlyminimal tnesseffectsduetochangesincopynumber (i.e.,lesspotentialtoupsetthebalanceofexpression;refs.65 67). Whensimulationswererunexcluding tnesscostsforCNV, clusteredarchitecturesstillevolvedatratescomparabletothose showninFig.3,providingtherateofgenedeletionwasatleast ashighastherateofrearrangement( Fig.S4 ). Discussion ImplicationsforGenomicIslandsofDivergence. Therehasbeen considerabledebateabouttheimportanceofdivergencehitch- hikingfortheevolutionofgenomicislandsofdivergence(16,20, 26),withrecentworknotingthattheeffectofdivergencehitch- hikingonestablishmentprobabilitymayoftenbequitelimited relativetotheeffectofselectionactingdirectlyonbene cial Spatially homogeneous Time ( g enerations x1000) Map position (cM) 0250500 0 10 20 30 40 50 Spatially heterogeneous Time ( g enerations x1000) 0250500 A B Fig.2. Thedistributionofselectedlocialonga singlechromosomeover500,000generationsof evolutionwhentheenvironmentisspatiallyhomo- geneous( A ;optimaare =+1and+1)andspatially heterogeneous( B ;optimaare =+1and 1).Pop- ulationsareoftenpolymorphicforarrangements, withselectedandneutrallocibothpresentindif- ferentindividualsatthesamechromosomallocation. Foragivenchromosomalpositionatagivenpointin time,blueindicatesthatmostindividualsinboth populationshaveaselectedlocus,redindicatesthat mostindividualsinonepatchhaveaselectedlocus whereasmostindividualsintheotherlocushave aneutrallocus,andwhiteindicatesthatmostindi- vidualsinbothpopulationshaveaneutrallocus; m =0.01, =0.75, N =10 4 , =10 6 . E1746 | www.pnas.org/cgi/doi/10.1073/pnas.1219381110 Yeaman 26.ViaS(2012)Divergencehitchhikingandthespr eadofgenomicisolationduringecological speciation-with-gene- ow. PhilosTransRSocLondBBiolSci 367(1587):451 460. 27.BartonNH,HewittGM(1989)Adapt ation,speciationandhybridzones. Nature 341(6242):497 503. 28.YeamanS,WhitlockMC(2011)Thegeneticarchitectureofadaptationunder migration-selectionbalance. Evolution 65(7):1897 1911. 29.CharlesworthB,NordborgM,CharlesworthD(1997)Theeffectsoflocalselection, balancedpolymorphismandbackgroundselectiononequilibriumpatternsof geneticdiversityinsubdividedpopulations. GenetRes 70(2):155 174. 30.RiesebergLH(2001)Chromosomalrearrangementsandspeciation. TrendsEcolEvol 16(7):351 358. 31.NoorMAF,GramsKL,BertucciLA,R
2 eilandJ(2001)Chromosomalinversionsandthe
eilandJ(2001)Chromosomalinversionsandthe reproductiveisolationofspecies. ProcNatlAcadSciUSA 98(21):12084 12088. 32.KirkpatrickM,BartonN(2006)Chromosomeinversions,localadaptationand speciation. Genetics 173(1):419 434. 33.FederJL,NosilP(2009)Chromosomalinversionsandspeciesdifferences:whenare genesaffectingadaptivedivergenceandreproductiveisolationexpectedtoreside withininversions? Evolution 63(12):3061 3075. 34.FederJL,RoetheleJB,FilchakK,NiedbalskiJ,Romero-SeversonJ(2003)Evidencefor inversionpolymorphismrelatedtosympatrichostraceformationintheapple maggot y, Rhagoletispomonella . Genetics 163(3):939 953. 35.ChengC,etal.(2012)EcologicalgenomicsofAnophelesgambiaealongalatitudinal cline:Apopulation-resequencingapproach. Genetics 190(4):1417 1432. 36.JonesFC,etal.;BroadInstituteGenomeSequencingPlatform&WholeGenome AssemblyTeam(2012)Thegenomicbasisofadaptiveevolutioninthreespine sticklebacks. Nature 484(7392):55 61. 37.NeafseyDE,etal.(2010)SNPgenotypingde nescomplexgene- owboundaries amongAfricanmalariavectormosquitoes. Science 330(6003):514 517. 38.StrasburgJL,Scotti-SaintagneC,ScottiI,LaiZ,RiesebergLH(2009)Genomicpatterns ofadaptivedivergencebetweenchromosomallydifferentiatedsun owerspecies. MolBiolEvol 26(6):1341 1355. 39.RanzJM,CasalsF,RuizA(2001)Howmalleableistheeukaryoticgenome?Extremerate ofchromosomalrearrangementinthegenus Drosophila . GenomeRes 11(2):230 239. 40.PringleEG,etal.(2007)Syntenyandchromosomeevolutioninthelepidoptera: Evidencefrommappingin Heliconiusmelpomene . Genetics 177(1):417 426. 41.d AlençonE,etal.(2010)Extensivesyntenyconservationofholocentricchromosomes inLepidopteradespitehighratesoflocalgenomerearrangements. ProcNatlAcad SciUSA 107(17):7680 7685. 42.HeliconiusGenomeConsortium(2012)Butter ygenomerevealspromiscuousex- changeofmimicryadaptationsamongspecies. Nature 487(7405):94 98. 43.BackströmN,etal.(2008)Agene-basedgeneticlinkagemapofthecollared ycatcher( Ficedulaalbicollis )revealsextensivesyntenyandgene-orderconservation during100millionyearsofavianevolution. Genetics 179(3):1479 1495. 44.EllegrenH(2010)Evolutionarystasis:Thestablechromosomesofbirds. TrendsEcol Evol 25(5):283 291. 45.PevznerP,TeslerG(2003)Humanandmousegenomicsequencesrevealextensive breakpointreuseinmammalianevolution. ProcNatlAcadSciUSA 100(13):7672 7677. 46.MurphyWJ,etal.(2005)Dynamicsofmammalianchromosomeevolutioninferred frommultispeciescomparativemaps. Science 309(5734):613 617. 47.OnoS(1973)Ancientlinkagegroupsandfrozenaccidents. Nature 244(5414): 259 262. 48.CoghlanA,EichlerEE,OliverSG,PatersonAH,SteinL(2005)Chromosomeevolution ineukaryotes:Amulti-kingdomperspective. TrendsGenet 21(12):673 682. 49.LynchM(2007)Thefrailtyofadaptivehypothesesfortheoriginsoforganismal complexity. ProcNatlAcadSciUSA 104(Suppl1):8597 8604. 50.KooninEV(2009)Evolutionofgenomearchitecture. IntJBiochemCellBiol 41(2): 298 306. 51.KimuraM(1962)Ontheprobabilityof xationofmutantgenesinapopulation. Genetics 47:713 719. 52.YeamanS,OttoSP(2011)Establishmentandmaintenanceofadaptivegenetic divergenceundermigration,selection,anddrift. Evolution 65(7):2123 2129. 53.BartonNH(2000)Genetichitchhiking. PhilosTransRSocLondBBiolSci 355(1403): 1553 1562. 54.PetesTD,HillCW(1988)Recombinationbetweenrepeatedgenesinmicroorganisms. AnnuRevGenet 22:147 168. 55.BishopAJR,SchiestlRH(2000)Homologousrecombinationasamechanismfor genomerearrangements:Environmentalandgeneticeffects. HumMolGenet 9(16): 2427 334. 56.RichardsonC,JasinM(2000)FrequentchromosomaltranslocationsinducedbyDNA double-strandbreaks. Nature 405(6787):697 700. 57.ArguesoJL,etal.(2008)Double-strandbreaksassociatedwithrepetitiveDNAcan reshapethegenome. ProcNatlAcadSciUSA 105(33):11845 11850. 58.ManiR-S,ChinnaiyanAM(2010)Triggersforgenomicrearrangements:insightsinto genomic,cellularandenvironmentalin uences. NatRevGenet 11(12):819 829. 59.WoodhouseMR,PedersenB,FreelingM(2010)Transposedgenesin Arabidopsis are oftenassociatedwith ankingrepeats. PLoSGenet 6(5):e1000949. 60.TonzetichJ,HayashiS,GrigliattiTA(1990)ConservatismofsitesoftRNAlociamong thelinkagegroupsofseveralDrosophilaspecies. JMolEvol 30(2):182 188. 61.FreelingM,etal.(2008)Manyormostgenesin Arabidopsis transposedafterthe originoftheorderBrassicales. GenomeRes 18(12):1924 1937. 62.LaiJ,LiY,MessingJ,DoonerHK(2005)Genemovementby Helitron transposons contributestothehaplotypevariabilityofmaize. ProcNatlAcadSciUSA 102(25): 9068 9073. 63.JiangN,BaoZ,ZhangX,EddySR,WesslerSR(2004)Pack-MULEtransposable elementsmediategeneevolutioninplants. Nature 431(7008):569 573. 64.WangW,etal.(2006)Highrateofchimericgeneoriginationbyretropositionin plantgenomes. PlantCell 18(8):1791 1802. 65.SpringerM,WeissmanJS,KirschnerMW(2010)Agenerallackofcompensationfor genedosageinyeast. MolSystBiol 6:368. 66.CooperGM,etal.(2011)Acopynumbervariationmorbiditymapofdevelopmental delay. NatGenet 43(9):838 846. 67.BirchlerJA,VeitiaRA(2012)Genebalancehypothesis:Connectingissuesofdosage sensitivityacrossbiologicaldisciplines. ProcNatlAcadSciUSA 109(37):14746 14753. 68.DavisMB,ShawRG(2001)RangeshiftsandadaptiveresponsestoQuaternary climatechange. Science 292(5517):673 679. 69.LowryDB,WillisJH(2010)Awidespreadchromosomalinversionpolymorphism contributestoamajorlife-historytransition,localadaptation,andreproductive isolation. PLoSBiol 8(9):e1000500. 70.JoronM,etal.(2011)Chromosomalrearrangementsmaintainapolymorphic supergenecontrollingbutter ymimicry. Nature 477(7363):203 206. 71.RenautS,etal.(2013)Genomicislandsofdivergencearenotaffectedbygeography ofspeciationinsun owers.Inpress. 72.HurstLD,PálC,LercherMJ(2004)Theevolutionarydynamicsofeukaryoticgene order. NatRevGenet 5(4):299 310. 73.WilliamsEJB,BowlesDJ(2004)Coexpressionofneighboringgenesinthegenomeof Arabidopsisthaliana. GenomeRes 14(6):1060 1067. 74.SchmidM,etal.(2005)Ageneexpressionmapof Arabidopsisthaliana development. NatGenet 37(5):501 506. 75.ZhanS,HorrocksJ,LukensLN(2006)Islandsofco-expressedneighbouringgenesin Arabidopsisthalianasuggesthigher-orderchromosomedomains. PlantJ 45(3):347 357. 76.LeeJM,SonnhammerELL(2003)Genomicgeneclusteringanalysisofpathwaysin eukaryotes. GenomeRes 13(5):875 882. 77.Al-ShahrourF,etal.(2010)Selectionupongenomearchitecture:conservationof functionalneighborhoodswithchanginggenes. PLOSComputBiol 6(10):e1000953. 78.BatadaNN,UrrutiaAO,HurstLD(2007)Chromatinremodellingisamajorsourceof coexpressionoflinkedgenesinyeast. TrendsGenet 23(10):480 484. 79.BatadaNN,HurstLD(2007)Evolutionofchromosomeorganizationdrivenby selectionforreducedgeneexpressionnoise. NatGenet 39(8):945 949. 80.LynchM,BobayLM,CataniaF,GoutJF,RhoM(2011)Therepatterningofeukaryotic genomesbyrandomgeneticdrift. AnnuRevGenomicsHumGenet 12:347 366. 81.CharlesworthD,CharlesworthB(1975)TheoreticalgeneticsofBatesianmimicryII. Evolutionofsupergenes. JTheorBiol 55(2):305 324. 82.FisherR(1930) TheGeneticalTheoryofNaturalSelection (Clarendon,Oxford). 83.KimuraM(1956)Amodelofageneticsystemwhichleadstocloserlinkageby naturalselection. Evolution 10:278 287. 84.NeiM(1967)Modi cationoflinkageintensitybynaturalselection. Genetics 57(3): 625 641. 85.
3 TurnerJRG(1967)Whydoesthegenotypenotcong
TurnerJRG(1967)Whydoesthegenotypenotcongeal? Evolution 21:645 656. 86.OttoSP(2009)Theevolutionaryenigmaofsex. AmNat 174(Suppl1):S1 S14. 87.CharlesworthD,CharlesworthB(1980)Sex-differencesin tnessandselectionfor centricfusionsbetweensex-chromosomesandautosomes. GeneticalResearch 35:205 214. 88.WrightS(1941)Ontheprobabilityof xationofreciprocaltranslocations. AmNat 75:513 522. 89.LandeR(1985)The xationofchromosomalrearrangementsinasubdivided populationwithlocalextinctionandcolonization. Heredity(Edinb) 54(Pt3):323 332. 90.ColosimoPF,etal.(2004)Thegeneticarchitectureofparallelarmorplatereduction inthreespinesticklebacks. PLoSBiol 2(5):E109. 91.AlbertAYK,etal.(2008)Thegeneticsofadaptiveshapeshiftinstickleback: pleiotropyandeffectsize. Evolution 62(1):76 85. 92.RogersSM,etal.(2012)Geneticsignatureofadaptivepeakshiftinthreespine stickleback. Evolution 66(8):2439 2450. 93.LarkinDM,etal.(2009)Breakpointregionsandhomologoussyntenyblocksin chromosomeshavedifferentevolutionaryhistories. GenomeRes 19(5):770 777. 94.WoodhouseMR,TangH,FreelingM(2011)Differentgenefamiliesin Arabidopsis thaliana transposedindifferentepochsandatdifferentfrequenciesthroughoutthe rosids. PlantCell 23(12):4241 4253. 95.LanctôtC,CheutinT,CremerM,CavalliG,CremerT(2007)Dynamicgenome architectureinthenuclearspace:regulationofgeneexpressioninthreedimensions. NatRevGenet 8(2):104 115. 96.WijchersPJ,deLaatW(2011)Genomeorganizationin uencespartnerselectionfor chromosomalrearrangements. TrendsGenet 27(2):63 71. 97.MartinG,OttoSP,LenormandT(2006)Selectionforrecombinationinstructured populations. Genetics 172(1):593 609. 98.Hart eldM,OttoSP,KeightleyPD(2010)Theroleofadvantageousmutationsin enhancingtheevolutionofarecombinationmodi er. Genetics 184(4):1153 1164. 99.CharlesworthB,CharlesworthD(1993)Directionalselectionandtheevolutionofsex andrecombination. GenetRes 61(3):205 224. 100.BürgerR(1999)Evolutionofgeneticvariabilityandtheadvantageofsexand recombinationinchangingenvironments. Genetics 153(2):1055 1069. 101.GuillaumeF,RougemontJ(2006)Nemo:Anevolutionaryandpopulationgenetics programmingframework. Bioinformatics 22(20):2556 2557. 102.TurelliM(1984)Heritablegeneticvariationviamutation-selectionbalance:Lerch s zetameetstheabdominalbristle. TheorPopulBiol 25(2):138 193. 103.GoutJ-F,KahnD,DuretL;ParameciumPost-GenomicsConsortium(2010)The relationshipamonggeneexpression,theevolutionofgenedosage,andtherateof proteinevolution. PLoSGenet 6(5):e1000944. 104.TongAHY,etal.(2001)Systematicgeneticanalysiswithorderedarraysofyeast deletionmutants. Science 294(5550):2364 2368. 105.CarpenterAE,SabatiniDM(2004)Systematicgenome-widescreensofgene function. NatRevGenet 5(1):11 22. Yeaman PNAS | PublishedonlineApril22,2013 | E1751 EVOLUTION PNASPLUS exceptionstothispatternoccuroververynarrowrangesofmi- grationratefallingjustbelowthecriticalthresholdabovewhich noadaptivedivergenceoccurs(Fig.1,grayshading).Inthese cases,thedivergencehitchhikingadvantageishighbecausethe establishmentprobabilityofsmallunlinkedmutationsiscloseto0. Undertheassumptionthatthelocithatcanpotentiallycon- tributetolocaladaptationareuniformlydistributedthroughout thegenome,theratio linked / unlinked willbeafunctionofthe fractionofthegenomethatoccurswithin50cMoneithersideof thefocallocus(i.e.,theregionoverwhichthedivergence hitchhikingadvantageiscalculatedinFig.1),relativetothe fractionofthegenomethatisunlinked.Todevelopasimple heuristicunderassumptionsmostfavorabletothedivergence hitchhikinghypothesis,thisratiocanbeapproximatedas clustered / unlinked =1/2/( k 1/2)=1/(2 k 1),where k isthe numberofchromosomes,assumingchromosomesareallof equivalentsize,andtheregionmaintainedasaclusterspans50% ofasinglechromosome.Ifweassumethedivergencehitchhiking advantage, EP clustered / EP unlinked 3(fromFig.1)and k =21(as inthethreespinestickleback),thenusingEq. 1 ,wewouldexpect atleast13unlinkedallelestoestablishforeveryclusteredallele thatestablishes.Notethatinmostcasesthisratiowouldbeeven larger,becausetheadvantageofdivergencehitchhikingissel- domashighasthreefold(Fig.1),andtheregionmaintainedas aclusterinmostempiricalcontextsistypicallymuchsmallerthan 50%ofachromosome. Basedontheaboveheuristic,veryfewoftheallelesthat contributetolocaladaptationareexpectedtoresideinagiven islandofdivergenceexceptunderverylimitedregionsofpa- rameterspace.Itmaybearguedthatashortcomingofthis heuristicisthatitconsidersonlytheformationofagenomic islandaroundasinglefocallocus,buttherecouldbemanysuch focallociindifferentlocationsthroughoutthegenome,eachof whichcouldformagenomicisland(effectivelyincreasingthe mutationaltargetand clustered ).However,thisissueisunlikelyto beimportantbecauseboththeincreaseinestablishmentprob- abilityandthesizeofthegenomicislandthatcanbeformedby divergencehitchhikingareproportionaltothestrengthofse- lectiononthefocaldivergentlocus.Ifmanysuchfocallociexist throughoutthegenome,thenthemaximumstrengthofselection ateachfocallocuswilltypicallybelower,therebyreducingboth thepotentialsizeofthecluster[dueto s vs. r (19,53)]andthe divergencehitchhikingadvantageforanylinkedmutations. Thus,divergencehitchhikingisunlikelytobeabroadlyimpor- tantexplanationfortheexistenceofislandsofdivergencewith- outsomegenomicmechanismthatdramaticallyreducestherate ofrecombinationaroundtheinitiallydivergentlocus,suchas alargeinversion(24,32)ormodi erscreatingarecombination coldspot.Alternatively,ifgenomicrearrangementsbringto- getherthelocithatcontributetolocaladaptation(therebycre- atingagenomicisland),thiswouldeffectivelyincreasetheratio of clustered / unlinked ,andcouldfacilitatedivergencehitchhikingin subsequentboutsoflocaladaptation(iftheallelesformingthe genomicislandhappenedtobelostowingtogenetichomoge- nizationamongpopulations). RearrangementsLeadtoClusteringofLocallyAdaptiveLoci. Clusters oflocallyadaptivelocimightbeexpectedtoevolveundermi- gration selectionbalancethroughthespreadofgenomicrear- rangements,providedtherateofrearrangementsandtheir probabilityof xationarebothsuf cientlyhigh.Ifthelocithat contributetolocaladaptationareinitiallyrandomlydistributed throughoutthegenome,thenthetimerequiredtoevolveacon- gurationwheretheselociareorganizedinasingletightlylinked clustercanbeapproximatedbyaccountingfortherateofoc- currenceofsuitablerearrangementsandtheirprobabilityof xation[basedonamodelbyKirkpatrickandBarton(32)].If L isthenumberofbasepairsoverwhich r 0, G isthenumberof basepairsinthegenome, istherateofrearrangementper locus,pergeneration, N p isthetotalmetapopulationcensussize, and m isthemigrationrate,thenthetime(ingenerations)to evolveaclusteredarchitecturecanbeapproximatedas t G = N p mL [2] ( Methods givesthederivation).Thiscoarseapproximationwill rangeovermanyordersofmagnitude,dependingontheparam- eters,andprovidessomeintuitionabouttheirrelativeimpor- tance.Itisinsensitivetothenumberoflociinvolvedinlocal adaptation( n ),becausewhen n increases,thelongertimere- quiredtorearrangealargernumberoflociisoffsetbyincreases intherealizedrateofrearrangement,becausetherearemore locithatcanpotentiallyberearrangedintoacl
4 uster.However, thetimerequiredtoevolveap
uster.However, thetimerequiredtoevolveapartiallyclusteredarchitecture(by Eq. 4 )maybeseveralordersofmagnitudelessthanthetimeto evolveafullyclusteredarchitecture(byEq. 2 ; Fig.S1 ).These equationscanbeusedtoprovidecoarsepredictionsofwhether clusteringofgenomicarchitecturemightbeexpectedinagiven taxon.Forexample,if N p =10 5 , L =10 5 , G =10 9 ,and m = 0.001,thencompleteclusteringmightbeseenatbiologicallyre- alistictimescalesif 10 5 (as t 10 7 ,byEq. 2 ),andpartial clusteringof n x =5outof n =50locimightoccurif 10 7 (as t 2 × 10 7 ,byEq. 4 ).Thismodelillustrateshowslightvariations insomeparameterscanyieldlargeeffectsontherateofgenome evolution,butasameansofmakingquantitativepredictionsit shouldbeseenasacoarseapproximation,owingtotheassump- tionsinvolvedinitsderivation(detailsaregivenin Methods ). Toexploretheevolutionofgenomearchitecturemoreex- plicitly,withfewersimplifyingassumptionsabouttheevolu- tionarydynamics,Iusedindividual-basedsimulationsofatwo- patchmodelundermigration selectionbalancesimilartothat usedinref.28.Thesesimulationsareinitializedwith10phe- notype-affecting(hereafter,selected)lociequallyspacedalong asinglechromosome,eachseparatedby49neutralloci.Geno- micrearrangementsoccurbymovingasingleselectedlocusto anewlocationaccordingtoeitheracut-and-pasteoraduplica- tionmodel( Methods ),ataper-locusrateof .Becauserear- rangementscanresultinchangesinthenumberofcopiesofeach locus,smalladditional tnesscostswereincurredforindividuals thatdidnothaveexactlytwocopiesofeachofthe10original selectedloci(detailsaregivenin Methods ). Whenbothpopulationsinhabitthesametypeofenvironment, thereisnobene ttohavingatightlylinkedgenomicarchitecture, andthedistributionofselectedlocionthesimulatedchromosome evolvesveryslowly,mainlyunderthein uenceofrandomgenetic drift(Fig.2 A ).Bycontrast,whenpopulationsinhabitdifferent environmentsandexperienceatensionbetweenmigrationand divergentselection,thearchitectureevolvesrapidlythroughthe successive xationofsingle-locusrearrangements,eventually leadingtoclustersoftightlylinkedselectedlocithatare xed acrossbothpopulations(Fig.2 B ).Inthiscase,populationsrapidly evolvegenotypicdivergencethroughtheestablishmentoflocally adaptedallelesateachlocus(usuallywithinseveralthousand generations).Oncegenotypicdivergenceisestablished,rear- rangementsthatmoveoneselectedlocuswithalocallyadapted alleleclosetoanothertendtoinvadeandreplacelesstightly clusteredarchitectures,becauseoftheadvantageofreducedre- combinationbetweenmoretightlylinkedloci.InFig.2 B ,anearly fullyclusteredarchitectureevolvesinanamountoftimesimilar tothatpredictedbyEq. 5 (for m =0.01, n =10, L =2, G =490, =10 6 , N p =20,000,then t =8.4 × 10 5 ). Inmostcases,therateofevolutiontowardclusteredarchi- tecturesincreaseswiththestrengthofdivergentselection(Fig. 3 A )andmigrationrate(Fig.3 B ).However,whenmigrationrates arehighandjustbelowthecriticalthreshold(24,52),therateof Yeaman PNAS | PublishedonlineApril22,2013 | E1745 EVOLUTION PNASPLUS sites (45,46).Studiesofgeneregulationhavefoundevidence thatneighboringgenessometimeshavehighlycorrelatedex- pressionpro les(72 75),ortendtobefunctionallyrelated(76, 77).Evolutionaryexplanationsforthesepatternshavebeenvar- iedbutareoftenrelatedtothebalancebetweenratesofrear- rangement,geneticdrift,andpurifyingselection(50,72).Where explanationsinvokepositiveselection,the tnessbene tsare typicallybasedonchangesthatareendogenoustotheorganism andenvironment-independent,suchasfacilitatedregulationof functionallyrelatedgenes(78)orreducedexpressionnoise(79). Thegeneralperspectivethatarisesfromthesestudiesisthat thearchitectureofthegenomeevolveswithlittleconnectionto theecologyofthespecies,beyondtheimpactofdemographyon geneticdrift(50,72).Lynchandcoworkers(49,80)havepre- sentedcompellingevidenceshowingthatspecieswithsmaller effectivepopulationsizestendtohavelargergenomes,duethe reducedef ciencyofnaturalselectiontoeliminateinsertionsthat increasetheglobalmutationhazard.Beyondthislinktoecology throughdemography,inmostcases,thepatternsthathavebeen studiedincomparativegenomicsaretestablewithoutgoingout intonaturetostudytheecologyofthespecies.Bycontrast,the mechanismofgenomeevolutiondescribedinthisstudyisex- plicitlytiedtotheecologyofthespecies.Thismechanismcanbe seenasanextensionoftheinversion-basedmechanismsoflocal adaptationandspeciationthathavereceivedconsiderablestudy inrecentyears(30 32)andmodelsfortheevolutionofsuper- genes(81),whichbothhavemuchincommonwithprevious studiesontheevolutionofrecombination(82 86).Thismecha- nismisalsosimilarinmanywaystomodelsfortheevolutionof sexchromosomes;CharlesworthandCharlesworth(87)showed thatrearrangementsmovingautosomallocitosexchromosomes wouldbefavoredwhendifferentalleleswerefavoredineachsex. Bycontrast,inthemechanismdiscussedhere,rearrangementsare favoredbecausetheyreducerecombinationbetweenlociwith allelesthatareadaptedtodifferentenvironments.Although earliermodelsofchromosomeevolutionalsonotedthatpositive selectionduetolinkagecouldresultinthespreadofrearrange- ments,theytypicallyfocusedontheeffectofdriftandpopulation structureonthespreadoflargerearrangementswithdeleterious effectsintheheterokaryotype(e.g.,88,89). Becausetherearrangement-basedmechanismdiscussedhereis explicitlytiedtoecology,studyingitmayrequireadifferentap- proachthanistypicallyusedincomparativegenomics.Inmany cases,localadaptationinvolvesgenesfromnumerouspathways. Forexample,adaptationtofreshwaterinthreespinesticklebacks includesresponsestonovelwaterchemistry,temperatureregimes, visualenvironments,predatorcommunities,andforageavailability (89 93).Therefore,analysesofclusteringbasedongeneontology (e.g.,77)mayfailtodetectclustersoflociiftheyinvolvegeneswith suchdifferentproximatefunctions.Also,clustersthatevolvein responsetoadaptationwithgene owmightspanphysicaldis- tancesmuchlargerthanthosecommonlyanalyzedinthecoex- pressionclusterstudies,whichtypicallyexamineslidingwindowsof 2to15genes(e.g.,73 75).Forexample,agenomicislandspanning 10cMhasbeenobservedinpeaaphidsadaptingtodifferent hosts(5),whichpresumablyincludestenstohundredsofgenes. OneintriguingcomparativegenomicstudyofamniotesbyLarkin etal.(93)foundthatrearrangementbreakpointstendedtocluster inregionscontaininggeneswithgeneontologytermsrelatingto in ammationandmuscularcontraction,whereasconservedre- gionswithfewbreakpointstendedtoincludegenesfordevel- opment.Theysuggestedthatbecausein ammationandmuscular contractiongenesaremoredirectlyinvolvedininteractionswith theenvironment,evolutionmayhavefavoredthespreadofrear- rangementsinresponsetonovelenvironmentalchallenges.Simi- larly,Freelingetal.(61)foundthatgeneswithproductsthat interactmoredirectlywithrapidlychangingbioticandabiotic environments(e.g.,diseaseresistanceor owering)hadhigher Average distance between selected loci (cM) 0 10 20 30 40 Time (generations x1000) 0 250 500 Time (generations x1000) 0250 500 0 10 20 30 40 Local optimum -2 -1 0 1 2 period ( T ) Time interval period ( T ) Local optimum -2 -1 0 1 2 Time AB CD Fig.4. Evolutionofclusteredarchitecturesunderspatialandtemporalheterogeneity( A ,as
5 shownin B andspatialheterogeneitypunctua
shownin B andspatialheterogeneitypunctuatedwithperiodsof homogeneity( C ,asshownin D ).InC,theintervallasts1,000generations,duringwhich m =0.5;theblacklineshowstheamountofclusteringforaconstant spatiallyheterogeneousenvironmentwithnoperiodsofhomogeneity.AllotherparametersaresetasinFig.3. E1748 | www.pnas.org/cgi/doi/10.1073/pnas.1219381110 Yeaman areunlinkedtoanarchitecturewhere n x oftheselociarearrangedin asingleclustercanbeapproximatedas t ¼ 4 N p m L = G n n 1 1 þ n x 1 i ¼ 2 4 N p m L = G n i 1 : [4] If n x = n ,thisreducesto t ¼ 4 N p mL = G 1 n 2 i ¼ 1 i 1 þ 1 = n n 1 ¼ G ð H n þ 1 = ð n ð n 1 ÞÞÞ 4 N p mL ; [5] where H n isthe( n 2) th harmonicnumber(if n =2, H n =0).Because1 H n 10formostbiologicallyrealisticvaluesof n ,thiscanbeapproximatedtoan orderofmagnitudeas t G /( N p mL ).Giventheassumptionsinvolvedin applyingthemodelfromKirkpatrickandBarton(32),parameterizingthis modelrequiresthat L representthenumberofbasepairssurroundingeach locusoverwhich r 0.Notethat N p representsthetotalsizeofthemeta- population,becauserearrangementswouldbefavoredwithinwhichever patchtheyoccurandwouldlikelyeventually xthroughthemetapopulation (again,thedynamicsinvolvedhereareoversimpli ed).Becausethismodel ignoresthecontributionsofrearrangementsthatyieldless-tightlylinked clusters(with0 r 0.5),itwilltendtooverestimatethetimerequiredto evolveaclusteredarchitecture.Amor ecomplexmodelbyBürgerandAkerman (24)suggeststhattheKirkpatrickandBarton(32)modelmayunderestimate theinvasionrate,furtherindicatingthatEqs. 4 and 5 shouldtendtoover- estimatethetimerequiredfortheevolutionofclusteredarchitectures. Individual-BasedSimulations. SimulationswererunusingtheversionofNemo (101)thatwasmodi edforref.28(sourcecodedepositedinSourceforge). Twopopulationsexchangingmigrantsatrate m experiencedivergentse- lection,wherethe tnessofanindividualwithagivenphenotype( Z )is afunctionofthelocaloptimum( = ± 1),strengthofselection( ),andthe curvatureofthe tnessfunction( =2): w ¼ 1 ðj Z j = j 2 jÞ (backward andforwardmigrationratesareequal,giventheconstantdensities). Unlessotherwiseindicated, N =10 4 , =0.75, m =0.01,andtheperlocus mutationrate =10 4 .Unlikeinref.28,mutationsfollowahouse-of-cards model(102),withthenewvalueofamutationreplacinganyexistingvalue. Thismutationmodelwasusedsothatnoallelesoflargeeffectcouldbuild upthroughmultiplemutationsatasinglelocus.Mutationeffectsizesare drawnfromaGaussiandistributionwithmean=0andSD=0.05andare truncatedsothatallmutationsizeswere | 0.05 | .Thephenotypeiscalcu- latedbyaddingthemutationeffectsacrossallselectedloci,sothatalocally optimalphenotypeinthepatchwith =1couldbebuiltwith10homozy- gousmutationsof0.05. Genomicarchitectureismodeledbyinterspersingeachof n =10locithat affectthephenotype(selectedloci)with49neutrallocialongasingle chromosome(foratotalof500loci)andsettingtherecombinationrate betweenadjacentlocito0.002,suchthatonaveragetherewouldbeone recombinationeventperindividualpergeneration.Genomicrearrange- mentsoccuratrate perlocus(setto10 6 unlessotherwisenoted), accordingtooneoftwomodelsofrearrangement.Forthecut-and-paste model,rearrangementsoccurredbymovingasingleselectedlocustoanew locationoccupiedbyaneutrallocus;theselectedlocusreplacestheneutral locusinitsnewlocationandisreplacedbyanewneutrallocusinitsoriginal location,tomaintainchromosomesize(thisprocesswasconstrainedsothat oneselectedlocusneverreplacedanotherselectedlocus).Fortheduplica- tionmodel,theprocesswasidenticaltothecut-and-pastemodel,exceptthe parent locusretaineditsstateasaselectedlocus.Selectedlocicouldbe lost(i.e.,converttoneutralloci)ataper-locusrateof ,whichrepresentsthe processofpseudogenizationorgenedeletion.Unlessotherwisenoted,all simulationswererunusingthecut-and-pastemodelwith =0. Torepresentthe tnesscostsarisingfromCNV,anadditional tnesscost ( s c )wascalculatedforeachindividualbasedonaGaussianfunctionwith s c ¼ n i ¼ 1 ð 1 exp ½ ð 2 c i Þ 2 = V c Þ ,summedacrosscontributionsfrom n loci, with c i representingthenumberofcopiesofthe i th locus.Whenthewidthof thisfunction, V c ,issmall,selectionagainstCNVisstrong(e.g.,for V c =50and c =3atonelocus, s c =0.0198);when V c islarge,selectionisweak(e.g.,for V c =5,000and c =3atonelocus, s c =0.0002);for V c = ,thereisnoselection againstCNV.Thisapproachissimilartoamorecomplexmodelproposedby Goutetal.(103),whichalsoassumesthat tnessdecreaseswithincreasing deviationsfromoptimallevelsofgeneexpression,whichinturnoccurdue tochangesingenecopynumber.Thecompletedeletionofagenewill sometimes,butnotalways,resultinlethality(104,105);forsimplicity,inall simulationsanyindividualwith c =0foranylocuswasgivena tnessof 0(irrespectiveof V c ).Unlessotherwisestated,thesimulationsinthispaper wererunusing V c =5,000. Unlessotherwiseindicated,simulationswererunfor500,000generations, withstatisticscalculatedevery500generations,basedonatleast20replicates foreachparametercombination.Theenvironmental uctuationsshownin Fig.4 A weremodeledusingasinewavewithamplitude=2andperiod T , andthese uctuationswereaddedontothelocaloptimaineachpopulation (i.e.,+1/ 1;asshowninFig.4 B ).Theenvironmental uctuationsshownin Fig.4 C weremodeledbychangingthelocaloptimafrom+1/ 1to0/0and themigrationratefrom0.01to0.5for1,000generations,every T gen- erations(asshowninFig.4 D ).ForallothersimulationsshowninFig.4, migrationratewasheldconstantat m =0.01. ACKNOWLEDGMENTS. IthankMikeWhitlockandSallyOttoforworking withmeontheprojectsthatprecededthismanuscriptandprovidinghelpful feedbackandsuggestionsthroughout.DiscussionswithL.Rieseberg, S.Aitken,J.Mee,D.Schluter,F.Guillaume,J.Mell,theAitken,Rieseberg, andWhitlocklaboratorygroup,andtheEvolutionofGenomicandGenetic Architecturejournalclubalsomadeveryimportantcontributionstothede- velopmentofthispaper;helpfulcommentsoftwoanonymousreviewersare alsoacknowledged.SpecialthankstoJ.Meefornotinganerrorinequation1 inref.28.ComputationswererunontheWestGridGlacierHighPerformance ComputingConsortium.ThisworkwassupportedbyGenomeCanadaand GenomeBCfundingtoAdapTree(co-ProjectleadersS.AitkenandA.Hamann). 1.AllenOrrH(2001)Thegeneticsofspeciesdifferences. TrendsEcolEvol 16(7):343 350. 2.BartonNH,KeightleyPD(2002)Understandingquantitativegeneticvariation. Nat RevGenet 3(1):11 21. 3.EhrenreichIM,PuruggananMD(2006)Themoleculargeneticbasisofplant adaptation. AmJBot 93(7):953 962. 4.TurnerTL,HahnMW,NuzhdinSV(2005)Genomicislandsofspeciationin Anopheles gambiae . PLoSBiol 3(9):e285. 5.ViaS,WestJ(2008)Thegeneticmosaicsuggestsanewroleforhitchhikingin ecologicalspeciation. MolEcol 17(19):4334 4345. 6.NosilP,FunkDJ,Ortiz-BarrientosD(2009)Divergentselectionandheterogeneous genomicdivergence. MolEcol 18(3):375 402. 7.NadeauNJ,etal.(2012)Genomicislandsofdivergenceinhybridizing Heliconius butter iesidenti edbylarge-scaletargetedsequencing. PhilosTransRSocLondB BiolSci 367(1587):343 353. 8.RenautS,etal.(2012)Genome-widepattern
6 sofdivergenceduringspeciation:the lakewh
sofdivergenceduringspeciation:the lakewhite shcasestudy. PhilosTransRSocLondBBiolSci 367(1587):354 363. 9.SmadjaCM,etal.(2012)Large-scalecandidategenescanrevealstheroleof chemoreceptorgenesinhostplantspecializationandspeciationinthepeaaphid. Evolution 66(9):2723 2738. 10.FederJL,EganSP,NosilP(2012)Thegenomicsofspeciation-with-gene- ow. Trends Genet 28(7):342 350. 11.RogersS,BowlesW,MeeJA(2013)Theconsequencesofgenomicarchitectureon ecologicalspeciationinpostglacial shes. CurrentZoology 59(1):53 71. 12.StrasburgJL,etal.(2012)Whatcanpatternsofdifferentiationacrossplantgenomestellus aboutadaptationandspeciation? PhilosTransRSocLondBBiolSci 367(1587):364 373. 13.SmithJM,HaighJ(1974)Thehitch- hikingeffectofafavourablegene. Genet Res 23(1):23 35. 14.HermissonJ,PenningsPS(2005)Softsweeps:Molecularpopulationgeneticsof adaptationfromstandinggeneticvariation. Genetics 169(4):2335 2352. 15.PrzeworskiM,CoopG,WallJD(2005)Thesignatureofpositiveselectiononstanding geneticvariation. Evolution 59(11):2312 2323. 16.NoorMAF,BennettSM(2009)Islandsofspeciationormiragesinthedesert? Examiningtheroleofrestrictedrecombinationinmaintainingspecies. Heredity (Edinb) 103(6):439 444. 17.NachmanMW,PayseurBA(2012)Recombinationratevariationandspeciation: Theoreticalpredictionsandempiricalresultsfromrabbitsandmice. PhilosTransR SocLondBBiolSci 367(1587):409 421. 18.PetryD(1983)Theeffectonneutralgene owofselectionatalinkedlocus. Theor PopulBiol 23(3):300 313. 19.BartonN,BengtssonBO(1986)Thebarriertogeneticexchangebetweenhybridizing populations. Heredity 56:357 376. 20.FederJL,NosilP(2010)Theef cacyofdivergencehitchhikingingeneratinggenomic islandsduringecologicalspeciation. Evolution 64(6):1729 1747. 21.CharlesworthD,CharlesworthB(1979)Selectiononrecombinationinclines. Genetics 91(3):581 589. 22.PylkovKV,ZhivotovskyLA,FeldmanMW(1998)Migrationversusmutationinthe evolutionofrecombinationundermultilocusselection. GenetRes 71(3):247 256. 23.LenormandT,OttoSP(2000)Theevolutionofrecombinationinaheterogeneous environment. Genetics 156(1):423 438. 24.BürgerR,AkermanA(2011)Theeffectsoflinkageandgene owonlocal adaptation:Atwo-locuscontinent-islandmodel. TheorPopulBiol 80(4):272 288. 25.FederJL,GejjiR,YeamanS,NosilP(2012)Establishmentofnewmutationsunder divergenceandgenomehitchhiking. PhilosTransRSocLondBBiolSci 367(1587): 461 474. E1750 | www.pnas.org/cgi/doi/10.1073/pnas.1219381110 Yeaman exclusive.Veryfewstudieshavebeenabletoidentifywhetherloci withsignaturesofdivergenceareinfactcontributingtolocal adaptationorjustneutralalleles alongfortheride. Several empiricalstudieshaveidenti edmultiplesignaturesofdiver- gencethatmaptochromosomalinversions(34 37),butothers havefoundthatinversionswereonlyweaklyimplicatedin adaptivedivergence(e.g.,ref.38).Beyondunderstandingthe originsandmaintenanceoflocaladaptation,determiningthe importanceofthesevariousmechanismsalsohasimplications forunderstandingthenatureofgenomeevolution.Ifislandsof divergenceareactivelybuiltbythe xationofgenomicrearrange- ments(mechanism iv ),thenadaptation-with-gene owcouldbe animportantandunderappreciateddriverofevolutionof genomicarchitecture. This competitionamonggenomicarchitectures mechanism couldprovideapartialexplanationforthedramaticvariationin theratesatwhichrearrangementsaccumulateinthegenomesof differenttaxa.Ratesofcoarse-scalechromosomalrearrangement aremorethanthreeordersofmagnitudehigherin Drosophila thaninmammals(39).Althoughrelativelylowratesofchangein geneorderhavebeenfoundinLepidopteransbasedoncoarse- scalemappingdata(40), ne-scalegenomicanalyseshaveshown ratesofrearrangementashighasorhigherthanthosein Dro- sophila ,largelyduetonumeroussmall-scaletranspositionsand inversions(41,42).However,genomicanalysesofbirdsreveal ratesofgene-orderevolutionevenlowerthanthoseinmammals atboththemicro-andmacroscale(43,44).Ratesofrearrange- mentaccumulationalsovaryacrossthegenome;somehighly conservedregionshaveverylowratesofaccumulation,relativeto so-calledfragileregionsthatexperiencehigherratesofaccumu- lation(45,46).Explanationsforthisdramaticvariationinratesof genomeevolutionhavelargelyfocusedonnonadaptiveargu- mentsbasedonrearrangementrate,theactivityoftransposable elements,ordemographicdrift-basedmechanismsthatmodify theef cacyofpurifyingselectiontopurgemildlydeleterious rearrangements(39,47 50).Ifrearrangementstendtobein- volvedinlocaladaptation,thenecologicallymediatedpositive selectionmayplayaroleindrivingsomeofthesedifferencesin ratesandpatternsofgenomeevolution. Here,Ijointlyconsidertheevolutionoflocaladaptationand genomearchitectureinpopulationsinhabitingheterogeneous environments.Idevelopaheuristicbasedonpreviouslyestab- lishedanalyticalmethodstoexplorethelikelihoodofgenomic islandsofdivergenceevolvingunderthestandarddivergence hitchhikinghypothesis(mechanism i ).Usingthisheuristic,I showthatincreasesinestablishmentprobabilityduetodi- vergencehitchhikingshouldrarelybelargeenoughtoexpect strongstatisticalsignaturesofislandsofdivergence.Ithenusean analyticalapproximationandindividual-basedsimulationsto explorethepotentialimportanceofsmallgenomicrearrange- mentsasawaytocreateclustersoflocallyadaptiveloci,which maybefavoredowingtoreducedratesrecombinationbetween loci(mechanism iv ).I ndthatclusteringoflocallyadaptiveloci evolvesreadilyunderawiderangeofparameterspace,sug- gestingthatsmall-scalegenomicrearrangementsmayplayan importantroleintheevolutionofislandsofgenomicdivergence. Results LikelihoodofIslandFormationbyDivergenceHitchhiking. Ifgenomic islandsofdivergencecommonlyevolvebydivergencehitchhik- ing,thenwewouldexpectastatisticalenrichmentofthenumber oflocallyadaptedallelesthatoccurinclusters,relativetothe numberofunlinkedlocallyadaptedallelesscatteredthroughout thegenome.Asimpleexpressionforthisratiocanbeapproxi- matedby n clustered n unlinked ¼ EP clustered EP unlinked clustered unlinked ; [1] where n clustered and n unlinked representthenumberoflocallyadap- tedallelesthatestablishinaclusterextendingto50cMoneither sideofthefocallocus(i.e., r =0.5)vs.anywhereintherestofthe genome, EP clustered and EP unlinked representtheaverageestablish- mentprobabilitiesforclusteredandunlinkedmutations,and clustered and unlinked representthegenomicmutationratesaver- agedoverallclusteredandunlinkedloci.Theestablishment probabilities, EP clustered and EP unlinked ,canbeestimatedbyusing thecontinent-islandmodelofBürgerandAckerman(24)to calculatetherelative tnessoflinkedandunlinkedmutations, andsplicingthisintoKimura s(51)equationfor xationproba- bility,asshownbyYeamanandOtto(52).Thisapproach assumesthatagenomicislandbeginsasasingledifferentiated allele(ata focallocus )andexpandsasotherlinkedlocally adaptedmutationsestablish(detailsaregivenin Methods ). Althoughtheadvantageduetodivergencehitchhikingmaybe quitelargeformutationsthathappentobeverytightlylinkedto theinitiallydivergedfocallocus,mostgenomesareverylarge, andthechancethatamutationwouldoccurfortuitouslyintight linkageisquitesmall,relativetothechanceofamutationoc- curringelsewhereinthegenome.Ifweassumethatthelocithat couldpotentiallycontributetolocalad
7 aptationarerandomly distributedthroughou
aptationarerandomly distributedthroughoutthegenome,thenrelativelyfewofthese lociwouldbefoundintightlinkagetothefocallocus.Thetotal advantageduetodivergencehitchhikingcanbequanti edby averagingtheratio EP clustered / EP unlinked overallpossiblelinked sites(from r =0.5to r 0,aroundthefocallocus).Asshown inFig.1,forthismodeltheaveragedivergencehitchhikingad- vantageisrarelyhigherthanthreefold(andusuallymuchlower) overawiderangeofbiologicallyrealisticparameters.Theonly 0246 8 10 Migration rate Divergence hitchiking advantage 0.0010.010.10.5 = 0.001 = 0.01 = 0.1 Fig.1. Onaverage,mutationslinkedtoalocuswithanestablishedlocally adaptedallelehaveonlyaslightadvantageoverunlinkedmutationsin termsofincreasedestablishmentprobabilityduetodivergencehitchhiking, (divergencehitchhikingadvantage= EP clustered / EP unlinked ),exceptatvery highmigrationrates,justbelowthecriticalthresholdforthemaintenanceof localadaptation(grayshading).Curvesshowthedivergencehitchhiking advantageforthreestrengthsofselectiononthenewmutation( ),aver- agedovermutationsoccurringatarangeofdistancesfromtheinitialdi- vergentlocus,from0.001cMto50cM,usingEq. 3 .Inallcases,selectionon theestablishedalleleis =0.9;similarresultsarefoundforlowervaluesof , butpeakstendtobenarrower. E1744 | www.pnas.org/cgi/doi/10.1073/pnas.1219381110 Yeaman mutations(25).Here,Ipresentaheuristicmodelthatexplicitly quanti estheeffectofdivergencehitchhikingonestablishment probability(Fig.1)andshowthatitisunlikelytoyieldstrong signaturesofislandsofdivergenceundertheassumptionthat thelocithatcancontributetolocaladaptationarerandomly scatteredthroughoutthegenom e.Instead,foreachlocally adaptedallelethatestablishesinagenomicislandduetodi- vergencehitchhiking,therewilllikelybemanyallelesthates- tablishatrandomlocationsthroughoutthegenome.Thisoccurs becausethemutationaltargetthatresideswithinapotential genomicislandisverylimitedrelativetothemutationaltarget acrossthewholegenome,the tnessadvantageforbeingtightly linkedissmall,andunlinkedmutationsstillhaverelativelyhigh probabilitiesof xation. Althoughthe tnessadvantageofbeingtightlylinkedmay usuallybetoosmalltohavemucheffectontheevolutionofge- nomicislandsontheshorttimescaleofasingleboutoflocalad- aptation,thesimulationresultsIpresentshowthatcompetition amonggenomicarchitecturescanhavepronouncedeffectsover longerevolutionarytimescales.Highlyclusteredgenomicarchi- tecturesevolvereadilygivensuf cientlylargepopulations,rear- rangementrates,strengthsofselectionandratesofmigration(Fig. 3).Withinthiscontext,themutationaltargetitselfisrearranged, whichwouldincreasetheproportionofclusteredrelativetoun- linkedmutationsthatcouldestablishinanysubsequentbouts ofadaptation. Divergencehitchhikingcouldpotentiallyplayanimportant roleinforminggenomicislandsofdivergenceinconjunction withthisrearrangementmechanism,ifaspeciesexperiences temporalenvironmentalordemographic uctuationsthatresult inrepeatedepisodesofdivergentselectionandadaptivere- sponsefollowedbyhomogenization.Thecyclingofglacialand interglacialperiodsovermillionsofyearsprovidesanempirical exampleofonesuchsourceofenvironmental uctuation,which haslikelyaffectedadaptationinnumerousspecies(68).Here,I foundthatclusteredgenomicarchitecturesevolveunderarange ofscenarioswithenvironmental uctuations(Fig.4 A ),andthese clusteredarchitecturespersistthroughperiodsofhomogeniza- tionamongpopulations(Fig.4 C ),unliketheclusteredallelic architecturesdescribedbyYeamanandWhitlock(28).The amountofclusteringtendedtovarywiththeperiodoftemporal uctuations(Fig.4 A ),suggestingthatglacialcyclesmighthave quitedifferenteffectsonthegenomicsofadaptationinlong- livedspeciessuchasforesttreescomparedwithshort-lived speciessuchasannualplants.Giventhecyclicalnatureofcli- matic uctuationsandaccompanyingrangeshifts,itseemspos- siblethatforeveryspeciationeventinthehistoryofsomegenera therehavebeenmultipleepisodesofdivergenceandhomoge- nization.Althoughtheprocessofecologicalspeciationmaybe conceptualizedasacontinuumfromearlystagesofdivergenceto good specieswithcompletereproductiveisolation(10),move- mentalongthiscontinuumisnotunidirectional.Thevagariesof historymayresultinrepeatedtransitionsbackandforthalong thiscontinuum,increasingthewindowofopportunityforcom- petitionamongarchitecturesanddivergencehitchhikingtoplay aroleinadaptionand,eventually,speciation. Studiesofthegeneticsoflocaladaptationandecologicalspe- ciationthat ndevidenceofgenomicislandsshouldstrivetotest whetherthelociinvolvedhavebeenbroughttogetherbyrear- rangements.Thiscouldbeachievedbycomparinghoworthologs associatedwithlocaladaptationaredistributedalongchromo- somesinanumberofsisterspeciesspanningincreasingtaxonomic distances.Numerousstudieshavefoundevidencethatsegregating inversionsareofteninvolvedinadaptivegeneticdivergence(36, 69,70),butlessfocushasbeenplacedonothertypesofrear- rangements.Itisworthnotingthatfortheinversion-basedmech- anismofadaptationproposedbyKirkpatrickandBarton(32), inversionsmustbesegregatingtoreducerecombination,whereas rearrangementsdiscussedinthispaperreducerecombinationby changingthepositionoflocionchromosomes,andthereforetend to xacrosspopulations.Anotherimportantpredictionfromthis workisthatislandsofdivergencewithmultipleselectedalleles couldbeexpectedtoforminallopatry,providedthespecieshad aprevioushistoryofadaptationwithgene ow(duringwhichtime clustersoflocicouldhaveevolvedthroughrearrangements).In- deed,thiscouldprovidepartialexplanationforthesimilarityin sizeandnumberofgenomicislandsfoundinsympatricvs.allo- patricpairsofpopulationsof Helianthus (71). ImplicationsfortheEvolutionofGenomeArchitecture. Withcurrent advancesingenomesequencingtechnologies,considerablere- searchhasfocusedon ndingpatternsintheorganizationof genesonchromosomes.Comparativegenomicstudieshavefound thatthedistributionofrearrangementbreakpointstendstobe nonrandomthroughoutthegenome,withhighernumbersof rearrangementsat fragilesites andlowernumbersat solid 0 10 20 30 40 Average distance between selected loci (cM) 0 10 20 30 40 0 Time (generations x1000) 0250500 10 20 30 40 A B C Fig.3. Theaveragepairwisedistancebetweenselectedlocidecreasesover timeatdifferentrates,givendifferentstrengthsofphenotypicselection( A ), migrationrates( B ),andpopulationsizesandrearrangementrates( C ).Un- lessotherwisespeci ed, m =0.01, =0.75, N =10 4 , =10 6 ,andallother parametersareasdescribedin Methods . Yeaman PNAS | PublishedonlineApril22,2013 | E1747 EVOLUTION PNASPLUS likelihoodsoftranspositionin Arabidopsis .Althoughratherspec- ulative,theseinterpretationsofthedataareconsistentwiththe predictionsforthespreadofrearrangementsduetolocaladap- tationdiscussedhere,soitwouldbeinterestingtotestforsimilar patternsinothertaxa. Therateofrearrangementoccurrenceisoneofthemostim- portantfactorsdeterminingwhetherclusteredarchitectures evolve(Eq. 2 andFig.3 C ),butverylittleisknownaboutthe magnitudeof fordifferentrearrangementmechanisms,despite mountingevidencethatnumeroussmall-scalerearrangements haveaccumulatedwithinsomelineagesovertime(42,61).For example,in Arabidopsisthaliana ,itisestimatedthat21 27
8 %of allgeneshavebeenrearrangedsincetheco
%of allgeneshavebeenrearrangedsincethecommonancestorwith Papaya,almostallofwhichoccurredassingle-generearrange- ments(61,94).Whereasmanyofthesewerelikelycausedbyret- rotransposition(94),othermechanismsareimplicatedinatleast 55%oftherearrangementsthatoccurredsincethe A.thaliana Arabidopsislyrata split(59).Althoughhighlyspeculative,itis possiblethatthe3Dorganizationofchromosomesinthenucleus couldfacilitatetheevolutionofclustering.Spatialproximityof genesinthenucleushasbeenshowntoin uencebothgenereg- ulation(95)andratesofrearrangement(96),soifgenesthattend tobeinvolvedinlocaladaptationarealsobroughtintoclose proximityinthenucleustofacilitatecoregulation,theymightmore readilyberearrangedontothesamechromosome.Furtherre- searchwillhelpidentifyhowcommonlythevariousrearrangement mechanismsoccur,andhowthisvariesamongspecies. Itisimportanttonotethatwhereasspatialenvironmentalhet- erogeneitywilltendtofavortightclusteringofthelociinvolvedin localadaptation,someformsoftemporalheterogeneitymayfavor lessclustering.Whenapopulationevolvestowardanewoptimum bythe xationofnovelbene cialmutations,tightlinkagebetween selectedlociwilltendtoslowtheresponsetoselection,duetothe slowerrateatwhichbene cialmutationsondifferentchro- mosomesarebroughttogetherbyrecombination[i.e.,theHill Robertsoneffect(86,97,98)].Similarly,currentlybene cial combinationsofallelesmightbecomedetrimentalfollowingan environmentalchange,favoringincreasedrecombinationbetween thelociinvolvedinlocaladaptation(23,86,99,100).Theresultsin Fig.4 A showthattherelationshipbetweenperiodlength( T )and amountofclusteringisnonmonotonic,suggestingthatfurtherre- searchisneededtodeterminehowthesedifferentformsofselec- tionwillinteractandwhethersomeintermediaterecombination ratemaybeoptimalundercertaintypesofenvironmentalvariation. Moregenerally,anyadvantageofreducedrecombinationbetween locallyadaptedlocimaybecounteredbyarangeoffactorsthat favorincreasedratesofrecombination(86).Thisbringsusbackto Turner(85),whoasked, Whydoesthegenotypenotcongeal? and suggestedthatclustersoflocimightevolveduetolocus-speci c bene tsoftighterlinkage,despiteglobalbene tsofrecombination. Itseemslikelysomeintermediateamountofclusteringwouldyield ratesofrecombinationlowenoughtomaintainlinkagedisequi- libriumbetweencombinationsoflocallyadaptedalleles,buthigh enoughtofacilitatethepurgingofdeleteriousallelesandthe spreadofnewbene cialmutationsintimesofenvironmental change.Giventheimportanceofmigrationratesfordetermining theadvantageoflinkage(32),andmutationratesfordetermining thedeleteriouseffectsoflinkage(86),itseemslikelythattheop- timalamountofclusteringwillbedeterminedtosomeextentbythe ratioofmutationtomigration.Regardlessofwhicharrangementof adaptivelociisselectivelyoptimal,driftandlimitedrearrangement ratesmightpreventtherealizationofsuchcon gurations.Com- parativegenomicstudiesofecologicaladaptationwillprovide much-neededdatatoevaluatetherelativeimportanceofselective andstochasticprocessesinshapingthearchitectureofthegenome. Methods AnalyticalApproximationsofEstablishmentProbability. Theestablishment probabilitiesofnewlinked( EP clustered )andunlinked( EP unlinked )mutations canbeapproximatedusingthesplicingapproachdescribedbyYeamanand Otto(52)withthetwo-locuscontinent-islandmodelofBürgerandAkerman (24).Considerapopulationexperiencingnaturalselectionandimmigration, withtwoloci( A and B )thatrecombineatrate r .Ifthereisastablepoly- morphismsegregatingatlocus B ( B 1 and B 2 experienceselectionof+ and , respectively),andamaladaptedalleleatlocus A ( A 2 experiencesselection of ),thenanewlocallyadaptedallele, A 1 (experiencingselectionof+ ), willestablishwhentheratesofrecombinationandimmigrationofthe maladaptedgenotype( m ; A 2 B 2 )aresuf cientlylow(24).Basedonthe derivationsdescribedbyBürgeran dAkerman(24),therateatwhichthis allelewillinvadecanbeapproximatedbyevaluatingtheJacobianmatrixfor theirequation2atfrequenciesof A 1 B 1 =0, A 2 B 1 =0, A 1 B 2 =1 ( m / ),and A 2 B 2 = m / (equilibriafromequation3.9inref.24).AsshownbyYeaman andOtto(52),theleadingeigenvalueofthismatrixcanbeusedtoap- proximatethenetstrengthofselectionfavoringtheinvasionoftheallele A 1 ,andcanbesubstitutedfor s inpopulationgeneticapproximationsfor xationprobability.Becausethe xationprobabilityofanewmutationis 2 s inlargepopulations, A 1 willestablishwithprobability EP ¼ 2 r þ 2 þ 2 r 4 mr þ r 2 q : [3] Therefore,theadvantageduetodivergencehitchhikingforanewmutation linkedatsomerecombinationrate r = x canthenbequanti edastheratioof EP r = x / EP r =0.5 . ToparameterizethisequationforFig.1,theaveragevalueof EP clustered wascalculatedacross50,000loci,withrecombinationratesranginglinearly from0.001cMto50cMinincrementsof0.001cM(i.e.,numericalin- tegration).Thisapproachassumesthatgenedensityandrecombination ratesareuniformacrossthegenomeandisintendedasameansofap- proximating EP clustered / EP unlinked toparameterizeEq. 1 ,ratherthananat- tempttomakehighlyaccuratepredictions.Althoughtheaverageof EP clustered / EP unlinked wouldbehigherifcalculatedoverasmallerrangeof recombinationrates,thiswouldalsoimplyasmallermutationaltarget(and therefore clustered ),leadingultimatelytoalowervaluefor n clustered . EvolutionofClusteredGenomicArchitectures. Ifall n locithatcontributeto localadaptationbegininanunlinkedcon gurationdistributedrandomly throughoutthegenome,thetimerequiredtorearrange n x oftheselociinto asingletightlylinkedclustercanbeestimatedasfollows.Iftheper-locusrate ofrearrangementis andtheprobabilityofinsertionnearaspeci clocusis L / G (where L isthesizeoftheregionaroundthelocusand G isthesizeofthe genome),thenrearrangementsyieldinga rstpairofclusteredlociwilloccur atprobabilityof2 N p ( L / G ) n ( n 1)pergenerationinametapopulationofsize N p ,becausethereare n possiblelocithatcouldmoveintotightlinkagewith anyofthe( n 1)otherloci.Followingthis rststep,theprobabilityofrear- rangementsmovingoneoftheremaininglociintotightlinkagewiththis rstclusterwouldbe2 N p ( L / G )( n i ),forthe i th locusaddedtothecluster, where i rangesfrom2to( n x 1). Itissomewhatlessstraightforwardtoestimatetheprobabilitythatany suitablerearrangementswouldspreadthroughthepopulation,butthiscan beapproximatedbyconsideringonlythoserearrangementsthatresultin extremelytightlinkage,reducingtherateofrecombinationto 0.For amodeloflocaladaptationwithimmigration(atrate m ),Kirkpatrickand Barton(32)showedthatifanewchromosomalinversioncapturestwolo- callyadaptedloci,reducingtherateofrecombinationfrom r =0.5to r =0,it shouldspreadatarateof m .Importantly,thisrateisindependentofthe strengthofselectiononthelociwithintheinversion,soitshouldgeneralize toanyrearrangementsthatreducerecombinationtonearly0betweentwo previouslyunlinkedloci,regardlessoftheeffectofsizesoftheallelesin- volved(althoughathighmigrationrates,alleleswithsmall s wouldbelost). Thisapproachneglectsan
9 ydirect tnesscostscausedbytherearra
ydirect tnesscostscausedbytherearrangement itselfowingtodisruptedexpression,problemswithrecombination,orim- balancedgenenumber,so shouldbeinterpretedastherateofrear- rangementsthatdonotinvolvesigni cant tnesscosts.FollowingYeaman andOtto(52),theprobabilityof xationofsuchrearrangementsshouldbe 2 m ,andtheexpectedtimebeforetheoccurrenceofarearrangementthat xestoyieldthe rstpairofclusteredlociisthereciprocalofthesecom- binedprobabilities: t 1 ¼ð 2 m × 2 N p ð L = G Þ n ð n 1 ÞÞ 1 .Ignoringrearrange- mentsthatreduceclusteringonceithasevolved,becausethesewillbe disfavored,theexpectedtimetoevolvefromanarchitecturewhereall n loci Yeaman PNAS | PublishedonlineApril22,2013 | E1749 EVOLUTION PNASPLUS Genomicrearrangementsandtheevolutionofclusters oflocallyadaptiveloci SamYeaman a,b,1 a DepartmentofForestandConservationSciencesand b BiodiversityResearchCentre,UniversityofBritishColumbia,Vancouver,BC,CanadaV6T1Z4 EditedbyMichaelLynch,IndianaUniversity,Bloomington,IN,andapprovedApril1,2013(receivedforreviewNovember8,2012) Numerousstudiesofecologicalgeneticshavefoundthatalleles contributingtolocaladaptationsometimesclustertogether,form- ing genomicislandsofdivergence. Divergencehitchhikingtheory positsthattheseclustersevolvebythepreferentialestablishmentof tightlylinkedlocallyadaptedmutations,becausesuchlinkage reducestheratethatrecombinationbreaksuplocallyfavorable combinationsofalleles.Here,Iusecalculationsbasedonpreviously developedanalyticalmodelsofdivergencehitchhikingtoshowthat veryfewclusteredmutationsshouldbeexpectedinasingleboutof adaptation,relativetothenumberofunlinkedmutations,suggest- ingthatdivergencehitchhikingtheoryalonemayoftenbeinsuf - cienttoexplainempiricalobservations.Usingindividual-based simulationsthatallowforthetranspositionofasinglegeneticlocus fromonepositiononachromosometoanother,Ithenshowthat tightclusteringofthelociinvolvedinlocaladaptationtendsto evolveonbiologicallyrealistictimescales.Theseresultssuggestthat genomicrearrangementsmayoftenbeanimportantcomponentof localadaptationandtheevolutionofgenomicislandsofdivergence. Moregenerally,theseresultssuggestthatgenomicarchitectureand functionalneighborhoodsofgenesmaybeactivelyshapedbynat- uralselectioninheterogeneousenvironments.Becausesmall-scale changesingeneorderarerelativelycommoninsometaxa,compar- ativegenomicstudiescouldbecoupledwithstudiesofadaptation toexplorehowcommonlysuchrearrangementsareinvolvedin localadaptation. chromosomalrearrangement | geneticarchitecture | migration selectionbalance | synteny U nderstandingthegeneticbasisofadaptationisacentral problemofevolutionarybiology.Inlightofarapidlygrowing bodyofempiricaldata,therehasbeenconsiderableinterestin howmanygenescontributetolocaladaptation,thedistributionof theireffectsizes,andwheretheyarelocatedinthegenome(1 3). Numerousstudieshavefoundevidencethatthelocithataremost stronglydifferentiatedbetweenpopulationsaresometimesclus- teredtogether(i.e.,inclosephysicallinkage),inwhathavebeen termed genomicislandsofdivergence (4 11).However,other studieshavefoundlittleornoevidenceofdivergentallelesclus- teringinlargegenomicislands(6,10,12).Inlightofthesecon- ictingpatterns,furtherresearchisnecessarytoclarifywhenand howgenomicislandsshouldbeexpectedtoevolve. Genomicislandsmayevolveasabyproductofnaturalselection orasdirectresponsetonaturalselection(orsomecombinationof thetwo).Inthebyproductcase,selectionincreasesthefrequencyof afocalallele,causingallelesatlinkedneutrallocitoalsoincreasein frequency(13 17).Thisyieldsasignatureofdivergenceatadjacent locithatdecreaseswithincreasingratesofrecombination(i.e., distancefromtheselectedsite),andwillgraduallydisappearfol- lowingtheinitialselectivesweepinpopulationsthatdivergedin allopatry.Ifdivergentselectionoccursinthefaceofcontinuedgene ow,geneticdivergenceatlinkedneutrallociwilltendtopersist, owingtoareductionintheeffectiverateofmigration(18,19), whichisalimited neutral formofdivergencehitchhiking(20). Genomicislandscanalsoevolveasadirectadaptiveresponse todivergentselectionwithgene ow.Ifadaptationtoagiven environmentisbasedonchangesinallelefrequencyatmultiple loci,thenrecombinationbetweenthechromosomesofalocally adaptedparentandamaladaptedimmigrantwilltendtobreak upcombinationsofalleleswithhigh tnessthathavebeen establishedbyselection(21 24).Aslongasmigrationratesare lowenoughthatthedescendentsofalocallyadaptedallele spendalargeproportionoftheirtimeinthepatchwherethey are ttest,thenitisbene cialtobetightlylinkedtootheralleles thatarealsolocallyadapted. Thereareatleastfourdifferentmechanismsfortheevolutionof genomicislandsasadirectresponsetoselection,basedonthe generaladvantageofreducedrecombinationbetweenlocally adaptedalleles:( i )increasedestablishmentprobabilityoflinked locallyadaptedalleles[divergencehitchhiking(5,25,26)],( ii ) increasedpersistencetimeoflinkedlocallyadaptedallelesfol- lowingsecondarycontact(6,27),( iii )competitionamongcombi- nationsofalleleswithsimilarphenotypiceffectsbutdifferent linkagerelationships[competitionamonggeneticarchitectures (28)],and( iv )the xationofgenomicrearrangementsthatmove locallyadaptedlociintoclosegeneticlinkage[competitionamong genomicarchitectures(28)].Forallfourofthesemechanisms,the tnessbene tsduetolinkagetendtodecreasewithincreasing ratesofrecombinationbetweenlocallyadaptedalleles(21,24,25, 28,29).Asaresult,genomicislandsaremorelikelytoevolvein areasofthegenomewithlowratesofrecombination,suchas coldspotsorregionswithsegregatinginversions(17,30 33).It shouldbenoted,however,thattheseregionsarealsomorelikelyto havehighlevelsofdivergenceduetothebyproductmechanism discussedabove(17). Unfortunately,itisoftenverydif culttoevaluatewhichof thesemechanismsmaybeinvolvedinbuildinggenomicislandsof divergence,especiallybecausetheyarenotnecessarilymutually Signi cance Genomescansoften ndthatthelociinvolvedinlocaladap- tationtendtoclustertogetheronchromosomes.Aleading explanationsuggeststhatclustersevolvebecausetheproba- bilityofanewmutationestablishingishigherwhenoccurring nearanotherlocallyadaptedmutation,becausesucharchi- tecturesareseldomdisruptedbyrecombination.Ishowthat thistheoryisunlikelytoexplainempiricallyobservedclusters. Instead,simulationsshowthatclustersaremorelikelytoform throughgenomicrearrangementsthatbringcoadaptedloci closetogether.Thissuggeststhatecologicalselectionmayplay animportantroleinshapinggenomearchitecture,incontrast tomanynonadaptiveexplanations. Authorcontributions:S.Y.designedresearch,performedresearch,contributednewre- agents/analytictools,analyzeddata,andwrotethepaper. Theauthordeclaresnocon ictofinterest. ThisarticleisaPNASDirectSubmission. FreelyavailableonlinethroughthePNASopenaccessoption. Datadeposition:ThesourcecodeforthesimulationshasbeendepositedintheSource- forgeRepository, https://sourceforge.net/projects/nemorearrange/ . 1 E-mail:yeaman@zoology.ubc.ca. Thisarticlecontainssupportinginformationonlineat www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1219381110/-/DCSupplemental . www.pnas.org/cgi/doi/10.1073/pnas.1219381110PNAS | PublishedonlineApril22,2013 | E1743 E1751 EVOLUTION PNASPL