Article

Molecular Phylogeny Biogeography and Habitat PreferenceEvolution of MarsupialsKieren J Mitchell1 Renae C Pratt2 Laura N Watson3 Gillian C Gibb4 Bastien Llamas1 Marta Kasper1

Janette Edson1 Blair Hopwood1 Dean Male1 Kyle N Armstrong1 Matthias Meyer5 Michael Hofreiter6

Jeremy Austin1 Stephen C Donnellan17 Michael S Y Lee17 Matthew J Phillips8 and Alan Cooper1

1Australian Centre for Ancient DNA School of Earth and Environmental Sciences University of Adelaide Adelaide SA Australia2Department of Evolution Ecology and Genetics The Australian National University Canberra ACT Australia3Department of Obstetrics and Gynaecology Research Centre for Reproductive Health University of Adelaide Adelaide SAAustralia4Institute of Agriculture and Environment Massey University Palmerston North New Zealand5Department of Evolutionary Genetics Max Planck Institute for Evolutionary Anthropology Leipzig Germany6Department of Biology University of York York United Kingdom7South Australian Museum Adelaide SA Australia8School of Earth Environmental and Biological Sciences Queensland University of Technology Brisbane QLD Australia

Corresponding author E-mail kierenmitchelladelaideeduau alancooperadelaideeduau

Associate editor Emma Teeling

Abstract

Marsupials exhibit great diversity in ecology and morphology However compared with their sister group the placentalmammals our understanding of many aspects of marsupial evolution remains limited We use 101 mitochondrialgenomes and data from 26 nuclear loci to reconstruct a dated phylogeny including 97 of extant genera and 58 ofmodern marsupial species This tree allows us to analyze the evolution of habitat preference and geographic distributionsof marsupial species through time We found a pattern of mesic-adapted lineages evolving to use more arid and openhabitats which is broadly consistent with regional climate and environmental change However contrary to the generaltrend several lineages subsequently appear to have reverted from drier to more mesic habitats Biogeographic recon-structions suggest that current views on the connectivity between Australia and New GuineaWallacea during theMiocene and Pliocene need to be revised The antiquity of several endemic New Guinean clades strongly suggests asubstantially older period of connection stretching back to the Middle Miocene and implies that New Guinea wascolonized by multiple clades almost immediately after its principal formation

Key words supermatrix ancestral state reconstruction mammal mitochondrion

IntroductionModern marsupials originated in the Late Cretaceous andtoday include over 300 known species distributed acrossthe Americas and Australasia (Wilson and Reeder 2005)These species vary from gliding possums to hopping kanga-roos and include obligate carnivores specialized insectivoresomnivores and herbivores and both diurnal and nocturnalforms They also occupy diverse habitats ranging from the wettropical rainforests of New Guinea and the Amazon to thearid Australian interior Marsupials occupy virtually every ter-restrial nonvolant niche occupied by placental analogs onother continents However despite the great diversity of ad-aptations displayed by living marsupials studies attemptingto explain the drivers and constraints behind their evolutionhave been limited compared with their sister group This ispartly due to extensive gaps in the marsupial fossil recordwhich obscure several critical periods in their evolutionaryhistory (Archer et al 1999)

Where the fossil record is depauperate molecular phylog-enies can be particularly useful tools for reconstructing thetempo and mode of evolution Time-calibrated phylogeniesare increasingly used to reconstruct biogeographic historyand test dispersalvicariance hypotheses (Avise et al 1987Avise 2000) Such phylogenies can also be used to modelthe evolution of morphological and ecological traits in a phy-logenetic context (Felsenstein 1985 Huey 1987 Pagel 1997)Researchers have even attempted to use phylogenies of livingtaxa to infer changes in the rate of speciation and extinctionthrough time (Alfaro et al 2009 Stadler 2011 Hugall andStuart-Fox 2012) To fully investigate these issues and under-stand the origin of extant marsupial diversity it is necessary todevelop a comprehensive phylogeny as a template for explor-ing marsupial macroevolution

Since the turn of the century great progress has beenmade in resolving the relationships among marsupial taxaEomarsupialia (Archer 1984 Phillips et al 2006 Beck 2008

The Author 2014 Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution All rights reserved For permissions pleasee-mail journalspermissionsoupcom

Mol Biol Evol doi101093molbevmsu176 Advance Access publication May 30 2014 1

MBE Advance Access published June 19 2014 at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

Meredith et al 2011) has been identified as a clade comprisingthe four Australasian orders Diprotodontia (the kangaroospossums wombats and koala) Peramelemorphia (bandi-coots and bilbies) Dasyuromorphia (including theTasmanian devil and the thylacine) and Notoryctemorphia(marsupial moles) Eomarsupialia is nested within a paraphy-letic collection of three endemic American lineages (Beck2008 Meredith Westerman Case et al 2008 Meredithet al 2011) Didelphimorphia (opossums) Paucituberculata(shrew-opossums) and Microbiotheria (the monito delmonte) The South American monito del monte(Dromiciops gliroides) is particularly interesting as its closeaffinity with the Australasian orders (together formingAustralidelphia) has implications for understanding the bio-geography and dispersal of ancestral marsupials in the LateCretaceous and Paleocene (Beck 2012)

Unfortunately the majority of past molecular studies haveonly been able to test a limited range of hypotheses Studiesseeking to resolve relationships among marsupial lineagesusing nucleotide sequence data have either included dispa-rate exemplar higher taxa but relatively few individual species(Phillips et al 2006 Beck 2008 Meredith Westerman Caseet al 2008 Meredith et al 2011 Phillips and Pratt 2008Nilsson et al 2010) or comprehensive species sampling butonly within restricted subclades (Meredith et al 2008a 2008b2010 Voss and Jansa 2009 Westerman et al 2012) Oftenthese individual data sets have involved different geneticloci making their results difficult to combine when usingsupermatrix approaches To circumvent this problem someauthors have turned to supertree methods (Cardillo et al2004 Bininda-Emonds et al 2007) However this approachhas several disadvantages including difficulties in quantifyingstatistical support for inferred topologies (de Queiroz andGatesy 2007 von Haeseler 2012) As a result phylogeniesfocusing on marsupials have generally been rather poorlyresolved and inadequately dated at least compared with pla-cental mammals

We have constructed a comprehensive phylogeny ofmodern marsupials to better understand their evolutionbiogeography and ecological history We sequenced 69 mi-tochondrial genomes using next-generation sequencingtechnology and synthesized this with existing nuclear andmitochondrial genomic nucleotide data to generate a super-matrix that includes 97 of extant genera and 58 of modernmarsupial species

Results and DiscussionMaximum likelihood (ML) Bayesian and parsimony analysesof the supermatrix resulted in well-resolved and concordantphylogenetic trees (fig 1 and supplementary fig S1 and dataset S1 Supplementary Material online) with 85 ofnodes in the ML tree supported by ML bootstrap support(MLBS) values 70 (fig 1 and Supplementary fig S1Supplementary Material online) Relationships among marsu-pial orders and families are consistent with most recent mo-lecular studies and generally receive high statistical support(MLBS 97) There is currently uncertainty surroundingthe precise affinities of two major clades Macropodiformes

(the kangaroos and wallabies) and Notoryctemorphia (themarsupial moles) We robustly recover Macropodiformesand Petauroidea (petauroid possums) as sister taxa(MLBS = 85) concordant with a recent study of majormammal lineages (Meredith et al 2011) this placement isrelatively novel as several previous studies of mitochondrial(Phillips and Pratt 2008) and nuclear loci (MeredithWesterman Case et al 2008 Meredith et al 2009) as wellas some aspects of morphology (Szalay 1994) favor an alter-native grouping of Macropodiformes with Phalangeroidea(phalangeroid possums) Additionally our results suggestthat Notoryctemorphia is sister to a clade consisting ofPeramelemorphia and Dasyuromorphia although supportfor this grouping is moderate (MLBS = 68) Relationshipsbelow the family level are largely concordant with previousestimates (eg Raterman et al 2006 Beck 2008 Meredith et al2008b 2009 2010 Phillips and Pratt 2008 Westerman et al2008 Voss and Jansa 2009 Westerman et al 2012)

To estimate divergence times among marsupials we em-ployed a set of 14 established fossil-based node age con-straints (supplementary table S1 Supplementary Materialonline) (see Meredith Westerman Case et al 2008Meredith et al 2008a 2008b 2009 2010 2011 Westermanet al 2012) for which taxonomic assignment is well justifiedThese constraints range in age from a Late Cretaceous (712Ma) maximum bound for the divergence betweenEomarsupialia and Microbiotheria to a Middle Pliocene(362 Ma) minimum bound for the divergence between thebandicoot genera Isoodon and Perameles Divergence datesinferred in this study are broadly consistent with past esti-mates based on molecular data (Beck 2008 MeredithWesterman Case et al 2008 Meredith et al 2008a 20092010 2011 Westerman et al 2012) The mean age of themarsupial crown group is 868 Ma (95 highest posteriordensity [HPD] 797ndash946) whereas the four Australasianorders diverge within a relatively brief window closely associ-ated with the KPg boundary 67ndash64 Ma The divergence be-tween the Australasian stem lineage and Microbiotheria onlyslightly predates this period (~69 Ma) but is in turn precededby an approximately 14 Ma internode Such a tight clusteringof divergences relatively deep in the tree is unlikely under astandard birthndashdeath diversification model This may indicaterapid diversification of ancestral australidelphians catalyzedby ecological opportunity following the KPg mass extinctionor colonization of AntarcticaAustralia from South AmericaThe latter hypothesis implies that the current distribution ofthe monito del monte may reflect a back migration fromAntarcticaAustralia to South America in the earlyCenozoic (Beck 2012)

Several previous studies have constructed comprehensivemarsupial phylogenies with similar aims to this study (Cardilloet al 2004 Bininda-Emonds et al 2007) However these usedsupertree approaches and do not recover several widely ac-cepted clades that are well supported by our analyses (egPotoroidae Echymiperinae and Eomarsupialia) Similarlymany marsupial genera in these previous studies are repre-sented by polytomies with resolution between congenericspecies completely lacking Lack of resolution makes the

2

Mitchell et al doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

dating analyses of Bininda-Emonds et al (2007) inaccurate formany clades which is exacerbated by the fact that approxi-mately 25 of node ages in their marsupial phylogeny wereinterpolated rather than empirically estimated Perhaps as aresult mean node ages inferred by Bininda-Emonds et al(2007) are substantially older for many clades (egDidelphimorphia) than those inferred by more recent studies(eg Jansa et al 2014) including our own Consequently the

direct dating and high resolution of our phylogeny make it amore robust scaffold for testing macroevolutionary hypothe-ses than existing supertrees

Extant marsupials inhabit a diversity of habitat types rang-ing from rainforest to desert Figure 2 illustrates that habitatpreference is highly phylogenetically conserved UsingBayesian ancestral state reconstruction we calculated theprobability that each node on the phylogeny (representing

Caenolestidae

Hypsiprymnodontidae

Didelphidae

Microbiotheriidae

Notoryctidae

Peramelidae

Thylacomyidae

Dasyuridae

Macropodidae

Burramyidae

Acrobatidae

Vombatidae

Phascolarctidae

Potoroidae

Phalangeridae

Pseudocheiridae

Petauridae

Tarsipidae

MyrmecobiidaeThylacinidae

Time (Ma)

Didelphimorphia

Paucituberculata

Microbiotheria

Notoryctemorphia

Peramelemorphia

Dasyuromorphia

Diprotodontia

Eomarsupialia

Australidelphia

80 70 60 50 40 30 20 10 0

PiMioceneOligoceneEocenePaleoceneLate Cretaceous Pt

1

2

3

FIG 1 Time-calibrated phylogeny of 193 marsupial species based on ML topology (supplementary data set S1 Supplementary Material online) Allnodes are strongly supported by all three methods (ML bootstrap90 Bayesian posterior probability095 and parsimony bootstrap90) unlessindicated by a dot Black dot indicates a node appears in trees inferred using all three methods but is not strongly supported in at least one tree Blue dotindicates node absent in Bayesian tree green dot indicates node absent in parsimony tree and red dot indicates node absent in both Bayesian andparsimony trees Shading behind branches reflects ancestral distribution as estimated using Lagrange light blue for the Americas orange for Australiaand red for New Guinea and Wallacea Lineages that are inferred by Lagrange to be codistributed between Australia and New GuineaWallacea areshaded purple Numbers indicate New GuineanWallacean ldquoold endemicrdquo clades (see main text) 1) New GuineanWallacean bandicoots(Echymiperinae and Peroryctinae) 2) ldquoMurexiardquo dasyures and 3) cuscuses (phalangerin possums) Geological epochs are marked on the x-axis LateCretaceous Paleocene Eocene Oligocene Miocene Pliocene (Pi) and Pleistocene (Pt) The Holocene (~11 ka to the present day) is not visible at thisscale Parsimony optimization of distribution based on the ML topology gives comparable overall results (supplementary data set S1 SupplementaryMaterial online) Individual species names and locus coverage are presented in supplementary figure S1 Supplementary Material online

3

Marsupials doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

an ancestral species) was associated with a given habitattype wet-closed (eg rainforest) mesic-open (eg sclerophyllor grassland) or arid A large proportion of the posteriorprobability for the most basal nodes in the phylogeny wasassociated with wet-closed environments (fig 2) This sug-gests that most of the primary divergences among extantmarsupial groups occurred in a rainforest context consistentwith many paleontological studies (Travouillon et al 2009)However the ancestral habitat on nodes at the base of theaustralidelphian radiation is unclear with intermediate(mesic-open) habitat often having the highest probability

Our results also suggest that adaptation to open and aridhabitats has evolved independently numerous times amongmarsupials and that these events occurred relatively recentlyAmong South American marsupials transitions from wet-closed to drier and more open environments appear unidi-rectional and begin no earlier than the Miocene (Jansa et al2014) (fig 2) The timing of these transitions follows theOligocene retraction of rainforest and expansion of woodedsavanna and grassland in southern South America (Ortiz-Jaureguizar and Cladera 2006 Barreda and Palazzesi 2007Iglesias et al 2011) Concordantly extant open-habitat-adapted opossum lineages (eg Thylamys and Lestodelphys)are distributed primarily in southern Brazil ParaguayUruguay Chile and Argentina whereas their respectivesister taxa are generally distributed in rainforests to the

north In Australia rainforest was the dominant habitattype across most of the continent during the earlyCenozoic (Frakes et al 1987) Open sclerophyll and grasslanddid not become widespread until after the Middle Mioceneand true arid zones did not develop until the Pliocene (Martin2006 White 2006) Given this sequence of events it is gener-ally thought that the modern Australian arid biota evolvedfrom ancestral mesic forms and that the ancestral state forthese clades should consequently optimize as wet-closed(Byrne et al 2008 2011) However departures from this ex-pected direction of habitat evolution are observed for severalAustralasian taxa For example several clades and individualspecies appear to have undergone reversions from mesic-open to wet-closed habitats (fig 2) particularly withinMacropodidae and Dasyuridae Additionally several groupsincluding Dasyuromorphia Vombatiformes and Trichosuriniappear to have made the transition from wet-closed to mesic-open environments prior to the onset of widespread Miocenearidification One explanation for this observation is thatthese older inferred transitions reflect lineages that survivedfrom drier periods in the Oligocene when limited open foresthabitats may have been available (Travouillon et al 2009)

A limitation of molecular analyses is that they are basedpurely on modern species This may lead to inference of ar-tificially old wet-closed to mesic-open transitions throughfailure to sample extinct rainforest-adapted lineages For

Didelphidae(opossums)

Caenolestidae(shrew opossums)

Macropodidae(inc kangaroos and wallabies)

Potoroidae(inc potoroos and bettongs)

Hypsiprymnodontidae(musky rat-kangaroo)

Pseudocheiridae(inc ringtail possums)

Petauridae(inc wrist-winged

gliders)

Tarsipidae(honey possum)

Acrobatidae (inc feathertail glider)

Phalangeridae(cuscuses and brushtail possums)

Burramyidae (pygmy possums)

Vombatidae (wombats)

Phascolarctidae (koala)

Dasyuridae (inc quolls and Tasmanian devil)Myrmecobiidae (numbat)Thylacinidae (thylacine)

Peramelidae(bandicoots)

Thylacomyidae (bilby)

Notoryctidae (marsupial moles)

Microbiotheriidae(monito del monte)

Arid

Mesic-open

Wet-closed

80

70

60

50

40

30

20

10

Time (Ma)

FIG 2 Ancestral habitat preference reconstruction using Bayesian optimization methods on the ML topology (fig 1 and supplementary fig S1 and dataset S1 Supplementary Material online) Boxes at tips reflect the modern state whereas pie charts reflect the inferred probability of a particular state ateach ancestral node Modern taxa inhabiting multiple habitat types are represented by an appropriate color gradient Parsimony optimization of habitatstates gives comparable results (supplementary data set S1 Supplementary Material online)

4

Mitchell et al doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

instance support for an early origin of mesic-open habitats indasyuromorphs decreases when their close relatives the arid-adapted marsupial moles (Notoryctemorphia) are omittedfrom the analysis The two extant marsupial mole species areassociated exclusively with arid habitats but extinct mesicnotoryctids are known from Miocene Riversleigh layers(Archer et al 2010) Additionally extinct dasyurid and vom-batiform taxa are known from Pleistocene rainforest assem-blages (Cramb et al 2009 Black et al 2012) However even ifwe were to assume that no single wet-closed to mesic-opentransition is truly older than the Middle Miocene these tran-sitions are still temporally dispersed over a long period Thissuggests that evolution toward mesic-open and arid habitatshas been a gradual process rather than a concerted shiftacross many Australasian clades in association with someabrupt climatic event This observation along with evidencefor reversions from mesic-open to wet-closed habitats indi-cates a high level of ongoing plasticity among marsupials withregard to habitat adaptation

A large proportion of Australasian wet-closed forest line-ages inhabit New Guinea and Wallacea rather than Australia(Flannery 1995) The colonization timeline of these islands isuncertain long-distance overwater dispersal is improbable formost marsupial taxa and due to the geological complexity ofthe region evidence for the presence and duration of landconnections is equivocal (Metcalfe et al 2001) Previous stud-ies have attempted to use the age and distribution of marsu-pial lineages to infer periods of biotic connection betweenAustralia and New GuineaWallacea (Aplin et al 1993) Weimplemented a similar approach on our phylogeny usingbiogeographic reconstructions to identify the temporalorigin of New GuineanWallacean marsupial taxa (fig 1)For taxa inferred to have colonized New GuineaWallaceafrom Australia the oldest probable dispersal time is repre-sented by the divergence between the ancestor of a NewGuineanWallacean clade and its Australian sister taxonwhereas the most recent probable dispersal time is markedby the age of the New GuineanWallacean crown cladeInferred dispersal dates for any individual clade may be arti-ficially young or old due to extinction of Australian lineages orparallel colonization events respectively Thus the power ofour study for testing biogeographical hypotheses comes fromthe sheer number of independent dispersal events encom-passed by our data set

A vicariant origin of the New GuineanWallacean marsu-pial fauna has previously been advanced based on early geo-logical reconstructions (Flannery 1988) Under this scenario asubstantial part of New Guinea was emergent and connectedto Australia until the Oligocene when the intervening PapuanBasin was inundated (Dow 1976) thereby severing land con-nection and preventing dispersal Our inferred dates are uni-formly too young to be consistent with this hypothesisIndeed more recent geological reconstructions indicate thatin fact no substantial part of New Guinea was emergent untilthe Miocene (van Ufford and Cloos 2005) Subsequently ithas been suggested that the ancestors of New GuineanWallacean marsupials dispersed no earlier than the latestMiocene (Westerman et al 2012) when a substantial drop

in global sea levels may have permitted overland dispersal(Hodell et al 1986) Most of our inferred dispersal dates canbe reconciled with this scenario andor more recent periodsof low sea level in the Pliocene and Pleistocene However theage of several clades is unexpectedly old given this hypothesis

The crown ages of the cuscuses (phalangerin possums)New Guinean bandicoots (Peroryctinae and Echymiperinae)and ldquoMurexiardquo dasyures suggest a period of accessibility be-tween Australia and New GuineaWallacea at the beginningof the Late Miocene (fig 1) The youngest inferred date for thepresence of cucuses in New GuineaWallacea is 1065 Ma(95 HPD minimum bound of the crown group) the analo-gous values for New Guinean bandicoots and dasyures are984 and 928 Ma respectively Although inference of suchearly colonization dates could also be explained by multipleparallel dispersals within each clade at the end of the Mio-cene the tight concordance among the crown ages of thesethree clades strongly suggests a dispersal window 11ndash9 MaSimilarly sweepstakes long-range overwater dispersal is un-likely to have generated three relatively simultaneous eventsThis suggests that dispersal has been at least intermittentlypossible between mainland Australia and New Guinea viadiffuse connection since the Middle Miocene A period ofconnection 11ndash9 Ma is broadly consistent with the originof the bulk of modern New Guinea which formed as aresult of orogeny approximately 12 Ma (van Ufford andCloos 2005) and is consistent with the suggestion of a similarperiod of dispersal based on microcomplement fixation(Aplin et al 1993)

This study presents a comprehensive molecular phylogenyof marsupials and highlights substantial macroevolutionaryheterogeneity The widely hypothesized trend of ancestralwet-closed forest lineages evolving into open and aridforms since the Middle Miocene appears to be overly simplis-tic Several lineages have evolved in the opposite directionAdditionally the taxonomic breadth of our phylogeny allowsus to compare the timing of numerous phylogeneticallyindependent dispersals from Australia to New GuineaWallacea The age and concordance of these events providessubstantial evidence of biological connectivity between NewGuinea and Australia coincident with New Guinearsquos principalformation approximately 12 Ma Ultimately our results sug-gest that marsupial evolution in the Cenozoic was dynamicand characterized by on-going ecological plasticity and op-portunistic dispersal

Materials and Methods

Data Set

We sequenced mitochondrial genomes (mitogenomes)for 69 extant marsupial species (supplementary table S2Supplementary Material online) Sixty-three mitogenomeswere sequenced according to a previously published protocol(Lerner et al 2011) using a combination of multiplexed 454pyrosequencing (Meyer et al 2008) and traditional capillaryelectrophoresis First we used long-range polymerase chainreaction (see supplementary table S3 SupplementaryMaterial online for primer sequences) to amplify each

5

Marsupials doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

mitogenome in several fragments Each fragment was thensheared barcoded and sequenced on a Roche454 GS FLXResulting sequencing reads were de novo assembled using apreviously published pipeline (Lerner et al 2011) Sequenceswere demultiplexed using ldquountagrdquo (httpbioinfevampgdepts last accessed June 5 2014) and consensus sequenceswere created using ldquonewblerrdquo (v 200020) SeqMan Pro(Lasergene suite 8 DNASTAR) and MIA (httpsourceforgenetprojectsmia-assembler last accessed June 5 2014)Remaining gaps and areas of low coverage were completedusing big dye terminator chemistry on an ABI 3130 (see sup-plementary table S3 Supplementary Material online forprimer sequences) The remaining six mitogenomes were ob-tained via shotgun sequencing of raw extract on an IlluminaGenome Analyzer II (one lane per sample) Sequencing readswere mapped to published mitogenomes of closely relatedspecies using TMAP v322 (httpsgithubcomnh13TMAPlast accessed June 5 2014) and samtools v14 (Li et al 2009)according to a previously published pipeline (Mitchell et al2014) duplicate reads were removed with PicardTools v179(httppicardsourceforgenet last accessed June 5 2014) anda consensus sequence was generated in Geneious v612(Biomatters httpwwwgeneiouscom last accessed June 52014)

New sequence data were combined with 32 previouslypublished mitogenomes and data from 26 nuclear loci toform a single nucleotide supermatrix (supplementary fig S1and data set S2 Supplementary Material online) The finaldata set included 193 marsupial species and ten outgrouptaxa Two monotremes and eight placental mammals Thesupermatrix (43616 bp) was 39 complete (excluding out-groups) with the proportion of missing taxa per locus varyingfrom 125 to 88 (0ndash15 at the family level and 0 at theorder level) We divided this alignment into 72 discrete binsCodon positions of each protein-coding nuclear locus firstand second codon positions of all H-strand mitochondrialprotein-coding genes (concatenated into one locus) individ-ual introns and stem and loop positions of mitochondrialRNA-coding loci (all rRNAs and tRNAs were concatenated)Mitochondrial third codon positions were discarded to min-imize bias in branch length estimation arising from saturationThese bins were analyzed using partitionfinder v092 (Lanfearet al 2012) to determine the most appropriate partitioningscheme and substitution models (supplementary table S4Supplementary Material online) for downstream analysis inRAxML v728 MrBayes v321 and MCMCtree (PAML v46)

PhylogeneticsModel-Based MethodsTree topology was estimated under ML and Bayesian frame-works using RAxML v728 (Stamatakis 2006) and MrBayesv321 (Ronquist and Huelsenbeck 2003) respectively BothRAxML and MrBayes gave very similar trees (fig 1 and sup-plementary fig S1 and data set S1 Supplementary Materialonline) All analyses were repeated with and without out-group taxa (placentals and monotremes) to assess the sensi-tivity of the splits within marsupials to outgroup sampling

these analyses produced very similar trees (supplementarydata set S1 Supplementary Material online) Our RAxMLanalysis comprised an ML search for the best-scoring treefrom 1000 bootstrap replicates MrBayes analyses wereundated (clock free) as dated (clock) analyses did not con-verge There were three runs Each individual run employedfour Markov chains (one cold and three incrementallyheated) with default priors Each chain ran for 107 genera-tions sampling every 500 Convergence in topology was as-sessed using the average standard deviation of splitfrequencies (lt002) whereas convergence in individual pa-rameter values was assessed through broadly overlapping dis-tributions in Tracer v15 (httptreebioedacuksoftwaretracer last accessed June 5 2014) and effective samplesizes more than 200 Trees generated by the three runswere pooled before being summarized and the first 25 oftrees from each run were discarded as burn-in

Parsimony MethodsParsimony analyses used PAUP (Swofford 2002) with mostparsimonious trees found via heuristic searches involving 100random addition searches followed by a strict consensusbootstrapping was performed with 100 replicates of simpleheuristic searches (due to time constraints) followed by amajority-rule consensus All analyses were again repeatedwith and without outgroup taxa with similar results (supple-mentary data set S1 Supplementary Material online)

Molecular Dating

We implemented 14 node calibrations across the marsupialtree (supplementary table S1 Supplementary Materialonline) Most constraints followed previous studies (egMeredith Westerman Case et al 2008 Meredith et al2008a 2008b 2009 2010 2011 Westerman et al 2012)with minimum bounds being chosen based on the oldestknown occurrence of fossil crown-group taxa Maximumbounds were determined using either stratigraphic bounding(Benton and Donoghue 2007) or phylogenetic bracketing(Reisz and Muller 2004 Muller and Reisz 2005) To allow forthe patchiness of the marsupial fossil record we followed theconservative philosophy of Meredith Westerman Springer(2008) when determining maxima Stratigraphic boundingmaxima were judged based on absence of the lineage of in-terest from two (rather than one) preceding fossil bearinglayers whereas phylogenetic bracketing maxima werejudged based on the oldest known occurrence of at leastthe second closest (rather than first closest) outgroup tothe target clade

Molecular dating was performed using MCMCtree withinthe PAML v46 software package (Yang 2007) on the RAxMLtree topology All calibrations (supplementary table S1Supplementary Material online) were implemented as uni-form priors with hard minima and soft maxima (975) Tomaximize computational efficiency we used the likelihoodapproximation approach implemented in MCMCtree andperformed our analysis on ingroup taxa only For the purposesof our MCMCtree analyses we defined individual time unitsas 10 Ma such that most node ages fell between 10 and 01

6

Mitchell et al doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

units Among-partition rate variation was modeled using agamma distribution with the parameter values = 1 and= 56 After a burn-in of 104 iterations each chain was runfor 8 105 iterations sampling every 80 for a total of 104

samples To ensure convergence and adequate samplingtwo separate runs were performed and each parameterwas monitored in Tracer v15 (httptreebioedacuksoftwaretracer last accessed June 5 2014) Parameter valuesfor one partition (out of 24) consistently failed to convergedue to a low number of informative sites this partition wasconsequently omitted from the final analysis (see supplemen-tary table S4 Supplementary Material online)

Ancestral State Reconstruction

Bayesian trait reconstruction was performed on the finaldated phylogeny using the MCMC option for discrete stateevolution in BayesTraits v10 (Pagel et al 2004) The analysiswas run for 106 iterations sampling every 103 The first 10 ofsamples were discarded as burn-in To ensure convergenceand adequate sampling each parameter was monitored inTracer v15 (httptreebioedacuksoftwaretracer lastaccessed June 5 2014) All modern taxa were coded as wet-closed mesic-open or arid adapted (or a combination ofthese) according to their distribution and habits (supplemen-tary table S5 Supplementary Material online) An ordered(wet-closed to mesic-open to arid) single-rate model wasused for evolution between the three character states Atwo-rate ordered model allowing a separate rate for transi-tions toward more arid environments and transitions towardmore mesic environments provided no significant improve-ment in fit as determined by Bayes factor comparison(BFlt 05) we thus used the simpler model in our analysisto avoid overfitting Both ordered models provided a substan-tially better fit to the data than corresponding unorderedmodels (BFgt 5) To test the robustness of our resultsunder different optimization methods we performed strictparsimony state reconstructions (again using both orderedand unordered models) in Mesquite (Maddison WP andMaddison DR 2011) with concordant results (supplementarydata set S1 Supplementary Material online)

Geographical Range Evolution

We used Lagrange v20120508 (Ree and Smith 2008) to re-construct ancestral distribution and infer dispersal times Themodel employed by this program attempts to account fordispersal fusion and fission of geographic regions speciationand local extinction Distribution was coded as one (or acombination) of three states (supplementary table S5Supplementary Material online) America Australia or NewGuineaWallacea Individual species (including ancestral spe-cies) were permitted to simultaneously inhabit the Americasand Australia (to reflect potential connection scenarios in thePaleocene) and Australia and New GuineaWallacea Themodel we employed allowed range shifts to occur bidirection-ally but only between adjacent areas (Australia and NewGuineaWallacea or America and Australia) because adirect connection between the Americas and New Guinea

Wallacea is not plausible The most probable inherited rangefor each node on the phylogeny is reflected in figure 1 To testour results we performed a strict parsimony state reconstruc-tion in Mesquite (Maddison WP and Maddison DR 2011)with concordant results (supplementary data set S1Supplementary Material online)

Supplementary MaterialSupplementary tables S1ndashS5 figure S1 and data set S1 areavailable at Molecular Biology and Evolution online (httpwwwmbeoxfordjournalsorg)

Acknowledgments

This study was funded by the Australian Research CouncilHigh-performance computing was provided by eResearch SAFor access to tissue and DNA samples the authors thank JPatton and C Cicero (Museum of Vertebrate Zoology) RVoss S Jansa and E Westwig (American Museum ofNatural History) J Cook (Museum of SouthwesternBiology) W Foster (Adelaide Zoo) J Graves (AustralianNational University) South Australian Museum and LJoseph and R Palmer (Australian National WildlifeCollection)

ReferencesAlfaro ME Santini F Brock C Alamillo H Dornburg A Rabosky DL

Carnevale G Harmon LJ 2009 Nine exceptional radiations plus highturnover explain species diversity in jawed vertebrates Proc NatlAcad Sci U S A 10613410ndash13414

Aplin K Baverstock PR Donnellan SC 1993 Albumin immunologicalevidence for the time and mode of origin of the New Guineanterrestrial mammal fauna Sci New Guinea 19131ndash145

Archer M 1984 The Australian marsupial radiation In Archer MClayton G editors Vertebrate zoogeography and evolution inAustralia Carlisle (Australia) Hesperian Press p 633ndash708

Archer M Arena R Bassarova M Black K Brammall J Cooke B CreaserP Crosby K Gillespie A Godthelp H et al 1999 The evolutionaryhistory and diversity of Australian mammals Aust Mammal 211ndash45

Archer M Beck R Gott M Hand S Godthelp H Black K 2010 Australiarsquosfirst fossil marsupial mole (Notoryctemorphia) resolves controver-sies about their evolution and palaeoenvironmental origins Proc RSoc B 2781498ndash1506

Avise JC 2000 Phylogeography the history and formation of speciesCambridge (MA) Harvard University Press

Avise JC Arnold J Ball RM Bermingham E Lamb T Neigel JE Reeb CASaunders NC 1987 Intraspecific phylogeographymdashthe mitochon-drialndashDNA bridge between population-genetics and systematicsAnnu Rev Ecol Syst 18489ndash522

Barreda V Palazzesi L 2007 Patagonian vegetation turnovers during thePaleogenendashEarly Neogene origin of arid-adapted floras Bot Rev 7331ndash50

Beck RM 2012 An lsquoameridelphianrsquo marsupial from the early Eocene ofAustralia supports a complex model of Southern Hemisphere mar-supial biogeography Naturwissenschaften 99715ndash729

Beck RMD 2008 A dated phylogeny of marsupials using a molecularsupermatrix and multiple fossil constraints J Mammal 89175ndash189

Benton MJ Donoghue PCJ 2007 Paleontological evidence to date thetree of life Mol Biol Evol 2426ndash53

Bininda-Emonds ORP Cardillo M Jones KE MacPhee RDE Beck RMDGrenyer R Price SA Vos RA Gittleman JL Purvis A 2007 Thedelayed rise of present-day mammals Nature 446507ndash512

Black K Archer M Hand S Godthelp H 2012 The rise of Australianmarsupials a synopsis of biostratigraphic phylogenetic

7

Marsupials doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

palaeoecologic and palaeobiogeographic understanding In Talent Jeditor Earth and life Global biodiversity extinction intervals andbiogeographic perturbations through time Dordrecht (TheNetherlands) Springer p 983ndash1078

Byrne M Steane DA Joseph L Yeates DK Jordan GJ Crayn D Aplin KCantrill DJ Cook LG Crisp MD et al 2011 Decline of a biomeevolution contraction fragmentation extinction and invasion ofthe Australian mesic zone biota J Biogeogr 381635ndash1656

Byrne M Yeates DK Joseph L Kearney M Bowler J Williams MAJCooper S Donnellan SC Keogh JS Leys R et al 2008 Birth of abiome insights into the assembly and maintenance of theAustralian arid zone biota Mol Ecol 174398ndash4417

Cardillo M Bininda-Emonds ORP Boakes E Purvis A 2004 A species-level phylogenetic supertree of marsupials J Zool (Lond) 26411ndash31

Cramb J Hocknull S Webb GE 2009 High diversity Pleistocene rain-forest dasyurid assemblages with implications for the radiation ofthe Dasyuridae Austral Ecol 34663ndash669

de Queiroz A Gatesy J 2007 The supermatrix approach to systematicsTrends Ecol Evol 2234ndash41

Dow DB 1976 A geological synthesis of Papua New Guinea Vol 201Canberra (Australia) Bureau of Mineral Resources Geology andGeophysics p 41

Felsenstein J 1985 Phylogenies and the comparative method Am Nat1251ndash15

Flannery T 1995 Mammals of New Guinea Chatswood (Australia)Reed Books

Flannery TF 1988 Origins of the Australo-Papuan land mammal faunaAust Zool Rev 115ndash24

Frakes LA McGowran B Bowler JM 1987 Evolution of Australian en-vironments In Dyne GR Walton DW editors Fauna of AustraliaVol 1A Canberra (Australia) Australian Government PublishingService p 1ndash16

Hodell DA Elmstrom KM Kennett JP 1986 Latest Miocene benthicd18O changes global ice volume sea-level and the Messinian salinitycrisis Nature 320411ndash414

Huey RB 1987 Phylogeny history and the comparative method InFeder ME Bennett AF Burgen W Huey RB editors New directionsin ecological physiology Cambridge (United Kingdom) CambridgeUniversity Press p 76ndash98

Hugall AF Stuart-Fox D 2012 Accelerated speciation in colour-poly-morphic birds Nature 485631ndash634

Iglesias ARI Artabe AE Morel EM 2011 The evolution of Patagonianclimate and vegetation from the Mesozoic to the present Biol J LinnSoc 103409ndash422

Jansa SA Barker FK Voss RS 2014 The early diversification of didelphidmarsupials a window into South Americarsquos ldquosplendid isolationrdquoEvolution 68684ndash695

Lanfear R Calcott B Ho SY Guindon S 2012 PartitionFinder combinedselection of partitioning schemes and substitution Mol Biol Evol 291537ndash1719

Lerner HR Meyer M James HF Hofreiter M Fleischer RC 2011Multilocus resolution of phylogeny and timescale in the extantadaptive radiation of Hawaiian honeycreepers Curr Biol 211838ndash1844

Li H Handsaker B Wysoker A Fennell T Ruan J Homer N Marth GAbecasis G Durbin R Subgroup GPDP 2009 The sequence align-mentmap (SAM) format and SAMtools Bioinformatics 252078ndash2079

Maddison WP Maddison DR 2011 Mesquite a modular system forevolutionary analysis Version 275 [Internet] [cited 2014 Jun 5]Available from httpmesquiteprojectorg

Martin HA 2006 Cenozoic climatic change and the development of thearid vegetation in Australia J Arid Environ 66533ndash563

Meredith RW Janecka JE Gatesy J Ryder OA Fisher CA Teeling ECGoodbla A Eizirik E Simao TLL Stadler T et al 2011 Impacts of theCretaceous terrestrial revolution and KPg extinction on mammaldiversification Science 334521ndash524

Meredith RW Mendoza MA Roberts KK Westerman M Springer MS2010 A phylogeny and timescale for the evolution of

Pseudocheiridae (Marsupialia Diprotodontia) in Australia andNew Guinea J Mamm Evol 1775ndash99

Meredith RW Westerman M Case JA Springer MS 2008 A phylogenyand timescale for marsupial evolution based on sequences for fivenuclear genes J Mamm Evol 151ndash36

Meredith RW Westerman M Springer MS 2008a A phylogeny andtimescale for the living genera of kangaroos and kin(Macropodiformes Marsupialia) based on nuclear DNA sequencesAust J Zool 56395ndash410

Meredith RW Westerman M Springer MS 2008b A timescale and phy-logeny for ldquoBandicootsrdquo (Peramelemorphia Marsupialia) based onsequences for five nuclear genes Mol Phylogenet Evol 471ndash20

Meredith RW Westerman M Springer MS 2009 A phylogeny ofDiprotodontia (Marsupialia) based on sequences for five nucleargenes Mol Phylogenet Evol 51554ndash571

Metcalfe I Smith JMB Morwood M Davidson I 2001 Faunal and floralmigrations and evolution in SE Asia-Australasia Lisse (TheNetherlands) Swets amp Zeitlinger Publishers

Meyer M Stenzel U Hofreiter M 2008 Parallel tagged sequencing onthe 454 platform Nat Protoc 3267ndash278

Mitchell KJ Wood JR Scofield RP Llamas B Cooper A 2014 Ancientmitochondrial genome reveals unsuspected taxonomic affinity ofthe extinct Chatham duck (Pachyanas chathamica) and resolvesdivergence times for New Zealand and sub-Antarctic brown tealsMol Phylogenet Evol 70420ndash428

Muller J Reisz RR 2005 Four well-constrained calibration points fromthe vertebrate fossil record for molecular clock estimates Bioessays271069ndash1075

Nilsson MA Churakov G Sommer M Van Tran N Zemann A Brosius JSchmitz J 2010 Tracking marsupial evolution using archaic genomicretroposon insertions PLoS Biol 8e1000436

Ortiz-Jaureguizar E Cladera GA 2006 Paleoenvironmental evolution ofsouthern South America during the Cenozoic J Arid Environ 66498ndash532

Pagel M 1997 Inferring evolutionary processes from phylogenies ZoolScr 26331ndash348

Pagel M Meade A Barker D 2004 Bayesian estimation of ancestralcharacter states on phylogenies Syst Biol 53673ndash684

Phillips MJ McLenachan PA Down C Gibb GC Penny D 2006Combined mitochondrial and nuclear DNA sequences resolve theinterrelations of the major Australasian marsupial radiations SystBiol 55122ndash137

Phillips MJ Pratt RC 2008 Family-level relationships amongthe Australasian marsupial ldquoherbivoresrdquo (Diprotodontia Koalawombats kangaroos and possums) Mol Phylogenet Evol 46594ndash605

Raterman D Meredith RW Ruedas LA Springer MS 2006 Phylogeneticrelationships of the cuscuses and brushtail possums (Marsupialia Phalangeridae) using the nuclear gene BRCA1 Aust J Zool 54353ndash361

Ree RH Smith SA 2008 Maximum likelihood inference of geographicrange evolution by dispersal local extinction and cladogenesis SystBiol 574ndash14

Reisz RR Muller J 2004 Molecular timescales and the fossil record apaleontological perspective Trends Genet 20237ndash241

Ronquist F Huelsenbeck JP 2003 MrBayes 3 Bayesian phylogeneticinference under mixed models Bioinformatics 191572ndash1574

Stadler T 2011 Mammalian phylogeny reveals recent diversification rateshifts Proc Natl Acad Sci U S A 1086187ndash6192

Stamatakis A 2006 RAxML-VI-HPC maximum likelihood-based phylo-genetic analyses with thousands of taxa and mixed modelsBioinformatics 222688ndash2690

Swofford DL 2002 PAUP Phylogenetic analysis using parsimony (andother methods) Sunderland (MA) Sinauer Associates

Szalay FJ 1994 Evolutionary history of the marsupials and an analysis ofosteological characters Cambridge (MA) Cambridge UniversityPress

Travouillon KJ Legendre S Archer M Hand SJ 2009 Palaeoecologicalanalyses of Riversleighrsquos Oligo-Miocene sites implications for Oligo-

8

Mitchell et al doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from

Miocene climate change in Australia Palaeogeogr PalaeoclimatolPalaeoecol 27624ndash37

van Ufford AQ Cloos M 2005 Cenozoic tectonics of New GuineaAAPG Bull 89119ndash140

von Haeseler A 2012 Do we still need supertrees BMC Biol 1013Voss RS Jansa SA 2009 Phylogenetic relationships and classification of

didelphid marsupials an extant radiation of new world metatherianmammals Bull Am Mus Nat Hist 3221ndash177

Westerman M Kear BP Aplin K Meredith RW Emerling C Springer MS2012 Phylogenetic relationships of living and recently extinct ban-dicoots based on nuclear and mitochondrial DNA sequences MolPhylogenet Evol 6297ndash108

Westerman M Young J Krajewski C 2008 Molecular relationships ofspecies of Pseudantechinus Parantechinus and Dasykaluta(Marsupialia Dasyuridae) Aust Mammal 29201ndash212

White ME 2006 Environments of the geological past In Merrick JRArcher M Hickey GM Lee MSY editors Evolution and biogeogra-phy of Australasian vertebrates Sydney (Australia) Auscipub PtyLtd

Wilson DE Reeder DM 2005 Mammal species of the world a taxo-nomic and geographic reference Baltimore (MD) The JohnsHopkins University Press

Yang Z 2007 PAML4 a program package for phylogenetic analysis bymaximum likelihood Mol Biol Evol 241586ndash1591

9

Marsupials doi101093molbevmsu176 MBE at Q

ueensland University of T

echnology on August 14 2014

httpmbeoxfordjournalsorg

Dow

nloaded from