Mol.Biol.Evol.19(10):1769 - PDF document

Mol.Biol.Evol.19(10):1769±1781.20022002bytheSocietyforMolecularBiologyandEvolution.ISSN:0737-4038RootingPhylogeneticTreeswithDistantOutgroups:ACaseStudyfromtheCommelinoidMonocotsSeanW.Graham,RichardG.Olmstead,andSpencerC.H.Barrett*DepartmentofBiologicalSciences,UniversityofAlberta,Canada;²DepartmentofBotany,UniversityofWashington;and³DepartmentofBotany,UniversityofToronto,Canada 1770Grahametal.melinoidmonocots,averylargeanddiversearrayoftaxathatincludesthegrasses,sedges,gingers,andpalms(reviewedinGrahamandBarrett1995).SeveralsubtlenonmolecularcharactersindicatearelativelycloserelationshipofPontederiaceaewithHaemodora-ceaeandPhilydraceae(orboth)(e.g.,Simpson1987;SteineckeandHamann1989;Tillich1994,1995;Giv-nishetal.1999),whereasvariousmolecularstudiessug-gestthatCommelinaceaemaybethesister-grouptoPontederiaceaeandthatHanguanaceaeisrelatedtoallfourfamilies(e.g.,Chaseetal.1995,2000;Givnishetal.1999).Themostrecentclassi®cationschemeofthemonocots(Chaseetal.2000)includesthese®vefami-liesintheorderCommelinales,asthesister-groupofZingiberales(thegingersandrelatives;seealsoSteven-sonetal.2000).However,thesehigher-orderrelationships®ndonlyweaksupportfromtheavailabledata.The®vefamiliesoftheCommelinalesareallquitedistinctfromonean-otherfrommorphologicalandmolecularperspectives(e.g.,Dahlgren,Clifford,andYeo1985,pp.149±150,323±344,374±387;Duvalletal.1993),andestimatesoftheiragebasedonLdatasuggestthattheydi-vergedfromoneanotherinthelateCretaceous(Bremer2000).InthecaseofPontederiaceaeatleast,thismaypredatesubstantiallythediversi®cationoftheextantmembersofthefamily(seeBarrettandGraham1997).Thisevolutionarydistinctnessmayexplainwhynumer-ouspermutationsofrelationshipsamongthemembersofCommelinalesandrelativeshavebeenobservedinphylogeneticstudiesbasedondifferentsetsoftaxaandvariousmolecularandmorphologicalmarkers(e.g.,Chaseetal.1993,1995,2000;GrahamandBarrett1995;Davisetal.1998;Givnishetal.1999;Stevensonetal.2000;Neyland2002).Mostofthesedatasetspro-videverypoorbootstrapsupportforanyparticularre-lationshipintheorder,andtheyneitherindicatestrongsupportforthesister-groupstatusofanyparticulartaxontoPontederiaceaenorofanyparticularrootpositionwithinthefamily(Kohnetal.1996;BarrettandGraham1997;Grahametal.1998).ItisthereforeofinteresttoaskhowmuchsignalthenearestoutgroupsprovideforrootingPontederi-aceae,comparedwiththoselesscloselyrelated,andtoinvestigatewhethertheoptimalrootingsdeterminedus-ingthenearestoutgroupsareaconsequenceoflong-branchattraction.ArelatedgoalistoclarifythepositionofPontederiaceaeinmonocotphylogeny.Severalau-thors(e.g.,Hillis1996;Graybeal1998;Soltisetal.1998;SwoffordandPoe1999)havealsonotedthatadd-ingtaxaandcharacters(orboth)toaphylogeneticanal-ysiscanimprovetheaccuracyofphylogeneticestima-tion,andsoitwouldbevaluabletoaddresstheextenttowhichaddingdata(genesoroutgroups)improvesourabilitytoassesswheretherootsplitofthefamilylies.Weaddressthesequestionsbysamplingmultipleout-grouptaxatoPontederiaceaefortwochloroplastgenesacrossabroadsampleofmonocots.Althoughourstudyfocusesexclusivelyonthissmallfamilyofcommelinoidmonocots,theinsightsgainedfromexploringrootinginthisfamilyarelikelytobebroadlyapplicabletoanyphylogeneticstudywhereambiguityintreerootingmaybeafunctionofdistantlyrelatedoutgroups.MaterialsandMethodsDataSourceandMatrixConstructionPartialcodingsequencesforthetwochloroplastgenesexamined(FandL)wereobtainedbyman-ualandautomatedDNAsequencingofPCRproducts,usingmethodsandprimersoutlinedinOlmsteadandSweere(1994),Grahametal.(1998)andGrahamandOlmstead(2000).Atotalof1343bpofDNAsequencewasobtainedfromL,andapproximately490bpofsequencewasobtainedfromaroundthe3-endofthemostvariablepartofthisgene(seeGrahametal.1998).TheregionofFsequencedcorrespondstobp1457±1946ofOryzasativandhF(GenBankaccessionnumberX15901).TheFandLsequencesfor24taxainPontederiaceaewereobtainedbyGrahametal.(1998).MostoftheFsequencesforoutgrouptaxawereobtainedforthecurrentstudy,andmostoftheLsequenceswereobtaineddirectlyfromGenBankorwereprovidedbyotherworkers;collectiondetailsandalistofthesequencesourcesareprovidedintable1.ManualsequencealignmentwasperformedusingcriteriaprovidedinGrahametal.(2000).Align-mentgapswerenotrequiredforL,butsixteenindelswereinferredinF.Thesewereallparsimonyunin-formative,apartfromasingleindelsharedbytwova-rietiesofPontederiacordata,andanothersharedbyAlignmentgapsweretreatedasmissingdata.In®vecases(table1)acomposite``placeholder''taxonwasrepresentedbysequencesfromtwodifferentspeciesinthesamegenus.Thetwogeneswereconsideredsepa-ratelyandinconcertfortherootingexperimentsde-scribedbelow.InferenceofMonocotPhylogenyThelocalpositionofPontederiaceaeamongthesampledmonocottaxawasexaminedinamaximumparsimonyanalysisusingPAUP*4,betaversions4±10(Swofford1999).Heuristicsearcheswereperformedwithallcharacterandcharacter-statechangesequallyweightedandwith``MulTrees''and``SteepestDescent''optionsactivated.Tominimizetheriskof®ndingonlylocaloptima,100randomadditionreplicateswereper-formed(Maddison1991).Branchsupportamongtheoutgrouptaxawasassessedusingbootstrapanalysis(Felsenstein1985),with100bootstrapreplicatesandonerandomadditionsequenceperbootstrapreplicate.RootingExperimentsAnunrooted,most-parsimonioustopologyof24taxaofPontederiaceaederivedfromthreechloroplastdatasets(®g.2inGrahametal.1998)waschosentoperformaseriesofrootingexperimentsforrealandran-domoutgroupsequencesusingPAUP*.Thisunrootedtreeisoneofthetenmost-parsimonioustreesfoundusingachloroplastrestriction-sitedataset(Kohnetal. RootingPhylogeneticTrees17711996)andisnearlyidenticalintopologytotreesin-ferredfromacombinedanalysisofFandL(Gra-hametal.1998).Thesametopologywasalsoinferredfromparsimonyanalysisofvariouscombinationsofthethreechloroplastdatasetsthatincludedtherestriction-sitedataset,andmostbranchesonitwererobustlysup-portedbyparsimonybootstrapanalysesofthethreedatasetscombined(Grahametal.1998).Rootingexperimentswereperformedusingvariousoutgroups(bothindividuallyandincombination)cho-sentorepresentcloselyanddistantlyrelatedtaxainthecommelinoidmonocotsandelsewhere.Arti®cialout-groupsequenceswerealsoconstructedtomimicrealoutgroupsthathavehadtheirphylogeneticsignalcom-pletelyerodedbythepassageoftime.TheserandomsequencesweregeneratedusingMacCladeversion3.07(Maddison,W.P.andMaddison,D.R.,1992),withbasefrequenciesdeterminedfromabroadsampleofmono-cots(A,28%;C,18%;G,21%;T,32%).Nosigni®cantheterogeneityinbasefrequencieswasdetectedacrosspairwisecomparisonsofthetaxa(0.999,df144).Areducedtwo-genematrixforPon-tederiaceaecomprisingthecharactersvariableinthefamilywasappendedtorandomsequencesofthesamelength.Thesecharacterswereconsideredbecausetheyallhavethepotentialtobeparsimony-informativeinthecontextofdatasetsthatincludeoutgrouptaxa.Onesetofparsimonyanalyseswasperformedtoexploretheextenttowhichrandomoutgroupsequencespreferentiallyrootonlongbranches.Theunrootedto-pologyofPontederiaceaethatwechosefromGrahametal.(1998)wasenforcedasabackboneconstraintinaseriesofparsimony-basedBranch-and-Boundsearchesthatpermittedindividualrandomoutgroupstoattachtooneormoreoptimalrootposition(s).Thefrequencyoffavoredrootingsoneachbranchintheunrootedingrouptopologywasnoted,withfractionalscoresassignedwhenarandomsequencehitmultipleoptimalbranches.Variationinthedegreeofsuboptimalityacrossthepossiblerootsfortheingrouptopologywasalsoas-sessedforavarietyofrealandrandomoutgroupcom-binations.Treescoresfordifferentrootingswereesti-matedforindividualrandomoutgroups(maximumpar-simonyonly)andusingavarietyofrealoutgroups(maximumparsimonyandlikelihood).The45possiblerootsofthe24-taxoningrouptopologyweregeneratedusingthe``Allrootings''optioninMacClade.There-sultingNEXUStext®leswereeditedtoincludedifferentoutgrouppossibilities.Fortheparsimonyanalyses,treescoresforthedifferentpossiblerootpositionswerede-terminedwithallcharactersequallyweightedandun-ordered.Modelsforthelikelihood-basedrootinganal-yseswerechosenusingthehierarchicallikelihoodratiotestsdescribedinHuelsenbeckandCrandall(1997).ThehierarchyofmodelstestedwasthesameastheexamplegiveninHuelsenbeckandCrandall(p.453;®g.4),ex-ceptthat(1)the``Generaltime-reversible''(GTR)mod-elwassubstitutedfortheHKY85modelusedthere;(2)amodelwithaclockwasnotassessed;and(3)the®nalpairedhypothesescomparedhereconsideredmodelswithandwithoutinvariablesites.Modelparameterswereestimatedfromthedataus-ingtheunrootedtopologyofPontederiaceaeforthecombinedtwo-genedatasetandalsoseparatelyforeachindividualgene.Thesemodelsandparameterestimateswereusedinallsubsequentlikelihoodanalysesthatin-cludedthe(real)outgroups.Themodelschosenusingthelikelihoodratiotestswerethe``GTRI''modelfortheLdatasetandthecombinedLandFdataset,andthe``GTR''modelforthedataset(0.01forallsigni®cantlydifferentcom-parisonswithBonferonnicorrectionsperformed;resultsnotshown).Theformermodelaccommodatesunequalbasecomposition,theproportionofinvariablecharac-ters(I),andnonuniformsubstitutionrateswithinandbetweennucleotidecharacters(theGTRmatrixwasusedtoaddressunevencharacter-statetransitionrates,andthegamma[]shapedistributionparameter[]wasusedtoaccommodateamong-siteratevariation);thelat-termodeldiffersonlybynotdirectlyaccountingforinvariablesites.Evidencefortheexistenceofhistoricalsignalinrealoutgroupsconcerningthepositionoftherootoftheingrouptreehascomefromcomparisonsamongrealandrandomoutgroupsequencesofthedecreaseinparsi-monyobservedbetweentheoptimalandnextbestroot-ings(MiyamotoandBoyle1989).Weextendthislogicbyexaminingallsuboptimalrootingsforrandomout-groupsequences.Foreachoutgroupsequence,parsi-monyscoresforallpossiblerootingswerecomparedwiththatoftheoptimalroot.Theresultingrootingpen-altiesrequiredtoplacetherootinsuboptimallocations(thescoreofthesuboptimalrootlocationminusscoreoftheoptimalrootlocation)wereranked,withtiedranksbeingbrokenarbitrarily.Themeanandstandarddeviationintherootingpenaltyacrosstherandomout-groupsforeachrankwerecomparedwiththepenaltyforthecorrespondingrankfortherealoutgroups.Foreachrealoutgrouporoutgroupcombinationconsideredhere,Shimodaira-HasegawatestswithRELL(resam-plingestimatedlog-likelihood)estimatesofthetestdis-tribution(ShimodairaandHasegawa1999andseeGold-man,Anderson,andRodrigo2000)werealsoperformedtoassessthenullhypothesisthatall45possiblerootingsofthePontederiaceaesubtreewereequallygoodexpla-nationsofthedata.PlacementofPontederiaceaeintheMonocotsThephylogenyofthemonocotsinferredhereusingtwochloroplastgenes(®g.1)indicatesrelationshipsthatarebroadlysimilartothoseseeninotherrecentstudies(e.g.,Chaseetal.1995,2000;Davisetal.1998).Aswiththosestudies,itsuffersfrompoorbootstrapsupportforthemajorityofthebackboneoftheinferredphylog-eny.Thereisgoodsupportformostofthesampledmonocotorders,asde®nedbyAPG(1998)andChaseetal.(2000),withintheverycoarselimitsofourtaxonsampling(table1and®g.1).ThetwosampledmembersofAsparagalesaredepictedasthesister-groupofthecommelinoidmonocots.Inlinewiththesestudies,our 1772Grahametal.Table1SourcesofMonocotDNASequencesEmployedintheCurrentStudy Order(Chaseetal.2000),ExemplarFamilyVoucher(orcitationdetails)GenBankAccessionNo. AcoruscalamusAcoraceaeDenverBotanicGard.,CO,novoucher;RGO97-149French232,CHRBF(AY007647)L(M91625GymnostachysancepsSagittarialatifoliaSpathiphyllumwallisiiWJKress42-3417Howard4325,FTGSCHBarretts.n.,TRTDHLess.n.,CONNMWChase210,NCUF(AY007653)L(M91629F(AY007657)L(L08767F(AY007658);L(AJ235807CaryotamitisArecaceaeSWGraham1020,TRT(Gautetal.1992)F(U79227)L(M81811CyanastrumcordifoliumNarcissuselegansAmaryllidaceaeSWGraham&SCHBarrett2,TRTSCHBarrett1434,TRTMWChase617,KF(U79228);L(U41572F(U79216);L(AF11697224taxaAnigozanthossp.A.¯avidusHanguanamalayanaPhilydrumlanuginosumTradescantiazebrina(VoucherinformationinKohnetal.1996)MGSimpson6-VI-95A,SDSUMWChase159,NCUSirirugsas.n.PJRudalls.n.,KSWGraham&SCHBarrett1,TRTSWGraham1030,TRTJCortesi7324,GMUFF(U41598-U41622L(U41573-U41597F(U79226)L(AJ404843F(AY007654)L(AJ404842F(U41622L(U41596F(U79229)L(L05042DioscoreabulbiferaDioscoreaceaeEPOBiology,U.Colorado,Boulder,CO,novoucher;RGO97-151(Katoetal.1995)F(AY007652)L(D28327Taccachantrieri...........TaccaceaeMWChase175,NCUF(AY007659);L(AJ235810CyclanthusbipartitusPandanusutilisP.veitchiiHammel15585Hahn6898Duvall19920301,UCRF(AY007650);L(AY007660)F(AY007656)L(M91632LiliumsuperbumLiliaceaeMWChase112,NCUF:(AY007655);L(L12682AnanascomosusCarexechinochloeSWGraham1000,TRT[Duvalletal.1993]AMuasya1051,K(A.M.Muasyaetal.,personalcommunication)F(U79225)L(L19977F(AF191810L(Y12997 RootingPhylogeneticTrees1773Table1 Order(Chaseetal.2000),ExemplarFamilyVoucher(orcitationdetails)GenBankAccessionNo. FlagellariaindicaFlagellariaceae(LGClarketal.,personalcommunication)MWChase206,NCUF(U22008L(L12678OryzasativaPoaceae(Hiratsukaetal.1989)Genesfromcompletechloroplastgenome(X15901TyphaangustifoliaT.latifoliaTyphaceaeSWGraham1040,TRTBradley24974,GMUFF(U79230)L(M91634ZeamaysPoaceae(Maieretal.1995)Genesfromcompletechloroplastgenome(X86563Cannasp.C.indicaCannaceaeSWGraham1010,TRTRKGodfrey57928F(AY007649)L(L.Prince,unpublishedsequence)UnassignedcommelinoidsCalectasiaintermediaDasypogonaceaeMWChase456,KF(AY007648);L(Chaseetal.1995,nonumberDasypogonhookeriD.bromelifoliusDasypogonaceaeMWChase430PJRudalls.n.,KF(AY007651)L(Chaseetal.1995,nonumber Previouslypublishedsequences.datasupporttherecentinclusion(APG1998)ofDasy-pogonaceaeandHanguanaceaewithinthecommelinoidmonocots,withmoderatebootstrapsupport(72%).Theformerfamilyispoorlysupportedasthesister-groupofZingiberales,thelatterasthesister-groupofCommeli-naceaeintheCommelinales.Asinotherstudieswithbroadertaxonandgenesampling,wedidnotinferstronglysupportedrelationshipsoftaxaatthebaseofthecommelinoidmonocots,includingthoseinvolvingthepalms,pineapplesandcattails(Arecaceae,Brome-liaceae,andTyphaceae,respectively).TheorderCom-melinales,whichincludesPontederiaceaeandfourotherfamiliesinthemostrecentordinaltreatmentofChaseetal.(2000),®ndsfurthercorroborationhere,butwithpoorsupport(33%).Philydraceae,representedherebyPhilydrumlanuginosum,isresolvedasthesister-groupofPontederiaceae,withonly38%bootstrapsupport.LongBranchesandtheRootofPontederiaceaePontederiaceaeisstronglysupportedasmonophy-letic(®g.1),aswasfoundpreviouslybyGrahamandBarrett(1995).RelationshipswithinthefamilyintheanalysisofmonocotphylogenyareessentiallyidenticaltothosefoundbyGrahametal.(1998)usingthesamegenesandfeweroutgrouptaxa.Mostbrancheswithinthefamilyarewellsupportedbybootstrapanalysiswhenoutgroupsareexcludedfromanalysis(seeGrahametal.1998,®g.2).However,theoptimalpositionoftherootofthefamilyintheparsimonysearch,betweenandtheothertaxainthefamily,ispoorlysupportedbybootstrapanalysis(®g.1;supportvaluesexcludedformostbranchesabovetherootinPontederiaceaeforclarityÐseeGrahametal.1998forbootstrapvalues),asinpreviousstudiesofthefamilywithfeweroutgrouptaxa.Morethan70%ofthe``hits''byrandomoutgroupswereonthethreelongestbrancheswithintheunrootedingroup,whichrangeinlengthfrom16±20stepsunderACCTRANoptimization(thesethreebranchesattracted16%,20%and35%ofthehits,respectively;®g.2).Wheeler(1990)predictedthattheprobabilityofaran-domsequencerootingonagiveningroupbranchshouldbeproportionaltoitslength.Thisrelationshipholdsfortheterminalbranchesin®gure2(correlationbetweenlengthandnumberofhits:0.001,df[24terminalbranches22).However,therela-tionshipbreaksdowncompletelyforthenonterminalbranches(0.142,dfdf22]519).Mostnonterminalbranchesthatwereofthesameorderoflengthasthelongestthreehadfewornohitsfromtherandomsequencessampledhere(®g.2).Thesedis-favoredbranchesincludetheonepreferredbytherealoutgroupsundermaximumparsimony(®g.1)andtheneighboringbranchfavoredundermaximumlikelihoodformostoutgroupcombinations(seebelow).Thesetwobrancheshaveaparsimony-basedlengthof15and6steps,respectively,ontheunrootedtreeunderACCT-RANoptimization.Arangeofotherbrancheswerealsofavoredasoptimalrootinglocationsbytherandomout-groupsequences,buteachoftheseaccountedforless 1774Grahametal. .1.ÐPhylogenyofthemonocotyledonsbasedonparsimonyanalysisofFandLsequencedatacombined.Thetree(oneofeightmost-parsimonious)isrepresentedasaphylogram(length2743steps;consistencyindex0.457;retentionindex0.632)withbranchlengthscomputedusingACCTRANoptimization.Twoofthreebranchesnotfoundacrossallmost-parsimonioustreesinvolveshortbranchesinPontederiaceae(notshown);athird(indicatedwithanarrow)involvestwodifferentarrangementsnearthebaseofthecommelinoidmonocotclade.BootstrapvaluesareindicatedbesidebranchesbutareexcludedfromallbranchesinPontederiaceae,exceptforthetwobranchesaroundtheroot.TaxonnamesfollowChaseetal.(2000).than5%ofthetotalhits.Intotal,lessthan10%ofthebranchesfavoredasrootlocationsbyrandomoutgroupsoccurredonnonterminalbranchesontheunrootedto-pology,despitetheirrepresentingmorethan40%ofthetotaltreelengthandcomprisinghalfofthetoptenbranchlengthsontheunrootedtopologyunderACCTRANRootingExperimentsUsingthetwo-genedataset,parsimony-andlike-lihood-basedmethodslargelyconcurredontheextentofsignalintheoutgrouptaxaforrootingtheingroupwhenthefourclosestoutgrouptaxawereusedtoroottheingrouptree(®gs.3and4;notethattheoutgroupsub-treeofCommelinalestaxausedisthatimpliedin®g.1).Rootingpenaltieswerehighlycorrelatedbetweentheparsimonyandlikelihoodoptimalitycriteria,atleastforthecaseconsideredhere(thefourCommelinalestaxaandbothgenescombined;table2).Mostoftheout-groupsconsideredintable2hadthesameoptimalrootlocationundermaximumlikelihood(indicatedin®gs.2and5),butforseveraloutgroupswherethiswasnottheoptimalrootundermaximumlikelihooditwasonlymarginallysuboptimal(resultsnotshown).Thisrootwasalsoonlymarginallysuboptimalforthemaximumparsimonycaseconsideredintable2(threestepslonger,thethirdbestrootunderparsimony).AstrongerparsimonypenaltyforrealoutgroupscomparedwithrandomoutgroupswasusedbyMiya-motoandBoyle(1989)asevidenceofrootingsignal,althoughtheyfocusedonlyontheoptimaland®rstsub-optimalrootlocations.Here,theparsimonypenaltiesre-quiredtoforcetheroottothetwenty-secondranked(andworse)rootingsfortherandomoutgroupsequenceswereonaveragemorethantwostandarddeviationsclos-ertooptimallengththanthoseobservedusingthefourrealoutgroupstogether(®g.3).However,notwelldistinguishedfromrandomsequences(®g.3).Theotherrealoutgroupsconsideredhere(table2)large-lyperformedbetweenthesetwoextremes.Oneinter-mediateexample,acompositeoutgroupinvolvingCom-melinaceaeandHanguanaceae,isshownin®gure3. RootingPhylogeneticTrees1775 .2.ÐPropensityforrandomoutgroupstorootondifferentbrancheswithinPontederiaceae.BranchlengthsforPontederiaceaewerecalculatedfromthecombinedFandLdata;thebarsateachnodeshowtheminimum,average,andmaximumbranchlengthsacrossthemost-parsimoniousreconstructions.Circledvaluesindicatethepercentageofoptimalrootingsby100randomoutgroupsontheunrootedtopologyofPontederiaceae.Fractionalvalueswereappliedtorandomoutgroupsthatrootedonmultiplebranches.Emptycirclesindicatethatsomehitswerefound,butwithafrequency0.5%.Thethreelabeledvalues(a,b,andc)arethetophitsbytherandomoutgroups.Therootingistheoptimaloneunderparsimony;theoptimalrootinvariousmaximumlikelihoodanalyses(seetext)isalsoindicated.Outgrouptaxawereexcludedfromthe®gureforclarity;thebranchsubtendingtheingroupisfromtherootofPontederiaceaetothe®rstoutgroupnodeonly. .3.ÐRankeddecreaseinparsimonywhendifferentoutgroups(realandrandom)areforcedtosuboptimalrootlocationsinPontederi-aceae.Therealoutgroupsconsideredare([([Anigozanthos,[Hanguana,Tradescantia]]]),([]),and(Allcasesarerankedindependently.Fortherandomoutgroups,themeanpenalty(plusorminustwostandarddeviations)wascalculatedacrosstreescoresateachrank.Thesamegeneralpatternsheldinthelikelihoodanalyses.Ofthoseoutgroupcombinationsexamined,theexperimentinvolvingthecompositeCommelinalesout-group(Commelinaceae,Philydraceae,HaemodoraceaeandHanguanaceae)rejectedasubstantialfractionofsuboptimalrootlocations(20of44suboptimalroots;®gs.4and5;table2).Ingeneral,therootsrejectedbythedifferentoutgrouppermutationsconsideredherewerealsosubsetsofthoserejectedusingthecompositeCommelinalesoutgroupsandbothgenescombined(bracketedvaluesintable2).Someotheroutgroupsper-formedaswellasthisorbetter(summarizedin®gs.4and5;table2),includingoneofthefourCommelinalestaxawhenusedindividuallyasanoutgroupfamily(,representingHaemodoraceae).FortheotherthreefamiliesofCommelinales(Commelinaceae,Han- 1776Grahametal. .4.ÐRankeddecreaseinlikelihoodwhenvariousoutgroupcombinationsareforcedtosuboptimalrootlocationsinPontederiaceae.Allcasesarerankedindependently.AsterisksindicatethelnLscoresabovewhichrootsaresigni®cantlysuboptimalforthatoutgroupset,asinferredusingaseriesofShimodaira-Hasegawatests(table2and®g.5).Thetopthreebranchesfavoredbytherandomoutgroupsundertheparsimonycriterion(labelsa,b,andc;®g.2)alsoareindicatedforthreeoftheoutgroupsets.Table2ScoreCorrelationandAbilitytoDiscriminateSuboptimalRootsofPontederiaceaeforVariousOutgroups,DataSetsandOptimalityCriteria CriterionDataSetNo.of )..............................Ani,[Han,Tra]])........Ani,[Han,Tra]])........Ani,[Han,Tra]])....MP(Fitch)13(13)20(18)5(3)4(4)6(6)24(18)2(2)14(14) Correlation(acrossrootpositions)ofscoreswiththoseseenusingthetwo-genedataset,theoutgroupcombinationcombinationPhi,[Ani,[Han,Tra]]]),andtheGTRIlikelihoodmodel.Twentyrootpositionswereworsethantheoptimalrootforthelattercase.(Commelinaceae);MLMaximumlikelihood;andMPMaximumparsimony.0.001.df(45ingroupbranches43.Bonferonnicorrectionswereperformedforthesecorrelationtests.AssessedusingShimodaira-Hasegawatests(seetext).Numbersinparenthesesindicaterootlocationsthatarerejectedbythatcaseandalsobythecasewiththetwo-genedataset,theoutgroupcombination([([Ani,[Han,Tra]]]),andtheIlikelihoodmodel.);AsparagalesguanaceaeandPhilydraceae,representedby,respectively),onlyahandfulofcandidaterootsinPontederiaceaecouldberejectedusingtheexemplargeneraindividuallyasout-groups(table2).However,incombination,theexem-plarsfromthesefamiliescanrejectasmanyormorebranchesasthefouroutgroupscombined:oneexample,thecombinationofCommelinaceaeandHanguanaceae,isshownin®gures4and5andtable2.Twoofthemostdistantlyrelatedoutgroupcasesexamined(byitselfandthetworepresentativesofPoaceaeconsideredtogether)werenotabletorejectanyrootcandidates(table2and®g.4).ThetwoPoaceaetaxaareonrelativelylongbrancheswithinthecom-melinoidmonocots(®g.1),andispresumedtobethesister-groupoftherestofthemonocotshere.However,onenoncommelinoidoutgroupcaseconsid-eredhere(thetworepresentativesofAsparagalestakentogether)wasabletorejectasubstantialfractionoftherootlocations(13of20)rejectedbythefouroutgroupsusedtogether(table2and®g.4).WhentherootingexperimentwiththefourCom-melinalestaxawasrepeatedusingindividualgenes,asmallersubsetofthesecandidaterootscouldberejectedthanwaspossiblewiththetwogenesconsideredto-gether(table2).Thepenaltiesobservedusingeachgene RootingPhylogeneticTrees1777 .5.ÐBranchesthataresigni®cantlysuboptimalinalikelihoodanalysis(Shimodaira-Hasegawatests;0.05;table2and®g.4)ofthecombinedDNAsequencedatawhenoutgroupsareforcedtodifferentrootlocationsinPontederiaceae.Thethreedarkcirclesindicatelocationsrejectedwhenthecompositeoutgroup((Anigozanthos,[Tradescantia,Hanguana]])wasemployed;the®velightercirclesindicatethoserejectedwhenthecompositeoutgroup()wasemployed;branchesintheintersectionofcircleswererejectedinbothinstances.Branchlengthsareparsimony-based.Theoptimalparsimony-andlikelihood-basedrootsforthecasewiththefouroutgrouptaxa,andthethreebranchesfavoredbythemajorityofrandomoutgroupswithmaximumparsimonyanalysis(labelsa,b,andc;®g.2)alsoareindicated.individuallywerenonethelesshighlycorrelatedtothoseobservedusingbothgenestogether(table2).Asimilarlystrongcorrelationtothiscasewasalsoseenwithregardtorootingpenaltiesforallpermutationsofoutgrouptaxa,genes,andoptimalitycriteriaconsidered.Thus,thiscorrelationwasapparentevenwherenoorfewrootscouldberejectedusinglikelihood(e.g.,forthefourCommelinalesconsideredtogetherbutusingLalone,orfororPoaceaeusingbothgenes;table2).Incomparisontootheroutgroups,theslopesofthelikeli-hood-penaltysurfacewere¯attestforthelattertwocas-es(Poaceaeand;®g.4).MonocotPhylogenyandthePlacementofOurdatacorroboratethebroadpatternofphylo-geneticrelationshipsindicatedinrecentstudies(e.g.,Givnishetal.1999;Chaseetal.2000;Stevensonetal.2000).Pontederiaceaebelongsinthecommelinoidmonocots,nearCommelinaceae,Haemodoraceae,Han-guanaceaeandPhilydraceae.Aswasfoundintheseear-lierstudies,thepreciseinterrelationshipsofthese®vefamiliestooneanotherwaspoorlysupported.SignalandBiasintheClosestRelativesofPontederiaceaeforRootingtheFamilyPoorsupportfortherootofPontederiaceaeisnotafunctionofuncertainrelationshipswithinthefamilybecausethetwoingroupbranchescreatedattherootsplitusingtherealoutgroups(®g.1)correspondtoasingleverystronglysupportedbranchintheunrootedversionofthetree(Grahametal.1998).Thislackofrobustrootingsupportissimilartothatobservedwithrandomoutgroups.Todemonstratethis,asubsetofran-domoutgroupsequenceswereemployedasoutgroupsinparsimony-basedbootstrapanalyses.Forsevenoutof10randomsequencesexamined,nosingleingroupbranchwasfavoredby50%ofbootstrapreplicates(resultsnotshown).Supportfortheremainingthreecas-eswasinthe60%±85%range,representingonlymod-eratelyrobustrootingsbyrandomoutgroups.Thus,al-thoughlong-branchattractionmayin¯uencerootingde-cisionsinPontederiaceae(®g.2),itisunlikelytoresultinarobustlysupportedwronganswer,atleastatthecurrentlevelofnucleotidesampling.ThecomparablebootstrapsupportforrootingsseenwithrealandrandomoutgroupsbegsthequestionofwhetheranysignalremainsintherealoutgroupsforrootingPontederiaceae.Comparisonsofthedegreeofsuboptimalityofcorrespondinglyrankedalternativerootingsforrealandrandomoutgroupsequencesdem-onstratethattherealoutgroupsdopossesssigni®canthistoricalsignal(®g.3).TheideaofinvestigatingthevariousrootingpossibilitiesofoutgroupsonunrootedtreesgoesbacktoLundberg(1972).ThisapproachwasrecommendedbyNixonandCarpenter(1993)andSwoffordetal.(1996)whenitissuspectedthatlong 1778Grahametal.branchesconnectingtheingroupandoutgroupsmaydis-torttheestimationofphylogeneticrelationshipswithintheingroup,aboveandbeyonduncertaintyoverthepointofrootattachment.MiyamotoandBoyle(1989)werethe®rsttoexploretherootingpotentialofrandomoutgroups(using``Lundbergrootingexperiments'')asabaselineforcomparingthebehaviorofdivergentrealsequences.TheyciteRohlfandFisher(1968)inthisregard,butthelatterauthorsusedrandomsequencesforaquitedifferentpurpose.RecentstudiesthatexaminetherootingpotentialofrandomversusrealsequencesincludeSullivanandSwofford(1997)andStillerandHall(1999).Theparsimony-basedtestusedheretoas-sessrootingsignalisbasedontheonesuggestedbyMiyamotoandBoyle(1989),butwehadtoexaminerootingssubstantiallylessoptimalthanthe®rstsubop-timalroottodemonstratephylogeneticsignalamongtherealoutgroupsforrootingPontederiaceae(®g.3).Likelihood-basedexplorationsofthepossibleroot-ings(table2and®gs.4and5)furthercon®rmthatmanycandidaterootlocationsinthefamilyaresigni®cantlypoorerexplanationsofthedatathantheoptimalcases(®gs.4and5).Bothparsimony-andlikelihood-basedmethodsusingthefourclosestoutgroupfamiliesintan-demsuggestthatroughlyhalfofallpossiblerootloca-tionswithinthefamilycanberejectedwithcon®dencebytheseoutgroups.Severallinesofreasoninglendcredencetotheop-timalrootsfavoredbytherealoutgroups,despitetheirlackofrobustsupportbybootstrapanalysisortheShi-modaira-Hasegawatests.First,theoptimalrootsfavoredbytherealoutgroupsundermaximumparsimonyandlikelihood(®gs.2and5)arepreferredbynoneoftherandomoutgroupsconsideredhere(®g.2).Second,al-thoughrealoutgroupsdonotrejectthethreerootsfa-voredbymostoftherandomoutgroups,atanalphalevelof0.05,theserootingpossibilitiesaregenerallynotinthetop20%ofrootingcandidatesfavoredbyrealoutgroups(®g.4).Third,asubstantialfractionofthehomoplasyaccumulatedindependentlyindifferentout-groups(thatfractionofhomoplasynotsharedthroughcommonancestry)shoulderodesignalconcerningtherootofPontederiaceaebutindifferentwaysineachter-minaloutgrouplineage.Althoughthesignalthatwede-tectedwasweak(sometimestothepointofnotbeingabletorejectanyalternativerootingposition),thepat-ternofrootingpreferencewasstillhighlyconsistentamongrealoutgroups(seethecorrelationslistedintable2),despitethesesubstantialandpresumablyindependentopportunitiesineachlineagefortheaccumulationofmisinformativecharacters(e.g.,Swoffordetal.1996),andforthelossofinformativecharactersduetomultiplehits.Becausetherehasbeenlittledetectableerosionintherootingcorrelationsduetosubstitutionsontheter-minalCommelinalesbranches,itseemsreasonabletoassumethattherootingsignalintotoamongtheseout-groupshasnotbeensubstantiallybiasedbylong-brancheffectsinthesenseofFelsenstein(1978)andHendyandPenny(1989).OnepreviouslypublishedrootofPontederiaceaebasedonarestriction-sitedataset(Kohnetal.1996)wasonabranchfavoredbymanyoftherandomout-groups(labeled``c''in®g.2).ThepossibilitythatthiswasanartifactualrootingwasnotedbyKohnetal.Ithasbeensuggestedthatrestriction-sitedatagenerallyhaveminimalutilityatdeeperlevelsofphylogeneticanalysis(OlmsteadandPalmer1994;SoltisandSoltis1998),andthismaybethecasewithregardtooutgroup-basedrootingofPontederiaceae.ForthepurposesoftheevolutionaryreconstructionsthatKohnetal.wereex-amining,theiroverallresultsdidnotdiffermarkedlyfromthoseperformedusingtherootindicatedwiththeDNAsequencedata(Kohnetal.1996;BarrettandGra-ham1997).However,evenifrootestimationhasbeennudgedonlyslightlyoffcourse,thismayreducetheutilityofthephylogeneticinferenceinotherreconstruc-tionsofcharacterevolution,orforotherpurposes,suchasthegenerationofclassi®cationsthatbetterre¯ectphylogeny.Pursuingamorecon®dentinferenceofrootplacementforPontederiaceaeisthusacrucialgoalforfuturestudy.FutureProspectsUsingthefouroutgrouprepresentativesinCom-melinalesasacompositeoutgroupwasmoreusefulforrejectingsuboptimalrootsthanmostoftheseoutgroupfamiliesusedindividually(table2).Forexample,whenexemplarsrepresentingCommelinaceaeorHanguana-ceaewereusedindividually,wewereabletorejectonlyafewsuboptimalrootpositions.Incombinationhow-ever,thediscriminatorypowerofthesetwofamilieswasamongtheverybestofthoseexamined(table2and®gs.4and5).BreakinglongbranchesbyfurtheradditionoftaxashouldthereforeimproveourabilitytoinfertherootofPontederiaceae.Thisgeneralresulthasbeennot-edbymanyworkers(e.g.,Graybeal1998;Hillis1998).However,Smith(1994;andseeHendyandPenny1989)pointedoutthatitisbettertohaveadenserrepresen-tationofoutgroupsinthesister-groupthantosampleheavilyinlesscloselyrelatedtaxa.Althoughweagreeinprinciplethatmultipletaxashouldbesampledinthesister-group(wherepracticalorpossible),samplingsomewhatmoredistantlyrelatedtaxaisimportanttoobecausethisshouldhelpbreakupthebranchbetweentheingrouprootandthesister-group.Samplingoutgroupsbeyondthesister-groupservesafurtherpurposeÐithelpstesttheideathatthesister-groupreallyisthesister-group.Itmaynotalwaysbeclearinadvancewhatthesister-groupis,aswasthecasehere.ThesequencesexaminedherefortheexemplarofCommelinaceae,thefamilysuggestedbypreviousmo-lecularstudiestobethesister-groupofPontederiaceae(e.g.,Chaseetal.1995)retainedlittlerobustsignalwhenusedindividuallytorootthefamily,althoughtheyperformedverywellincombinationwithanotherpoorlyperformingoutgroupfamily(Hanguanaceae).Beforewecanaddadensersamplingoftaxawithinthesister-groupandothercloserelativesofPontederiaceae,weneedtomorecon®dentlyidentifythesister-groupofthefamily.Anongoingstudyofcommelinoidmonocotphy-logenybasedonmultiplegenesisaimedataddressing RootingPhylogeneticTrees1779thisquestion.Preliminaryresults(S.W.Graham,un-publisheddata)provideextremelystrongsupportforHaemodoraceaeasthesister-groupofPontederiaceae(andintriguingly,thesinglerepresentativeofthisfamilyexaminedinthecurrentstudy,,rejectedasmanyrootpositionsasthefourCommelinalesoutgroupscombined;table2).Addingfurthergenesshouldfurtherimproveourabilitytorejectsuboptimalroots,asthecombineddiscriminatorypoweroftwogeneswassub-stantiallygreaterthanforeithergeneindividually,intermsofthenumberofingrouprootsthatcouldbere-jectedwithcon®dence(table2).FuturetaxonsamplingwithinPontederiaceaeshouldfocusonsamplingthosebranchesthatcouldnotberejectedwithcon®dence.Currentlyunsampledmem-bersofPontederiaceaearelikelytoplaceinsectorsofthetreethathavebeenrejectedasrootingcandidateshere(seeBarrettandGraham1997),andsotheseun-sampledlineagesareunlikelytoincludethepointofattachmentoftheroot.Itisnoteworthythatthecandi-datebranchesthatcannotberejectedwithcon®denceaspointsofattachmentfortherootofPontederiaceaearelargelyrestrictedtotheinteriorbackboneoftheun-rootedtree.Thisbackboneisgenerallystronglydisfa-voredbyrandomoutgrouproots,suggestingthatifthetruerootofacladeistobefoundonitsnonterminalbranches(asmaybethecasehere),long-branchattrac-tionisunlikelytogenerateanartifactualrootingthatsuggestsotherwise.Thetendencyobservedhereforran-domlong-branchestodisfavorinteriorpartsoftheun-rootedtree,andtohavenorootingpreferenceontheinteriorbackbonebasedonbranchlength,hasnotbeennotedpreviously.Itwouldbevaluabletoexplorethispropertyinotherphylogeniestodetermineifitisawidespreadphenomenon.AnalyticalSimpli®cationsandConsequencesTofacilitatetree-scoreestimationinareasonabletimeframe,parametersforthelikelihoodmodelwereestimatedusingthePontederiaceaesubtreeonly,andthesevalueswereusedinallsubsequentlikelihoodanal-yses.Toinvestigatewhetheritwouldmakeasubstantialdifferencetoestimatemodelparametersdirectly,theShimodaira-Hasegawaanalyseswererepeatedforoneoftheoutgroupcases(,withbothgenesconsid-eredsimultaneously),butwiththe``GTRI''modelparametersestimatedseparatelyforeachrootpo-sition.Estimatedmodelparameters(notshown)wereneitherfoundtodiffersubstantiallyacrossrootingsnortothoseestimatedusingthePontederiaceaesubtreealone.Thenumberandidentityofsigni®cantlysubop-timalrootsaccordingtotheShimodaira-HasegawatestswerealsofoundtobeverysimilartothoseobtainedwhenparameterswerederivedfromthePontederiaceaesubtreealone(resultsnotshown).Theuseofthisana-lyticalshortcutthereforelikelyhaslittleornoeffectonouroverallconclusions.Appropriatecorrectionstosigni®cancelevelsaremadeinShimodaira-Hasegawa'stesttoaccountforthemultiplecomparisonsbeingperformed(Goldman,An-derson,andRodrigo2000),inourcaseacrossthe45possiblerootpositions.However,multiplesetsofShi-modaira-Hasegawatestswereperformedusingdifferentoutgroupcombinations,anditwouldalsobevaluabletocorrectthesigni®cancelevelstotakeaccountofthislevelofhypothesistesting.However,becausetestsusingrelatedoutgroupsarelikelytobestronglycorrelated(ta-ble2),adjustingthealpha-leveltoaccountformultipletestsisprobablynotappropriate,anditisnotclearwhattheappropriatecorrectionwouldbe.Goldman,Anderson,andRodrigo(2000)empha-sizetheimportanceofanhonestchoiceofapriorihy-pothesistopologiestotheconclusionsgeneratedusingtheShimodaira-Hasegawatest.Weerredonthesideofbeingconservativehereandconsideredallpossibleroot-edversionsoftheunrootedtopologyofPontederiaceae.However,thetestassumesthatthetopologiesconsideredincludethetrueone(Goldman,Anderson,andRodrigo2000),andourresultsshouldbeviewedwiththecaveatthatalthoughthetrue,unrootedchloroplastphylogenyofPontederiaceaeislikelyverysimilarinshapetotheunrootedtopologyconsideredhere,itisnotguaranteedtobeidenticaltoit.Incorrectlyrootedtreesmayresultinprofoundlymisleadingevolutionaryandtaxonomicinferences,andthismaybearelativelywidespreadphenomenoninphy-logeneticstudies.Theapproachespresentedheremaybeusefulinanystudywheredistantlyrelatedoutgroupsmayleadtoartifactualorambiguousrootingsoftheingroupsubtree.Forthecasestudyexaminedhere,theavailabledatadonotyetpermitaconclusiverootingofPontederiaceae,butthegeneralpatternofrootingpref-erencesfortheoutgroupsinthecommelinoidmonocots(andbeyond)hasapparentlynotbeendegraded,eveninthefaceofthesubstantialerosionofphylogeneticsignalonlongoutgroupbranches.Ourresultshighlightsomegeneralareasforfutureresearch,includingthedifferentrootingbehaviorofrandomoutgroupsequencesonter-minalversusnonterminalingroupbranches.Thevarietyofapproachesemployedhereconcuronthenatureandstrengthofthesignalinreal,distantoutgroupsforin-ferringthepositionoftherootofPontederiaceae,andtheabilitytodiscriminateagainstsuboptimalrootlo-cationsisshowntobesubstantiallyimprovedbyaddingoutgrouptaxaandcharacters.WethankMichaelDonoghueforbringingMiya-motoandBoyle'sworkontreerootingtoourattention,MarkChase,WilliamHahn,JohnKress,ThomasLem-ieux,LindaPrince,MichaelSimpson,AlanYen,themanfromDelMonteandothersforaccesstoplantma-terialorunpublishedsequences,andtworeviewersforhelpfulsuggestionsonthemanuscript.ThisworkwasfundedinpartbyNSFgrantDEB9727025toR.G.O.andbyresearchgrantstoS.C.H.B.andS.W.G.from 1780Grahametal.LITERATURECITEDAPG(ANGIOSPERM).1998.Anordinalclassi®cationforthefamiliesof¯oweringplants.Ann.Mo.Bot.Gard.,T.J.,G.C,J.R.M,J.L,W.J.E,G.M,A.D.W,andC.W..2000.Independentandcombinedanalysesofsequencesfromallthreegenomiccompartmentsconvergeontherootof¯oweringplantphylogeny.Proc.Natl.Acad.Sci.USA,S.C.H.,andS.W.G.1997.Adaptiveradi-ationintheaquaticplantfamilyPontederiaceae:insightsfromphylogeneticanalysis.Pp.225±258T.J.GandK.J.S,eds.Molecularevolutionandadaptiveradiation.CambridgeUniversityPress,Cambridge,U.K.,K.2000.EarlyCretaceouslineagesofmonocot¯ow-eringplants.Proc.Natl.Acad.Sci.USA:4707±4711.,M.W.,M.R.D,H.G.Hetal.(11co-authors).1995.MolecularphylogeneticsofLilianae.Pp.109±137P.J.R,P.J.C,D.F.C,andC.J.H,eds.Monocotyledons:systematicsandevolution.RoyalBotanicGardens,Kew,U.K.,M.W.,D.E.S,R.G.OLMSTEADetal.(43co-authors).1993.Phylogeneticsofseedplants:ananalysisofnucleotidesequencesfromtheplastidgeneL.Ann.Mo.Bot.Gard.,M.W.,D.E.S,P.S.Setal.(13co-authors).2000.Higher-levelsystematicsofthemonocotyledons:anassessmentofcurrentknowledgeandanewclassi®cation.Pp.1±24K.L.WandD.A.M,eds.Mono-cots:systematicsandevolution.CSIRO,Melbourne.,R.M.T.,H.T.C,andP.F.Y.1985.Thefamiliesofthemonocotyledons:structure,evolutionandtaxonomy.Springer-Verlag,Berlin.,J.I.,M.P.S,D.W.S,andJ.F.WDEL.1998.Datadecisiveness,dataquality,andincongru-enceinphylogeneticanalysis:anexamplefromthemono-cotyledonsusingmitochondrialAsequences.Syst.Biol.,M.J.,andJ.A.D.2000.Seedplantphylog-eny:demiseoftheanthophytehypothesis?Curr.Biol.,M.R.,M.T.C,M.W.Cetal.(11co-authors).1993.Phylogenetichypothesesforthemonocot-yledonsconstructedfromLsequencingdata.Ann.Mo.Bot.Gard.:607±619.,J.E.,andS.C.H.B.1986.Phyloge-neticsystematicsofPontederiaceae.Syst.Bot.,J.1978.Casesinwhichparsimonyandcompat-ibilitymethodswillbepositivelymisleading.Syst.Zool.:401±410.ÐÐÐ.1985.Con®dencelimitsonphylogenies:anapproachusingthebootstrap.Evolution,B.S.,S.V.M,W.D.C,andM.T.C1992.Relativeratesofnucleotidesubstitutionatthelocusofmonocotyledonousplants.J.Mol.Evol.,T.J.,T.M.E,J.C.P,andK.J.S1999.PolyphylyandconvergentmorphologicalevolutioninCommelinalesandCommelinidae:evidencefromsequencedata.Mol.Phyl.Evol.:360±385.,N.,J.P.A,andA.G.R.2000.Likelihood-basedtestsoftopologiesinphylogenetics.Syst.:652±670.,S.W.,andS.C.H.B.1995.PhylogeneticsystematicsofPontederiales:implicationsforbreeding-sys-temevolution.Pp.415±441P.J.R,P.J.CD.F.C,andC.J.H,eds.Monocotyledons:systematicsandevolution.RoyalBotanicGardens,Kew,,S.W.,J.R.K,B.R.M,J.E.EWALDER,andS.C.H.B.1998.Phylogeneticcon-gruenceanddiscordanceamongonemorphologicalandthreemoleculardatasetsfromPontederiaceae.Syst.Biol.,S.W.,andR.G.OLMSTEAD.2000.Utilityof17chloroplastgenesforinferringthephylogenyofthebasalangiosperms.Am.J.Bot.,S.W.,P.A.R,A.C.E.B,andR.G..2000.MicrostructuralchangesinnoncodingchloroplastDNA:interpretation,evolution,andutilityofindelsandinversionsinbasalangiospermphylogeneticin-ference.Int.J.PlantSci.,A.1998.Isitbettertoaddtaxaorcharacterstoadif®cultphylogeneticproblem?Syst.Biol.:9±17.,M.D.,andD.PENNY.1989.Aframeworkforthequantitativestudyofevolutionarytrees.Syst.Zool.,D.M.1996.Inferringcomplexphylogenies.NatureÐÐÐ.1998.Taxonomicsampling,phylogeneticaccuracy,andinvestigatorbias.Syst.Biol.,J.,H.S,R.Wetal.(16co-au-thors).1989.Thecompletesequenceoftherice(Oryzasa-)chloroplastgenome:intermolecularrecombinationbe-tweendistincttRNAgenesaccountsforamajorplastidDNAinversionduringtheevolutionofthecereals.Mol.Gen.Genet.,J.P.,J.P.B,andA.M.L.2002.Inferringtherootofaphylogenetictree.Syst.Biol.,J.P.,andK.A.C.1997.Phylogenyestimationandhypothesistestingusingmaximumlikeli-hood.Ann.Rev.Ecol.Syst.:473±466.,H.,R.T,F.H.U,andS.K.1995.MolecularsystematicsoftheTrilliaceaesensulatoasin-ferredfromLsequencedata.Mol.Phyl.Evol.:184±,J.R.,S.W.G,B.M,J.J.D,andS.C.H.B.1996.ReconstructionoftheevolutionofreproductivecharactersinPontederiaceaeusingphyloge-neticevidencefromchloroplastDNArestriction-sitevari-ation.Evolution:1454±1469.,J.G.1972.Wagnernetworksandancestors.Syst.:398±413.,J.,andG.A.H.1997.EscapingfromtheFelsensteinZonebydetectinglongbranchesinphylo-geneticanalysis.Mol.Phyl.Evol.,J.,G.A.H,andR.J.T.1996.RelativeApparentSynapomorphyAnalysis(RASA)I:thestatisticalmeasurementofphylogeneticsignal.Mol.Biol.:749±757.ÐÐÐ.1998.Optimaloutgroupanalysis.Biol.J.Linn.Soc.,D.R.1991.Thediscoveryandimportanceofmul-tipleislandsofmost-parsimonioustrees.Syst.Zool.,D.R.,M.R,andD.L.S.1992.GeographicoriginsofhumanmitochondrialDNA:phylo-geneticevidencefromcontrolregionsequences.Syst.Biol.,W.P.,andD.R.M.1992.MacClade:anal-ysisofphylogenyandcharacterevolution.Version3.0. RootingPhylogeneticTrees1781ComputerprogramanddocumentationdistributedbySin-auer,Sunderland,Mass.,R.M.,K.N,G.L.IGLOI,andH.K1995.Completesequenceofthemaizechloroplastgenome:genecontent,hotspotsofdivergenceand®netuningofge-neticinformationbytranscriptediting.J.Mol.Biol.614±628.,M.M.,andS.M.B.1989.ThepotentialimportanceofmitochondrialDNAsequencedatatoeuthe-rianmammalphylogeny.Pp.437±450B.F,K.,andH.J,eds.Thehierarchyoflife.El-sevierPress,Amsterdam.,R.2002.Aphylogenyinferredfromlarge-subunit(26S)ribosomalDNAsequencessuggeststhatBurmanni-alesarepolyphyletic.Aust.Syst.Bot.:19±28.,K.C.,andJ.M.C.1993.Onoutgroups.Cla-:413±426.,R.G.,andJ.D.P.1994.ChloroplastDNAsystematics:areviewofmethodsanddataanalysis.Am.J.,R.G.,andJ.A.S.1994.Combiningdatainphylogeneticsystematics:anempiricalapproachusingthreemoleculardatasetsintheSolanaceae.Syst.Biol.467±481.,H.,andP.F.1999.Therootingoftheuni-versaltreeoflifeisnotreliable.J.Mol.Evol.:509±523.,Y.-L.,J.L,B.A.W,F.BRONI,andO.D.2001.WastheANITArootingoftheangiospermphylogenyaffectedbylong-branchat-traction?Mol.Biol.Evol.,F.J.,andD.R.F.1968.Testsforhierarchicalstructureinrandomdatasets.Syst.Zool.:407±412.,H.,andM.HASEGAWA.1999.Multiplecompar-isonsoflog-likelihoodswithapplicationstophylogeneticinference.Mol.Biol.Evol.:1114±1116.,M.P.,C.P.R,J.V.F,andJ.W..2002.LimitationsofRelativeApparentSynapo-morphyAnalysis(RASA)formeasuringphylogeneticsig-nal.Mol.Biol.Evol.:14±23.,M.G.1987.PollenultrastructureofthePontederi-aceae:evidenceforexinehomologywiththeHaemodora-ceae.Grana,A.B.1994.Rootingmoleculartrees:problemsandstrategies.Biol.J.Linn.Soc.:279±292.,D.E.,andP.S.S.1998.Choosinganapproachandanappropriategeneforphylogeneticanalysis.Pp.1±P.S.SOLTIS,D.E.S,andJ.J.D,eds.MolecularsystematicsofplantsII:DNAsequencing.Klu-wer,Boston,Mass.,D.E.,P.S.S,M.E.M,M.W.C,V.,S.B.H,andC.M.M.1998.In-ferringcomplexphylogeniesusingparsimony:anempiricalapproachusingthreelargeDNAdatasetsforangiosperms.Syst.Biol.:32±42.,H.,andU.H.1989.Embryologisch-syste-matischeUntersuchungenanHaemodoraceen.Bot.Jahrb.,D.W.,J.I.D,J.V.F,C.R.,M.P.S,andC.D.S.2000.Aphy-logeneticanalysisofthemonocotyledonsbasedonmor-phologicalandmolecularcharactersets,withcommentsontheplacementofandHydatellaceae.Pp.17±24K.L.WandD.A.M,eds.Monocots:sys-tematicsandevolution.CSIRO,Melbourne.,J.W.,andB.D.H.1999.Long-branchattractionandtherDNAmodelofearlyeukaryoticevolution.Mol.Biol.Evol.:1270±1279.,J.,andD.L.S.1997.Areguineapigsrodents?Theimportanceofadequatemodelsinmolecularphylogenetics.J.Mammal.Evol.,D.L.1999.PAUP*4.0(betaversion).ComputerprogramanddocumentationdistributedbySinauer,Sun-derland,Mass.,D.L.,G.J.O,P.J.W,andD.M..1996.Phylogeneticinference.Pp.407±514D.M.,C.M,andB.K.M,eds.Molecularsys-tematics.2ndedition.Sinauer,Sunderland,Mass.,D.L.,andS.P.1999.Taxonsamplingrevisited.:299±300.,H.-J.1994.UntersuchungenzumBauderKeimp¯an-zenderPhilydraceaeundPontederiaceae(Monocotyledo-neae).Sendtnera,H.-J.1995.Seedlingsandsystematicsinmonocoty-ledons.Pp.303±352P.J.R,P.J.CRIBB,D.F.,andC.J.H,eds.Monocotyledons:sys-tematicsandevolution.RoyalBotanicGardens,Kew,U.K.,W.C.1990.Nucleicacidsequencephylogenyandrandomoutgroups.CladisticsLIZABETHELLOGG,reviewingeditorAcceptedJune3,2002