Impact of the terrestrial-aquatic transition on disparity and rates of evolution in the carnivoran skull

20 Pages·2015·2.84 MB·English

by  Katrina E Jones

Impact of the terrestrial-aquatic transition on
disparity and rates of evolution in the carnivoran
skull
Citation
Jones, Katrina E, Jeroen B Smaers, and Anjali Goswami. 2015. “Impact of the terrestrial-aquatic
transition on disparity and rates of evolution in the carnivoran skull.” BMC Evolutionary Biology
15 (1): 8. doi:10.1186/s12862-015-0285-5. http://dx.doi.org/10.1186/s12862-015-0285-5.
Published Version
doi:10.1186/s12862-015-0285-5
Permanent link
http://nrs.harvard.edu/urn-3:HUL.InstRepos:14065412
Terms of Use
This article was downloaded from Harvard University’s DASH repository, and is made available
under the terms and conditions applicable to Other Posted Material, as set forth at http://
nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA
Share Your Story
The Harvard community has made this article openly available.
Please share how this access benefits you. Submit a story .
Accessibility
Jonesetal.BMCEvolutionaryBiology (2015) 15:8
DOI10.1186/s12862-015-0285-5
RESEARCH ARTICLE Open Access
Impact of the terrestrial-aquatic transition on
disparity and rates of evolution in the carnivoran
skull
Katrina E Jones1,5*, Jeroen B Smaers2 and Anjali Goswami3,4
Abstract
Background: Whichfactorsinfluencethedistributionpatternsofmorphologicaldiversityamongclades?Theadaptive
radiationmodelpredictsthatacladeenteringnewecologicalnichewillexperiencehighratesofevolutionearlyinits
history,followedbyagradualslowing.HerewemeasuredisparityandratesofevolutioninCarnivora,specifically
focusingontheterrestrial-aquatictransitioninPinnipedia.Weanalyzefissiped(mostlyterrestrial,arboreal,and
semi-arboreal,butalsoincludingthesemi-aquaticotter)andpinniped(secondarilyaquatic)carnivoransasacasestudy
ofanextremeecologicaltransition.Weused3Dgeometricmorphometricstoquantifycranialshapein151carnivoran
specimens(64fissiped,87pinniped)andfiveexceptionally-preservedfossilpinnipeds,includingthestem-pinniped
Enaliarctosemlongi.Range-basedandvariance-baseddisparitymeasureswerecomparedbetweenpinnipedsand
fissipeds.Todistinguishbetweenevolutionarymodes,aBrownianmotionmodelwascomparedtoselectiveregime
shiftsassociatedwiththeterrestrial-aquatictransitionandatthebaseofPinnipedia.Further,evolutionarypatternswere
estimatedonindividualbranchesusingbothOrnstein-UhlenbeckandIndependentEvolutionmodels,toexaminethe
originofpinnipeddiversity.
Results:Pinnipedsexhibitgreatercranialdisparitythanfissipeds,eventhoughtheyarelesstaxonomicallydiverse
and,asacladenestedwithinfissipeds,phylogeneticallyyounger.Despitethis,thereisnoincreaseintherateof
morphologicalevolutionatthebaseofPinnipedia,aswouldbepredictedbyanadaptiveradiationmodel,anda
Brownianmotionmodelofevolutionissupported.Insteadbasalpinnipedspopulatednewareasofmorphospace
vialowtomoderateratesofevolutioninnewdirections,followedbylaterburstswithinthecrown-group,potentially
associatedwithecologicaldiversificationwithinthemarinerealm.
Conclusion:Thetransitiontoanaquatichabitatincarnivoransresultedinashiftincranialmorphologywithoutan
increaseinrateinthestemlineage,contratotheadaptiveradiationmodel.Insteadthesedatasuggestareleasefrom
evolutionaryconstraintmodel,followedbyaquaticdiversificationswithincrownfamilies.
Keywords:Disparity,Carnivora,Fissiped,Pinniped,Cranialmorphology,Radiation,Ecologicaltransition,Aquaticmammal
Background organisms enter a new adaptive zone, that is a niche
Understanding factors which influence tempo and mode with relatively few competitors, there will be an initial
in evolution is an important theme in evolutionary biol- burst in evolutionary rate [6,7]. This ‘early burst’ may be
ogy[1-3].Onefactorwhichmayinfluencetheevolution- followed by a slowing of morphological diversification as
ary patterns is ecology [4,5]. In particular, the ‘adaptive the niche begins to become filled [2,8,9]. This model
radiation’ model of Simpson (1944) suggests that when provides a potential link between ecological transitions
andevolutionary rate.
Support for adaptive radiations has been found using
*Correspondence:[email protected]
1CenterforFunctionalAnatomyandEvolution,JohnsHopkinsUniversity, empirical data [10-17], the best-known examples includ-
Baltimore,MD,USA ing Darwin’s finches [18-20], Hawaiian silverswords [21]
5DepartmentofOrganismicandEvolutionaryBiology,MuseumofComparative
and African lake cichlids [22,23]. Further, in terms of
Zoology,HarvardUniversity,26OxfordStreet,Cambridge,MA02138,USA
Fulllistofauthorinformationisavailableattheendofthearticle marine mammals, cetaceans underwent a rapid increase
©2015Jonesetal.;licenseeBioMedCentral.ThisisanOpenAccessarticledistributedunderthetermsoftheCreative
CommonsAttributionLicense(http://creativecommons.org/licenses/by/4.0),whichpermitsunrestricteduse,distribution,and
reproductioninanymedium,providedtheoriginalworkisproperlycredited.TheCreativeCommonsPublicDomain
Dedicationwaiver(http://creativecommons.org/publicdomain/zero/1.0/)appliestothedatamadeavailableinthisarticle,
unlessotherwisestated.
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page2of19
in body size disparity early in the clade’s history despite evidence supports a closer relationship between muste-
thelack ofarapidinitialtaxonomic diversification [24]. loids and pinnipeds. Pinnipeds were also previously
The mammalian order Carnivora is an ecologically thought to be diphyletic, with otariids linked to ursids
and taxonomically diverse group of mammals which and phocids linked to mustelids [48,49]. However, a
have been the source for many studies of morphological monophyletic originof pinnipeds is now well supported
variation, though they display relatively low cranial dis- bybothmorphologicalandmolecularevidence[36,50].
parityrelativetoothermammalorders[25-33].Arguably To investigate the influence of the terrestrial-aquatic
the largest ecological transition in carnivoran evolution transition on morphological diversity, we tested evolu-
wastheshiftfromterrestrialtoaquaticlifestyleintheevo- tionary modelsonthecarnivoranphylogeny.Bycombining
lution of the Pinnipedia (seals, sea lions and walruses) a phylogeny with species’ information, statistical models
[34]. This study investigates how this extreme ecological canbeusedtoinfertheevolutionarypast[51].Suchstatis-
shift has influenced disparity (morphological diversity) ticalmodelsusespecificparameterstodeterminehowtraits
andratesofevolutioninpinnipedskulls,incomparisonto change in phylogenetic space. The most commonly used
their fissiped relatives. The Carnivora provides an ideal statisticalmodelisBrownianMotion(‘BM’),whichassumes
case study for the influence of the terrestrial-aquatic thattraitsevolveineachinstantofunitoftimewithamean
transition on disparity and rates of evolution because changeofzeroandunknownandconstantvariance.Within
both aquatic pinniped carnivorans (seals, sea lions, and BM, the evolution of a continuous trait X, along a branch
walruses) and terrestrial (fissiped) carnivorans (dogs, bears, over time increment t, is quantified as dX(t) = σdB(t),
weasels, cats, hyaenas, mongooses and allies) have a large whereσconstitutesthemagnitudeofundirected,stochastic
extant taxonomic diversity[33,35-37]. Moreover, pinnipeds evolution (σ2 is generally presented as the BM rate par-
are less divergent from their closest living fissiped relatives ameter)anddB(t)isGaussianwhitenoise.
in cranial morphology than other marine mammal groups Althoughgenerallyagreedtobeanunrealisticassump-
(e.g., cetaceans or sirenians) [38] and thus may be directly tion for most analyses, the advantage of BM is that it is
compared with fissipeds. Though increased rates of body mathematically tractable. Recent phylogenetic compara-
size evolution in pinnipeds relative to fissipeds were not tive methods have continued using BM as a baseline
supported [39], rates of evolution in the cranium have not model, but incorporate additional parameters to reflect
beencompared. more nuanced assumptions about the evolutionary
In order to test hypotheses of cranial evolution in the process. Recent advances include the development of
Carnivora, we employed analyses that work within a methods based on Ornstein-Uhlenbeck (‘OU’) assump-
phylogeneticframework.Thephylogeneticrelationshipsof tions. The OU model incorporates stabilizing selection as
fissiped and pinniped carnivorans used here are shown in a constraint and quantifies the evolution of a continuous
Figure 1 [40,41]. There are 21 genera and 34–36 species traitXasdX(t)=α[θ–X(t)]dt+σdB(t),whereσcaptures
divided into the three pinniped families: Phocidae (seals), the stochastic evolution ofBM, and α determines the rate
Otariidae (fur seals and sea lions) and Odobenidae of adaptive evolution towards an optimum trait value θ
(walruses), which diverged ~29ma [40,42-44]. Molecular [52,53]. This standard OU model can be modified into
data generally indicate that odobenids are most closely multiple-optima OU models that allow optima to vary
related to otariids, forming the Otarioidea (Figure 1) across the phylogeny [54]. In these implementations, the
whereas morphological data links the odobenids with optima are defined a priori to allow testing of alternative
phocidsinthecladePhocomorpha(Figure1)[37,44,45], parameterizations and therefore alternative biological
sobothhypotheseswereemployedhere.Themostbasal hypotheses [54]. An added advantage of these OU
pinnipeds were the enaliarctines, a stem radiation first model fitting approaches is that they allow proper
known from California around 28ma [45], of which one multivariate models rather than fitting variables one at
representative fossil with excellent cranial preservation atime[55].Otherrecentmethodsdonotfixthenumberof
isincludedinthisstudy(seeMethods). shifts or their locations on the phylogeny, but instead im-
The paraphyletic fissipeds are a group of mainly ter- plement algorithms that estimate them (e.g. reversible-
restrial (including arboreal and fossorial) carnivorans jump),whilejointlysamplingOUparameters[56].Suchan
which consist of ten families, 105 genera and over 241 approach was recently implanted in the R package bayou
species (Figure 1); [46]. Pinnipeds are caniform carnivor- [57], and allows an inference of the location, magnitude,
ans, and molecular and morphological evidence support and number of adaptive shifts for univariate models. We
theplacementofpinnipedswithinthearctoids(bears,rac- hypothesizethattheevolutionofanaquaticlifestyleinpin-
coons, weasels, and allies). Within Arctoidea, there has nipeds will cause a selective regime shift at the basal node
beenlong-standingdisagreementoverwhethermusteloids ofpinnipedswhichcouldbedetectedbytheseOUmodels.
or ursids are the sister group of pinnipeds (Figure 1) Although OU-based methods are a powerful tool for
[35,36,45]. Most [35,40], but not all [37,47], recent testing alternative biological interpretations and ‘painting’
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page3of19
Figure1(Seelegendonnextpage.)
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page4of19
(Seefigureonpreviouspage.)
Figure1Compositephylogenyusedinthisstudy.Extantrelationshipsandbranchlengthsfrom[40],placementoffossilsaccordingto[41].
ThisshowstheOtarioideatopology,withOdobenidaeassistertaxontotheOtariidae.Analyseswerealsorunonthesametreebutwitha
Phocidae-Odobenidaesistergrouping,followingthePhocomorphahypothesis.Branchcolors:Feliformia,orange;non-pinnipedCaniformia,red;
stempinnipedsandallodesmines,darkblue;Phocidae,mid-blue;Odobenidae,teal;Otariidae,paleblue.
regime shifts onto branches of a phylogeny [54,57], shift moved into new regions of morphospace. Combin-
they do not provide estimations of ancestral states nor ing analyses of rates and morphological disparity thus
of variable rates for individual branches. To overcome allows for a more complete analysis of tempo and mode
this empirical hurdle while avoiding increased model in the evolution of diversity. Under a model of diffusive
parameterization, [58] and [59] developed an approach evolution, we would predict that fissipeds would have
(‘Independent Evolution’) that relies on similar assump- greater disparity than fissipeds because fissipeds are the
tions as a multiple regime OU model, but requires fewer more taxonomicallydiverseandancient clade.
parameters. This approach assumes that population phe-
notypes are affected by the wandering adaptive peaks of Methods
adaptive surfaces (aligning with an Adaptive Peak model Dataset
of evolution), and is therefore similar to an OU model This study included 151 specimens of adult carnivorans,
with shifting locations in assuming that different regimes including 64 fissiped and 87 pinniped specimens. These
may occur at different locations in phylogenetic space. specimens constituted 34 fissiped species and 28 extant
However, the formalizationof IE assumptions differscon- pinniped species (Table 1). Fissipeds are more taxonom-
siderably tothe OU approaches, as the IEmethod utilizes ically diverse than pinnipeds (241 and 36 species re-
a geometric approach that consists of two main steps: (1) spectively), so fissipeds were relatively less densely
quantifyinganexpectedvalueforeachinternalnodebased sampled [46]. Specifically, we measured 14% of fissiped
onlyonphylogeneticandtraitdata,assumingapuregrad- species, and 30% of fissiped genera. In contrast, 77% of
ual mode of evolution; (2) quantifying the deviation of pinniped species and all genera were included [46]. We
internalnodestotheexpectedvaluebasedongradualevo- were testing if the greater taxonomic diversity of fissi-
lution using a triangulation between the expected value peds was reflected in a larger morphological disparity, so
and the two descendant values of the internal node in we sampled as broadly as possible within the group.
question. The triangulation between the gradual mode Fissiped species were selected to encompass their full
expectation of the ancestor andthe observed descendants range of phylogenetic diversity, including representatives
results inarescaling of branches suchthatthebarycentre from every extant family and major ecological group.
among them provides the best fit for the data [58]; a pro- Although fossil fissipeds were unavailable for this study,
cedure akin to Farris optimization [60,61]. The rate par- previousstudiessuggestthatmostfossilfissipeds(except
ameter is the distance between the barycentre of the sabre-toothed cats) fall within the range of morphospace
triangulation procedure and the descendant value of the of extant clades [33]. The difference in sampling be-
branch. This method therefore provides variable rate esti- tween fissipeds and pinnipeds may influence the results
mates for individual branches, as well as ancestral values of variance-based disparity analyses as follows. Variance-
for individual nodes. Under an adaptive radiation model, based disparity measures may be overestimated in fissi-
we would predict an increase in evolutionary rates on the peds relative to pinnipeds, because more dissimilar taxa
branchesatthebaseofPinnipedia,relativetothosehigher were sampled for fissipeds, in order to cover their extant
upthepinnipedphylogeny. phylogenetic and ecological breadth without fully sam-
Measuring rates of evolution captures an important pling their species diversity, than for pinnipeds, for
aspect of shape or trait evolution, but high rates of whichmostextantandsomefossilspeciesweresampled.
evolution may not translate simply into high diversity, or However, as our null hypothesis is that fissiped disparity
vice versa. For example, if taxa are constrained develop- should exceed pinniped disparity (due to their greater
mentally or ecologically to a particular range of shapes, alpha diversity), this will not lead to a false rejection of
they mayshow high ratesofevolution andhigh amounts the null hypothesis (type 1 error), but will make it more
of convergence in form, but low overall morphological difficult to reject (type 2 error). Further, rarefaction will
disparity [17,62,63]. In this scenario, analyses could ac- be used to statistically account for the differences in
curately recover high rates of evolution but this would sampling(see below fordetails).
not show that the taxa of interest are repeatedly explor- Several fossil pinnipeds were also available for study.
ing the same range of morphospace rather than expand- Stem pinniped fossils were particularly important for the
ing into new morphologies. Alternatively, a clade could comparative analyses (disparity analyses wereextant-only)
achieve high disparity through slow evolution if each because they provide information about the ancestral
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page5of19
Table1Specieslist,samplingandtaxonomicassignment Table1Specieslist,samplingandtaxonomicassignment
Group Family Species N (Continued)
Feliformia Pinniped Otariidae Arctocephaluspussillus 3
Fissiped Eupleridae Cryptoproctaferox 1 Pinniped Otariidae Arctocephalustropacalis 2
Fissiped Eupleridae Eupleresgoudotii 2 Pinniped Otariidae Callorhinusursinus 2
Fissiped Eupleridae Fossafossana 2 Pinniped Otariidae Eumetopiasjubatus 3
Fissiped Eupleridae Galidiaelegans 2 Pinniped Otariidae Neophocacinerea 1
Fissiped Felidae Acinonyxjubatus 2 Pinniped Otariidae Otariaflavescens 3
Fissiped Felidae Felisbengalensis 2 Pinniped Otariidae Zalophuscalifornianus 1
Fissiped Felidae Felisvivverina 2 Pinniped Phocidae †Acrophocalongirostris 1
Fissiped Felidae Lynxrufus 2 Pinniped Phocidae †Piscophocapacifica 1
Fissiped Herpestidae Cynictispenicillata 1 Pinniped Phocidae Cystophoracristata 3
Fissiped Herpestidae Ichneumiaalbicauda 1 Pinniped Phocidae Erignathusbarbatus 1
Fissiped Hyaenidae Crocutacrocuta 2 Pinniped Phocidae Halichoerusgrypus 5
Fissiped Hyaenidae Protelescristatus 2 Pinniped Phocidae Histriophocafasciata 7
Fissiped Nandinidae Nandiniabinotata 2 Pinniped Phocidae Hydrurgaleptonyx 2
Fissiped Viverridae Civettictiscivetta 2 Pinniped Phocidae Leptonychotesweddelli 3
Fissiped Viverridae Genettagenetta 2 Pinniped Phocidae Lobodoncarcinophagus 4
Fissiped Viverridae Paradoxurushermaphroditus 2 Pinniped Phocidae Miroungaleonina 3
Caniformia Pinniped Phocidae Monachusmonachus 2
Fissiped Canidae Canislupus 2 Pinniped Phocidae Ommatophocarossi 2
Fissiped Canidae Cerdocyonthous 2 Pinniped Phocidae Pagophilusgroenlandica 1
Fissiped Canidae Otocyonmegalotis 2 Pinniped Phocidae Phocahispida 4
Fissiped Canidae Vulpesvulpes 2 Pinniped Phocidae Phocalargha 4
Fissiped Mephitidae Mephitismephitis 2 Pinniped Phocidae Phocavitulina 2
Fissiped Mustelidae Enhydralutris 2 Pinniped Phocidae Pusacaspica 4
Fissiped Ailuridae Ailurusfulgens 1 Pinniped Phocidae Pusasibirica 4
Fissiped Mustelidae Gulogulo 1 Speciesmeanswereusedinallanalyses.Daggerindicatesfossiltaxa.
Fissiped Mustelidae Martespennanti 2
Fissiped Mustelidae Melesmeles 2 morphology of the group. These taxa may improve
Fissiped Procyonidae Nasuanasua 2 estimations of rates on the pinniped stem branches
Fissiped Procyonidae Potosflavus 2 which spanned the terrestrial-aquatic transition.Cranial
material of stem musteloids (sister taxa to pinnipeds)
Fissiped Procyonidae Procyoncancrivorous 2
would have also been useful in this capacity, but were
Fissiped Procyonidae Procyonlotor 2
unavailable.
Fissiped Ursidae Ailuropodamelanoleuca 2
Enaliarctos emlongi (USNM 250345) is a stem pinni-
Fissiped Ursidae Melursusursinus 1 ped from the Miocene (~20 mya) of California and rep-
Fissiped Ursidae Tremarctosornatus 2 resents an early radiation of enaliarctine pinnipeds.
Fissiped Ursidae Ursusamericanus 2 Allodesmus sp. (USNM 335445) is also from the Mio-
Pinniped Desmatophocidae †Allodesmussp. 1 cene of California but this specimen has not yet been
carefully assessed in terms of its phylogenetic and taxo-
Pinniped Enaliarctinae †Enaliarctosemlongi 1
nomic position. It is currently referred to the genus
Pinniped Odobenidae Odobenusrosmarus 3
Allodesmus,whichisamemberofDesmatophocidae,an
Pinniped Odobenidae †Pontolismagnus 1 extinct pinniped family that has been variably related
Pinniped Otariidae Arctocephalusaustralis 2 to either the Otarioidea or the Phocoidea [45,64-66].
Pinniped Otariidae Arctocephalusgalapogoensis 1 Pontolis magnus (USNM 335567) is an extremely large
fossil odobenid from the lower Pliocene Empire Forma-
Pinniped Otariidae Arctocephalusgazella 5
tion of California and is thought to be closely related to
Pinniped Otariidae Arctocephalusphilippi 1
the extinct Dusignathinae [67,68]. The two fossil phocids,
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page6of19
Acrophoca longirostris (USNM 421632) and Piscophoca data) are described in [71,72] and [31]. From the data-
pacifica (USNM 360406) are both from the Pisco Forma- sets detailed in those studies, a subset of 11 overlap-
tionofthePlioceneofPeruandarethoughttoberelatives ping landmarks, observable and with clear homologyin
ofthemonachineseals[69,70]. both fissipeds and pinnipeds, and identifiable in all the
Specimens were obtained from the University of fossils, was selected. These cranial landmarks are shown
CambridgeZoologyMuseum,theNaturalHistoryMuseum in Figure 2 and described in Table 2, and the complete
(London), the United States National Museum of Natural datasetis available atwww.goswamilab.com.This number
History (Washington, D.C.), the Field Museum of Natural is reduced from the original datasets because of the need
History (Chicago), and the American Museum of Natural tocaptureequivalentlandmarksondisparateskullshapes
History (New York). Specific details on landmark and incomplete material. However, the common land-
collection, unification, and mirroring (to fill in missing marks still include information from most regions of
Figure211landmarksusedindataanalysis,shownonaskullofArctocephalusgazella.Landmarks4–7weretakenbilaterally.Landmark
descriptionscanbefoundinTable2.WireframeusedtopresentshapevariationfromPCAshowninred.
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page7of19
Table2Descriptionoflandmarkstaken Comparisonsofdisparity
Number Landmark The variation in shape within a group can be expressed
in many different ways, and the disparity measure
1 Nasalmidline
chosen may impact interpretation of the data. Hence,
2 Nasal-Frontalmidlinesuture
multiple measures of disparity should be used to fully
3 Parietal-Occipitalmidlinesuture
understand the patterns of variation observed [3,75].
4 Caninelabial* Range-based methods identify the maximal morpho-
5 Jugal-Maxillaposteroventralsuture* space occupation of each group, and therefore are more
6 Jugal-squamosalposteroventralsuture* sensitive to outliers and sample size bias. However they
do inform about the maximum differences in shape
7 Auditorybullaanteromedialextreme*
within a group. Here, the sum of ranges across all PCs
*bilaterallandmarkweretakenonbothsides.
wasusedasameasureofrange.Variance-baseddisparity
metrics also include information about the spread of
the skull and provide much more information about species within the morphospace, but are much more
skull proportions than would equivalent linear robust to sampling bias [3]. For example, a tightly clus-
measures. tered group with a few divergent species might have the
Two specimens of Ailuropoda melanoleuca were digi- same range as a widely dispersed group, but would have
tized by both observers to test inter-observer error. The a lower variance. The mean multivariate distance to the
average Procrustes distance between repeats (0.055) was group centroid was used as the variance-based disparity
smaller than the than the distance between different metric. Estimates of the total variance of each group
specimens of A. melanoleuca (0.081), and much less should be less sensitive to uneven sampling than the
than the average Procrustes distance between fissiped convex hull area[75,76].
specimensandthemean (0.114). Thesemetrics were calculated using the MDApackage
inMatlab[76,77]forfissipedsandextantpinnipeds.Fos-
sil pinnipeds were excluded from this analysis to provide
PCA a fairer comparison to fissipeds, for which fossil speci-
Three-dimensional landmark data were analyzed using mens were unavailable. Disparity analyses were carried
thesoftwareMorphoJ[73].Specimenswerebroughtinto outusingallprincipalcomponent scoresinordertotake
the same shape space (removing all non-shape elements) into account all of the variation in the dataset. To com-
via translation, rotation and scaling by Procrustes super- pute standard deviations and confidence intervals on the
imposition. Further, the Procrustes co-ordinates were disparity measures, a bootstrap procedure was used.
symmetrized and the symmetric component of variation Specimens were re-sampled randomly with replacement
used in subsequent analyses. Previous work indicates 1000 times, and the mean (disparity value) and deviation
that shape variation of the carnivoran cranium is limited calculated. In addition, to take into account the uneven
enough to apply the tangent-space approximation [74]. sampling between fissipeds and pinnipeds, rarefaction
Species averages were then calculated from specimen was used. The pinniped sample was rarefied to a
data so that each tip on the phylogenetic tree is repre- sample-size of five using bootstrapping. This represents
sented by a single shape point [3]. Though most species anequivalent species-samplingtothatoffissipeds.
were represented by multiple individuals sampling
ranged from 1 to 7 individuals. Quantifying intraspecific Evolutionaryrates
variation is not critical here because we are measuring The phylogenetic relationships and branch lengths used
large-scale cross-taxonomic morphological variation. for estimating evolutionary rate were taken from [40].
Any fossil specimens with high asymmetry were re- Fossil species were then added based on positions indi-
moved prior to analysis, as unusually high asymmetry cated by [41] and [70], with branch lengths equivalent to
may reflect distortion that occurred during or after their earliest appearance in the fossil record (Figure 1).
fossilization. Principal components analysis was used to Due to controversy over the position of the walrus, and
reduce the dataset to orthogonal principal components the potentially large impact of this key node to our in-
(PCs) for subsequent analysis. These PC data were used terpretations, rate analyses were run using both the
in disparity and rate analyses, as opposed to Procrustes OtarioideaandPhocomorphahypotheses.
co-ordinate scores, to provide more convenient and inter- A summary of comparative analyses run in this study
pretablegraphingoftheevolutionarymorphospacethrough canbefoundinTable3.Toinvestigatetherelativelikeli-
time, and to overcome computing limits on variable num- hood of alternative evolutionary scenarios of shifts in
bersinsomeanalyses.However,both data setsprovidethe cranial morphology in the carnivoran phylogeny, we
sameshapedatawhenallthePCsareincluded. used the ouch R-package [78] to fit Ornstein-Uhlenbeck
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page8of19
Table3Rateandcomparativeanalysessummary
Analysis Data Model Implementation Presentedin
MultivariateIE AllPCs IndependentEvolution evomap,Euclideandistances Figure7,Additionalfile10
Multivariatehypothesistest PC1-9 Ornstein-Uhlenbeck ouch Table5
UnivariateIE PC1-4 IndependentEvolution evomap Figure6,Additionalfiles1,2,3,4,5,6,7,8and9
UnivariateOU PC1-4 Ornstein-Uhlenbeck bayou Additionalfiles11,12,13,14,15,16,17and18
modelstomultivariatedata.Toinvestigatethelikelihood an estimate of total evolutionary change along each
of an evolutionary scenario in which pinnipeds indicate branchofthetree(multivariateIEanalysis).
a higher rate of change at their basal branch, we tested The method implemented by bayou allows the infer-
four alternative models. The first model is a neutral ence of the location, magnitude and number of adaptive
Brownian motion model, which assumes that evolution- shifts within a multiple-optima OU framework, hereby
ary change and selection follow a random walk, and providing a detailed inference of selective shifts in the
therefore does not align with adaptive radiation assump- phylogeny (Univariate OU analysis). We applied this
tions. The second model considers a single optimum analysis to the top 4 PC’s as an independent validity
along the entire phylogeny, consistent with a shared se- checkofresultsobtained bytheIEmethod.
lective regime across carnivorans. The third model con-
siders two separate optima, for terrestrial and aquatic Results
carnivorans. This hypothesizes regime shifts both at the Cranialshapeoffissipedandpinnipedcarnivorans
base of Pinnipedia and on the branch leading to the Figure 3 depicts the first and second principal compo-
otter. Finally, the fourth model considers an adaptive nents which account for 60.12% of the total variance.
shiftatthebaseofpinnipedia,inconcertwiththerecon- Illustration of shape variation is provided by the wire-
structed position of the terrestrial-aquatic transition. frames in Figure 4 and the comparison plate of skull
The analysis was run on the top 9 PCs which together photographs in Figure 5. PC1 distinguished phocid and
account for 95.2% of variation (multivariate hypothesis odobenid pinnipeds from otariids and fissipeds. Phocids,
test). The analyses included the top 9 PCs rather than with high scores on PC1, displayed large nasal openings,
all PCs due to the extensive time required to run the dorso-ventrally tall and mediolaterally wide crania. This
models. is in comparison to the relatively dorsoventrally flat and
To describe morphological changes along individual mediolaterally narrow terrestrial skull that has more
branches of the phylogeny we employed the R packages anteriorly positioned nasal bones. The aquatic mustelid
evomap [79] and bayou [57]. The IE method (available Enhydra lutris (sea otter, superior-most orange star) fell
in evomap) provides both ancestral states and variable close to otariid space and in the most positive position
rates, allowing for a detailed description of how mor- on PC1 among fissipeds, although some ursids (orange
phospace changed through evolutionary time along trefoil) also displayed relatively high PC1 scores. The
individual branches of the phylogeny. These changes stem pinniped E. emlongi and the desmatophocid Allo-
can be visualized into an evolutionary morphospace desmus sp. were similar to the otariids on this axis,
that captures changes through time by taking snap- whereas the fossil phocids P. pacifica and A. longirostris
shots of morphospace along intervals of time. This fellwithin orclose toextantphocids.
approach is different from the more widely used PC2 reflected snout elongation across Carnivora.
‘phylo-morphospace’ approach [17] in that the evo- Short-faced (brachycephalic) species, such as Enhydra,
morphospace approach consists of visualizing the evo- displayed positive scores on PC2, whereas long-snouted
lutionary changes in the morphospace through time, (dolichocephalic) species, like most canids, had negative
rather than the projection of a phylogeny into a mor- scores. The fossil taxa A. longirostris (the long-necked
phospace. The evolutionary morphospace approach, seal), and Pontolis magnus (a stem odobenid), as well as
available in evomap [79], thus fully captures evolution- the extant leopard seal, Hydrurga leptonyx, had crania
ary trends by displaying how morphospace is inferred thatwere verydolichocephalic.
to have changed over time through phylogenetic space. Figure 3 also illustrates PC3 and 4, which combined
Rates and reconstructed node values were calculated represent 21.06% of the variance in the analyses. PC3
for all PCs individually using evomap, of which the top distinguished otariids from other pinniped and fissiped
4 are presented (univariate IE analysis). From the node groups. Otariids differed in the more anterior placed
values, the Euclidean distance between ancestor–des- parietal-occipital suture relative to the bullae, longer
cendant pairs across all PCs was calculated, providing ventral portion of the jugal, and more anteriorly place
Jonesetal.BMCEvolutionaryBiology (2015) 15:8 Page9of19
Figure3ScatterplotsshowingvariationonPC1-PC4.These
axesrepresent39.76%,20.36%,14.92%and6.14%ofvariance.
Basedonspeciesmeans.Fossilpinnipedsareasfollows:En,Enaliarctos
emlongi;Al,Allodesmussp.;Po,Pontolismagnus;Pi,Piscophocapacifica;
Ac,Acrophocalongirostris.Polygonsconnectfissipeds(red)andextant
pinnipeds(blue)andreflectgroupingsusedinthedisparityanalyses.
ExtremalshapesareshowninFigure4.
nasofrontal relative to the jugal-maxillary suture. Both E.
emlongi and Allodesmus sp. were near to, but just out-
side of, otariid morphospace on this axis. PC4 represents
Figure4WireframesshowingshapevariationonPC1-PC4in
around 6% of variation and did not distinguish at all fis-
lateralanddorsalviews.Anterioristotherightoftheimage.
sipeds from pinnipeds,which overlapfully onthisaxis. LandmarksthewireframewasbasedonareshowninFigure2.
Disparityanalyses
Resultsofdisparity analyses offissipedandpinnipedcar-
nivorans can be found in Table 4. For pinnipeds, results difference indisparity between the two groups. However,
are shown both for the full number of specimens, and when pinnipeds are bootstrapped the range-based dis-
for a sample which has been rarefied down to the parity falls drastically, such that it is below the lower
equivalent fissiped sampling-level (proportional to their confidence limit of fissipeds. The mean distance to the
taxonomic diversity) using bootstrapping (five speci- centroid, a variance-based disparity metric, was much
mens). The sum of ranges, a range-based disparity more robust to sampling changes. In this case, both
metric, is more sensitive to sample size. When all pinni- original- and rarefied-pinniped disparity was above the
peds are included, pinniped disparity was larger, but the 95% confidence interval for fissipeds, indicating that
pinniped confidence interval overlapped with the fis- pinnipeds had higher variance, which was robust to
siped value, indicating that there was not a significant sampling.