(Pybus&Harvey2000;FitzJohn,Maddison&Otto2009).Finally,ifthemissingspec - PDF document

(Pybus&Harvey2000;FitzJohn,Maddison&Otto2009).Finally,ifthemissingspeciescanbeassignedtosubclades,thesesubcladescanbeaugmentednumerically(Alfaroetal.2009;FitzJohn,Maddison&Otto2009).Whilereasonable,noneoftheseapproachesmakefulluseofavailableinformation.Wedescribeandimplementageneralapproachthatincorporatespriorinformationontheplacementofmissingtaxaandthatissuitedtoanytaxonomicgroupwithmissinggeneticorphenotypicinformation.Weprovidenewsoftware(PASTIS;availablefromhttp://cran.r-project.org/web/packages/pastis/)writtenintheRenvironmentto(i)generateMrBayesinputlesfortreeinferenceand(ii)conductpost-MrBayesassessmentoftheplacementofmissingtaxa.Ourapproachhasrecentlybeenusedtoinfercompletespeciesphylogeniesforbirds(Jetzetal.2012).Weuseabirth-deathpriorontheedgelengths,whichmeansthat(i)edgelengthsforallspecies(andclades,includingthosewithnogeneticdata)aresampledunderacommonframeworkand(ii)theinclusionofmissingspe-ciesisunbiasedwithrespecttottingrate-heterogeneousdiversicationmodels.Weimposenomodelpriorontopologies,whichmeansthatalllabelledtreesareequallylikelybeforethedataandtopologicalconstraintsareimposed.Suchapriorisconsistentwithourcurrentunder-standingofphylogenetictreeshape(see,e.g.Blum&Francois2006).Theresultingtreedistributionisexpectedtobebroad,andtheconsensustreecreatedfromitshouldreconstructtheconstrainttreeplusalltheaddedspeciesattachedaspolytomiestotheirmostexclusiveclade.How-ever,thefulldistributionalsorepresentsallavailabletaxo-nomicandgeneticinformationandmeaningfulbifurcatingtopologiesandedgelengths.ThePASTISmethodRUNNINGPASTISOurobjectiveistointegratemissingtaxaintoaposteriordistri-butionoftreesthatincludesalltaxainthecladeandisconsis-tentwithrawsequencedataandtaxonomicinformation.Toachievethis,thePASTISmethodtakesadvantageofexibletopologyconstraintsrecentlyimplementedinMrBayesversion3.2(Ronqvistetal.2012)andusesfunctionsintheRpackagesape(Paradis,Claude&Strimmer2004)andcaper(Ormeetal.2012).Topologyconstraintsarestatementsdictatingwherespeciescanandcannotbeplacedwhileinferringaphylogenetictree(e.g.Swoord&Beagle1993;Day,Cotton&Barraclough2008;Thomas2008;Lanfear&Bromham2011).MrBayesallowsconstraintstobehard(enforcingmonophylyonataxonnegative(preventingmonophylyonataxonset)orpartialPartialconstraintsdenetaxonsetsthatmustbemonophyleticwithrespecttoasecondtaxonsetwhileallowingtaxanotdenedineithersettomovefreelyaboutthetree.Useofmulti-plecomplementaryandhierarchicalconstraintsprovidesapowerfulapproachwithwhichtoincorporatetaxawithnogeneticdataintoposteriordistributionsofphylogenetictreesbytheapplicationofaseriesofrulesandassumptionsontheplacementofmissingtaxa.OurgeneralapproachisoutlinedinFig.1.Thekeycomponentsare(i)aconstrainttreethatdenesrelationshipsamongtaxawithsequencedataand(ii)ataxonomyorotherexternaldataprovidingpriorinformationontheanitiesoftaxawithnosequencedata.Generatingsetsoftensorhundredsofconstraintsmanuallyisnon-trivial,tediousandpronetoerror.PASTISautomatestheprocessbycombiningthetwoprimarysourcesofinformationintoa Geneticdata[Consensus+Taxonomy] Fig.1.Aschematicofthetreeinferenceprocess.Aconstrainttree(typicallyaconsensustopology)isgeneratedfromthegeneticdatasetorprevi-ouslypublishedphylogenetichypotheses.ThisconstrainttreeandthetaxonomicdataassociatedwithmissingspeciesdictateplacementconstraingeneratedbyPASTIS.AposteriordistributionofultrametrictreescontainingalltaxaisthengeneratedinMrBayes.Asterisksdenotetheconstrainfromtheconsensustopology.Intheexamplebelow,genusEhasnosequencedataandisfreetomoveaboutthetree,butnottoenteranyothergenus.Forsimplicity,thesearedepictedwithnovariationindepth;inafullposterior,depthswillalsovary.2013TheAuthors.MethodsinEcologyandEvolution2013BritishEcologicalSociety,MethodsinEcologyandEvolution,1011…1017G.H.Thomasetal. MrBayesinputlewithapotentiallyextensivesetoftopologyconstraints.Wedenethreecategoriesofspecies:type1specieshavegeneticinformation;type2specieshavenogeneticinformationbutarecongenersofaspecieswithgeneticinformation;andtype3specieshavenogeneticdataandaremembersofagenusthatdoesnothavegeneticdata.Tointegratethethreetypesofmissingspecies,wemaketwoimportantassumptions:(i)taxonomicgroups(e.g.genera,subfamilies)aremonophyleticunlessthereispositiveevidence(i.e.geneticdata)thatsuggestsotherwiseand(ii)reasonableedge-lengthandtopologypriors(i.e.birth-deathmodels)exist.WeusetheavianfamilyAccipitridaetodemonstratehowweintegratemissingspeciesintothephylogeneticpipelinewithPASTIS.Weprovidealldata,detailsofdatasourcesandRscriptsongshare(http://dx.doi.org/10.6084/m9.gshare.692180).TheAccipitridaeconsistsof243speciesforwhichweobtainedsequencedataon175speciesfromGenBank(Bensonetal.2009).Werefertotheseastype1species.Theremaining68specieshavenosequencedata,ofwhich60havecongenerswithsequencedata(type2species)andeightspeciesfromsixgeneradonothavecongenerswithsequencedata(type3spe-cies).Weconstructedaconstrainttree(Accipitidriae.tree,Table1)basedonanalignmentincludingmitochondrial,nuclearcodingandnuclearnon-codinggenesforthe175type1species,plustwooutgroupsusingMrBayes(Ronqvistetal.2012).Wecollapsednodeswith95%posteriorprobabilitiestopolytomies.ModelparametersfortheconstrainttreeareasdenedinthesupplementaryleAccipitridae.templateong-share,andfurtherdetailsofconstructioncanbeobtainedfromJetzetal.(2012).Theconstrainttreecouldalternativelyhavebeenderivedfrompublishedsourcesorinferredfromdierentdataormethods.However,inpractice,wehavefoundthat,inordertoavoidconictbetweendataandimposedconstraints,theconstrainttreeshouldpreferentiallybegeneratedwiththesameunderlyinggeneticdata(e.g.newdataorpreviouslypub-lishedsequencesoralignments)thatwillbeusedinthegenera-tionofthefulltree.Tointegratethe68type2andtype3species,weconstructasimpletaxondenitionle,Accipitridae.taxa(Table1),thatlistsallspecies(types1,2and3)alongwithacladename.Thecladenameinthisexampleissimplythegenusnamebutcouldinprinciplebeanyhighertaxonormoreinclusiveclade.Theleisincsvformat,andtherstfewlinesareastaxon,cladeTyto_alba,OutgroupCathartes_aura,OutgroupAccipiter_albogularis,AccipiterAccipiter_badius,AccipiterAccipiter_bicolor,AccipiterAccipiter_brachyurus,AccipiterPASTISintegratestheinformationinthetaxalewiththeconstrainttreetoformulatethesimplestpossibletopologycon-straintsthatcombinetaxonomicdata(type2,3)andgeneticdata(type1).TogenerateaMrBayesinputleusingPASTIS,wesimplyrun:library(pastis)pastis_main(constraint_tree=Accipitridae.treetaxa_list=Accipitridae.taxasequences=Accipitridae.sequencesoutput_template=Accipitridae.templateoutput_file=Accipitridae.nexusTheleAccipitridae.templateprovidesaneditabletemplatefortheMrBayesinputleandisoptional:PASTISusesasim-pledefaulttemplateifthisleisomitted.Intheabovecalltopastis_main,PASTISwillassumethatallcladesdenedinthetaxalearemonophyleticunlessthereisconictingevi-dencetothecontrary.Figure2describestheplacementsofexemplartype2speciesfromtheAccipitridaetree.Thetype2speciesButasturliventerbelongstoaclade()thatismonophyleticontheconstrainttree.Withnoevidencetothecontrary,PASTISdenesconstraintsthatrestrictsButasturtotheButasturcladebutallowittomovefreelyamongbrancheswithinthatclade.Incontrast,thetype2speciesButeoarcheriismemberofagenusforwhichtheconstrainttreepro-videspositiveevidenceofnon-monophylyandimpliesthatispartofamoreinclusivecladeincludingthegeneraLeucopternisGeranoaetus.Here,Buteoarcheriisconstrainedtothecladethatincludesthemostrecentcom-monancestorofalltype1speciesbelongingtoandtheadditionalthreegeneralistedabove.Werefertothisasa Table1.AccipitridaeinputusedinPASTISPASTISargumentsuxRequiredDescriptionconstraint_treeAccipitridae.treeYesAconstrainttreecontaining177species(175Accipitridaeandtwooutgroups).Thetreeisa95%majority-ruleconsensustreederivedfromanunconstrainedMrBayesanalysisofthealignmentinAccipitridae.sequences.Thisstructurewillbeincludedinalloutputtrees.taxalistAccipitridae.taxaYesThisisalist(.csvformat)ofall245taxa(includingoutgroups)tobeincludedinthecompleteAccipitridaetree.Eachspeciesisassignedmembershiptoitsgenus.Accipitridae.missingcladesOptionalThisisalistofsixgenerathatarenotrepresentedintheconstrainttreeandwheretheymaybeplacedinthetree.Accipitridae.sequencesOptionalThisisthealignedsequencedataforthe177speciesintheconstrainttree.PASTISexpectsthealignmentinFASTAformat.Thisisoptional,butwillbepresentinmosttypicalanalyses.output_templateAccipitridae.templateOptionalThisisatemplatelefortheMrBayesoutputle.Itoutlinesoptionssuchasthedatapartitions,numberofiterations,burninperiod,etc.2013TheAuthors.MethodsinEcologyandEvolution2013BritishEcologicalSociety,MethodsinEcologyandEvolution,1011…1017Model-basedcompletephylogenies supragenus.ButeoarchericanmovefreelywithinthisbroadcladebutcannotbreakthemonophylyofHarpyhaliaetus,allofwhicharemonophyleticgeneranestedwithinthesupragenus(Fig.2).Withoutfurtherinformation,type3speciesareconstrainedtobemonophyletic(e.g.thetwomembersofHarpagusareforcedtobesistertaxa),theycannotbreakthemonophylyofanygenusorsupragenusbutcanotherwisemovefreelythroughoutthetree.Theoutputle(Accipitridae.nexus)containsfullsetsofconstraintsthatmeetthesecriteriaandcanbeexecutedinMrBayes.Intheaboveexample,weallowedtype3taxatomovethroughoutthewholetree.However,wemayhavepriorinfor-mationthatallowstype3taxatobeconstrainedfurther.Fortheeighttype3taxa,wedeneadditionalconstraintsinasec-ondconstraintle,Accipitridae.missingclades(Table1).Forbrevity,theexamplebelowprovidesahypotheticalsetofconstraintsforagenus,A,withnosequencedataforanymem-berspecies(butseetheleAccipitridae.missingcladesforamorecompleteexample).A,include,B,C,D,EA,exclude,B,CHere,thegenusAisconstrainedtobeincludedinacladecontainingthegeneraB,C,DandE.However,taxonomyorothersourcessuggestthatAismorecloselyaliatedtogeneraDandEthantoBandC,buttheconstraintphylogenysuggeststhatDandEarenotmonophyletic.Inparentheticalformat,therelationshipbetweenB,C,DandEisdenedas:(D,(E,(C,B))).TheexcludeconstraintpreventsgenusAfromenteringtheclade(C,B).Inpractice,type3speciescanhavemultipleexcludeconstraintsbutrequireonlyasingleincludeconstraint(seeexamplesinAccipitridae.missingclades).Figure3ahighlightsoneexample,Megatriorchisdoriae,show-ingthebranchestowhichthegenuscanattach.ToobtainMrBayesinputrun:pastis_main(constraint_tree=Accipitridae.treetaxa_list=Accipitridae.taxa,missing_clades-Accipitridae.missingcladessequences=Accipitridae.sequencesoutput_template=Accipitridae.templateoutput_file=Accipitridae.nexusAlternatively,weprovideawrapperfunctiontopastis_mainpastis_simplewhichsearchesforandautomaticallyloadsrelevantles.Ifalllesareinthesamelocation(intheexamplebelowPASTISwillsearchthecurrentworkingdirectorybutafulllepathcanbespecied)withtheappropriatesux(tree,taxa,missingclades,sequences,tem-plate),thefunctionpastis_simpleisrunasfollows:pastis_simple(AccipitridaeIneitherexample,PASTISwillgenerateaninputlewiththenexussuxreadyforMrBayesexecution. (a)(b) Fig.2.Placementofmissing(type2andtype3)speciesonasubcladeoftheAccipitridaephylogeny.(a)Thetype2speciesButasturliventeriscon-nedtotheareaofthetreecontainingthegenus(highlightedinblue)andisallowedtoattachedtoanybranchmarkedwitharedline.Theconstructionofotherconstraintsismorecomplex.ThegenusButeohasseveralmissing(type2)species.isnotamonophyleticgenusandisinferredtoformacladewithmembersofParabuteoLeucopternisandGeranoaetus(taxonnamesinblackinpanela).Werefertothisasasuprage-nus.ThesupragenusisfurthercomplicatedbythebroadspreadoftheincludedgenusLeucopterniswhichformspartofapolytomyattherootofthesubclade.Toresolvethepolytomy,weincludetheminimumpossiblesetofgeneradenedbythemostrecentcommonancestoroftype1spe-ciesbelongingtoButeo,ParabuteoLeucopternisGeranoaetus.ThisadditionallyincludesthegeneraButeogallusHarpyhaliaetus.Type2speciescanattachtobrancheswithinthissupragenusshownasblackbranchespanel(b).However,withintheresolvedsupragenus,type2speciescannotbreakthemonophylyofanynestedgenusorsupragenusandsocannotattachtothegreybranchesbelongingtothegenusHarpyhaliaetus2013TheAuthors.MethodsinEcologyandEvolution2013BritishEcologicalSociety,MethodsinEcologyandEvolution,1011…1017G.H.Thomasetal.