牛科（哺乳纲偶蹄目）动物与生境利用有关的适应形态模式

牛科（哺乳纲：偶蹄目）动物与生境利用有关的适应形态模式动物52(6):971987,2006ActaZoologicaSinicaCharacterizingadaptivemorphologicalpatternsrelatedtohabitatuseandbodymassinBovidae(Mammalia:Artiodactyla)ManuelMENDOZA,PaulPALMQVISTDepartamentodeEcologiaYGeologia,FacultaddeCiencias,UniversidaddeMdlaga,Mdlaga29071,SpainAbstractAmultivariateanalysisofthepostcranialskeletonofextantbovidsrevealspatternsofosteologicalfeaturesindicativeofecologicaladaptationsforhabitatuseandbodysizeThemorphologicalpatternsthatcharacterizethepostcranialanatomyofbovidspeciesfromeachhabitattypewereidentifiedwithstepwisecanonicaldiscriminantanalysisanddecisiontrees.atechniquebasedonmachinelearning.Theanalyseswerecarriedoutusing43measurementsfrom110extantbovidspecies.Thediscriminantfunctionsanddecisiontreesobtainedal1owaperfectdiscriminationamongbovidsadaptedtoopenplains,forestsandmountainousareas(100%ofcorrectreclassificationsobtainedinallcomparisons),usingsetsofvariablesmeasuredinallmajorlimbbonesaswellascombinationsofvariablesderivedexclusivelyfromsinglelimbdements.Giventhattheadjustedalgorithmsinvolvesmallsetsofpostcranialmeasurements.theycanalsobeappliedtononcompletespecimenspreservedinarchaeologicalandpaleontologicalassemblages,thusbeingusefulforestimatingthelocomotorperformancesofancienttaxa.Thesealgorithms.indicativeofecologicaladaptationsforhabitatuse,combinedwiththoseadjustedwithcraniodentalmeasurementsforestimatingthedietarypreferencesofbovidspecies.havethepotentialforcharacterizingthepaleoautecologyofextincttaxaandmaybeusedinpaleoenvironmentalreconstruction.Wealsoanalyzeifmultipleregressionequationsshowhigherpredictiveabilityforestimatingbodymassthansimpleregressionequations,andproposethebestalgorithmsobtainedfrompostcranialmorphologicalvariablesmeasuredineachsinglemajorlimbbone【ActaZoologicaSinica52(6):971987,2006.KeywordsBovidae,Ecomorphology,Habitatuse,Bodymass,Discriminantanalysis,Decisiontrees牛科(哺乳纲:偶蹄目)动物与生境利用有关的适应形态模式*ManuelMENDOZAPaulPALMQVISTDepartamentodeEcologiaYGeologia,FacultaddeCiencias,UniversidaddeMdlaga,Mdlaga29071,Spain摘要对广义牛科动物颅后骨骼的多元变量分析揭示了牛科生境利用和体型之间的骨学特征.利用逐步分辨分析方法和一个基于机器学习的决策树方法鉴别了每种生境中牛科动物颅后解剖结构的形态特征.从110个广义牛科动物测量了43个指标进行了这项分析.利用所有主要肢骨测量值和以单根肢骨测量为主的测量值获得的分辨函数和决策树可以完美地区分适应开阔生境,森林和山地的牛科动物(在所有分析中得到了100%正确的再分类).由于调整的函数仅涉及到很小的颅后骨骼测量集,这些函数可以应用于研究考古学和古生物学发掘物中保存的不完整标本.这些表征生境利用的生态适应函数与那些用颅齿部性状建立,用于推测牛科动物食物选择的函数结合,具有刻画已灭绝的分类类群的古个体生态学和重建古环境的潜力.我们还分析了多元回归是否较单一因子回归表现出较高的预测能力,并提出了从每一种单根主要肢骨测量的颅后形态变量得到的最好代数函数动物52(6):971987,2006.关键词牛科生态形态学生境利用体重分辨分析决策树Associationsofpostcranialstructureinextantfiedinordertocharacterizemorphologicaladaptationsbovidsrelatedtolocomotorperformancesareidentitogrosshabitattypes(flatgrasslands,forests,andReceivedApr.22,2006;acceptedSep.18.2006ThisresearchwasfundedbyprojectCGL200401615/BTE.ManuelMENDOZAwasfundedbyapostdoctoralgrantfromtheSpanishCICYTandtheFulbrightVisitingScholarProgram.*Correspondingauthor.E-mail:ppbuII1a.es2006动物ActaZoologicaSinica972动物52卷hillyandmountainousareas)usingamultivariateapproach.Theseassociationsmaybeusedtoreconstructthelocomotorbehaviorandhabitatpreferencesofancientbovidspreservedinfossilassemblages,andthusallowtheirautecologytobeestimated(e.g.,Palmqvisteta1.2003;Mendozaeta1.,2005).Inaddition,thebodysizeofextantbovidsiscorrelatedtothedimensionsofthepostcranialskeletonandmultipleregressionfunctionsareprovidedforpredictingthemassofancientspecies(e.g.,Mendozaeta1.,2006).Postcranialfeaturescorrelatedwithhabitattypeandlocomotoradaptationsincludetheshapeofmajorlimbbones(Scott,1985;Kappelmaneta1.,1997;PlummerandBishop,1994)andoftheankleandfoot(KShlerandMoy6So16,2001;DeGustaandVrba,2003,2005a,b).Thisstudyonpostcranialadaptationsofbovidsdiffersfrompreviousonesintheuseofamultivariateanalysistoevaluateabroadsetofmeasurementsfromdifferentlimbelementsofvariousextantbovidstoidentifymorphologicalcorrelatesforhabitat.Manyauthorshaveusedsimplelogtransformedbivariateleastsquaresregressionequationstoestimatethebodymassofextinctspeciesbasedonsingleanatomicalmeasurements.suchastheareaofthefirstlowermolar(e.g.,Beardeta1.,1996;Gagnon,1997;Kayeta1.,1998),thevolumeofthefemoralhead(CartelleandHartwig,1996;Kappelmanetal_,1997),ortheareaoftheeyeorbit(KordosandBegun,2001).Asmightbeexpected,nosinglemorphologicalvariableshowsaperfectcorrelationwithbodymass,hencereducingtheprecisionandreliabilityofpredictionsderivedfromthem,apparentinthelargestandarderrorsandwideconfidenceintervalsaroundsuchestimates.Arelativelycommonstrategyforavoidingsuchproblemsistocombinepredictionsfromseveralallometricequations,basedondifferentanatomicalstructures.tocalculatetheaveragebodymassforextincttaxa(e.g.,MacFadden,1986;Gingerich,1990;Anyonge,1993,1996;Walkereta1.,1993;McCrossin,1994;Viranta,1994;Flynneta1.,1995;Geboeta1.,1997;Farifiaetal_,1998;Christiansen,1999;Delsoneta1.2000;K6hlerandMoy&So16,2004;ChristiansenandHarris,2005).Becausemultipleregressionexploitsthecomplementaryinformationcontainedinthemorphologicalvariablesstudied,itisprobablythebestsuitedmethodologytocompensatefortheinfluenceofthephylogenyorspecificfunctionaladaptations(Andersoneta1.,1985;Damuth,1990;Jungers,1990;HammerandFoley,1996;Biknevicius,1999;Palmqvisteta1.,1999,2002;Payseureta1.,1999;Mendozaeta1.2006).Forthestudyoflocomotorbehavior,deviationsfromtheexpectationsofgeometricscalingareusedtoinfertheecologicaladaptationsofspeciesintermsoflocomotorperformanceandhabitatchoice(e.g.,morecursoriallyadaptedbovidslivinginplainsandsavannahaveproportionallylongerdistalelements,thosespecializedforclimbinginmountainouscountryhaveshortermetapodials.whereasthoselivinginpartlyopenwoodlandorhillycountrytendtohavebonesofintermediatelengths).Forthestudyofbodymass,however,theeffectsofthesedeviationshavetobeminimizedandthebestestimatorsofbodymassarethedimensionsoftheproximallimbsegments(Kappelmaneta1.,1997).1MaterialsandmethodsAmultivariateapproachisusedtoanalyzemorphologicalfeaturesinthepostcranialskeletonofextantbovids,featureswhichareseentobeindicativeofbodysizeandecologicaladaptationsforhabitatuse.Theanalysesareperformedusing43measurementstakenfrom110extantspeciesinScotts(1985)database.ThesemeasurementsaredescribedinAppendixIandrepresentedinFig.1.Fordetailsonthespecimensusedforcalculatingthespeciesmeans,seeScott(1985).?1.1HabitatadaptationsAssociationsofmorphologicalfeaturesrelatedwithhabitatadaptationsinbovidswereidentifiedbasedon42sizeadiustedvariables.Thesesizetransformedvariablesareobtainedbydividingeachmeasurementbythewidthoftheproximalarticularsurfaceoftheradius(Rd,),avariablethatscalesisometricallywithbodvsizeandishighlycorrelatedwithmass(rz=0.98).0bviously,Rd,wasnotsizeadiustedanditwasnotdirectlyusedasavariable.Thespeciesweredividedintothreediscretegroupsfortheanalysis,correspondingtothreegeneraltypesofhabitat:flatgrasslands,forests,andhillyandmountainousareas.Accordingtotheinformationavailable,only71outofthe110bovidspeciescouldbeunequivocallyclassifiedascharacteristicofoneofthesegrosshabitattypes(seeAppendix1I):46speciesareassociatedwithopenplains,13aretypicalofforestedareas,andtheother12speciesareadaptedtomountainousterrain.Thesegeneraltypesofhabitatrepresentaratherheterogeneousmosaicofhabitatandlocomotoradaptationsinbovids.However.ifthehabitattypeswerefurthersubdivided,thiswouldre.sultinadecreaseofthenumberofspeciesassociatedwitheachofthem,whichwouldleadtoahigherprobabilityofobtainingagooddiscriminationeitherbychanceorbyclosephylogeneticrelationship(seebelow).Thereare39bovidspecieswhichhavetincertainhabitatpreferencesorthatlireinmorethanonehabitattypethatwerenotincludedinthestatist/6期Manue1MENDOZAeta1.:Habitatadaptationsandbodysizeinbovids973calanalyses.ThespeciesofthegenusOryxareagoodexampleofthosebovidsthatcouldnotbeunequivocallyassignedtoonehabitatcategory:thefringe-earedoryx(0.gazella)livesinaridgrasslands.forestedsavannas,semi.desertplains,thickbrush,andnearrockyhillsides.Therefore,0.gazellacouldbeclassifiedinanyofthethreetypesofhabitatcompared(i.e.,openplains,forestandmountainousareas).Similarly,theformerhabitatoftheArabiansoryx(0.1eucoryx)wastheflatandundulatinggravelplainsintersectedbyshallowwadisanddepressions,andthedunesedgingsanddeserts,withadiversevegetationoftrees,shrubs,herbs,andgrasses.Finally,thescimitar-hornedoryx(0.tao)inhabitsthesub.desertlandsandisfoundinrollingdunes,grassysteppesandwoodedinter-dunaldepressions.Thediscriminantfunctions,however,wereemployedforpredictingthehabitatpreferencesofthe39speciesnotusedintheiradjustment.Associationsofmorphologicalfeaturescharacter.isticofeachtypeofhabitatwereidentifiedusingstepwisecanonicaldiscriminantanalysis(SCDA)(Man.dozaandPalmqvist,2006)ratherthanprincipalcomponentsanalysis(PCA)becausethelattertechniqueisnotappropriateforidentifyingparticularmorphologicalpatternsindiscretegroupingsofspecies.Incontrast,SCDAisspecificallydesignedtoidentifythosevariablesinvolvedinthedifferencesbetweenthegroupscomparedanditpermitstheidentificationofcombinationsofmeasurementsthathavegreaterpotentialforspecificapplications(e.g.,intheanalysisofboneremainsfromarchaeologicalandpaleontologicalassemblages)thanPCA.Althoughtheanalysiswasconductedoncompleteelements,thealgorithmscanalsobeappliedtoincompletespecimens,giventhattheyinvolvesmallsetsofpostcranialvariables(seebelow).Speciesmeanswereusedratherthanvaluesforindividualspecimens,becausealthoughtherearemorphologicaldifferencesrelatedtointraspecificvariability(e.g.,sexualdimorphism),weexpectthatmostbovidspeciestheywillshareacommonmorphologicalpat-ternintheirpostcranialstructure.indicativeoftheirlocomotorperformance.Thealgorithmsproposedhereincludeonlyafewvariableswhosecontributiontothediscriminationamonghabitatgroupsisespeciallyrelevant.Indoingso,ourapproachdiffersfromtheoneusedinMandozaeta1.(2002),whichwasorientedtotheobtainingofcomplexalgorithmsthatallowustoinfertheecologicaladaptationsofancientspecies.Themainadvantageofthenewalgorithmsisthattheyidentifyro.bustmorphologicalpatternswhichenableustounder.standtherelationshipbetweenthepostcranialstruc.tureofbovidsandtheiradaptationstoliveindifferenthabitattypes.Minimizingthenumberofvariablesincorporatedinthealgorithms,wealsominimizetheprobabilityofobtainingagooddiscriminationmerelybychance,becausethisprobabilityincreaseswiththenumberofvariablesincludedinthediscriminantfunctions(seediscussioninMendozaeta1.,2002,2005).Althoughtheresultingalgorithmsmaymisclassifysomespecies,theyaremoreaccuratethanthosebasedonmanymeasurements,becausetheyincludeonlythosemorphologicaltraitsmoreclearlyrelatedwiththeadaptivepatterns.Athirdadvantageofourapproachisthatthesesimplemorphologicalpatternscanoftenbedepictedinbivariatescatterplots,whichareausefultoolfordeterminingthehabitatadaptationsofextinctbovids.Itisworthnotingthatifthediscriminantfunc.tionsareappliedtoanextinctbovidthatshowsapostcranialmorphologyidenticaltoanyofthe39modernspecieswithnoclearhabitatpreferences,thisspecieswillbeclassifiedinoneofthethreehabitatgroupscompared(open/forested/mountain),asthediscriminantfunctionsaredesignedfordoingso.T,hisdoesnotmeanthattheextinctspeciesonlylivedwithinthepredictedhabitat;rather,whatthismeansisthatsuchhabitatistheoneforwhichthepostcranialanatomyoftheancientspecieswasbetteradapt.ed.Anotherimportantaspectofthenewmethodologicalapproachisthatthespeciesanalyzedareweightedaccordingtothediversityoftheirtaxonomicgroupings,whichhelpstoavoidtheobtainingofphylogeneticallyconstrainedpatterns(seebelowandMendozaandPalmqvistinthisvolume).Overweigh.ingthespeciesofthosegroupsunderrepresentedinthedatabaseforcestheanalysistotakeintoaccounttheinformationcontributedbythesefewspeciesatthesamelevelthanthatprovidedbythespeciesfromothergroupsmoreabundantlyrepresented.Inthisway,thetaxonomicevennessofthedatabaseismaxi.mizedandthemorphologicalpatterncapturedbytheresultingalgorithmsismoregenera1.However,phy.1ogenyisnottheonlyfactorthatmaydisruptecologi.calinferences.becausebovidsmaymoveintonewhabitatsorhavealternativemodesoflocomotionnotexpressedmorphologicallyinthesamewayasinotherbovids(e.g.,seethepeculiaradaptationsofthepostcranialskeletoninMyotragusbalearicus,anandemicfossilgoatfromthePleistoceneoftheBalearicIslandsintheMediterraneanSea;K6hlerandMoyflSo16,2001,2004).Twotypesofalgorithmswereobtained:1)thoseinvolvingmeasurementsfromacombinationofallmajorlimbbones;and2)algorithmsobtainedfromsetsofmorphologicalvariablesestimatedinasinglelimbbone.Inthesecondcase,thevariablesweresize-ad974动物52卷justed.Althoughalgorithmsdevelopedforsinglelimbboneswillhavealowerdiscriminationpowerthanthosebasedonmeasurementsfromseveralbones,theformercanbeusefultoapplytoskeletalremainsfromdifferentindividualsortodisarticulatedandfracturedspecimens(seereviewsinArribasandPalmqvist,1998;PalmqvistandArribas,2001).Thealgorithmscanbeusedtocharacterizethee.cologicaladaptationsofbovidspeciesnotincludedintheoriginaldataset.Inordertoidentifymorecomplexcombinationsofalgorithms,amachinelearningprogram.calleddecisiontrees(Quinlan,1985),wasused.Decisiontreesrepresentatypeofmachinelearning,wherebycomputersystemsacquireknowledgeinductivelyfromtheinputofalargenumberofsamples(e.g.,bovidspeciesandpostcranialmeasurementsinourcase).Theproductofthislearningisapieceofproceduralknowledgethatcanassignanobject(e.g.,anextinctbovidwhosefeedingpreferencesareunknown)tooneofaspecifiednumberofdiointclasses(e.g.,habitatcategories)basedontheiterativedivisionofthemultidimensionalspacedefinedbytheinputvariables.Theresultisabifurcatingtreepatternordichotomouskeywithdecisionsateachbranchingpointthatcombinestheinformationprovidedbythepostcranialmeasurementsinalogicalway,makingthecharacterizationofanyextinctbovideasier.1.2BodysizeThevaluesofthemorphologicalvariablesinagivenspeciesnotonlydependonitsbodysize,butal一80onitsphylogeneticlegacyandecologicaladaptations(K6hler,1993;K6hlerandMoyflSo16,2004).Therefore,thepredictionsofmassderivedfromsinglemeasurementsarenotalwayspreciseandaccurate(e.g.,Palmqvisteta1.,1999,2002).Ourhypothesisestablishesthatthisbiascanbecompensatedforbytheuseofseveralmeasurementscombinedinamultipieregressionequation.Thepredictionsarethatthesecombinationsofmorphologicalvariableswillallowabettercharacterizationofbodymass.andthattheresultantalgorithmswillshowahigherpredictivepowerwhenappliedtonewspecies(seeMendozaeta1.,2006).Bovidsareoneofthemostsexuallydimorphicgroupsofhoofedmammals,withthemalesofseveralspeciesweighingtwiceasmuchasthefemales.However,inmostspeciesthedimorphismratio(i.e.,massofmales:massoffemales)fluctuatesbetween1.05and1.25(e.g.,seedatainMysterud,2000)andinsomegroupssuchasthedwarfantelopes(neotragines)thesevaluesmaybeevenreversed.Inaddition.?