Open Access
How to translate text using browser tools
28 June 2022 Craniodental Morphology and Phylogeny of Marsupials
Robin M.D. Beck, Robert S. Voss, Sharon A. Jansa
Author Affiliations +
Abstract

The current literature on marsupial phylogenetics includes numerous studies based on analyses of morphological data with limited sampling of Recent and fossil taxa, and many studies based on analyses of molecular data with dense sampling of Recent taxa, but few studies have combined both data types. Another dichotomy in the marsupial phylogenetic literature is between studies focused on New World taxa and those focused on Sahulian taxa. To date, there has been no attempt to assess the phylogenetic relationships of the global marsupial fauna based on combined analyses of morphology and molecular sequences for a dense sampling of Recent and fossil taxa. For this report, we compiled morphological and molecular data from an unprecedented number of Recent and fossil marsupials. Our morphological data consist of 180 craniodental characters that we scored for 97 terminals representing every currently recognized Recent genus, 42 additional ingroup (crown-clade marsupial) terminals represented by well-preserved fossils, and 5 outgroups (nonmarsupial metatherians). Our molecular data comprise 24.5 kb of DNA sequences from whole-mitochondrial genomes and six nuclear loci (APOB, BRCA1, GHR, RAG1, RBP3 and VWF) for 97 marsupial terminals (the same Recent taxa scored for craniodental morphology) and several placental and monotreme outgroups. The results of separate and combined analyses of these data using a wide range of phylogenetic methods support many currently accepted hypotheses of ingroup (marsupial) relationships, but they also underscore the difficulty of placing fossils with key missing data (e.g., †Evolestes), and the unique difficulty of placing others that exhibit mosaics of plesiomorphic and autapomorphic traits (e.g., †Yalkaparidon). Unique contributions of our study are (1) critical discussions and illustrations of marsupial craniodental morphology including features never previously coded for phylogenetic analysis; (2) critical assessments of relative support for many suprageneric clades; (3) estimates of divergence times derived from tip-and-node dating based on uniquely taxon-dense analyses; and (4) a revised, higher-order classification of marsupials accompanied by lists of supporting craniodental synapomorphies. Far from the last word on these topics, this report lays the foundation for future research that may be enabled by the discovery of new fossil taxa, better-preserved material of previously described taxa, novel morphological characters (e.g., from the postcranium), and improved methods of phylogenetic analysis.

INTRODUCTION

FRONTISPIECE.

The honey opossum (Tarsipes rostratus, illustrated by Gould, 1863) has aptly been described as “a paragon of autapomorphic specialization” (Aplin and Archer, 1987).

img-z2-1_01.jpg

Marsupialia (marsupials) comprises one of the three major extant mammalian clades, together with Placentalia (placentals) and Monotremata (monotremes). The name Marsupialia is now often used to refer to the crown clade—the last common ancestor of living marsupials and all its descendants (e.g., Rougier et al., 1998; Flynn and Wyss, 1999; Horovitz and Sánchez-Villagra, 2003; Sereno, 2006; Asher et al., 2007; Sánchez-Villagra et al., 2007; Horovitz et al., 2008, 2009; Forasiepi, 2009; Williamson et al., 2012, 2014; O'Leary et al., 2013; Beck et al., 2014; Forasiepi et al., 2014a; Wilson et al., 2016; Velazco et al., 2022). However, other recent studies have used the name Marsupialia in a more inclusive sense by incorporating fossil taxa that appear to lie outside the crown clade (e.g., Kielan-Jaworowska et al., 2004; Case et al., 2005; Fox and Naylor, 2006; Davis, 2007; Beck et al., 2008b). Beck et al. (2014) provided the following crown-based definition of Marsupialia, which will be followed here: the least inclusive clade containing Didelphis marsupialis Linnaeus, 1758; Caenolestes fuliginosus (Tomes, 1863); and Phalanger orientalis (Pallas, 1766).

Marsupialia sits within the more inclusive (total, or stem-based) clade Metatheria, which comprises marsupials together with all taxa more closely related to Marsupialia than to Placentalia (Sereno, 2006; O'Leary et al., 2013). We follow Sereno's (2006: table 10.1) stem-based phylogenetic definition of Metatheria, namely the most inclusive clade containing Didelphis marsupialis Linnaeus, 1758, but not Mus musculus Linnaeus, 1758.

Living marsupials include >400 currently recognized species (∼6% of modern mammalian diversity; Burgin et al., 2018; Eldridge et al., 2019; Mammal Diversity Database, 2021) belonging to seven clades that are currently ranked as orders in the Linnean hierarchy (table 1; Aplin and Archer, 1987; Wilson and Reeder, 2005; Burgin et al., 2018; Eldridge et al., 2019): Didelphimorphia (opossums), Paucituberculata (shrew opossums), Microbiotheria (the monito del monte, Dromiciops gliroides), Dasyuromorphia (predominantly carnivorous forms such as quolls, antechinuses, dunnarts, the Tasmanian devil, the numbat, and the recently extinct thylacine), Diprotododontia (possums, gliders, kangaroos, wallabies, rat kangaroos, wombats, koalas, etc.), Notoryctemorphia (marsupial moles), and Peramelemorphia (bandicoots and bilbies).

Dromiciops gliroides (the sole extant microbiotherian) and the three genera of modern paucituberculatans (Caenolestes, Lestoros, and Rhyncholestes, all members of the family Caenolestidae) are exclusively South American in distribution (Gardner, 2008). Fossil members of both orders are known from South America (e.g., Marshall, 1980; Marshall, 1982; Abello, 2013; Goin and Abello, 2013; Goin et al., 2016), and fossil microbiotherians are also known from Antarctica (Goin and Carlini, 1995; Goin et al., 1999, 2007c; Gelfo et al., 2019).

Extant didelphimorphians (all of which belong to the family Didelphidae) are predominantly Neotropical in distribution (Gardner, 2008; Voss and Jansa, 2009; Castro et al., 2021), with the unique exception of Didelphis virginiana, whose range extends from Central America to southern Canada. Interpretation of the fossil record of didelphimorphians is complicated by the fact that numerous dentally plesiomorphic metatherians, including many Mesozoic taxa, have often been referred to “Didelphimorphia” (sensu lato) by various authors (e.g., Kielan-Jaworowska et al., 2004; Case et al., 2005; Martin et al., 2005; Davis, 2007). However, phylogenetic analyses consistently indicate that “Didelphimorphia” in this latter sense is nonmonophyletic because it includes both marsupials and nonmarsupial metatherians (Rougier et al., 1998; Luo et al., 2003; Sánchez-Villagra et al., 2007; Horovitz et al., 2008, 2009; Luo et al., 2011; Williamson et al., 2012, 2014; Wilson et al., 2016; Maga and Beck, 2017; Bi et al., 2018).

By contrast, monophyly of Didelphidae (crown-clade Didelphimorphia; Voss and Jansa, 2009) is strongly supported by both morphological and molecular data (Jansa and Voss, 2000; Voss and Jansa, 2003, 2009; Steiner et al., 2005; Jansa et al., 2014; Amador and Giannini, 2016), and didelphids may have diversified only within the past 20–30 million years (Jansa et al., 2014; Beck and Taglioretti, 2020; but see Steiner, et al., 2005; mitchell, et al., 2014; Vilela, et al., 2015). Fossil didelphids are known almost exclusively from South America (Goin and Pardiñas, 1996; Goin, 1997a, 1997b; Cozzuol et al., 2006; Goin et al., 2009b, 2013; Voss and Jansa, 2009; Goin and de los Reyes, 2011; Antoine et al., 2013, 2016, 2017; Suárez Gómez, 2019; Beck and Taglioretti, 2020; Castro et al., 2021), with the exception of fossil remains of Didelphis from Pleistocene sites in North America (Bell et al., 2004; Morgan, 2018). South American fossil “sparassocynids” have also been generally accepted members of Didelphimorphia sensu stricto (i.e., more closely related to didelphids than to any other Recent family; Goin, 1991, 1995; Forasiepi et al., 2009; Engelman and Croft, 2014). However, Beck and Taglioretti (2020) concluded that “sparassocynids” fall within Didelphidae, as a lineage sister to Monodelphis, and assigned them tribal status (as Sparassocynini), which we follow here (table 2).

Of the remaining four modern orders, species of Dasyuromorphia, Diprotodontia, and Peramelemorphia today occur in mainland Australia, Tasmania, New Guinea, and surrounding islands, but the two extant notoryctemorphian species (Notoryctes caurinus and N. typhlops) occur only in the deserts of mainland Australia (Flannery, 1995a, 1995b; Van Dyck and Strahan, 2008). The known fossil record of these four orders is likewise restricted to this region (Archer, 1984c; Archer et al., 1999; Long et al., 2002; Archer and Hand, 2006; Black et al., 2012b), with the possible exception of isolated tarsals from the Eocene of southern Argentina that may be diprotodontian (Lorente et al., 2016).1

TABLE 1

Higher Classification and Geographic Distribution of Recent Marsupialsa

img-AVC_01.gif

Most phylogenetic analyses published over the past 15 to 20 years, particularly those using molecular or total-evidence datasets, have supported monophyly of the supraordinal clade that Szalay (1982a; 1982b; 1994) named Australidelphia, which includes the four Australian orders together with Microbiotheria (e.g., Phillips et al., 2001, 2006; Amrine-Madsen et al., 2003b; Horovitz and Sánchez-Villagra, 2003; Asher et al., 2004; Nilsson et al., 2004; Beck, 2008a, 2012; Beck et al., 2008a; Nilsson et al., 2010; Meredith et al., 2011; Mitchell et al., 2014; Lorente et al., 2016; Duchêne et al., 2018). Other superordinal relationships have proven more difficult to resolve confidently (Eldridge et al., 2019). However, retroposon insertion studies by Nilsson et al. (2010) and Gallus et al. (2015a) found relatively strong support for a clade comprising the four Australian orders (Eomarsupialia sensu Beck et al., 2014) to the exclusion of Microbiotheria and for positioning the root between Didelphimorphia and the other six orders. Recent molecular-sequence analyses have also consistently supported a clade comprising the orders Dasyuromorphia, Peramelemorphia, and Notoryctemorphia (e.g., Beck, 2008a; Meredith et al., 2011; Mitchell et al., 2014; Duchêne et al., 2018), as have some total-evidence analyses (Beck et al., 2008a; 2014; 2016; Beck, 2017a; Maga and Beck, 2017); Beck et al. (2014) named this clade Agreodontia. Gallus et al. (2015a) found that different retroposon insertions support different branching patterns within Agreodontia and suggested that the reason for this is that the three orders diverged in quick succession, with phenomena such as independent lineage sorting, hybridization, and/or gene flow resulting in conflicting phylogenetic signals. However, the recent phylogenomic study of Duchêne et al. (2018) found strong support for a sister-taxon relationship between Notoryctemorphia and Peramelemorphia to the exclusion of Dasyuromorphia.

Intraordinal relationships are also not fully resolved, with particular uncertainty surrounding the relationships among didelphid subfamilies (Voss and Jansa, 2009) and among phalangeridan diprotodontians (Phillips and Pratt, 2008; Duchêne et al., 2018). Another problem is that relatively few fossil taxa have been included in published phylogenies of higher-level marsupial relationships (Sánchez-Villagra, 2001; Sánchez-Villagra et al., 2007; Beck et al., 2008a, 2014, 2016; Horovitz et al., 2008, 2009; Abello and Candela, 2010, 2019; Beck, 2012, 2017a; Maga and Beck, 2017; Zimicz and Goin, 2020), limiting the usefulness of such phylogenies for comparative analyses of trait evolution (Slater et al., 2012), diversification (Rabosky, 2010, 2016), and biogeography (Crisp et al., 2011).

In contrast with the currently restricted distribution of Recent marsupials, fossil metatherians are known from every continent, including Antarctica (Weisbecker and Beck, 2015; Eldridge et al., 2019). The oldest metatherian was previously thought to be †Sinodelphys szalayi, from the Early Cretaceous (Barremian; ∼126 Mya) of China (Luo et al., 2003), but this taxon has subsequently been reinterpreted as a eutherian (Bi et al., 2018). Thus, the oldest metatherian fossils currently known are fragmentary dental remains from the Aptian-Albian of North America (Davis et al., 2008; Davis and Cifelli, 2011; Williamson et al., 2014; Cifelli and Davis, 2015). However, the description of a probable eutherian, †Juramaia sinensis, from the Middle Jurassic of China suggests that the Eutheria-Metatheria split might be as old as 160 Mya (Luo et al., 2011), congruent with several recent molecular estimates for this divergence (Meredith et al., 2011; dos Reis et al., 2012; Tarver et al., 2016).2

Current evidence suggests that many fossil metatherians are either stem taxa or only questionably members of Marsupialia (Rougier et al., 1998, 2004, 2015; Luo et al., 2003; Goin et al., 2006; Ladevèze and Muizon, 2007; Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Forasiepi, 2009; Ladevèze and Muizon, 2010; Luo et al., 2011; Beck, 2012, 2017a; Williamson et al., 2012, 2014; Engelman and Croft, 2014; Forasiepi et al., 2014a; Wilson et al., 2016; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Bi et al., 2018; Carneiro, 2018, 2019; Carneiro et al., 2018; Engelman et al., 2020; Muizon et al., 2018; Ladevèze et al., 2020; Muizon and Ladevèze, 2020).3 This is the case for all known Mesozoic forms (e.g., †Asiatherium, deltatheroidans,”alphadontids,” ”pediomyids,” stagodontids), the predominantly Laurasian Cenozoic taxa traditionally referred to the families †Herpetotheriidae and †Peradectidae (neither of which was recovered as monophyletic in the phylogenetic analyses of Williamson et al., 2012, 2014), and South American Cenozoic groups such as pucadelphyids, protodidelphids, caroloameghiniids, derorhynchids, sparassodon-tans, and polydolopimorphians (Eldridge et al., 2019). Definitive (crown-clade) marsupials are known only from the Cenozoic and (with the exception of living North American species descended from South American immigrants; Jansa et al., 2014) are entirely Gondwanan in distribution (Beck, 2017a). Of these, nearly all can be identified as belonging to one or another of the seven Recent orders, with a handful of exceptions (Beck et al., 2008a, 2014; Beck, 2012).

To date, available morphological and total-evidence phylogenies focusing on the higher-level relationships of Marsupialia have suffered from limited taxon sampling (Horovitz and Sánchez-Villagra, 2003; Asher et al., 2004; Beck et al., 2008a, 2014, 2016; Horovitz et al., 2008, 2009; Beck, 2012, 2017a; Maga and Beck, 2017), whereas molecular phylogenies, some of which are taxonomically densely sampled (e.g., Kirsch et al., 1997; Meredith et al., 2009a; Mitchell et al., 2014; May-Collado et al., 2015; Álvarez-Carretero et al., 2021), are necessarily restricted to extant or recently extinct species. Thus, our understanding of the evolution and biogeography of marsupials has been hampered by the absence of a reasonably comprehensive, quantitative, character-based phylogeny of the group that includes a diverse sampling of extant and fossil taxa.

Estimates of divergence times are necessary to compute metrics such as absolute rates of lineage diversification and trait evolution from molecular phylogenies, and such estimates typically require the use of fossil evidence for calibration purposes (Ho and Phillips, 2009). Unfortunately, many marsupial fossils used for molecular node-dating (sensu Ronquist et al., 2012a; O'Reilly et al., 2015; O'Reilly and Donoghue, 2016) have not had their evolutionary relationships tested by formal phylogenetic analysis (e.g., some of those used by Beck, 2008a; Meredith et al., 2008b, 2009a, 2009c, 2011; Mitchell et al., 2014; Duchêne et al., 2018; Álvarez-Carretero et al., 2021). Tip-dating and tip-and-node dating are promising alternatives to traditional node-dating (but see O'Reilly et al., 2015) that involve simultaneous analyses of character and temporal data from Recent and fossil taxa (Ronquist et al., 2012b; O'Reilly and Donoghue, 2016; Zhang et al., 2016). Tip-and-node dating has been implemented on reasonably comprehensive total-evidence datasets for the marsupial orders Dasyuromorphia (Kealy and Beck, 2017), Peramelemorphia (Travouillon and Phillips, 2018), and Didelphimorphia (Beck and Taglioretti, 2020), but previously published tip-and-node dated phylogenetic analyses of total-evidence datasets for Marsupialia as a whole are characterized by sparse taxon sampling (Beck et al., 2016; Maga and Beck, 2017), limiting their usefulness.

The current study attempts to remedy such shortcomings. Our phylogenetic dataset includes 24.5 kb of DNA sequence data from whole mitochondrial genomes and six nuclear loci (APOB, BRCA1, GHR, IRBP, RAG1, and VWF) for 97 ingroup (marsupial) terminals representing at least one member of every currently recognized Recent genus. We analyze these DNA data separately and in combination with 180 craniodental characters. In the process of describing our craniodental characters, we provide a detailed overview of the comparative craniodental morphology of marsupials, including discussion of additional morphological variation that we were unable to consistently parse into discrete characters. Similar to the DNA dataset, we scored this craniodental dataset for 97 ingroup (marsupial) terminals representing at least one member of every currently recognized Recent genus, as well as a further 42 fossil ingroup terminals and five fossil outgroup terminals. We analyzed these data using maximum parsimony (craniodental dataset only), undated Bayesian (DNA, craniodental, and combined DNA and craniodental datasets), and Bayesian tip-and-node dating (combined DNA and craniodental dataset) approaches. Our results effectively summarize the evidence at hand for inferring marsupial relationships and provide a basis for future studies of metatherian evolution and biogeography.

MATERIALS AND METHODS

Ingroup Taxon Sampling

The designated ingroup for our study is Marsupialia, including the seven currently recognized extant orders together with crown-clade fossil taxa (table 2). To ensure a comprehensive sampling of Recent marsupials, we included at least one representative of every Recent marsupial genus recognized by Wilson and Reeder (2005), with exceptions as noted in the footnotes to table 2. To avoid future nomenclatural confusion in the event of future taxonomic splitting, we tried to score data from the type species of each genus.

In general, we avoided combining morphological data from multiple species, with three exceptions (see table S1 in the online supplement:  https://doi.org/10.5531/sd.sp.54): (1) We scored morphological characters for Caenolestes based on specimens of C. fuliginosus (the type species of Caenolestes), C. caniventer, C. convelatus, and unidentified material that preserves important morphological features (notably the deciduous dentition; see also Luckett and Hong, 2000) that were missing in all C. fuliginosus individuals examined; monophyly of Caenolestes relative to other extant caenolestids is strongly supported by both mitochondrial sequence data and combined mitochondrial and morphological data (Ojala-Barbour et al., 2013). (2) Morphological data for Chaeropus were obtained from specimens of C. ecaudatus, but also from specimens that were subsequently described as a second species, C. yirratji (see Travouillon et al., 2019). (3) Morphological data for Phalanger were obtained from specimens of P. orientalis (the type species of Phalanger), P. intercastellanus, and P. mimicus, which collectively appear to comprise a closely related species complex (Colgan et al., 1993; Norris and Musser, 2001; Kealy et al., 2019).

In compiling DNA data for our analyses, the following Recent terminals represent composites of more than one congeneric species (see table S1 in the online supplement): the didelphids Caluromys, Didelphis, Gracilinanus, Marmosops, Monodelphis, Philander, and Thylamys; the dasyurids Antechinus, Dasyurus, Myoictis, Ningaui, and Planigale; the peramelid Echymipera; the burramyid Cercartetus; the macropodoids Bettongia, Dendrolagus, Dorcopsis, Macropus, Notamacropus, Petrogale, and Thylogale; the phalangerid Phalanger; and the vombatid Lasiorhinus. Recent phylogenetic analyses indicate that all sets of congeneric species combined to form these Recent terminals represent monophyletic clusters relative to our other Recent and fossil taxa (Meredith et al., 2009a, 2009b; Voss and Jansa, 2009; Westerman et al., 2012, 2016; Mitchell et al., 2014; Dodt et al., 2017; Kealy and Beck, 2017; Nilsson et al., 2018; Travouillon and Phillips, 2018).

TABLE 2

Assumed Classification of Marsupial Ingroup Genera Scored for Phylogenetic Analysisa

img-AH8_01.gif

continued

img-z12-2_01.gif

continued

img-z13-2_01.gif

continued

img-z14-2_01.gif

continued

img-z15-2_01.gif

Our fossil ingroup taxa include 42 terminals that have been consistently recovered as crown-clade marsupials in previous phylogenetic studies. To maximize the phylogenetic informativeness of included fossils, we chose plesiomorphic members of speciose clades—on the assumption that these would be more likely to subdivide otherwise long branches than highly derived representatives—and taxa represented by relatively complete specimens. Additionally, we included the early Eocene peradectid †Mimoperadectes and the early Oligocene herpetotheriid †Herpetotherium, whose relationships to Marsupialia are not fully resolved, although they usually fall close to, but outside, the crown clade (Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Beck, 2012, 2017a; Beck et al., 2014, 2016; Wilson et al., 2016; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Carneiro, 2018, 2019; Carneiro et al., 2018; Muizon et al., 2018; Rangel et al., 2019; Ladevèze et al., 2020).

Insofar as possible, we scored each fossil terminal based on specimens of a single species (usually the type species of the genus; see appendix 1 and table S1 in the online supplement). However, the following terminals were scored based on individuals of multiple species, because we felt that the advantage of more complete character scoring outweighed the risk of potential nonmonophyly in these cases (see appendix 1 and table S1 in the online supplement): the paucituberculatans †Acdestis (†A. maddeni, †A. oweni, and †A. spegazzinii) and †Palaeothentes (†P. lemoinei and †P. minutus); the diprotodontids †Ngapakaldia (†N. tedfordi, †N. bonythoni, and †N. sp.) and †Silvabestius (†S. michaelbirti and †S. johnnilandi); the phascolarctid †Litokoala (†L. kutjamarpensis and †L. dicksmithi); and the wynyardiids †Muramura (†M. pinpensis and †M. williamsi) and †Namilamadeta (†N. albivenator, †N. crassirostrum, and †N. superior). In fact, monophyly of each of these sets of congeneric species seems highly likely based on the evidence at hand (Pledge, 2003; 2005; Abello, 2007; Black, 2008; Black, 2010; Black et al., 2012a; Abello, 2013; Black et al., 2014a). Appendix 1 provides further details regarding our fossil ingroup terminals, including locality information and probable age(s) of the specimens from which we recorded data, and additional remarks regarding their significance.

Rooting and Outgroup Taxa

For our molecular analyses, we used the same outgroups as Mitchell et al. (2014): the extant monotremes Ornithorhynchus anatinus and Tachyglossus aculeatus, together with the extant placentals Elephas maximus, Trichechus manatus, Bradypus tridactylus, Dasypus novemcinctus, Mus musculus, Homo sapiens, Bos taurus, and Canis lupus (see table S1 in the online supplement).

However, monotremes differ markedly from marsupials and stem metatherians in many aspects of their craniodental morphology (with Recent monotremes entirely lacking functional dentitions; Archer et al., 1993b; Kielan-Jaworowska et al., 2004; Crompton et al., 2018), as do placentals and stem eutherians (albeit to a lesser extent; Kielan-Jaworowska et al., 2004; Wible et al., 2005). Therefore, adding monotremes, placentals, or stem eutherians to our morphological matrix would have required the inclusion of many additional morphological characters and character states to account for such differences. Doing so would also have raised several problematic issues of homology, for example, regarding postcanine dental loci in eutherians and metatherians (table 12; Averianov et al., 2010; O'Leary et al., 2013). For this reason, we did not use monotremes, placentals, or stem eutherians as outgroups for our morphological or total-evidence matrices. Instead, we included three well-preserved Paleocene fossil metatherians: †Pucadelphys andinus, †Mayulestes ferox, and †Allqokirus australis. These three taxa consistently fall outside Marsupialia, and more distant from the crown clade than either †Herpetotherium or †Mimoperadectes, in most recent phylogenetic analyses (see Rougier et al., 1998, 2004, 2015; Luo et al., 2003; Ladevèze and Muizon, 2007; Sánchez-Villagra et al., 2007; Beck et al., 2008a, 2014; Horovitz et al., 2008, 2009; Forasiepi, 2009; Ladevèze and Muizon, 2010; Luo et al., 2011; Beck, 2012, 2017a; Williamson et al., 2012, 2014; Engelman and Croft, 2014; Forasiepi et al., 2014a; Suarez et al., 2015; Ni et al., 2016; Wilson et al., 2016; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Bi et al., 2018; Carneiro et al., 2018; Engelman et al., 2020; Muizon et al., 2018; Abello and Candela, 2019; Carneiro, 2019; Rangel et al., 2019; Ladevèze et al., 2020).However, the phylogenetic analyses of Goin et al. (2006) and Velazco et al. (2022) are notable exceptions to this (see the account for †Pucadelphys in appendix 1). All three are represented by exceptionally complete cranial material (Muizon, 1994, 1998; Marshall and Muizon, 1995; Ladevèze and Muizon, 2007; Ladevèze et al., 2011; Muizon et al., 2018).

In our dated total-evidence analysis, calibrated nodes needed to be constrained as monophyletic. To calibrate deep divergences within Marsupialia, we constrained the root between Didelphimorphia and the rest of Marsupialia; five uncontradicted retroposon insertions provide statistically significant support for this rooting position (Nilsson et al., 2010; Gallus et al., 2015a). We also constrained †Herpetotherium (the relationships of which are unresolved in our morphological and undated total-evidence analyses) to fall outside Marsupialia, in agreement with the majority of published phylogenetic analyses (Beck et al., 2008a, 2014, 2016; Horovitz et al., 2008, 2009; Forasiepi et al., 2014a; Suarez et al., 2015; Lorente et al., 2016; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Carneiro, 2018, 2019; Carneiro et al., 2018; Engelman et al., 2020; Muizon et al., 2018; Abello and Candela, 2019; Rangel et al., 2019; Ladevèze et al., 2020; Muizon and Ladevèze, 2020). A full justification for constraining and calibrating these and other nodes is given in appendix 2.

Morphological Character Selection and Scoring

The morphological characters used in this study are taken solely from the cranium, mandible, and dentition. Although the postcranium and soft tissues represent additional valuable sources of morphological characters (see, e.g., Szalay, 1982a; 1994; Szalay and Sargis, 2001, 2006; Horovitz and Sánchez-Villagra, 2003; Voss and Jansa, 2003, 2009; Flores, 2009; Kealy and Beck, 2017), most of the fossil taxa and some of the Recent terminals included in the current study are represented only by craniodental material, and hence would necessarily have to be scored as unknown for characters from other anatomical systems.

We initially identified a large pool of candidate craniodental characters by examining published morphological datasets focused on higher-level marsupial and metatherian relationships (e.g., Springer et al., 1997; Rougier et al., 1998; Sánchez-Villagra and Wible, 2002; Horovitz and Sánchez-Villagra, 2003; Luo et al., 2003; Goin et al., 2006; Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Beck, 2012), as well as those intended to resolve relationships among marsupial subclades such as macropodiforms (Prideaux, 2004; Kear et al., 2007), didelphids (Voss and Jansa, 2003; 2009), paucituberculatans (Abello, 2007; Goin et al., 2009a), peramelemorphians (Muirhead, 1994; Travouillon et al., 2010; Gurovich et al., 2014), dasyuromorphians (Wroe et al., 2000; Wroe and Musser, 2001; Murray and Megirian, 2006a), pseudocheirids (Roberts, 2008), phalangerids (Crosby, 2002b), and vombatiforms (Gillespie, 2007; Black, 2008; Black et al., 2012a). Studies that did not carry out explicit phylogenetic analyses of morphological characters but that nevertheless provided morphological data bearing on marsupial phylogeny were also examined for potential characters; these included anatomical descriptions (e.g., Aplin, 1987; 1990; Wroe et al., 1998; Cooke, 2000; Wible, 2003) and reviews of marsupial phylogeny (e.g., Archer, 1984b, 1984c; Aplin and Archer, 1987; Marshall et al., 1990).

We evaluated this initial pool of characters against our own observations of relevant taxa and eliminated all that were poorly defined or otherwise unclear, that seemed to exhibit continuous variation that could not easily be parsed into discrete states, or that clearly overlapped with others. Many characters were retained essentially unchanged from these previous studies. However, others were significantly modified or redefined to better correspond with our own observations and hypotheses of homology and to take full account of the range of variation seen in our taxon set. We also identified additional morphological characters that—as far as we are aware—have not been used in previous studies of marsupial phylogeny. Our final morphological dataset of 102 craniomandibular and 78 dental characters is restricted to just those that we could score consistently and confidently over our entire taxon set and for which homologies appear to be relatively straightforward. Although our morphological character set might be considered scanty in comparison to the number of taxa in our study, we believe it preferable to use only characters that have been carefully evaluated and can be scored unambiguously for all taxa.

We made a deliberate attempt to identify and score autapomorphies for three main reasons (see also Voss and Jansa, 2003: 8): First, such autapomorphies may become synapomorphies with denser taxon sampling. Second, the inclusion of autapomorphies is important for analyses that use the Mk model and its variants for discrete morphological data, particularly those employing tip- and tip-and-node dating (Lewis, 2001; Müller and Reisz, 2006; Wright and Hillis, 2014; Lee and Palci, 2015; Matzke and Irmis, 2018). Third, the supposed correction to the Mk model implemented by MrBayes for datasets that include only parsimony-informative characters (Nylander et al., 2004; Ronquist et al., 2005; Allman et al., 2010; Ronquist et al., 2011) does not appear to be computationally feasible for nonbi-nary characters (which are used extensively here), and so is unlikely to function as intended (Koch and Holder, 2012; dos Reis et al., 2016; Matzke, 2016; Matzke and Irmis, 2018).

Wherever possible, we scored morphological characters from multiple specimens of each taxon, ideally including juveniles (with incompletely erupted dentitions), both young and old adults (distinguished largely based on the degree of ossification and tooth wear) and both males and females. This strategy allowed us to score characters that were ascertainable only during particular ontogenetic stages (for example, features of the deciduous dentition) and to assess the overall impact of ontogeny (for example, the increased ossification, sutural fusion, and tooth wear that typically accompany advancing age) and sexual dimorphism. However, such inclusive sampling was not possible for most fossil taxa, with the notable exceptions of the diprotodontid †Nimbadon (known from a large ontogenetic series; Black et al., 2010) and the stem taxon †Pucadelphys (known from multiple putative males and females; Ladevèze et al., 2011). For most other fossil taxa, the absence of ontogenetic series and/or information about sex dictated conservative scoring. For example, dental characters that can be confidently determined only from unworn or lightly worn teeth (such as the presence or absence of a posterior lobe on i3; char. 150) were not scored if the only available specimens exhibited substantial dental wear. By contrast, the morphology of the palatal vacuities (if present; chars. 35–39) and the presence or absence of a sagittal crest (char. 27) were scored based on adult specimens only, because these features are fully developed only in mature individuals (see, e.g., Flores et al., 2003, 2006, 2010, 2013; Black et al., 2010). Characters that we scored based on specimens of a particular ontogenetic stage (or stages) are indicated in our character list as appropriate. We examined damaged or disarticulated skulls, where available, in addition to intact specimens, to facilitate scoring of characters of regions that are often concealed in intact skulls, such as the middle ear. We scored all our terminals based on firsthand examination of specimens, with the exception of a single outgroup taxon, †Allqokirus australis, for which we relied upon the richly illustrated monograph of Muizon et al. (2018). A full list of the specimens we used for scoring purposes, together with institutional abbreviations, is given in table S2 in the online supplement ( https://doi.org/10.5531/sd.sp.54).

Following Voss and Jansa (2009), we scored characters as polymorphic only when alternative states were about equally common among examined conspecific specimens. Otherwise, the taxon was scored based on the clearly modal condition. Polymorphisms were scored as variable (e.g., “0+1”), rather than as intermediate states (with transformations to and from the polymorphic state given a weight of 0.5—the “scaled” option of Wiens, 2000) to allow Bayesian analysis of our morphological dataset, because current versions of MrBayes do not allow the use of step matrices. In certain cases, a terminal could not be scored as exhibiting a specific state for a particular character, but at least one alternative state could be definitively ruled out (for example, a sagittal crest [char. 27] is clearly present in the phascolarctid †Litokoala and the dasyuromorphian †Mutpuracinus, but it is not clear based on available material whether this crest extended onto the frontals in either taxon; Murray and Megirian, 2006a; Louys et al., 2009); in such cases, we used uncertainty (either/or) coding, with all possible states scored (e.g., “1/2”).

We elected not to score sexual dimorphism of particular features (for example, root number and the presence or absence of accessory cusps on the upper canines of the extant caenolestids Caenolestes and Rhyncholestes) as distinct character states because sample sizes for many of our terminals were insufficient to determine whether sexual dimorphism was present; instead, such sexually dimorphic traits were coded as polymorphisms (following Voss and Jansa, 2009).4 Following standard practice in morphological phylogenetics, we coded missing/unknown data as “?” and inapplicable data as “-” although both were treated in the same way (as missing/unknown) in our phylogenetic analyses. As in Voss and Jansa (2009), characters representing putative morphoclines or transformation series were treated as ordered (additive) in all analyses (Wiens, 2001:691 presented a formal justification of this approach; but see Brocklehurst and Haridy, 2021).

Molecular Characters

We analyzed DNA sequences from the following six protein-coding nuclear loci (gene names and symbols follow the HUGO Gene Nomenclature Committee; Yates et al., 2017): exon 26 of apolipoprotein B (APOB); exon 11 of breast cancer 1, early onset (BRCA1); exon 10 of growth hormone receptor (GHR); exon 1 of retinol binding protein 3 interstitial (RBP3, also known as interphotoreceptor retinoid binding protein [IRBP] in other studies); the intronless recombination activating gene 1 (RAG1); and exon 28 of von Willebrand factor (VWF). We also analyzed sequence data from whole mitochondrial genomes, namely the 12 heavy-strand protein-coding genes, the light-strand protein-coding gene NADH-ubiquinone oxidoreductase chain 6 (MT-ND6), the mitochondrially encoded 12S (MT-RNR1) and 16S (MT-RNR2) ribosomal RNA genes, and 21 mitochondrially encoded tRNA genes (Janke et al., 1994). The mitochondrially encoded tRNA lysine gene (MT-TK) was excluded because it appears to be a pseudogene in marsupials (Janke et al., 1997; Dörner et al., 2001; Nilsson et al., 2003) and cannot be aligned with homologous nonmarsupial sequences.

For APOB, BRCA1, RAG1, VWF, and the mitochondrial genomes, DNA sequences were downloaded from GenBank (table S1 in the online supplement). Some RBP3 and GHR sequences were also obtained from GenBank, but we also obtained many new marsupial sequences for these loci (table S1). For most new sequences of GHR, exon 10 was amplified using forward primer GHRF1alt-5′GCCATGACAGCTAYAAGCCTCA paired with reverse primer GHRendAlt-5′GATTTTGTTCAGTTGGTCTGTGCTCAC. For some taxa, the forward primer was replaced with GHRF1mon-5′TTCTAYARYGATGACTCYTGGGT. This full-length transcript was reamplified as two smaller overlapping segments by pairing the forward primer with GHR750R-5′GTAAGGCTTTCTGTGGTGATRTAA and the reverse primer with GHR50F-5′TTCTAYARYGATGACTCYTGGGT. For unknown reasons, we were unable to amplify exon 10 of GHR from any didelphin taxon using these primers.

New RBP3 sequences were amplified from all non-dasyuromorphian taxa using forward primer IRBPA-5′ATGGCCAAGGTCCTCTTGGATAACTACTGCTT paired with reverse primer IRBPD1-5′ CATCATCAAACCGCAGATAGCCCA. This fragment was reaplified as two smaller overlapping segments by pairing the forward primer with IRBPF-5′CTCCACTGCCCTCCCATGTCT and the reverse primer with IRBPE1-5′AGCCTACATCCTCAAGAAGATGCG. For dasyuromorphians, primer IRBPA was replaced with IRBP_DasyA-5′TGCAAGAAGCGATTGAACAAGC and IRBPD1 was replaced with IRBP_ KraB-5′ AGGTGCTCCGTGTCCTGAA.

All PCR amplifications used AmpliTaq Gold DNA polymerase (Thermo Fisher Scientific) with recommended reagent concentrations. For GHR, full-length amplifications were performed using a modified touchdown program in which annealing temperature started at 58° C and decreased by 2° every 5 cycles until 52° C. Ream plifications were done using 25 cycles at an annealing temperature of 52° C. For IRBP, touchdown protocols started at 64° C and decreased by 2° every 5 cycles until 58° C; reamplifications used an annealing temperature of 58° C or 60° C for 25 cycles. Resulting PCR products were Sanger sequenced on an Applied Biosystems 3730xl DNA Analyzer using ABI BigDye Terminator ver. 3.1 chemistry.

Given their essential roles in cellular respiration and mitochondrial protein translation respectively, the mitochondrial protein-coding genes and ribosomal and tRNA genes are expressed ubiquitously in eukaryotic cells. Voss and Jansa (2009) provided detailed information regarding the genes BRCA1, RBP3, RAG1, and VWF, demonstrating that these loci appear to exist as single copies, are unlinked in Monodelphis domestica, are expressed in different tissues, and have translated protein products with very different functions. Here we provide similar information regarding APOB and GHR.

Apolipoprotein B (APOB): The human apolipoprotein B gene (APOB) consists of 29 exons spanning about 45 kb on the short arm of chromosome 2 (Blackhart et al., 1986; Chen et al., 1986; Cladaras et al., 1986; Amrine-Madsen et al., 2003a). Meredith et al. (2009b: accessory publication) reported that APOB maps to chromosome 1 in Monodelphis domestica, and it appears to exist as a single copy in both M. domestica and Macropus eugenii. In humans, the gene is expressed primarily in the liver, where the gene product is a very large (4536 amino acids) protein (APOB100), and in the small intestine, where editing of the mRNA transcript results in a smaller isoform (2152 amino acids, APOB48; Chen et al., 1986; Cladaras et al., 1986). The same editing process has been shown to occur in the small intestine of M. domestica (Fujino et al., 1999). The two APOB isoforms are the major apolipoproteins of chylomicrons and low-density lipoproteins (Young, 1990).

APOB was first used to investigate mammalian phylogeny by Amrine-Madsen et al. (2003a; 2003b), who sequenced part of exon 26 (which contains putative binding sites for the LDL receptor; Hospattankar et al., 1986; Yang et al., 1986; Ebert et al., 1988; Maeda et al., 1988). Subsequent studies focusing on relationships within Marsupialia (Meredith et al., 2008a, 2008b, 2009a, 2009b, 2009c, 2010) sequenced exon 26 for 65 of the 97 Recent ingroup terminals included in our study (table S1 in the online supplement). We downloaded these sequences from Genbank and performed our own alignment (see below).

Growth Hormone Receptor (GHR): The human growth hormone receptor (GHR) gene consists of 10 exons spanning about 87 kb on the short arm of chromosome 5 (Godowski et al., 1989). In Monodelphis domestica, the homologous gene is present on chromosome 3. In humans, the full-length (638 amino acids) GHR gene product (isoform 1) is a type I cytokine receptor that acts as a transmembrane receptor for growth hormone, but multiple alternatively spliced transcript variants are known (Urbanek et al., 1992; Pantel et al., 2000). GHR was first used to investigate mammalian phylogeny by Adkins et al. (2001), who sequenced exon 10 in rodents and nonrodent outgroups. We amplified and sequenced part of exon 10 of GHR for 53 of our 97 Recent terminals (details above), to which we added additional marsupial sequences downloaded from GenBank (table S1 in the online supplement).

Sequence Alignment

DNA sequences for the protein-coding genes were aligned separately using MUSCLE (Edgar, 2004) for codons in MEGA 6.0 (Tamura et al., 2013) and then subsequently adjusted manually to maintain an open reading frame. RBP3 is a pseudogene in Notoryctes (AF025385; see Springer et al., 1998) and APOB is a pseudogene in Tarsipes (FJ603116),5 but all other sequences for both genes could be aligned to an open reading frame. MT-RNR1 and MT-RNR2 were aligned manually with reference to the secondary structure model presented by Springer and Douzery (1996) and Burk et al. (2002), respectively, with stem and loop regions identified, and alignment-ambiguous regions removed. The tRNA genes were aligned using the reference alignments in the Mamit-tRNA database (Pütz et al., 2007). The final alignments are summarized in table 3. Each of our Recent terminals is represented by sequence data from at least one gene (table S1 in the online supplement), but we note that a few terminals do not overlap with others (e.g., the didelphid Caluromysiops is represented only by an RBP3 sequence, whereas the peramelemorphian Rhynchomeles is represented only by a MT-RNR1 sequence). The final alignments were concatenated into an alignment of mitochondrial genes (14,704 bases), an alignment of nuclear genes (9797 bases), and a single combined (mitochondrial and nuclear DNA) alignment (24,501 bases).

Total-Evidence Matrix

Concatenating our combined DNA and morphological matrices resulted in a total-evidence matrix comprising 24,681 characters (24,501 molecular, 180 morphological). The 10 nonmarsupial outgroup terminals used to root the molecular-only analyses were deleted from the total-evidence matrix, leaving an ingroup of 141 marsupial terminals (97 Recent, 42 fossil), plus †Herpetotherium and †Mimoperadectes, with †Pucadelphys, †Mayulestes, and †Allqokirus as the three fossil outgroup terminals. Relative completeness of all terminals is given in table 4.

TABLE 3

Summary of Molecular Loci

img-ApJJ_01.gif

Online Data Archives

All the new molecular sequences produced for this study have been deposited in GenBank with accession numbers ON677544–ON677610 (for new RBP3 sequences) and ON677611–ON677671 (for new GHR sequences); for a complete list of GenBank accession numbers of all analyzed sequences, old and new, see table S1 in the online supplement. All our datasets and associated trees have been deposited on Dryad (doi:10.5061/dryad.jdfn2z3dc).

Phylogenetic Analyses

Molecular: Phylogenetic analysis of the molecular matrix was carried out using Bayesian inference implemented in MrBayes 3.2.7a (Ronquist et al., 2011, 2012b) via the CIPRES Science Gateway (Miller et al., 2010). PartitionFinder v2.1.1 (Lanfear et al., 2017) was first used to identify an appropriate partitioning scheme and models for the molecular matrix. For the PartitionFinder analysis of our molecular matrix, we initially partitioned the nuclear protein-coding genes and MT-ND6 by gene and codon position, but we treated our 12 mitochondrial heavy-strand genes as a single locus partitioned by codon position only, and we combined the rRNA and tRNA genes, which were partitioned into stems and loops (i.e., two partitions) only. The PartitionFinder analysis employed the “greedy” search algorithm (Lanfear et al., 2012), with the assumption of linked branch lengths among partitions (Duchêne et al., 2020). We used the Bayesian Information Criterion (BIC) to identify the partitioning scheme and models that best fit our sequence data. The BIC penalizes extra parameters more than either the Akaike Information Criterion or corrected Akaike Information Criterion (Sullivan and Joyce, 2005; Luo et al., 2010; Lanfear, 2016: 18; Duchêne et al., 2020), so it is likely to prefer a partitioning scheme with fewer partitions, less complex models, or both. This is desirable because fewer partitions and simpler models reduce the computational burden of subsequent analyses. Comparisons were restricted to models implemented by MrBayes, except that we did not test models that combined a gamma distribution for rate heterogeneity with a proportion of invariant sites, following the recommendation of Stamatakis (2016; see also Yang, 2006: 113–114). The optimal partitioning schemes and models are given in table 5.

TABLE 4

Percentage Completeness of Terminal Taxaa

img-ALXv_01.gif

continued

img-z24-2_01.gif

continued

img-z25-2_01.gif

continued

img-z26-2_01.gif

continued

img-z27-2_01.gif

TABLE 5

Partitions for Bayesian Analysis Based on PartitionFinder Outputa

img-AUa-_01.gif

For our Bayesian analyses, we assumed linked branch lengths between partitions, with the number of gamma rate categories increased from the default number of four to eight (see Jia et al., 2014); the latter may be particularly important here as we did not use models that combined a gamma distribution for rate heterogeneity with a proportion of invariant sites (see above). The MrBayes analysis comprised four runs of four chains (each with one cold and three incrementally heated chains, with the temperature of the heated chains set at the default value of 0.1), which were run for 50 × 106 generations and sampling trees every 5000 generations. Stationarity and convergence between chains was checked using Tracer 1.6 (Rambaut et al., 2014). Tracer was also used to determine the burnin for for each individual run, and effective sample sizes (ESSs) from the combined post-burnin samples were checked to make sure that they were greater than 200 for most parameters. The post-burnin trees were then concatenated into a single file using the Burntrees v.0.3.0 perl script (available from  https://github.com/nylander/Burntrees) and then summarized using 50% majority rule consensus in MrBayes.

Morphological: The morphological matrix was analyzed using maximum parsimony, as implemented in TNT 1.5 (Goloboff et al., 2008; Goloboff and Catalano, 2016), and Bayesian inference, as implemented in MrBayes 3.2.7a (Ronquist and Huelsenbeck, 2003; Ronquist et al., 2012b). We note here that TNT recognizes cases where a particular taxon is scored as multiple states for a particular character but does not distinguish between those due to observed polymorphism within a terminal and those due to uncertainty (treating both as cases of polymorphism; Watanabe, 2015), whereas MrBayes treats all such scorings as missing/unknown (Ronquist et al., 2011).

The TNT analysis comprised an initial New Technology search (which combined Sectorial Search, Ratchet, Drift, and Tree Fusing), which was terminated after the same minimum length was found 1000 times; this was followed by a Traditional search with tree-bisection reconnection (TBR) within the trees saved from the first stage, saving a maximum of 100,000 trees. Support values were calculated using 10,000 standard (sampling-with-replacement) bootstrap replicates with a traditional search, and the results were output as absolute frequencies. The consistency index and retention index for the most parsimonious trees were calculated using the “STATS.RUN” script provided with TNT 1.5.

The MrBayes analysis employed the Lewis (2001) Mk model for discrete morphological characters under the assumption that only variable characters have been scored (i.e., the Mkv variant). A lognormal distribution with eight rate categories was used to model rate heterogeneity across characters following Harrison and Larsson (2015). As for the molecular analysis using MrBayes, the morphological analysis comprised four runs of four chains each, run for 50 × 106 generations and sampling trees every 5000 generations. To reduce the time required for chains to reach stationarity and convergence, we randomly selected 16 most-parsimonious trees from the equivalent TNT analysis and used these as starting topologies for each of the 16 chains. Tracer was again used to check for stationarity and convergence between chains, to determine the burnin for each chain, and to check that ESSs were greater than 100 for most parameters. The post-burnin trees were again concatenated into a single file using Burntrees, and then summarized using 50% majority rule consensus in MrBayes.

Total Evidence: We used only Bayesian inference (implemented by MrBayes 3.2.7a) for our total-evidence analysis. We carried out an undated analysis and a “tip-and-node dated” analysis (sensu O'Reilly and Donoghue, 2016). For both analyses, we used the same models for the molecular and morphological partitions as for the molecular-only and morphological-only Bayesian analyses (see above; table 5) and assumed a single set of linked branch lengths across all partitions. To reduce the time taken to reach stationarity and convergence in the undated analysis, we first analyzed the total-evidence matrix in TNT with the same search settings used for analyzing the morphological matrix (see above) and then used 16 of the most-parsimonious trees from this analysis as starting topologies for each of the 16 chains.

For the dated analysis, we used the Independent Gamma Clock (IGR) model (Lepage et al., 2007; Ronquist et al., 2012a, 2012b; Zhang et al., 2016), with one IGR model applied to the molecular partition and one to the morphological partition (see Beck and Taglioretti, 2020). For both partitions, we specified a normally distributed clock rate prior with a mean of 0.01 and a standard deviation of 0.1. Each of our Recent terminals was assigned a fixed age of 0 Mya, whereas each fossil terminal was assigned a uniform prior age range based solely on the age(s) of the fossil specimens used to score it (see appendix 1), even if older specimens have been reported, following the recommendations of Püschel et al. (2020). We also specified uniform prior age ranges for the age of the root and for 24 nodes that could be calibrated based on available fossil and phylogenetic evidence; a full justification for these calibrations is given in appendix 2. We specified a fossilized birth-death prior with sampled ancestors (Heath et al., 2014; Matzke and Wright, 2016; Zhang et al., 2016; Gavryushkina et al., 2017), and we assumed diversity sampling of our Recent terminals (which were selected to include at least one representative of each currently recognized Recent genus; Zhang et al., 2016). We included approximately a quarter (97/406) of the total number of Recent species currently recognized; we therefore specified the fraction of extant species sampled to be 0.25. We used the MrBayes default priors for the speciation, extinction, and fossilization rate. Because the dated analysis already included topological constraints (corresponding to the age-calibrated nodes mentioned above), we did not specify starting topologies.

TABLE 6

Summary of Posterior Age Estimates for Calibrated Nodes in Dated Total-Evidence Analysisa and Comparisons with the Results of Previous Molecular-Clock Studies

img-ADtgy_01.gif

continued

img-z31-2_01.gif

For both the dated and undated analysis, we again used four runs of four chains each—run for 50 × 106 generations and sampling trees every 5000 generations—with Tracer used to check for stationarity and convergence between chains, to determine the burnin for each chain, and to check that ESSs were greater than 200 for most parameters. The post-burnin trees were again concatenated into a single file using Burntrees and then summarized using 50% majority rule consensus in MrBayes. To determine how our specified priors on the ages of tips and selected nodes interacted with our other priors, we repeated our dated analysis without character data and then used the post-burnin trees from this analysis to calculate effective priors. The specified priors and effective priors for our calibrated nodes are given in appendix 2. Posterior age extimates for these calibrated nodes, as well as estimated ages for these same nodes (where available) taken from five recent molecular clock studies (Meredith et al., 2009a, 2011, 2014; Duchêne et al., 2018; Álvarez-Carretero et al., 2021), are given in table 6.

Nodal Support: As described in more detail above, nodal support was calculated using bootstrap values for our maximum parsimony analysis of morphological data and Bayesian posterior probabilities for all our Bayesian analyses. However, these two measures of nodal support are not equivalent, and they are difficult to compare directly (Suzuki et al., 2002; Douady et al., 2003; Erixon et al., 2003; Alfaro and Holder, 2006). Nevertheless, for the purposes of discussion, for our maximum parsimony analysis of morphological data, we refer to clades with <50% bootstrap support as “weakly supported,” those with 50%–69% bootstrap support as “moderately supported,” and those with ≥70% bootstrap support as “strongly supported” (Hillis and Bull, 1993). Likewise, for our Bayesian analyses, we refer to clades with posterior probabilities (BPP) of 0.5–0.74 as “weakly supported,” those with BPPs of 0.74–0.94 as “moderately supported,” and those with BPPs ≥0.95 as “strongly supported” (Erixon et al., 2003; Alfaro and Holder, 2006).

Visualization and Annotation of Phylogenetic Trees

We visualized our phylogenies (figs. 2733) directly from the tree files output by TNT and MrBayes using the ggtree package (Yu et al., 2016; Yu, 2020) in R 4.0.3 ( https://www.R-project.org/). We then used ggtree to annotate these with dots at nodes to indicate support values (bootstrap values for the maximum parsimony analysis of our morphological data, Bayesian posterior probabilities for the molecular and total-evidence analyses; see Nodal Support above) as follows: black dots for “strong support” (>70% bootstrap support or ≥0.95 Bayesian posterior probability); dark gray dots for “moderate support” (50%–69% bootstrap support or 0.75–0.94 Bayesian posterior probability); light gray dots for “weak support (<50% bootstrap support or 0.50–0.74 Bayesian posterior probability). For our dated total-evidence analysis (fig. 33), we used ggtree to add node bars representing 95% highest posterior density (HPD) intervals for the ages of nodes (for clarity, 95% HPD intervals were not included for the ages of fossil terminals), and we added a geological time scale using the R package MCMCtreeR (Puttick, 2019).

Identification of Craniodental Synapomorphies for Selected Clades

To identify craniodental synapomorphies of marsupial families and selected suprafamilial clades (see Taxonomic Accounts), we used the “DescribeTrees” command in PAUP* 4.0a169 (Swofford, 2003) to map craniodental characters onto our dated total-evidence topology (fig. 33). We only identify unambiguous synapomorphies (those that do not change between Accelerated and Delayed Transformation optimizations) together with their consistency indices (a consistency index less than 1 indicates some degree of homoplasy). A full list of unambiguous synapomorphies is given in file S3 in the online supplement ( https://doi.org/10.5531/sd.sp.54).

FIG. 1.

Dorsal and ventral cranial views of Marmosa murina showing principal osteological features mentioned in the text. Abbreviations: als, alisphenoid (coossified with basisphenoid); atp, alisphenoid tympanic process; bo, basioccipital (coossified with exoccipitals); bs, basisphenoid (coossified with alisphenoid); cc, carotid canal; ect, ectotympanic; fm, foramen magnum; fpj, frontal process of jugal; fro, frontal; gf, glenoid fossa; gpa, glenoid process of alisphenoid; ip, interparietal; jug, jugal; lac, lacrimal; max, maxillary; mp, mastoid process (of petrosal); nas, nasal; occ, occipital condyle (of exoccipital); of, orbital fossa; pal, palatine; par, parietal; pcf, paracanine fossa; pogp, postglenoid process (of squamosal); pop, postorbital process; pre(ap), premaxillary (alveolar process); pre(fp), premaxillary (facial process); prgp, preglenoid process (of jugal); pro, promontorium (of petrosal); ps, presphenoid; pt, pterygoid; rtp, rostral tympanic process (of petrosal); sq, squamosal; tcf, transverse canal foramen; tf, temporal fossa; za, zygomatic arch.

img-z33-1_01.jpg

CRANIAL AND MANDIBULAR CHARACTERS

Our survey of marsupial cranial morphology includes only taxonomically variable aspects of the cranium and mandible that do not require dissection, histology, or X-ray imaging for their study. Anatomical nomenclature in these accounts largely follows Wible (2003) and Voss and Jansa (2003, 2009), with exceptions as noted below. The principal cranial and mandibular features of Marmosa murina, a Recent didelphid that retains many plesiomorphic traits for Marsupialia, are illustrated in figures 1 and 2.

FIG. 2.

Left lateral cranial and mandibular views of Marmosa murina showing principal osteological features mentioned in the text. Abbreviations: als, alisphenoid; ap, angular process; atp, alisphenoid tympanic process; conp, condylar process; corp, coronoid process; ect, ectotympanic; exo, exoccipital; fpj, frontal process of jugal; fr, foramen rotundum; fro, frontal; hpp, hamular process of pterygoid; iof, infraorbital foramen; ip, interparietal (coossified with supraoccipital); jug, jugal; lac, lacrimal; lc, lambdoid crest; lf, lacrimal foramina; maf, masseteric fossa; mas, mastoid exposure of pars canalicularis (of petrosal); max, maxillary; mef, mental foramina; nas, nasal; pal, palatine; par, parietal; pcf, paracanine fossa; pco, pars cochlearis (of petrosal); pop, postorbital process; pre, premaxillary; rmf, retromolar fossa; sq, squamosal; ssf, subsquamosal foramen; sup, supraoccipital (coossified with interparietal); zps, zygomatic process of squamosal.

img-z34-1_01.jpg

Dorsolateral Rostrum

The dorsolateral rostrum forms the roof and sides of the nasal cavity. It includes the nasal bones, together with facial processes of the premaxillary and maxillary bones. With the conspicuous exception of the infraorbital foramen, the bones of the dorsolateral rostrum are imperforate in marsupials and other metatherians. Therefore, taxonomic variation in this region primarily concerns the shape, relative positions, and patterns of contact among adjacent elements, presence/absence of sutural fusion, and the development of processes for muscle attachment. A (nonexhaustive) list of anatomical synonyms that have been applied to dorsolateral rostral structures in the marsupial-focused literature is given in table 7.

Character 1. Nasals produced anteriorly beyond facial processes of premaxillae, concealing incisive foramina from dorsal view (0); or not produced beyond premaxillary facial processes, incisive foramina exposed in dorsal view (1). The nasal bones of most metatherians (e.g., Marmosa [figs. 1, 2]) are produced anteriorly beyond the vertically oriented facial processes of the premaxillae that contribute to the lateral walls of the nasal cavity (state 0). Such nasals cover the anterior part of the nasal orifice and effectively conceal the incisive foramina from dorsal view, with the undamaged tips of these bones typically also exhibiting a well-defined median apex. Relatively minor taxonomic variation in nasal length that was coded for phylogenetic analysis among Recent didelphids by Voss and Jansa (2003: char. 32; 2009: char. 41) is subsumed among the range of morphologies here assigned to state 0. The nasals extend anteriorly considerably beyond the alveolar processes of the premaxillae in a number of nondidelphid metatherians (e.g., Notoryctes [fig. 36], Hypsiprymnodon moschatus [fig. 52], and Potorous [fig. 54]), but a continuous range of intermediate morphologies prevents us from scoring this trait as a separate state.

TABLE 7

Selected Anatomical Synonyms for Structures of the Dorsolateral Rostruma

img-AavI_01.gif

A qualitatively distinct state, however, is seen in all examined dasyuromorphians that preserve intact nasal bones: the nasals are truncated anteriorly, terminating behind the anterior margins of the premaxillary facial processes, and the anterior floor of the nasal orifice (usually including the anterior part of the incisive foramina) is dorsally exposed (state 1; e.g., Thylacinus [fig. 40], Myrmecobius [fig. 41], and Pseudantechinus [fig. 42]). Typically, no median apex is developed; instead, the undamaged anterior nasal margins are prominently notched in the midline.

Character 2. Nasals very broad posteriorly, contacting lacrimals on each side (0); or maxillae and frontals in contact (1); or premaxillae and frontals in contact (2). Among the taxa included in this study, naso-lacrimal contact (state 0) only occurs consistently in †Allqokirus (see Muizon et al., 2018), †Mayulestes (see Muizon, 1998: figs. 5, 6), and Tarsipes ([fig. 49]), but this trait also occurs polymorphically in Lasiorhinus, Phalanger, Spilocuscus, and Trichosurus. In metatherians that lack naso-lacrimal contact, the maxillae usually contact the frontals (state 1; e.g., in Marmosa [figs. 1, 2, 34]. The relevant morphology of adult specimens of Notoryctes is obscured by sutural fusion, but CT-scans of a fluid-preserved juvenile (NMV C11082) reveal the presence of maxillo-frontal contact (Beck, in prep.). Although the nasals are very broad posteriorly in †Ngapakaldia, they do not prevent maxillo-frontal contact (based on SAM P13851, contra Stirton [1967: figs. 1, 2]). Vombatus also exhibits the widespread marsupial condition of maxillo-frontal contact, but the morphology of this taxon is distinctive: as the maxilla extends posterodorsally, it is reduced to a very thin strip interposed between the premaxilla and jugal, before widening again to contact the frontal (fig. 45). In Dactylopsila and Dactylonax, however, the premaxillae contact the frontals (and also the lacrimals and jugals; Tate, 1948: 241); this latter morphology, which may be connected with the presence of greatly enlarged upper incisors (e.g., Gingerich, 1971), has been scored as an additional state (“2”).

Because these states do not represent a clear morphocline, this character has not been ordered in any of our phylogenetic analyses.

Character 3. Nasals extend posteriorly between lacrimals (0); or nasals truncated anterior to lacrimals (1). The nasal bones extend posteriorly between the lacrimals in most metatherians (state 0; e.g., Marmosa [figs. 1, 2, 34]). In Recent peramelemorphians (e.g., Chaeropus [fig. 37], Macrotis [fig. 38], and Perameles [fig. 39]), however, the nasal bones are truncated anterior to the lacrimals (state 1; Filan, 1990: 618; Muirhead, 2000: 516; Travouillon et al., 2010: char. 31). This distinctive morphology does not occur in the fossil peramelemorphians †Yarala, †Galadi, or †Bulungu (Muirhead, 2000; Travouillon et al., 2010, 2013b; Gurovich et al., 2014), although it is present in the fossil †Madju (Travouillon et al., 2015b), which has not been included here. Among macropodiforms, it is seen in Potorous (fig. 54) as well as some specimens of Aepyprymnus, with the latter scored as polymorphic (“0+1”) here. Adults of Notoryctes in which remnants of the naso-frontal suture can still be seen (e.g., AMNH 198651 [fig. 36]) demonstrate that the nasals did not extend between the orbits in this taxon, but the lacrimal cannot be identified as a discrete ossification due to extensive fusion of the bones making up the rostrum. However, CT-scans of a fluid-preserved juvenile (NMV C11082) suggest that the nasals do indeed terminate anterior to the lacrimals in this taxon (Beck, in prep.), so Notoryctes has also been scored as state 1.

Character 4. Anteorbital vacuity between nasal, frontal, and maxillary bones on each side of rostrum absent (0); or present (1). In most metatherians, the posterodorsal rostrum is essentially imperforate (state 0; e.g., in Marmosa [figures 1, 2, 34]). In Recent caenolestids (e.g., Caenolestes [fig. 35]), however, a prominent unossified space—the anteorbital vacuity of Thomas (1895)—occurs between the nasal, frontal, and maxillary bones (state 1; Thomas, 1895; Osgood, 1921; 1924; Bublitz, 1987; Patterson and Gallardo, 1987; Goin et al., 2007b; Martin, 2013; Ojala-Barbour et al., 2013; González et al., 2020). Goin et al. (2003: 313; 2007b: 1272–1273) suggested that this feature might be connected with the elongate rostrum of these taxa. By contrast, the fossil pauctituberculatans †Acdestis (Engelman and Croft, 2016: fig. 6) and †Palaeothentes (Forasiepi et al., 2014b) both unequivocally lack an anteorbital vacuity.

We note some discrepancies between our observations and those of previous authors regarding the presence or absence of anteorbital vacuities in other fossil paucituberculatans. In agreement with Marshall and Pascual (1977) but contra Goin et al. (2003; 2007b), we consider anteorbital vacuities to be present in †Pichipilus: in MLP-68-I-17-204 (a skull of †P. centinelus), the posteromedial margin of the right nasal has a distinctly ragged appearance (Marshall and Pascual, 1977: fig. 4) that does not appear to be the result of breakage and that is very similar in morphology to the equivalent region in living caenolestids (where the nasal forms the anteromedial margin of the anteorbital vacuity [fig. 35]; Patterson and Gallardo, 1987: fig. 1). We therefore scored †Pichipilus as state 1. Goin et al. (2007b) stated that anteorbital vacuities are absent in the dentally plesiomorphic late Oligocene paucituberculatan †Evolestes hadrommatos, but we consider the only known skull (MNHN Bol-V-004017, previously MNHN-Bol 96-400) to be insufficiently well preserved to score this character confidently, so we scored †Evolestes as unknown (“?”). Based on specimens available to us, it is unclear whether anteorbital vacuities are present in †Stilotherium, which has also been scored as unknown.

Character 5. Infraorbital foramen enclosed by maxilla (0); or infraorbital foramen occupies maxillary-jugal suture (1). In most metatherians, the infraorbital foramen—the anterior opening of the infraorbital canal, which transmits the infraorbital nerve and accompanying blood vessels (Sánchez-Villagra and Asher, 2002; Wible, 2003)—is contained entirely in the maxillary bone (state 0; e.g., as in Marmosa [fig. 2], Monodelphis [Wible, 2003: fig. 2], and Macropus [Wells and Tedford, 1995: fig. 9]). Although relevant sutures are not discernible in adult specimens of Notoryctes, CT-scans of a fluid-preserved juvenile (NMV C11082) demonstrate that the infraorbital foramen is, in fact, entirely enclosed by the maxilla in this taxon (Beck, in prep.). Although we observed minor taxonomic variation in the position of this foramen within the maxilla (for example, it is located close to the premaxilla in Dactylopsila and Dactylonax, but close to the jugal in most Phascolarctos specimens), a qualitatively distinct condition is seen in all examined specimens of Thylacinus, where the infraorbital foramen occupies the maxillary-jugal suture (fig. 40; see also Murray and Megirian, 2006a: appendix 1, figs. 1A, 2B; Warburton et al., 2019: fig. 4). This is an autapomorphy of Thylacinus in the current analysis, although we also observed it occasionally in Phascolarctos (e.g., AMNH 173404).

We observed additional variation in the morphology of the infraorbital foramen that we decided not to score in the current analysis but that merits brief discussion. The infraorbital foramen is occasionally subdivided into multiple, closely spaced foramina—usually arranged dorsoventrally—in older specimens of a few taxa (notably Phalanger [fig. 47], Thylacinus, and Dactylopsila). Uniquely among the taxa included in this study, most examined specimens of Hypsiprymnodon moschatus (fig. 52) have two similar-sized external openings of the infraorbital canal, with the two foramina arranged anteroposteriorly within the maxilla (see also Johnson and Strahan, 1982). By contrast, only a single foramen appears to be present in the only known cranium of H.bartholomaii (QM F13051; see Flannery and Archer, 1987c: fig. 2A). A small accessory foramen posterior and slightly dorsal to the infraorbital foramen and adjacent to (or within) the maxillary-jugal suture is consistently present in Potorous (e.g., in AMNH 65332), whereas one or more small accessory foramina are usually present anterior to the infraorbital foramen in Bettongia.

Character 6. Masseteric process absent or, if present, not projecting ventrally below plane of molar alveoli (0); or present and projecting ventrally below plane of molar alveoli (1). Most examined metatherians have an indistinct-to-small rugose process at the base of the anterior root of the zygomatic arch, at or near the maxillary-jugal suture, which does not project ventrally below the level of the molar alveoli (state 0; e.g., in Marmosa [figs. 2, 34]). Dissections of Didelphis (e.g., Hiiemae and Jenkins, 1969; Turnbull, 1970) identify this structure as the attachment site for the tendon of origin of the superficial masseter. The attachment site for this tendon in most peramelemorphians (described by Filan, 1990) and a number of other metatherians (e.g., †Herpetotherium, Dactylopsila, Dactylonax, †Nimbacinus, and Sarcophilus) is difficult to distinguish osteologically. The degree of development of this process also appears to vary ontogenetically in other taxa, with older adults often having better-developed processes than younger individuals (e.g., as in Potorous). For these reasons, we do not distinguish here between presence of a small masseteric process and complete absence of such a process (subsuming both of these morphologies within state 0).

By contrast, a conspicuously different condition occurs in some metatherians, wherein the origin of the superficial masseter (dissected inter alia by Parsons, 1896: Abbie, 1939; Barbour, 1963; Sanson, 1980; and Warburton, 2009) extends along a large masseteric process that projects ventrally to, or below, the plane of the molar alveoli, as in the macropodids Macropus (Wells and Tedford, 1995: fig. 9) and Osphranter (Warburton, 2009: fig. 1). Among the taxa we examined, the latter condition (state 1) was observed in Trichosurus (but not in “Trichosurus” †dicksoni), †Litokoala, diprotodontids (e.g., †Nimbadon [Black and Hand, 2010: figs. 1, 2]), wynyardiids (e.g., †Namilamadeta [see Pledge, 2005: fig. 8A]), most Recent macropodids (e.g., Lagostrophus [fig. 53]), Caloprymnus (Finlayson, 1932), and Myrmecobius (fig. 41). Only one examined taxon, Wyulda, is polymorphic for this character. In some specimens of Glironia, the maxilla extends below the plane of the molar alevoli (Marshall, 1978: fig. 2; Voss and Jansa, 2009: fig. 37; Ardente et al., 2013: fig. 2; Arguero et al. 2017: fig. 2), but this is due to the large size of the orbit in this taxon; the masseteric process itself is weakly developed, so we scored Glironia as conforming to state 0.

TABLE 8

Selected Anatomical Synonyms for Structures of the Orbit and Zygoma

img-AAg0_01.gif

Character 7. Masseteric process formed at least partially by the maxilla (0); or maxillary process formed by the jugal alone (1). The maxilla contributes to the masseteric process of most metatherians in which it occurs (state 0; e.g., in Macropus [Wells and Tedford, 1995: fig. 9]). However, †Ngapakaldia has a masseteric process that is formed entirely by the jugal (state 1). Tedford and Woodburne (1987: 404) reported that the masseteric process is also formed entirely by the jugal in †Ilaria, but we observed that the maxilla does, in fact, make a clear contribution to this process.

This character was scored for all taxa in which a masseteric process is identifiable, whether or not the process extends ventrally below the plane of the molar alveoli (i.e., for some taxa scored as “0” for char. 6); only taxa in which no trace of a masseteric process could be identified (e.g., Dactylopsila, Dactylonax, Tarsipes, vombatids, most peramelemorphians, and †Herpetotherium; see char. 6) were scored as inapplicable (“-”).

Orbit and Zygomatic Arch

The osteology of the orbit and zygomatic arch provide a rich source of characters for mammalian phylogenetics. In particular, the taxonomically variable morphology of the orbital mosaic—the patchwork of bones that form the medial wall of the eye socket—has received much attention by systematists concerned with eutherian relationships (e.g., Muller, 1934; Cartmill, 1978; Cox, 2006), but this part of the skull has seldom been described in the metatherian literature (with some notable exceptions cited in the following accounts; see also Gregory, 1920: figs. 63–74). Unfortunately, certain elements of the marsupial orbital mosaic are misidentified in a number of published illustrations, including Dawson et al. (1989: fig. 17.3A), Hershkovitz (1992b: fig. 19; 1999: fig. 25), and Novacek (1993: fig. 9.4C), all of which mislabel the orbital process of the palatine as the “sphenoid” or orbitosphenoid.

The bones comprising this region include the lacrimal and jugal, together with orbital processes of the maxilla, palatine, frontal, and sphenoid complex. Anatomical nomenclature for this part of the skull is generally unproblematic, although multiple synonyms exist for several features (table 8). Whereas some structures in this part of the skull (notably those that resist mechanical forces associated with jaw movements) are robust and often well preserved in fossils, others are quite fragile; therefore, character information based on the latter is often missing for extinct taxa.

Character 8. Lacrimal exposure on orbital rim smooth (0); or with one or more distinct tubercles (1); or forming a distinct crest (2). The anterodorsal orbital exposure of the lacrimal is essentially smooth (state 0) in most metatherians, including †Pucadelphys, †Allqokirus, †Mimoperadectes, most didelphids (e.g., Caluromysiops [fig. 4A]), caenolestids, microbiotheriids, some peramelemorphians, some dasyurids, Notoryctes, acrobatids, petaurids, pseudocheirids, Tarsipes, and most phalangerids (e.g., Ailurops [fig. 3A]). However, distinct lacrimal tubercles (state 1) are present in †Sparassocynus (Beck and Taglioretti, 2020), all examined phascolarctids (e.g., Phascolarctos [fig. 3B]) and diprotodontids, all examined macropodiforms that preserve the lacrimal except Hypsiprymnodon moschatus, some phalangerids (Trichosurus [fig. 4B], Wyulda, and some specimens of Spilocuscus), and Recent vombatids. Wells and Tedford (1995: fig. 9) distinguished dorsal (supralacrimal) from ventral (infralacrimal) tuberosities in Macropus, in which such structures are sometimes distinct, but the homology of either with the single tuberosity exhibited by other taxa is problematic. Because supra- and infralacrimal tuberosities appear to coalesce in some macropodid specimens—for example, in AMNH 107374, a specimen of Osphranter robustus—and in the absence of recognition criteria that can be applied unambiguously from taxon to taxon, we scored lacrimal tuberosities as present or absent regardless of number or position. Based on dissections of “Halmaturus” (probably a species of Osphranter; M.D.B. Eldridge, personal commun.) by Boas and Paulli (1908: plate 13), Gregory (1920: 143) stated that “the orbital tubercle, or rim, serves dorsally for the attachment of the fascia covering the temporal mass and ventrally for the attachment of the palpebral ligament.”

A distinct crest extending along the lacrimal contribution to the orbital rim (state 2) is seen in †Herpetotherium, †Acdestis, Hypsiprymnodon moschatus, †Wakaleo, †Thylacoleo, †Namilamadeta, thylacinids, Myrmecobius (fig. 3C), many dasyurids (e.g., Phascogale, Sarcophilus, and Sminthopsis) and some peramelemorphians (e.g., Chaeropus [see Travouillon et al., 2019], Macrotis, and Perameles).

These states do not represent a clear morphocline, so this character has not been ordered here.

Character 9. Distinct lacrimal foramen or foramina present (0); or absent (1) in adult specimens. In the majority of metatherians examined for this study the lacrimal bone of adults is pierced by one or more distinct foramina that transmit the nasolacrimal ducts or canaliculi (state 0; figs. 3, 4). According to Archer (1976b: 221), the dasyurid lacrimal foramen transmits only the nasolacrimal duct, but Wible (2003: 79) stated that the two lacrimal foramina of Monodelphis each transmit a vein in addition to a nasolacrimal canaliculus. Adult specimens of Lasiorhinus uniquely lack a discrete lacrimal foramen or foramina (state 1; Murray et al., 1987: table 2, 454), although a lacrimal foramen is reportedly visible in juveniles (J. Wible, personal commun.). Instead, the lacrimal of this taxon is pierced by many small perforations that resemble the (presumably nutrient) foramina observable on the adjacent surfaces of neighboring bones.

Character 10. Two or more lacrimal foramina usually present (0); or one lacrimal foramen usually present (1). Most metatherians have two lacrimal foramina on each side (state 0; e.g., as in Phascolarctos [fig. 3B], Myrmecobius [fig. 3C], and Caluromysiops [fig. 4A]). Although †Nimbacinus (based on the single known skull of †N. dicksoni [QMF 36357; Wroe and Musser, 2001: 495]) and some specimens of Thylacinus (e.g., SAM M922 and M95) appear to have three lacrimal foramina, we chose to score these taxa as state 0 rather than create an additional state for this condition. Lestodelphys has two distinct lacrimal foramina close to the orbital rim (Voss and Jansa, 2009: fig. 51), but one or two much smaller foramina are also present within the orbital fossa; the presence of dried blood within the lumina of the latter perforations in some specimens (e.g., BMNH 1928.12.11.206, 1928.12.11.207) suggests that these are vascular pores, possibly similar to the vascular foramen of the lacrimal observed in the bat Pteropus by Giannini et al. (2006), rather than true lacrimal foramina (see also Wible et al., 2004: 47–49). Beck and Taglioretti (2020: 9) also reported the presence of probable nutrient foramina within the orbital exposure of the lacrimal in †Sparassocynus. However, we have not scored presence or absence of these additional small foramina as an additional character here.

By contrast, only a single distinct lacrimal foramen (state 1) is present in some specimens of †Pucadelphys, a few didelphids (Hyladelphys, Chironectes, and some specimens of Caluromysiops), paucituberculatans (e.g., Caenolestes [fig. 35]), most peramelemorphians (with the exception of Macrotis), some dasyurids, Notoryctes (fig. 36), and many diprotodontians (e.g., Ailurops [fig. 3A] and Trichosurus [fig. 4B]). Lasiorhinus was scored as inapplicable (“-”) for this character, based on the absence of any obvious lacrimal foramina in this taxon (see char. 9).

Character 11. Lacrimal foramen (or foramina) contained entirely within lacrimal bone (0); or at least one lacrimal foramen located within suture between lacrimal and maxilla (1); or at least one lacrimal foramen located within suture between lacrimal and jugal (2). The lacrimal foramen (or foramina) is (or are) usually entirely enclosed by the lacrimal bone in metatherians (state 0; e.g., as in figs. 24). However, at least one lacrimal foramen is consistently located within the suture between the lacrimal and maxilla (state 1) in Petauroides, Cercartetus, †Thylacoleo, †Yarala (Muirhead, 2000), Macrotis (fig. 38; see also Muirhead, 2000: 518), and Peroryctes (Aplin et al., 2010: 22). The latter morphology is also sometimes seen in Chaeropus, Rhynchomeles, Pseudocheirus, Osphranter, and Petrogale, all of which have been scored as polymorphic (“0+1”). Rhyncholestes is unique among the metatherians we surveyed in that the single lacrimal foramen is usually located within the lacrimal-jugal suture (state 2).

We scored Notoryctes as unknown (“?”) for this character, due to the absence of relevant sutures in the specimens we examined (fig. 36; see char. 12), whereas we scored Lasiorhinus as inapplicable (“-”) because it lacks lacrimal foramina (see char. 9). Although the lacrimal and maxillary bones are fused in all specimens of Burramys that we examined (see char. 12), the jugal suture is still identifiable and well separated from the lacrimal foramen (or foramina; fig. 46); therefore, Burramys has been scored using ambiguity coding (as “0/1”).

Character 12. Bones of orbital mosaic separate, unfused (0); or maxilla and lacrimal fused (1); or entire orbital mosaic fused (2). In most metatherians, the bones comprising the orbital mosaic (i.e., frontal, lacrimal, maxilla, and palatine) are unfused, with sutures between them clearly identifiable (state 0; figs. 3, 4). In Burramys, however, the maxilla and lacrimal are fused, both within the orbital mosaic and, external to the orbit, on the rostrum (state 1; fig. 46). In Notoryctes, the entire orbital mosaic is fused, with no sutures identifiable (state 2; fig. 36). As defined, these character states do not appear to represent an unequivocal morphocline (it is not obvious that the condition in Notoryctes must necessarily have evolved from a precursor morphology in which only the maxillary-lacrimal suture was fused), so this character was not ordered in any of the analyses presented here.

FIG. 3.

Lacrimal region of Ailurops ursinus (A, AMNH 152884), Phascolarctos cinereus (B, AMNH 65612), and Myrmecobius fasciatus (C, AMNH 155328). Alternative states for characters 8 and 10 (see main text for descriptions of these characters and character states) are illustrated as follows: Ailurops 8(0), 10(1); Phascolarctos 8(1), 10(0); Myrmecobius 8(2), 10(0). Abbreviations: dlf, dorsal lacrimal foramen; fro, frontal; jug, jugal; lac, lacrimal; lc, lacrimal crest; lf, lacrimal foramen; lt, lacrimal tubercle; max, maxillary; nas, nasal; pal, palatine, vlf, ventral lacrimal foramen. Specimens are not drawn to the same scale.

img-z41-1_01.jpg

Character 13. Palatine and lacrimal bones in contact on medial wall of orbit (0); or maxillary and frontal bones in contact on medial orbital wall (1). Two alternative patterns of contact among the lacrimal, maxillary, palatine, and frontal bones in the anterior part of the orbit have been reported among metatherians. Most of the taxa we examined exhibit palatine-lacrimal contact (e.g., as in Caluromysiops [fig. 4A] and †Pucadelphys [Marshall and Muizon, 1995: fig. 15]). However frontal-maxillary contact (Gregory, 1920; Springer et al., 1997: char. 50) is consistently present in phalangerids (e.g., Trichosurus [fig. 4B]), Dactylonax, most macropodiforms (with the notable exceptions of Potorous, some specimens of Dorcopsis [e.g., the left side of AMNH 198088], and one specimen of †Balbaroo [QM F30456]), Phascolarctos, †Nimiokoala, †Warendja, and †Namilamadeta. Ailurops (Crosby, 2002b) and †Nimbadon are polymorphic for this character, with specimens of both taxa presenting different states on the left and right sides (e.g., AMNH 196495 for Ailurops; QM F42677 for †Nimbadon). However, based on the specimens examined in this study, the modal condition in Ailurops appears to be frontal-maxillary contact (as in other phalangerids, contra Flannery et al., 1987; Crosby, 2002b), whereas in †Nimbadon it appears to be palatine-lacrimal contact. Other bones are occasionally exposed on the medial wall of the orbit, but we did not score these rare variants for phylogenetic analysis.

FIG. 4.

Anterior orbital region of Caluromysiops irrupta (A, AMNH 208101) and Trichosurus vulpecula (B, AMNH 65543). Alternative states for characters 8, 10, 13, and 14 (see main text for descriptions of these characters and character states) are illustrated as follows: Caluromysiops 8(0) 10(0[but note that C. irrupta is polymorphic for this character]), 13(0), 14(0); Trichosurus vulpecula 8(1), 10(1), 13(1), 14(1). Abbreviations: dlf, dorsal lacrimal foramen; fro, frontal; jug, jugal; lac, lacrimal; lf, lacrimal foramen; lt, lacrimal tubercle; max, maxillary; mf, maxillary foramen; nas, nasal; pal, palatine; vlf, ventral lacrimal foramen. Specimens are not drawn to the same scale.

img-z42-1_01.jpg

The orbital mosaic is not sufficiently well preserved to score this character in available specimens of †Allqokirus, †Mayulestes, †Mimoperadectes, and †Herpetotherium, all of which we coded as unknown (“?”). The lacrimal and maxilla appear to be fused within the orbital fossa in all specimens of Burramys that we examined, and the entire orbital mosaic is fused in Notoryctes (see char. 12); therefore, we also scored both of these taxa as unknown pending examination of juvenile specimens that might preserve evidence of sutures in this region.

FIG. 5.

Posterior orbital region of Potorous tridactylus (A, AMNH 65293) and Perameles gunnii (B, MVZ 127070). Alternative states for characters 15–17 (see main text for descriptions of these characters and character states) are illustrated as follows: Potorous 15(0), 16(1), 17(1); Perameles 15(1), 16(0), 17(0). Abbreviations: als, alisphenoid; atp, alisphenoid tympanic process; for, foramen rotundum; fro, frontal; jug, jugal; lac, lacrimal; max, maxillary; os, orbitosphenoid; pal, palatine; plpf, posterolateral palatal foramen; pt, pterygoid; sf, sphenorbital fissure; sq, squamosal. Specimens are not drawn to the same scale.

img-z43-1_01.jpg

Character 14. Maxillary foramen bordered by lacrimal (0); or contained entirely within the maxilla (1). The maxillary foramen (the posterior opening of the infraorbital canal) is bordered dorsally by the lacrimal bone in most metatherians (state 0), and a small anterior process of the palatine occasionally forms part of the medial border as well. This widespread condition is seen in †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, didelphids (e.g., Caluromysiops [fig. 4A]), paucituberculatans, peramelemorphians, dasyuromorphians, microbiotheriids, and some diprotodontians (Cercartetus, acrobatids, pseudocheirids, Tarsipes, Petaurus, Gymnobelideus, and Potorous). However, in other diprotodontians—namely, phalangerids (e.g., Trichosurus [fig. 4B]), phascolarctids, diprotodontids, thylacoleonids, †Warendja, †Namilamadeta, Dactylonax, most specimens of Dactylopsila, and all examined macropodiforms except Potorous—the maxillary foramen is contained entirely within the maxilla. Lasiorhinus and Vombatus are both polymorphic (“0+1”) for this character. As for the preceding character, fusion of the lacrimal and maxilla in Burramys and of the entire orbital mosaic in Notoryctes (see char. 12) compelled us to score both taxa as unknown (“?”).

This character is obviously at least partially correlated with character 13, because frontal-maxillary contact and full enclosure of the maxillary foramen by the maxilla both reflect a proportionally larger orbital exposure of the maxilla. However, some degree of independence between these characters is indicated by the observation that the maxillary foramen is entirely within the maxilla but the lacrimal and palatine remain in contact in some taxa (e.g., Dactylopsila, †Silvabestius, †Litokoala, †Lekanoleo, †Thylacoleo).

Character 15. Orbitosphenoid exposed to lateral view in medial orbital wall (0); or orbitosphenoid laterally inapparent or absent (1). Bilaterally paired orbitosphenoid ossifications, which always appear to be fused with the median presphenoid element, are prominently exposed to lateral view in the medial wall of the orbit in most metatherians (state 0; e.g., as in Potorous [fig. 5A]). Although the orbitosphenoid is fused to the frontal and alisphenoid in adult specimens of Caenolestes, it is identifiable as a separate ossification that is clearly exposed in lateral view in subadult specimens (e.g., FMNH 94948).

By contrast, in adult specimens of most Recent peramelemorphians (e.g., Perameles [fig. 5B]), the lateral exposure of the orbitosphenoid is much reduced or absent (state 1; Muirhead, 2000; Travouillon et al., 2010). In some specimens, this morphology appears to result from the orbitosphenoid being concealed behind outgrowths of the frontal and/or maxilla (as in AMNH 151939, a specimen of Peroryctes raffrayana), whereas in others the orbitosphenoid itself appears to be reduced in size (as in AMNH 190986, a specimen of Echymipera kalubu), but we have not distinguished between these apparently alternative conditions when scoring this character. However, the orbitosphenoid remains prominently exposed to lateral view in a few peramelemorphians, namely Macrotis, †Yarala (Muirhead, 2000), †Galadi (Travouillon et al., 2010), and †Bulungu (Gurovich et al., 2014), all of which have been scored as state 0. Isoodon and Chaeropus are both polymorphic (“0+1”) for this character.

Character 16. Maxillary and alisphenoid separate (0); or in contact on orbital floor (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 43; 2009: char. 51), who consistently observed maxillary-alisphenoid contact (state 1) only in Lutreolina and Monodelphis (Voss and Jansa, 2009: fig. 10B). We maintain Voss and Jansa's (2003, 2009) didelphid scoring here, except that we score Tlacuatzin as polymorphic (“0+1”) for this character. We also observed broad maxillary-alisphenoid contact in †Sparassocynus (Beck and Taglioretti, 2020) and †Thylatheridium. The maxillary and alisphenoid are separated by the palatine (state 0) in most other metatherians (e.g., Perameles [fig. 5B]), but the maxilla and alisphenoid broadly contact one another in all examined macropodiforms (e.g., Potorous [fig. 5A] and Macropus [Wells and Tedford, 1995: fig. 9]). Narrower maxillary-alisphenoid contact is also consistently present in Pseudochirops cupreus (fig. 51), Ps. archeri, and Petropseudes, and it is sometimes also present in Pseudochirulus and Burramys (which have been scored as polymorphic).

Character 17. Foramen rotundum laterally exposed and separate from sphenorbital fissure (0); or foramen rotundum concealed from lateral view, partly or wholly confluent with sphenorbital fissure in rear of orbit (1); or foramen rotundum ventral to sphenorbital fissure (2). The foramen rotundum (by which the maxillary branch of the trigeminal nerve exits the skull) is separate from the sphenorbital fissure and exposed in lateral view (state 0) in †Pucadelphys, †Allqokirus, †Mayulestes, †Herpetotherium, didelphids (e.g., Monodelphis [Wible, 2003: fig. 4]), paucituberculatans (with the exception of some specimens of Rhyncholestes), †Yalkaparidon, peramelemorphians (e.g., Perameles [fig. 5B]), dasyuromorphians (with the exception of some specimens of Neophascogale, in which the medial wall of the foramen rotundum fails to ossify; see, e.g., right side of AMNH 109520), and many diprotodontians (e.g., diprotodontids, phalangerids, Dactylopsila, Dactylonax, and thylacoleonids), all of which are scored as state 0. Although Novacek (1993: 458) stated that the foramen rotundum and sphenorbital fissure are confluent in didelphids, both openings are distinct and clearly identifiable in all the Recent and fossil didelphid material we examined.

By contrast, the foramen rotundum is concealed from lateral view within a common vestibule that it shares with the sphenorbital fissure in the back of the orbit (state 1) in Dromiciops, most macropodiforms (e.g., Potorous [fig. 5A]), pseudocheirids, burramyids, and Acrobates. Several other taxa—including Macropus, Phascolarctos, Distoechurus, Petaurus, Gymnobelideus, and Rhyncholestes—are polymorphic (“0+1”) for this character. A distinctive morphology is seen in Notoryctes, in which the foramen rotundum is ventral (rather than lateral) to the sphenorbital fissure (state 2), with the two openings sometimes appearing confluent (e.g., on the left and right sides of AMNH 202105; see also comments by Archer, 1976b: 270, 308). As noted by Voss and Jansa (2003: 27), the foramen rotundum has been misidentified in some published illustrations, notably in drawings of the skull of Dromiciops by Hershkovitz (1999: fig. 25), where the true foramen rotundum cannot be seen at all.

These states do not form a plausible morphocline, so this character has not been ordered in any analyses presented here.

Character 18. Postorbital process of frontal absent or indistinct (0); or present (1). The dorsal attachment point of the postorbital ligament (which delimits the orbital fossa from the temporal fossa; fig. 1) on the frontal lacks a distinct osteological marker in many metatherians (state 0), but in others (e.g., Marmosa, fig. 1) its position is indicated by a distinct, laterally directed postorbital process (state 1). Because the postorbital process (if present) often develops relatively late in postnatal ontogeny (see, e.g., Abdala et al., 2001; Flores et al., 2006, 2010), we scored this character only from adult specimens. Voss and Jansa (2003, 2009) reviewed the distribution of postorbital processes in Recent didelphids, and we largely followed their scoring criteria and decisions here. Following Voss and Jansa (2009), we do not distinguish between postorbital processes that are “flattened” from those that “not flattened” sensu Voss and Jansa (2003). In addition, we scored Recent didelphids with supraorbital “beads” sensu Voss and Jansa (2009) but lacking distinct, laterally projecting processes as state 0.

Among other metatherians, postorbital processes are absent or indistinct in †Pucadelphys (Marshall and Muizon, 1995; Ladevèze et al., 2011: fig. 2), †Mayulestes (Muizon, 1994: fig. 1a; 1998), †Allqokirus (Muizon et al., 2018), caenolestids (e.g., Caenolestes [fig. 35]), Notoryctes (fig. 36), most dasyurids (e.g., Pseudantechinus [fig. 42]), peramelemorphians (e.g., Perameles [fig. 39]), Dromiciops (fig. 43), some †diprotodontids (e.g., †Neohelos [Murray et al., 2000]), burramyids (e.g., Burramys [fig. 46]), acrobatids, Gymnobelideus, Tarsipes (fig. 49), most pseudocheirids, most phalangerids, and some macropodiforms (e.g., Lagostrophus [fig. 53]). By contrast, distinct postorbital processes are present in †Thylophorops (Goin et al., 2009b: fig. 5A, B), †Sparassocynus (Reig and Simpson, 1972: fig. 2), a few dasyurids (e.g., Sarcophilus, Myoictis, Dasyurus, Dasykaluta), thylacinids (e.g., Thylacinus [fig. 40]), Myrmecobius (fig. 41), and various diprotodontians (e.g., thylacoleonids, phascolarctids, †Namilamadeta, Lasiorhinus, †Silvabestius, Pseudochirulus, most petaurids, and some macropodiforms). Specimens of Macropus giganteus have a distinct bulge in the lateral wall of the frontal approximately level with the frontal process of the jugal (Wells and Tedford, 1995: fig. 9C), but this convexity is due to an enlarged frontal sinus and is not a true postorbital process (it is distinctly ventral to the dorsolateral margin of the frontal), so we scored Macropus as lacking distinct postorbital processes.

Although not amenable to scoring as discrete character states, postorbital processes (where present) vary considerably in size and shape within Metatheria. They are particularly prominent in †Mimoperadectes (Horovitz et al., 2009), the Recent didelphids Caluromys and Caluromysiops (Voss and Jansa, 2009: figs. 38, 39), and †Sparassocynus (in which they are large and hornlike [Reig and Simpson, 1972: fig. 2]). In †Thylacoleo, they closely approach or meet the frontal process of the jugal, forming a postorbital bar (see, e.g., Anderson, 1929). Nevertheless, these processes (where identifiable) are usually roughly triangular in outline when viewed dorsally, with the apex directed laterally. A notable exception is seen in Myrmecobius (fig. 41), where they form distinctly winglike outgrowths that extend along most of the dorsolateral margin of the frontal and almost contact the lacrimal crest anteriorly.

Character 19. Postorbital process formed by frontal and parietal (0); or by frontal only (1); or by frontal and lacrimal (2). The postorbital process is formed exclusively by the frontal (state 1) in most metatherians that possess such a structure (e.g., Marmosa, fig. 1), but the postorbital process (where identifiable) is formed by frontal and parietal moieties (state 0) in Glironia (Voss and Jansa, 2009: fig. 37), Distoechurus (fig. 48), Petaurus (fig. 50), Hemibelideus, Petauroides, Pseudochirulus, and some specimens of Pseudochirops archeri (e.g., AMNH 253664). Although postorbital processes are not always distinct in adult individuals of Dactylopsila and Dactylonax, where identifiable they are positioned far anteriorly, on the suture between the frontal and lacrimal, with both of these bones contributing (a remnant of the postorbital ligament is still attached to the right postorbital process of AMNH 194759, a specimen of Dactylonax palpator, confirming the identity of this structure); this unusual morphology has been assigned its own state here (state 2). The anterior position of the postorbital process in Dactylopsila and Dactylonax may be connected with their very dorsally directed orbits (Cartmill, 1974).

We scored this character as inapplicable (“-”) only for those taxa in which a postorbital process was not identifiable in any specimen examined (as in, e.g., Recent caenolestids [e.g., Caenolestes; fig. 35], Notoryctes [fig. 36], Dromiciops [fig. 43], Tarsipes [fig. 49], and Acrobates). However, taxa in which at least one specimen exhibited a distinct postorbital process were scored for this character, even if the modal condition was absence. The states comprising this character have been arranged to form a plausible morphocline, corresponding to a sequential anteroposterior shift in the position of the postorbital processes, and have been ordered in all our analyses.

Character 20. Facial exposure of jugal variously developed but not deeply bifid (0); or facial exposure of jugal deeply bifid, with distinct anterodorsal and anteroventral processes (1). The facial exposure of the jugal has a roughly crescentic, straight, or irregular suture with the maxilla in most metatherians (e.g., Marmosa; figs. 2, 34; state 0). However, in all peramelemorphians (including †Yarala, †Galadi, and †Bulungu; Muirhead, 2000; Travouillon et al., 2010; 2013b; Gurovich et al., 2014), the facial portion of the jugal is deeply bifid, with distinct anterodorsal and anteroventral processes, usually flanking a deep nasiolabial fossa (an excavation in the maxilla from which the maxillonasolabialis muscle originates) (state 1: figs. 3739; Filan, 1990; Muirhead, 2000). Although the anteroventral jugal process is absent in one examined specimen of Echymipera kalubu (AMNH 190973), we coded the modal condition (bifid jugal with both processes present) for Echymipera. The facial exposure of the jugal is weakly bifid in †Yalkaparidon (Beck, 2009: fig. 3; Beck et al., 2014), Dactylopsila, and some specimens of Dactylonax, but the bifurcation is much weaker than in peramelemorphians, and there is no distinct nasiolabial fossa between the anteroventral and posteroventral processes; therefore, we scored these taxa as corresponding to state 0. The maxillo-jugal suture is fused in adult specimens of Notoryctes, but CT-scans of a fluid-preserved juvenile (NMV C11082) demonstrate that this suture is not bifid prior to fusion (Beck, in prep.).

Character 21. Jugal extends posteriorly to glenoid region (0); or jugal terminates well anterior to glenoid region (1). In all examined metatherians except Tarsipes, the jugal extends posteriorly to the glenoid fossa (state 0), a probable therian plesiomorphy (Wible et al., 2005). In Tarsipes, however, the jugal terminates well anterior to the glenoid region (fig. 49; Rosenberg and Richardson, 1995; fig. 1c), an autapomorphic morphology that we scored as a distinct state (state 1).

Character 22. Jugal terminates posteriorly in a faceted preglenoid process (0); or in a ventrally expanded ectoglenoid crest (1); or jugal terminates posteriorly without a distinct process or crest (2). The jugal produces a knoblike preglenoid process that bears a distinct, posteriorly oriented articular facet (state 0) in most metatherians (e.g., Marmosa [fig. 1], Phascogale [fig. 10A]). In Myrmecobius, however, the jugal terminates posteriorly in a ventrally expanded lateral (ectoglenoid) crest that sometimes bears an indistinct medial articular facet (state 1 [fig. 41]). A similar ventrally expanded crest is seen in Phascolarctos and some macropodiforms (e.g., †Ekaltadeta [Wroe et al., 1998], Lagorchestes, Setonix, Wallabia, Lagostrophus), all of which have also been scored as state 1. By contrast, neither a distinct preglenoid process nor an ectoglenoid crest is present in most diprotodontians (e.g., Petaurus [fig. 10B]); in these taxa, the posterior extremity of the jugal tapers dorsoventrally and does not appear to restrict the anterior excursion of the mandibular condyle (state 2). As noted above (char. 21), the jugal of Tarsipes terminates well anterior to the glenoid region, so we scored this taxon as inapplicable (“-”).

In the absence of a clear transformational relationship among these three conditions, we treated this character as unordered (nonadditive) in all the phylogenetic analyses reported below.

Character 23. Glenoid process of alisphenoid absent (0); or present (1). In Myrmecobius and most diprotodontians—including diprotodontids, wynyardiids, vombatids (e.g., Vombatus [fig. 45]), thylacoleonids, burramyids (e.g., Burramys [fig. 46]), phalangerids (e.g., Phalanger [fig. 47]), and petaurids (e.g., Petaurus [figs. 9C, 10B, 50])—the posterior root of the zygomatic arch is formed exclusively by the squamosal with no contribution from the alisphenoid (state 0). By contrast, a distinct glenoid process (“entoglenoid process” sensu Muizon, 1998; 1999) of the alisphenoid, which may or may not actually participate in the temporomandibular joint, forms part of the posterior zygomatic root (state 1) in all the other metatherians we examined that could be scored for this character. The latter include †Allqokirus (Muizon et al., 2018: fig. 21), †Mayulestes (Muizon, 1998: fig. 8), †Pucadelphys (Muizon, 1999: char. 5), didelphids (e.g., Marmosa [figs. 1, 9A, 34]), pauctituberculatans (e.g., Caenolestes [fig. 35]), thylacinids (e.g., Thylacinus [fig. 40]), dasyurids (e.g., Phascogale [figs. 9B, 10A], Pseudantechinus [fig. 42)]), Phascolarctos (fig. 44), acrobatids (e.g., Distoechurus [fig. 48]), most macropodiforms (e.g., Lagostrophus [fig. 53]), and some specimens of the pseudocheirid Hemibelideus.

Dorsolateral Braincase

The dorsolateral braincase—including the parietals and the interparietal (if identifiable as a distinct ossification), together with parts of the frontals, alisphenoid, and squamosals—encloses and protects the brain and provides the surface of origin for the temporalis muscle. Taxonomically variable features of the dorsolateral braincase among Recent and extinct metatherians include the fusion or persistence of sutures, alternative patterns of contact among closely juxtaposed bones, the development of crests for muscle attachment, presence/absence of fenestrae, and pneumatization of braincase-roofing elements. For the most part, terminology for dorsolateral braincase features is not complicated by synonymous usage in the metatherian literature. Two exceptions that we note here are “post-parietal,” which has been used some authors (e.g., Marshall, 1976; see also Novacek, 1993: 472) to refer to the interparietal bone (= os interparietalis [International Committee on Veterinary Gross Anatomical Nomenclature, often cited as NAV]), and “temporal crest” (as in Macalister, 1872; Thomas, 1888) which is synonymous with the sagittal crest.

Character 24. Median suture between left and right frontals unfused in subadults (0); or median frontal suture at least partially fused in subadults (1). In most metatherians, the right and left frontals are separated by a median suture that persists throughout adult life (state 0). However, Voss and Jansa (2009) noted that the frontals are coossified early in postnatal life in several Recent didelphids (Chironectes, Didelphis, Lutreolina, and Philander), such that the median suture between them is partly or completely fused in all examined subadult and young adult specimens (and in most juveniles as well; state 1). Partial or complete fusion of the median frontal suture is also consistently seen in subadult caenolestids, Recent vombatids, †Sparassocynus, Thylacinus, Notoryctes, and Hemibelideus, all of which have been scored as corresponding to state 1. Subadult frontal fusion is also variably present in Ailurops, which has been scored as polymorphic (“0+1”).

Taxa in which the median frontal suture is unfused in subadults but fused in mid- to late-stage adults (e.g., †Nimbadon; Black et al., 2010: table 1) have been scored as state 0 because this fusion reflects the widespread mammalian trend of sutural obliteration during adulthood rather than the precocial fusion described above. Therefore, our scoring criteria are more restrictive than those employed by Voss and Jansa (2009), who scored the Recent didelphid Caluromysiops as exhibiting fusion of the median frontal suture based on the condition in most of the fully adult individuals they examined. Given that we were unable to examine any subadult specimens of Caluromysiops, and that one fully adult specimen retains an unfused median frontal suture (FMNH 84426; Voss and Jansa, 2009: 32), we prefer to score this taxon as unknown (“?”). Other taxa in which the median frontal suture is fused in adults but for which no subadult specimens could be examined (e.g., Dactylonax, and many fossil taxa) have also been scored as unknown. Taxa in which this suture remains completely unfused in most or all adult specimens could obviously be scored as state 0 regardless of the availability of younger individuals.

Character 25. Median suture between left and right parietals unfused in subadults (0); or median parietal suture at least partially fused in subadults (1). Resembling the widespread frontal morphology described above, the paired parietal bones are likewise separated by a median suture that persists to adulthood in most metatherians (state 0). However, Osgood (1921: 103) noted the absence of a median parietal suture in Caenolestes (fig. 35) and implied that a similar condition is seen in peramelemorphians. We observed complete fusion of this suture (state 1) in subadult Caenolestes and Rhyncholestes, but we were unable to examine subadult specimens of Lestoros, which is here scored as unknown (“?”). The median parietal suture is at least partially fused in all examined subadult peramelemorphians except Rhynchomeles, which is polymorphic for this character (the median parietal suture is unfused in BMNH 1920.7.26.35, an adult male). Other taxa in which this suture is fused in subadults include †Pucadelphys, †Allqokirus (Muizon et al., 2018), several phalangerids (Spilocuscus, Strigocuscus, and some specimens of Ailurops and Phalanger), †Nimbadon (Black et al., 2010), Recent vombatids, Thylacinus (Warburton et al., 2019: fig. 3), and some specimens of Myoictis (e.g., AMNH 152010). As for the preceding character, we scored taxa as corresponding to state 1 only if this suture is at least partially fused in subadult specimens; taxa were scored as unknown (“?”) if subadults were not examined. Taxa were also scored as unknown if the region of the median parietal suture is concealed by the sagittal crest in available specimens.

Character 26. Parietal and alisphenoid in contact on lateral aspect of braincase (0); or frontal and squamosal in contact (1). Descriptions of character states and an account of the distribution of this character among Recent didelphids were provided by Voss and Jansa (2003), whose character 36 is slightly rephrased above. As noted by Voss and Jansa (2003, 2009), most didelphids exhibit parietal-alisphenoid contact (state 0), with Metachirus uniquely exhibiting frontal-squamosal contact (state 1). Most nondidelphid metatherians also exhibit parietal-alisphenoid contact, notably including some fossil peramelemorphians (†Yarala, †Galadi, and †Bulungu [see Muirhead, 2000; Travouillon et al., 2010; Gurovich et al., 2014]) and the fossil vombatid †Warendja.

By contrast, frontal-squamosal contact occurs in Recent peramelemorphians (figs. 3739), some dasyurids (Sminthopsis and Antechinomys; Archer, 1981), Thylacinus (fig. 40), Dromiciops (fig. 43; contra Hershkovitz, 1999: fig. 25), Recent vombatids, diprotodontids, †Thylacoleo, †Namilamadeta, Petaurus (fig. 50), Gymnobelideus, and many nonmacropodid macropodiforms (e.g., Potorous [fig. 54]). Frontal-squamosal contact is also occasionally seen in Pseudantechinus (e.g., AMNH 162569), but alisphenoid-parietal contact is the modal condition (fig. 42). Phascolarctos, Neophascogale, and Phascolosorex are polymorphic (“0+1”) for this character. The sutural pattern of the lateral aspect of the braincase is not discernable in adult specimens of Notoryctes, but CT-scans of a fluid-preserved juvenile (NMV C11082) demonstrate that the frontal and squamosal are in contact (Beck, in prep.) contra Horovitz and Sánchez-Villagra (2003: char. 176) and subsequent studies that have used versions of their matrix (e.g., Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009).

In general, our observations are congruent with those of Wroe et al. (1998: 744), except that we score †Yalkaparidon as exhibiting parietal-alisphenoid contact: the squamosal and frontal are in contact in QMF13008 (the only known skull of Y. coheni), but much of the alisphenoid contribution to the lateral wall of the braincase appears to have flaked off; when intact, the alisphenoid would almost certainly have contacted the parietal in this specimen (Beck et al., 2014: fig. 7, 142).

Character 27. Scars of M. temporalis origin on braincase not fused middorsally to form sagittal crest in adults (0); or sagittal crest short in adults, extending from nuchal crest onto parietals, but not extending onto frontals (1); or sagittal crest long in adults, extending onto frontals (2). Voss and Jansa (2003: char. 35) summarized the morphology and taxonomic distribution of the sagittal crest among Recent didelphids, but we follow Voss and Jansa (2009: char. 57) in recognizing an intermediate state between absence of a sagittal crest and presence of a large crest extending to the frontals, namely a small sagittal crest that is present on the parietals but that does not extend onto the frontals. Voss and Jansa (2009: 32–33) briefly discussed the distribution of sagittal crests in selected nondidelphid metatherians, and we extend their observations here. A sagittal crest is consistently absent (state 0) in adult caenolestids (e.g., Caenolestes [fig. 35]), Notoryctes (fig. 36), some peramelemorphians (e.g., Chaeropus [fig. 37] and Perameles [fig. 39]), Myrmecobius (fig. 41), some dasyurids (e.g., Pseudantechinus [fig. 42]), Dromiciops (fig. 43), Recent vombatids (e.g., Vombatus [fig. 45]), some macropodiforms, and most other diprotodontians.

In most of the remaining taxa we examined, a sagittal crest is present in at least some adult specimens, but it is usually small, present only on the parietals (state 1). However, a large sagittal crest extending to the frontals (state 2) is seen in †Thylatheridium (Reig, 1958a: figs. 5, 6), †Thylophorops (Goin et al., 2009b: fig. 5A), †Yarala (Muirhead, 2000), †Galadi (Travouillon et al., 2010), a few dasyurids (e.g., Sarcophilus, Dasyurus, and †Barinya [Wroe, 1999]), thylacinids (e.g., Thylacinus [fig. 40]), a few diprotodontians (e.g., †Ekaltadeta [Wroe et al., 1998], most specimens of Dactylonax, some specimens of Dactylopsila), and some specimens of Macrotis ([fig. 38]). Ontogenetically variable, the sagittal crest is often absent in juveniles but present in conspecific adults (e.g., Abdala et al., 2001; Flores et al., 2006; Black et al., 2010), so we scored this character only from specimens with fully erupted permanent dentitions. Because the states of this character form a plausible morphocline, we treated it as ordered in all our phylogenetic analyses.

Character 28. Petrosal not exposed on posterolateral braincase (0); or exposed through a posterolateral fenestra (1). In most metatherians, the petrosal is not exposed on the posterolateral braincase (state 0), but a distinct fenestra exposes the underlying pars canalicularis of the petrosal on the posterolateral surface of the braincase anterior to the lambdoid crest (state 1) in some Recent didelphids (Voss and Jansa, 2003: char. 37; 2009: char. 58). Interestingly, we observed a similar morphology in several macropodids, namely Lagostrophus, Notomacropus, Osphranter, Onychogalea, Setonix, Wallabia, and some specimens of Macropus, Petrogale, and Thylogale; in Wells and Tedford's (1995: fig. 9B) cranial illustration of Macropus giganteus, this fenestra is labelled as the postsquamosal fissure. In all the didelphids and in most of the macropodids in which it is present, this fenestra is contained within the squamosal-parietal suture (Voss and Jansa, 2009: fig. 12B), but it is surrounded by the squamosal in Setonix (see char. 29).

Character 29. Fenestra exposing petrosal on posterolateral braincase within squamosal-parietal suture (0); entirely within squamosal (1). As noted above, the fenestra that exposes the petrosal on the posterolateral braincase in some didelphids and macropodids is usually within the squamosal-parietal suture (state 0), but the fenestra is entirely within the squamosal (state 1) of Setonix. This character was scored as inapplicable (“-”) in all taxa with an unfenestrated posterolateral braincase.

Character 30. Interparietal absent, very small, or polymorphic (0); or large and consistently present (1). There is significant disagreement in the literature concerning the occurrence of the interparietal (sometimes referred to as the postparietal) among metatherian mammals. Novacek (1993), for example, stated that marsupials lack an interparietal, whereas Rougier et al. (1998: char. 155) and subsequent studies that have used versions of this matrix (e.g., Wible et al., 2001; Rougier et al., 2015; Ni et al., 2016) scored an interparietal as absent in †Mayulestes but present in †borhyaenids, Didelphis, Marmosa, dasyurids, and Dromiciops. Because these authors did not explain the observational basis for their discrepant conclusions, the criteria by which an interparietal can be recognized as such clearly merit discussion. Our observations suggest that metatherians exhibit substantial parsimony-informative variation in interparietal morphology that has yet to be incorporated in any phylogenetic analysis. We also discuss Koyabu et al.'s (2012: supplementary information, p. 4) observations regarding the presence or absence of this bone in metatherians.

A single, large, wedge-shaped or oblong interparietal bone is unequivocally present (state 1) in thylacinids, †Barinya (Wroe, 1999), Myrmecobius (fig. 41), Dromiciops (fig. 6A; fig. 43), and in all our diprotodontian terminals that could be scored for this character except vombatids. In these taxa, the sutures that separate the interparietal from adjacent bones (supraoccipital and parietals) are clearly visible in all of the juvenile skulls we examined—with the notable exception of Dactylopsila and Dactylonax (see char. 31)—and they remain obvious in adult specimens of Myrmecobius, Dromiciops, and most diprotodontians. In some adult diprotodontians (e.g., phalangerids and pseudocheirids), however, the interparietal is almost completely overgrown by parietal scars (from which the temporalis muscle originates), which meet in the midline to form a sagittal crest that conceals the median parietal suture (see comments under char. 27). This may explain Murray et al.'s (1987: table 2) statement that the interparietal is “very small” in Trichosurus vulpecula: the interparietal does indeed appear to be small in older adults due to overgrowth by the parietals, but it is a large and obvious element separated by distinct sutures from surrounding bones in all of the juvenile, subadult, and young adult specimens of T. vulpecula that we examined. Although large and distinctly sutured in juvenile Thylacinus (e.g., SAM M1958; see also Newton et al., 2018: supplemental material), the interparietal is indistinguishably fused with both the supraoccipital and the parietals in older specimens (fig. 40), whereas the interparietal fuses with the parietals (but not with the supraoccipital) in subadult and adult specimens of Strigocuscus (e.g., AMNH 196498, 196503).

Our observations regarding diprotodontians exhibit some apparent discrepancies with those of Black et al. (2012a: char. 63), who scored the interparietal as absent in †Lekanoleo and †Nimbadon. We consider the interparietal to be clearly present and large in both of these taxa: in †Lekanoleo, the parietal-interparietal suture is identifiable in dorsal view on the left side of QM F23453 (contra Gillespie et al., 2020: 6), and the interparietal can be identified as a distinct ossification when viewed ventrally through the foramen magnum of this specimen. In †Nimbadon, the interparietal is clearly a large, distinct ossification in QM F53648 (a pouch young specimen; see Black et al., 2010: fig. 10A). Black et al. (2012a) also scored the interparietal as absent in †Namilamadeta and †Ngapakaldia, whereas we consider available specimens of both of these taxa to be insufficiently well preserved to determine whether or not this bone was present. However, Black et al. (2012a) used a slightly different scoring criterion to ours, namely whether or not a suturally distinct interparietal is visible in adult specimens (K.H. Black, personal commun.).

As discussed by Voss and Jansa (2009: 34–35), an interparietal is also unambiguously present in didelphids, all of which exhibit a large, unpaired element wedged between the right and left parietals anterior to the lambdoid crest (fig. 6B), in the same position as the sutured interparietals described above. The didelphid interparietal, however, is fused with the supraoccipital even in very young juveniles—including the youngest available postweaning juveniles (corresponding to age class 1 of Gardner [1973; see also Tyndale-Biscoe and MacKenzie, 1976; Tribe, 1990; van Nievelt and Smith, 2005a])—none of which show any trace of a suture (see also Koyabu et al., 2012: fig. S4A-D). Developmental studies of Didelphis (Nesslinger, 1956) and Monodelphis (Clark and Smith, 1993), however, demonstrate the presence of a distinct interparietal center of ossification that fuses to the supraorbital early in postpartum life (see char. 31). The didelphid interparietal does not, however, fuse with the parietals (as it does in Thylacinus and Strigocuscus), nor is it overgrown by them (as it is in some diprotodontians), the sutures between these bones persisting even in the largest adult specimens we examined (see also Koyabu et al., 2012: fig. 3A).

A large, wedge-shaped element that extends from the lambdoid crest between the left and right parietals and that is similar in size and position to the unequivocal interparietals discussed above is present in the only known skull of †Yalkaparidon (QM F13008; Beck et al., 2014). However, there is no clear suture separating this element from the supraoccipital in QM F13008, which is clearly a large adult (Beck et al., 2014). Nevertheless, given the large size of this element and its morphological resemblance to the undoubted interparietals of other metatherians in which these bones are present, we tentatively scored †Yalkaparidon as corresponding to state 1.

The interparietal is very small and polymorphic or absent (state 0) in all Recent dasyurids. Although at least some specimens of several dasyurids exhibit a minute sutured interparietal (e.g., Pseudantechinus [fig. 42]), we failed to identify a suturally distinct interparietal in dry skulls of comparably immature individuals of others (e.g., Myoictis), and it does not seem useful to distinguish these conditions (vestigial versus absent) in the context of this analysis. Frigo and Woolley (1996: figs. 2, 3) labeled a middorsal ossification center on the occiput of cleared-and-stained pouch young of Sminthopsis as the interparietal, but the absence of any other mineralized element between the foramen magnum and the parietals in their material strongly suggests that the structure in question is the supraoccipital. This supposition is supported by images of cleared and stained Sminthopsis pouch young provided to us by D. Koyabu (see also Koyabu et al., 2012), in which the large single middorsal ossification on the occiput is clearly the supraoccipital. The same images reveal the presence of several (probably three) much smaller ossifications between the supraoccipital and paired parietals, and we agree with Koyabu et al. (2012) that these are likely homologous with the interparietal. However, these ossifications are very small, and our failure to identify a distinct interparietal in dry skulls of immature Sminthopsis specimens (e.g., SAM M11536) suggests that they are variably present or absent; therefore, we scored Sminthopsis as state 0 for this study.

The interparietal is also very small or not identifiable as a suturally distinct ossification in Vombatus and Lasiorhinus. We were not able to ascertain the morphology of the interparietal in †Warendja based on specimens we examined. However, Brewer (2007: 164) described additional material of †Warendja, and stated that “there is no evidence of an interparietal process [sic] of the occipital,” implying that an interparietal was absent in this taxon as well; therefore, we score it as absent here. The interparietal of Dactylopsila and Dactylonax is apparently fused to the supraoccipital in even the youngest juvenile specimens that we examined, resembling didelphids in this respect (see char. 31).

Based on our observations of dry skulls, a distinct interparietal appears to be entirely absent in peramelemorphians, which show no trace of any bony element wedged between the parietals anterior to the lambdoid crest; instead, the median parietal suture extends all the way to the supraoccipital in juvenile specimens (e.g., Echymipera [fig. 6C]). Unfortunately, the neonatal specimens of Isoodon obesula and Perameles nasuta examined by Esdaile (1916) were too young to show any ossification centers behind the parietals, and Koyabu et al. (2012) were unable to find conclusive evidence for either the presence or absence of the interparietal in Perameles (D. Koyabu, personal commun.). Peramelemorphia may therefore represent an exception to Koyabu et al.'s (2012) general conclusion that the interparietal is present (at least embryologically) in all extant mammalian orders.

As discussed by Voss and Jansa (2009: 88), a specimen of †Herpetotherium (AMNH 127684) preserves the dorsal skull roof and reveals the presence of a median parietal suture that extends posteriorly as far as the supraoccipital, without any other intervening bony element. Thus, †Herpetotherium resembles peramelemorphians in lacking an obvious interparietal. Selva and Ladevèze (2016) identified a fragment of bone on the left side of a specimen of another herpetotheriid, †Peratherium cuvieri (MNHN GY679), as representing part of the interparietal, but this fragment appears to be located well lateral to the median parietal suture (Selva and Ladevèze, 2016: fig. 3), in which case it is more likely to be part of the squamosal.

We scored as missing (“?”) taxa in which all available specimens show partial or complete fusion of the median parietal suture. In such taxa, which include Recent caenolestids (e.g., Caenolestes [fig. 35]) and Notoryctes (fig. 36), it is impossible to distinguish between absence of the interparietal (as in peramelemorphians) versus fusion of the interparietal with the parietals (as in Thylacinus). In this context, we note some important discrepancies between our observations and published accounts of certain key fossil taxa. We examined multiple individuals of †Pucadelphys, including several subadults (MHNC 8265, 8384, 8388, 8391 [see Ladevèze et al., 2011]), but all of these specimens exhibit fusion of the sutures of the dorsal braincase. We therefore scored †Pucadelphys as unknown (“?”) contra Marshall and Muizon (1995: 50, fig. 12), who identified paired “postparietals” in this taxon. Although the interparietal of mammals apparently develops from four ossification centers—homologous with the paired postparietals and tabulars of plesiomorphic amniotes (Koyabu et al., 2012)—the interparietal is almost always a single ossification in the taxa we examined (exceptions include two specimens each of Myrmecobius [AM 3102, SAM M20184] and Vombatus [AM 7937, SAM M840]). Hence, the alleged presence of paired interparietals in †Pucadelphys (which our observations failed to confirm) would be highly unusual within Metatheria. Similarly, we could not identify the median parietal suture in the holotype (and only known) skull of †Mayulestes (MHNC 1249), which we also scored as unknown (contra Muizon 1998: fig. 6, which implies that an interparietal is absent in this taxon). Muizon et al. (2018: 399) inferred that †Allqokirus has an interparietal because “this bone is apparently universal for mammals (Koyabu et al., 2012)” (but note our comments regarding peramelemorphians and some dasyurids, above). However, fusion of the median parietal suture in the holotype and only known specimen of †Allqokirus (MHNC 8267) means that the interparietal cannot be identified as a suturally distinct ossification, so we also scored this taxon as unknown.

FIG. 6.

Posterior braincase of Dromiciops gliroides (A, UWBM 78641), Marmosa murina (B, AMNH 266418), and Echymipera kalubu (C, AMNH 190977). Alternative states for characters 30 and 31 (see main text for descriptions of these characters and character states) are illustrated as follows: Dromiciops 30(1), 31(0); Marmosa 30(1), 31(1); Echymipera 30(0), 31(-). Note that the incompletely mineralized lambdoid sesamoids (see Character 89) have fallen away in this juvenile specimen of Echymipera (the mature morphology is illustrated in fig. 11), fully exposing the parietal-supraoccipital suture. Abbreviations: exo, exoccipital; ip, interparietal; mas, mastoid; mf, mastoid fenestra; par, parietal; so, supraocciptal; sq, squamosal; ssf, subsquamosal foramen. Specimens are not drawn to the same scale.

img-z53-1_01.jpg

Muizon and Ladevèze (2020) described the cranial anatomy of another Tiupampan metatherian, †Andinodelphys cochabambensis (which has not been included here, but which appears to be a close relative of †Pucadelphys) and concluded that a large interparietal was probably present in this taxon. This inference was based on the presence of a thickened ridge on the left side of the dorsal skull roof of †A. cochabambensis (MHNC 8370) that Muizon and Ladevèze (2020) identified as a probable fused parietal-interparietal suture. If so, then the interparietal is anteriorly more extensive in †Andinodelphys than in any other metatherian we have examined (Muizon and Ladevèze, 2020: fig. 17A). However, this ridge is present on only one side of MHNC 8370, and it is not apparent on another specimen of †A. cochabambensis in which the dorsal skull roof is similarly well preserved (MHNC 8264), so we are not convinced that it is a fused parietal-interparietal suture; it may instead be a raised area of muscle attachment (temporal line) for the temporalis muscle. Frustratingly, it therefore remains unclear whether an interparietal was present or absent in any of the Tiupampan metatherians due to fusion of the sutures of the bones making up the posterior part of the dorsal skull roof in all available specimens that preserve this region.

Character 31. Interparietal separated from supraoccipital by an open suture, at least in juvenile specimens (0); or interparietal-supraoccipital suture fused in juveniles and adults alike (1). In most taxa in which an interparietal is identifiable, this bone is separated from the supraoccipital by an obvious suture in juveniles (state 0), although this suture sometimes fuses in adult specimens (e.g., Thylacinus). However, as discussed above (see char. 30), the large interparietal of didelphids is fused with the supraoccipital (state 1; e.g., in Marmosa [fig. 6B]) in even the smallest examined juvenile skulls. Although Voss and Jansa (2009: 39) stated that this morphology “appears to be unique among marsupials,” we additionally note its presence in Dactylopsila and Dactylonax: examination of very young juveniles of Dactylopsila (e.g., AMNH 154417, 101988) and Dactylonax (e.g., AMNH 109415, 221611) reveals no evidence of a suture separating the interparietal from the supraoccipital in these specimens. This suture is also fused in some juvenile specimens of Vombatus (e.g., MVZ 135036), which has been scored as polymorphic (“0+1”) here.

The suture separating the interparietal from the supraoccipital cannot be unequivocally identified in the only known cranium of †Yalkaparidon (QM F13008, the holotype of †Y. coheni). However, based on its large size compared with cranial fragments and isolated mandibles of other †Yalkaparidon specimens, QM F13008 appears to represent an old adult individual (Beck et al., 2014), and it is possible that the interparietal-supraoccipital suture fused relatively late in ontogeny, as occurs in Thylacinus; in the absence of cranial material of younger †Yalkaparidon individuals, we scored this taxon as unknown (“?”).

We scored this character as inapplicable (“-”) for peramelemorphians, in which a distinct interparietal appears to be absent (e.g., Echymipera; fig. 6C), and as unknown (“?”) in †Pucadelphys, †Allqokirus, †Mayulestes, caenolestids, Notoryctes, and other taxa in which the presence or absence of an interparietal cannot be determined with confidence (see char. 30).

Character 32. Parietal and mastoid in contact (0); or squamosal prevents parietal-mastoid contact (1). The parietal extends posteriorly to the lambdoid crest and contacts the mastoid exposure of the pars canalicularis of the petrosal in most didelphids (state 0; fig. 6B; Voss and Jansa, 2009: fig. 13A, B). In other didelphids, however, parietal-mastoid contact is prevented by contact between the squamosal and the interparietal (state 1; Voss and Jansa, 2009: fig. 13C, D). The parietal and mastoid are in contact in most other metatherians (e.g., Dromiciops [fig. 6A] and Echymipera [figs. 6C, 11]). However, parietal-mastoid contact is prevented in some taxa, either by contact between the squamosal and interparietal or between the squamosal and the supraoccipital; the latter morphology is seen in some taxa that entirely lack an interparietal, such as Lasiorhinus. Thus, we do not distinguish between the alternative ways that the squamosal prevents parietal-mastoid contact, but subsume them within state 1 and have modified the character description of Voss and Jansa (2009: char. 61) accordingly.

In Setonix, there is a large unossified vacuity dorsal to the mastoid exposure (the mastoid fenestra; see char. 91), so neither parietal-mastoid nor squamosal-supraoccipital contact is present; we therefore scored Setonix as inapplicable (“-”). Because parietal-mastoid contact is possible only when the mastoid is extensively exposed on the lateral occiput, we also scored this character as inapplicable for taxa in which the occipital exposure of the mastoid is restricted ventrally, namely phalangerids (fig. 12C; see char. 90). This character was scored as unknown (“?”) in acrobatids because the squamosal and mastoid of Acrobates and Distoechurus are seamlessly fused and cannot be distinguished as separate elements (see char. 88). Although this region is fused in adult specimens of Notoryctes, CT-scans of a fluid-preserved juvenile (NMV C11082) demonstrate that the parietal and mastoid are in contact in this taxon (Beck, in prep.).

Character 33. Extensive pneumatized endocranial sinuses within parietal absent (0); or present (1). Enlarged frontal sinuses are present in many marsupials (e.g., Spilocuscus [Helgen and Flannery, 2004], †Ekaltadeta [Wroe et al., 1998], and Thylacinus [Warburton et al., 2019: fig. 5]), but the parietal usually lacks extensive pneumatization (state 0). Although damaged and sectioned skulls reveal that the parietal is greatly thickened in Lasiorhinus and Vombatus (the “thickhead” condition sensu Murray, 1992), we did not score this condition as a distinct state. A qualitatively distinct morphology, however, is observable in damaged skulls of †Nimbadon (Black et al., 2010: fig. 11), †Neohelos (Murray et al., 2000a: fig. 38), †Silvabestius, and †Thylacoleo. In these taxa, extensive air-filled endocranial sinuses separate the inner table of the endocranial cavity from the outer table of the external skull surface throughout the parietal contribution to the dorsolateral braincase (the “airhead” condition sensu Murray, 1992), a morphology that we scored as state 1.

Palate

The bony palate separates the nasal cavity from the oral cavity, and it includes parts of only three bones—premaxillaries, maxillaries, and palatines. Despite this simplicity of function and composition, the palate exhibits remarkable taxonomic variation in several mammalian clades, notably Metatheria and Rodentia. Among metatherians, most noteworthy palatal variation involves foramina (which transmit nerves and blood vessels) and fenestrae. The latter term is used here for openings that are not known to transmit anything, as well as those that are much larger than the vessels or nerves that pass through them (thus resembling windows—fenestrae—rather than conduits). Unhappily, much terminological inconsistency is associated with metatherian palatal foramina and fenestrae; the nomenclature adopted in these accounts follows that of Voss and Jansa (2009) and is illustrated in figure 7; it is compared with synonymous usage in table 9.

FIG. 7.

Palatal morphology of Thylamys venustus (AMNH 261254) illustrating nomenclature for fenestrae, foramina, and other features described in the text (reproduced from Voss and Jansa, 2003: fig. 5; 2009: fig. 14). Dental loci (I1–M4) provide convenient landmarks for defining the size and position of palatal structures. Abbreviations: if, incisive foramen; m, maxillary fenestra; mp, maxillopalatine fenestra; p, palatine fenestra; plpf, posterolateral palatal foramen.

img-z55-5_01.jpg

Character 34. Paracanine fossa present (0); or absent (1). A distinct paracanine fossa occupies the diastema between the posteriormost upper incisor and C1, where it accommodates the apex of the lower canine (state 0) in fully adult specimens of †Pucadelphys (Marshall and Muizon, 1995: fig. 12; Muizon et al., 2018: fig. 4D), †Allqokirus (Muizon et al., 2018: fig. 4A), †Mayulestes (Muizon, 1998: fig. 6; Muizon et al., 2018: fig. 4B), most Recent didelphids (e.g., Marmosa [figs. 1, 2, 34]), †Thylatheridum,Thylophorops, most dasyuromorphians, and †Microbiotherium. A paracanine fossa is also present in †Sparassocynus, although it takes the form of an open notch rather than an enclosed fossa (Reig and Simpson, 1972). By contrast, a distinct paracanine fossa is absent (state 1) in Dromiciops, most peramelemorphians (†Galadi is a notable exception; Travouillon et al., 2010: fig. 3), a few dasyurids (Ningaui, Phascogale, and Antechinomys), Notoryctes, †Yalkaparidon, and all known paucituberculatans and diprotodontians.

TABLE 9

Selected Anatomical Synonyms for Structures of the Palate

img-Ayds_01.gif

Character 35. Accessory palatal fenestrae absent (0); or present (1). The bony palate of most metatherians is unperforated between the incisive foramina and the maxillopalatine fenestrae (state 0), but some peramelemorphians (e.g., †Yarala [Muirhead, 2000: fig. 1.1], †Bulungu [Gurovich et al., 2014: fig. 3], Chaeropus [fig. 37], Microperoryctes, Perameles [fig. 39], some specimens of Peroryctes) exhibit one or more pairs of nonvascular openings in the maxillary palate at the level of P1 and/or P2 (state 1). These appear to have first been named by Lyne and Mort (1981: 124), who called them “accessory vacuities.” They appear to lack obvious homologs in other examined taxa. By contrast, Osgood (1921: 107) implied that the long incisive foramina of Recent caenolestids (e.g., Caenolestes [fig. 35]) correspond to both the incisive foramina and the anteriormost palatal fenestrae of peramelemorphians. Although we have not dissected the soft tissues of any peramelemorphian palates, this equivalence seems implausible. Except in length, the well-formed incisive foramina of peramelemorphians and Recent caenolestids are structurally indistinguishable from each other and from the incisive foramina of didelphids, which are thought to transmit both the nasopalatine duct, blood vessels, and nerves (Wible, 2003: 175). By contrast, the accessory palatal fenestrae of peramelemorphians have the irregular aspect of nonvascular openings that result from osseous resorption; on incompletely cleaned crania they are covered by a thin membrane (presumably connective tissue) unperforated by ducts, and no trace of emerging vessels or nerves can be seen on surrounding bony margins (as in, e.g., AMNH 151937, a specimen of Peroryctes raffrayana). Pending histological examination, we conclude that these openings are analogous to the more posteriorly located palatal fenestrations long familiar to comparative anatomists and are not remnants (or precursors) of longer incisive foramina.

The incisive foramina of both Burramys (fig. 46) and Tarsipes (fig. 49) are also anteroposteriorly elongate (extending nearly as far back as P3), resembling caenolestids in this respect; in Burramys they are also markedly expanded posteriorly. As in caenolestids, the margins of these foramina are well formed, and there is no compelling evidence that any part of these openings is homologous with the accessory palatal fenestrae of peramelemorphians. Therefore, we score both Burramys and Tarsipes as state 0 for this character.

Character 36. Maxillopalatine fenestrae consistently absent (0); or present (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 38; 2009), who noted that maxillopalatine fenestrae are absent (state 0) only in Caluromys and Caluromysiops. Obvious fenestrae between the maxilla and palatine are present in most other metatherians (state 1), but they are absent in †Pucadelphys, †Allqokirus, †Mayulestes, Myrmecobius (fig. 41), diprotodontids (e.g., †Nimbadon [Black and Hand, 2010: figs. 1B, 2B]), Phascolarctos (fig. 44; see also char. 38), vombatids (e.g., Vombatus [fig. 45]), Dactylopsila, Dactylonax, Potorous (fig. 54), Osphranter, Macropus (Wells and Tedford, 1995: fig. 9E), some specimens of Notoryctes (e.g., AMNH 198651 [fig. 36]), and some specimens of Aepyprymnus. However, small foramina (probably homologous with the major palatine foramina of placental mammals; see Voss and Jansa, 2009: 36) are often present within the maxillopalatine suture of these taxa. Several authors (e.g., Reig and Simpson, 1972; Simpson, 1972; Engelman and Croft, 2014) have stated that †Sparassocynus lacks maxillopalatine fenestrae, but short, slotlike openings similar in size to the maxillopalatine fenestrae of some Monodelphis species (e.g., M. dimidiata; Pine et al., 1985: fig. 4; Chemisquy, 2015: fig. 1c) are present in at least one adult specimen (MMP 339-S) of †S. derivatus (see Beck and Taglioretti, 2020); thus, †Sparassocynus has been scored as state 1 for this character.

At least some palatal fenestrae are ontogenetically variable in didelphids, resulting from bone resorption in postweaning juveniles (Abdala et al., 2001), and a similar pattern has been observed in the dasyurids Dasyurus geoffroii (see Archer, 1984b: fig. 5) and D. albopunctatus (see Flores et al., 2006). Therefore, we scored this and subsequent fenestral characters from fully adult specimens only.

Character 37. Left and right maxillopalatine fenestrae separated by a median septum (0); or confluent, septum absent (1). Where present, the left and right maxillopalatine fenestrae are always separated by a median septum in didelphids (state 0; figs. 1, 7, 34). A dividing septum is present in most other metatherians that possess such vacuities, although the septum is very narrow in some taxa (e.g., Pseudochirops cupreus). However, this septum appears to be naturally incomplete or entirely absent, such that the maxillopalatine fenestrae are confluent (state 1), in Rhyncholestes (Patterson and Gallardo, 1987: fig. 1), some peramelemorphians (e.g., Macrotis [fig. 38]), some macropodiforms (e.g., Lagostrophus [fig. 53]), most Recent phalangerids (except Wyulda), acrobatids (e.g., Distoechurus [fig. 48]), †Thylacoleo, and Petropseudes. We scored this character based on adult specimens only, because a complete septum is sometimes present in juveniles or subadults but subsequently lost during ontogeny.

We noted a few discrepancies between our own observations and those of other authors. Wroe and Musser (2001: 493) stated that “the maxillary vacuities [= maxillopalatine fenestrae here] appear to have been large and conjoined” in †Barinya, but a complete median septum is clearly present in QM F31408 (Wroe, 1999: figs. 1.1, 2.1). Murray et al. (1987: 444) stated that the median septum that partially divides the maxillopalatine fenestrae of the only known skull of †Wakaleo vanderleuri (NTM P85553-4; CPC26604) is “naturally discontinuous,” but we do not consider this specimen to be sufficiently well preserved to score this character with confidence, so we scored †Wakaleo as unknown (“?”). Similarly, we scored this character as unknown for fossil taxa in which the palate of available specimens was damaged or incomplete.

This character was scored as inapplicable (“-”) for all taxa lacking maxillopalatine fenestrae (see char. 36).

Character 38. Palatine fenestrae absent (0); or present (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids (which exhibit either state 0 or state 1) were provided by Voss and Jansa (2003: char. 39; 2009). Palatine fenestrae are absent in most other metatherians, including †Pucadelphys, †Allqokirus, †Mayulestes, †Herpetotherium, †Sparassocynus, †Thylatheridium, Caenolestes (fig. 35), †Pichipilus, microbiotheriids, some peramelemorphians, Notoryctes (fig. 36), most dasyuromorphians, and most diprotodontians. However, distinct palatine fenestrae (vacuities fully enclosed by the palatines) are present in some peramelemorphians (Chaeropus [fig. 37], Macrotis [fig. 38], Perameles [fig. 39], some specimens of Isoodon [Lyne and Mort, 1981]), some dasyurids (e.g., Antechinomys [Archer, 1977], Dasycercus, Dasyuroides, Ningaui [Archer, 1975; Kitchener et al., 1983: fig. 2], Sminthopsis [Archer, 1981: fig. 60]) and some diprotodontians (e.g., Phascolarctos [fig. 44], Vombatus [fig. 45], Lasiorhinus, Hypsiprymnodon moschatus [fig. 52], Potorous [fig. 54], Hemibelideus, and some specimens of Cercartetus). Multiple tiny openings pierce the palatine of the pseudocheirid Pseudochirulus, but large vacuities are absent, so we scored this taxon as state 0. As for the preceding character, this character was scored from adult specimens only.

The palatal vacuities of phascolarctids warrants further discussion. The posterior region of the palate of †Litokoala is perforated by a single pair of fenestrae that are within the maxillary-palatine suture (Louys et al., 2009), and so are identified here as maxillopalatine fenestrae, following the nomenclature proposed by Voss and Jansa (2003, 2009; fig. 7). However, the single pair of palatal vacuities in Phascolarctos lie entirely within the palatine bone (fig. 44), and so represent palatine fenestrae following Voss and Jansa's (2003, 2009) nomenclature, and yet they are otherwise similar in relative position and size to the paired openings in †Litokoala (compare figs. 1A and 4 of Louys et al., 2009). Therefore, it is possible that the palatine fenestrae of Phascolarctos are, in fact, homologous with the maxillopalatine fenestrae of †Litokoala—differing only with respect to their position relative to the maxillopalatine suture—and not homologous with the palatine fenestrae of taxa in which both maxillopalatine and palatine openings are present (such as Thylamys [fig. 7], Chaeropus [fig. 37], and Perameles [fig. 39]). The same logic may apply to other taxa in which maxillopalatine fenestrae are absent but palatine fenestrae are present, such as Vombatus (fig. 45) and Potorous (fig. 54). Despite such perplexities, we follow Voss and Jansa's (2003, 2009) operational nomenclature to maximize consistency and repeatability of scoring for this study.

We have scored this character as inapplicable (“-”) in taxa where maxillopalatine fenestrae are present and extend posteriorly as far as the postpalatine torus—as in, for example, Rhyncholestes (Patterson and Gallardo, 1987: fig. 1), phalangerids (e.g., Phalanger [fig. 47]), and some macropodiforms (e.g., Lagostrophus [fig. 53]). In such cases, it is unclear whether or not these very large openings represent fused maxillopalatine and palatine fenestrae; indeed, young juveniles of Phalanger (AMNH 198728) and Spilocuscus (AMNH 157173) have separate maxillopalatine and palatine fenestrae, whereas these openings are confluent in all older specimens.

Character 39. Maxillary fenestrae absent (0); or present (1). As discussed by Voss and Jansa 2003, 2009), some Recent didelphids have an additional pair of nonvascular openings that are located entirely within the maxillary bone lateral to the maxillopalatine fenestrae, approximately level with M1 and/or M2 (state 1; fig.7). No nondidelphid metatherian that we examined has similar fenestrae, with the exception of some specimens of Chaeropus (e.g., AM 422, BMNH 1848.1.27.40, and BMNH 1848.1.27.41; see Travouillon et al., 2019: figs. 3c, 6c, j, k), which we scored as polymorphic (“0+1”).6 In Tarsipes an unossified space is present on each side between the anterior root of the zygomatic arch and the maxillopalatine fenestra, but close examination suggests that this opening represents an enlarged and highly modified maxillary foramen (i.e., the posterior opening of the infraorbital canal). However, the general palatal morphology of Tarsipes is difficult to decipher owing to weak ossification, so we scored this taxon as unknown (“?”). As for the preceding character, we scored this character based on adult specimens only.

Character 40. Posterolateral palatal foramina present, with complete posterior margins (0); or absent, posterior margins incomplete (1). Most metatherians have a foramen that transmits the minor palatine artery (Archer, 1976), and presumably also the minor palatine nerve (Evans, 1993; Wible, 2003), through the maxillo-palatine suture near the posterolateral corner of the palate (state 0; fig. 7). However, the posterolateral palatal margins are only notched (not perforated) by this artery (state 1) in †Allqokirus (Muizon et al., 2018), †Sparassocynus, †Nimbacinus, †Barinya, and most Recent dasyurids (Wroe, 1997b; Wroe et al., 2000) except Neophascogale (in which the posterolateral palatal foramen is consistently complete) and Myoictis (which is polymorphic [“0+1”] for this character). Thylacinus is also polymorphic for this character (with some individuals exhibiting different states on the left and right sides; e.g., SAM M922), as are Monodelphis (Voss and Jansa, 2009: fig. 42) and †Thylatheridium. Murray and Megirian (2006a: 252) argued that the posterolateral palatal foramen was incomplete in †Mutpuracinus, but the posterior palate is poorly preserved in the only known skull of this taxon (NTM P91168-5), which we scored as unknown (“?”) for this character.

Character 41. Posterolateral palatal foramina (or equivalent notches) small, posterior to M4 on each side (0); or very large, extending lingual to M4 protocones (1). We scored this character regardless of whether a complete posterolateral palatal foramen or an incomplete notch is present (see char. 40). Scoring criteria and information about the distribution of this character among Recent didelphids were provided by Voss and Jansa (2003: char. 41; 2009), who reported that very large posterolateral palatal foramina extend lingual to the protocone of M4 only in Lestodelphys and Thylamys. Of the nondidelphid taxa we examined in this study, only †Herpetotherium (based on AMNH 22304) has similarly enlarged posterolateral palatal foramina (see also Horovitz et al., 2008: plate 2H). The posterolateral palatal foramen opens level with the M4 protocone in †Ngapakaldia (Stirton, 1967: fig. 3), but it is very small, so we scored this taxon as exhibiting state 0. Although Acrobates, Distoechurus (fig. 48), and †Wakaleo all lack M4, we scored these taxa as state 0 because the posterolateral palatal foramen is small and located well posterior to M3. However, the absence of M2–4 in †Thylacoleo precludes meaningful scoring of this character, which is clearly inapplicable (“-”) for that taxon.

Character 42. Posterior palatal margin is anterior to presphenoid-basisphenoid suture (0); or palate extends posterior to presphenoid-basisphenoid suture (1). The bony palate terminates anterior to the presphenoid-basisphenoid suture (state 0) in most metatherians examined here (e.g., Caluromys; fig. 8A), but in Myrmecobius the palate is very long and extends posteriorly well behind the presphenoid-basisphenoid suture (state 1; fig. 41).

Nasopharyngeal and Postpalatal Regions

The nasopharyngeal and postpalatal regions (including the internal nares and the interpterygoid fossa) have been peculiarly neglected by systematists (but see Wible et al., 2018: for a recent exception). Five bones—the pterygoids, palatines, vomer, presphenoid, and basisphenoid—are closely juxtaposed in this part of the skull, and their alternative patterns of contact provide several characters of interest for metatherian researchers. A sixth bone, the parasphenoid, can also occur in this region, but in most metatherians it is a minor ossification that is variably present or absent and has been overlooked by most researchers (Wible et al., 2018); an apparent exception is the condition in Dromiciops, with Wible et al. (2018) reporting that in this taxon the parasphenoid is present and fused to the basisphenoid, where it forms part of an unusual midline crest that is continued anteriorly by what we identify as the pterygoids and vomer (char. 48), but which Wible et al. (2018: fig. 13, 128–129) identified as the presphenoid and vomer.

Immediately adjacent to the interpterygoid (or mesopterygoid) fossa are other postpalatal structures—sites of muscle attachment and vascular foramina—that we also treat in this section. We note in passing that, although the basisphenoid and the alisphenoid are invariably discussed as though they were separate bones, these elements are indistinguishably fused in all metatherians, including the youngest postweaning juveniles we examined (in Monodelphis domestica, they are fused by postnatal day 25; Wible, 2003: 151). Ontogenetic evidence provides an indication of the approximate boundary between basisphenoid and the alisphenoid, as the transverse canal foramen and carotid canal develop within the basisphenoid, whereas the primary foramen ovale and foramen rotundum develop within the alisphenoid (Maier, 1987; Clark and Smith, 1993; Wible, 2003: 151–153; Sánchez-Villagra and Forasiepi, 2017).

FIG. 8.

Postpalatal region of Caluromys philander (A, AMNH 267002), Dromiciops gliroides (B, UWBM 78641), Dactylonax palpator (C, AMNH 191042), and Potorous tridactylus (D, AMNH 66168). Alternative states for characters 43 and 45–50 (see main text for descriptions of these characters and character states) are illustrated as follows: Caluromys 43(0), 45(0), 46(0), 47(1; but note that C. philander is polymorphic for this character), 48(0), 49(0), 50(0); Dromiciops 43(1), 45(1), 46(3), 47(0), 48(1), 49(0), 50(0); Dactylonax 43(1), 45(1), 46(2), 47(1), 48(0), 49(0), 50(1); Potorous 43(-), 45(2), 46(1; but note that the modal condition for P. tridactylus is 0), 47(1), 48(0), 49(1), 50(0). Abbreviations: als, alisphenoid (purple, coossified with basisphenoid); apf, anterior pterygoid foramen; atp, alisphenoid tympanic process; bk, basisphenoid keel; bs, basisphenoid (purple, coossified with alisphenoid); cc, carotid canals; ecr, ectopterygoid crest; fo, foramen ovale; pal, palatine (green); plfp, posterolateral palatal foramen; ps, presphenoid (blue); pt, pterygoid (yellow); ptf, pterygoid fossa; tcf, transverse canal foramen; v, vomer (orange). Specimens are not drawn to the same scale.

img-z61-1_01.jpg

Two cases of synonymous usage for structures in this region are worth noting. The carotid canal (= canalis caroticus [NAV]) has been referred to as the “anterior entocarotid foramen” (e.g., Black and Hand, 2010), “carotid foramen” (e.g., Dederer, 1909), “entocarotid canal” (e.g., Owen, 1859), “entocarotid foramen” (e.g., Coues, 1872), “foramen carotidus” (e.g., Reig et al., 1987), “internal carotid foramen” (e.g., Murray, 1986), and “foramen lacerum medium” (e.g., Marshall 1976), although the last of these terms is usually applied to the foramen for the greater petrosal nerve, or to the confluence of the latter foramen and the primary foramen ovale (see Marshall and Muizon, 1995; Muizon, 1998, 1999; Wroe et al., 1998, 2000: char. 69; Forasiepi, 2009: 37). The transverse canal foramen is sometimes known as the “transverse canal” (e.g., Dederer, 1909), “transverse foramen” (e.g., Archer, 1976b), or the “foramen for transverse sinus canal” (e.g., Aplin et al., 2010).

Character 43. Presphenoid exposed in roof of nasopharyngeal fossa above posterior palate (0); or presphenoid concealed by vomer above posterior palate (1). In most metatherians, the nasopharyngeal fossa above the posterior palate is at least partially roofed by unpaired median bones. The presphenoid is typically the exposed median element in didelphids (e.g., Marmosa [figs. 1, 34], Caluromys [fig. 8A], Monodelphis [Wible, 2003: fig. 5]), thylacinids, burramyids, thylacoleonids, some macropodiforms (e.g., Caloprymnus, Dendrolagus, Dorcopsulus, Lagostrophus, Notamacropus, Setonix, Thylogale), Petaurus, Gymnobelideus, and Acrobates. In these taxa, the vomer (if visible) is represented by flanking lateral processes, one on either side of the presphenoid.

By contrast, in caenolestids, many diprotodontians (e.g., Dactylonax [fig. 8C], Phascolarctos, Tarsipes, †Nimbadon, Vombatus, Lasiorhinus, most pseudocheirids, some macropodiforms), most peramelemorphians, and most dasyurids, the presphenoid is typically concealed by the vomer, which extends posteriorly as an undivided element at least as far as the posterior palatal margin (state 1). We consider this morphology to be also present in Dromiciops (fig. 8B), based on examination of multiple specimens (e.g., FMNH 22675, 127445, 127453, 127465 and 129804; UWBM 78631, 78633 and 78640), all of which appear to us to demonstrate that the vomer conceals the presphenoid, so we have scored this taxon as state 1. However, Wible et al. (2018: fig. 13, 128–129) instead identified the presphenoid as an exposed element contributing to the midline crest of Dromiciops. These discrepant observations warrant further investigation. In Wallabia (e.g., AMNH 100112) and Macropus (e.g., AMNH 66173), the vomer extends beyond the level of the posterior palatal margin as an undivided element, but it is transversely narrow and spine-like (see also Wible et al., 2018: 128, fig. 15), such that the presphenoid remains clearly visible on either side; we also scored these taxa as exhibiting state 1. Several taxa are polymorphic (“0+1”) for this character, including Neophascogale, Chaeropus, Microperoryctes, Notoryctes, Distoechurus, Lagorchestes, Onychogalea, Osphranter, Petrogale and Aepyprymnus.

We scored this character as inapplicable (“-”) for taxa in which the palatines and/or the pterygoids are in midline contact, completely concealing the presphenoid above the posterior palate (e.g., Potorous [fig. 8D]; see chars. 45, 46). Myrmecobius has also been scored inapplicable (“-”), because the posterior margin of the palate extends posteriorly beyond the presphenoid-basisphenoid suture in this taxon (char. 42).

Character 44. Right and left choanae confluent posteriorly (0); or partially or completely divided by a bony septum (1). The internal choanae are confluent in most metatherians (state 0), but a bony septum formed by the palatines and the vomer partially or completely separates the left and right choanae (state 1) in several diprotodontians, including Vombatus, Lasiorhinus, †Nimbadon, Phascolarctos, Dendrolagus, and †Bohra. A few taxa—Aepyprymnus, Dactylonax, and Notoryctes—are polymorphic (“0+1”) for this character.

Character 45. Palatines in lateral contact with presphenoid (0); or separated from presphenoid by vomerine-pterygoid contact (1); or right and left palatines in midline contact ventral to presphenoid (2). Several different patterns of contact among the palatines, vomer, and pterygoids are present in the roof of the nasopharyngeal and interpterygoid (mesopterygoid) fossae. In some taxa (e.g., Acrobates, Tarsipes, burramyids, most macropodids, thylacoleonids, thylacinids, Myrmecobius, most caenolestids, and most didelphids [e.g., Marmosa [fig. 1]; Caluromys [fig. 8A]; Monodelphis [Wible, 2003: fig. 5]), the left and right palatines are in lateral contact with, and at least partially separated by, the presphenoid (state 0). In other taxa, the palatines are separated by vomerine-pterygoid contact (state 1) as in, for example petaurids (e.g., Dactylonax [fig. 8C]), pseudocheirids, phalangerids, most dasyurids, and most peramelemorphians. We also consider the vomer and pterygoids to be in contact in Dromiciops (fig. 8B), although we note that Wible et al. (2018: fig. 13, 128–129) interpreted the morphology of this region differently.

A third, qualitatively distinct morphology is seen in Bettongia, Caloprymnus, Potorous (fig. 8D), Hypsiprymnodon moschatus and H.bartholomaii, in which the left and right palatines are in midline contact (state 2). This distinctive morphology was remarked by Archer (1984c: fig. 216) as supporting the monophyly of Potoroidae sensu lato (i.e., including Hypsiprymnodon). However, we note (as did Ride, 1956) that it is absent in Aepyprymnus, and it is also present in the probable stem-macropodid †Ganguroo (Travouillon et al., 2014b: fig. 1).

The pterygoids are commonly lost or broken in skulls of fossil and extant taxa alike. When the pterygoids are missing or damaged, the presence of ridges on adjacent bones can sometimes give an indication as to their extent. However, we scored this and the following character only for those taxa in which at least one pterygoid bone appeared to be fully intact.

Character 46. Pterygoids in lateral contact with presphenoid (0); or separated from presphenoid by palatine-basisphenoid contact (1); or separated from presphenoid by vomerine-basisphenoid contact (2); or left and right pterygoids in midline contact ventral to presphenoid (3). In most metatherians—including most didelphids (e.g., Marmosa [fig. 1]; Caluromys [fig. 8A]; Monodelphis [Wible, 2003: fig. 5]), caenolestids, most dasyurids, most peramelemorphians, Petaurus, Gymnobelideus, and most macropodiforms—the pterygoids are in lateral contact with, and separated by, the presphenoid. However, we observed several alternative conditions among the taxa scored for this study.

In Acrobates, Burramys, several macropodiforms, Lasiorhinus, †Wakaleo, Myrmecobius, and Chironectes (Voss and Jansa, 2009: fig. 45), for example, pterygoid-presphenoid contact is prevented by palatine-basisphenoid contact (state 1; note that this morphology is illustrated by a specimen of Potorous in figure 8D, although the modal condition for this terminal is, in fact, pterygoid-presphenoid contact [state 0]). By contrast, in Pseudocheirus, Dactylonax (fig. 8C), and some specimens of Pseudochirops, Pseudochirulus, Ailurops, and various dasyurids (e.g., Dasycercus, Dasyurus, Murexia, Phascogale, and Sminthopsis), the vomer and basisphenoid are in contact (state 2). In yet other taxa, the right and left pterygoids contact one another under the presphenoid, which they conceal from ventral view (state 3); this is the condition seen in †Pucadelphys, Macrotis, Isoodon, †Yarala (Muirhead, 2000: fig. 1), and various diprotodontians (e.g., †Nimbadon, †Namilamadeta, Dactylopsila, most phalangerids, and some pseudocheirids). We also interpret the left and right pterygoids as contacting one another in Dromiciops (fig. 8B) based on several specimens that preserve a midline suture between these elements (e.g., FMNH 22675 and 127445; UWBM 78633 and 78640), as well as one in which the suture is located to the left of the midline (UWBM 78644); however, Wible et al. (2018: fig. 13) interpreted the pterygoids as separated by the presphenoid in this taxon.

Based on our observations, the pterygoids contribute to a sagittal keel in Dromiciops (fig. 8B) and certain other taxa (including †Microbiotherium), but pterygoid keels range from distinct to indistinct; therefore, we did not score such variation for phylogenetic analysis. The pterygoid keel in Dromiciops and †Microbiotherium is part of a continuous carina that is formed by the vomer anteriorly and by the basisphenoid posteriorly (see char. 48). Although Hershkovitz (1999), Giannini et al. (2004), and Wible et al. (2018) all stated that the postpalatal carina of Dromiciops is formed by the presphenoid and the basisphenoid, the presphenoid is not ventrally exposed in any of the specimens we examined (e.g., FMNH 22675, 127445, 127453, 127465, and 129804; UWBM 78631, 78633, and 78640). We have not, however, examined the specimen described and illustrated by Wible et al. (2018) and acknowledge that our discrepant interpretations warrant further investigation.

As for the previous character, we scored this character only for taxa preserving at least one fully intact pterygoid.

Character 47. Pterygoids short, not extending posteriorly to the carotid canal on each side (0); or long, extending posteriorly to sheath the ventral margin of the carotid canal (1); or very long, extending beyond the carotid canal to contact the petrosal (2). In most metatherians, the pterygoids do not extend posteriorly to the external opening of the carotid canal (state 0), which is entirely enclosed by the basisphenoid (Wible, 2003) (e.g., in Dromiciops [fig. 8B]). However, in nearly all diprotodontians (with the exception of Petropseudes, some specimens of Pseudochirops cupreus, and some specimens of †Thylacoleo) the pterygoids contribute to the ventral border of the external opening of the carotid canal (state 1; e.g., in Dactylonax [fig. 8C] and Potorous [fig. 8D]; note that the left pterygoid is missing from the specimen of Petaurus illustrated in fig. 9C). This morphology is uncommon among nondiprotodontian metatherians, but we also observed it in Caluromysiops, Myrmecobius (fig. 41), and in all examined peramelemorphians with intact pterygoids (e.g., Perameles [fig. 39]; unfortunately, †Yarala and †Galadi could not be scored for this character). It is also present in some, but not all, specimens of Caluromys (including the specimen illustrated in fig. 8A), which we scored as polymorphic (“0+1”). In Onychogalea, the pterygoid extends posteriorly beyond the external opening of the carotid canal to contact the petrosal, a distinctive morphology that we did not observe in any other examined taxon (state 2).

As defined, these states represent a plausible morphocline, reflecting progressive posterior extension of the pterygoid, so we ordered this character in all our analyses. As for characters 45 and 46, we scored this character only for taxa represented by specimens with at least one intact pterygoid bone.

Character 48. Basisphenoid unkeeled (0); or with a distinct sagittal keel (1). In most metatherians, the basisphenoid forms a relatively flat plate in ventral view (state 0). However, the basisphenoid of Dromiciops (fig. 8B) and †Microbiotherium (based on MACN A 8505) develops a distinct midline keel (state 1), which we identify as being continued anteriorly by the pterygoids (see char. 46). According to Wible et al. (2018), the posterior part of the keel in Dromiciops is formed by a coossified parasphenoid (which in other metatherians is present only as a very minor element, if at all) fused to the basisphenoid; if so, then the same presumably applies to †Microbiotherium.

Character 49. Fossa of origin for internal pterygoid muscle absent or small and indistinct (0); or pterygoid fossa large, deeply excavated (1). In most marsupials, the origin of the internal pterygoid muscle (dissected inter alia by Coues, 1872; Barbour, 1963; Hiiemae and Jenkins, 1969; Turnbull, 1970; Filan, 1990) is not marked by conspicuous osteological features (state 0). However, in Burramys and in all examined macropodiforms (e.g., Potorous [fig. 8D]) this muscle originates from a distinct fossa (state 1) that is deeply excavated anteriorly and bounded anterolaterally by a well-developed ectopterygoid crest of the alisphenoid (Winge, 1941: 92; Aplin, 1990; Tomo et al., 2007; Warburton, 2009). Curiously, as far as we are aware this distinctive feature has not previously been scored for phylogenetic analysis, and it is not included in any recent list of macropodiform synapomorphies that we have encountered. A well-excavated pterygoid fossa bound laterally by an ectopterygoid crest is also present in the fossil phascolarctids †Litokoala and †Nimiokoala (but not Phascolarctos; Louys et al., 2009), the wynyardiids †Muramura and †Namilamadeta, and the vombatid Lasiorhinus, all of which have been scored as state 1; however, in these taxa (and also Burramys) the ectopterygoid crest does not extend as far ventrally, nor is it as posteriorly extensive, as it is in macropodiforms. Although the ectopterygoid crest is well developed in Dactylopsila and Dactylonax (fig. 8C), the pterygoid fossa itself is not deeply excavated, so these two taxa have been scored as state 0 for this character.

Character 50. Anterior pterygoid foramen absent (0); or present (1). The anterior surface of the pterygoid fossa of Dactylopsila is perforated by a large venous foramen that communicates with the foramen rotundum within the alisphenoid bone (state 1). Aplin (1990: 327) suggested that this foramen may transmit “an aberrant venous connection, perhaps linking the orbital sinus and the pterygoid plexus.” We observed an apparently identical perforation in Dactylonax (fig. 8C) and Gymnobelideus, both of which have been scored as state 1, but it is absent in Petaurus. The anterior pterygoid region of other metatherians is usually imperforate, but is sometimes pierced by tiny foramina that are highly variable in both position and number. These minute foramina differ qualitatively from the large and conspicuous anterior pterygoid foramen seen in Dactylopsila, Dactylonax, and Gymnobelideus, so we scored taxa in which they are sometimes present (such as many macropodiforms) as corresponding to state 0.

Character 51. Transverse canal foramen absent (0); or present (1). When present, the transverse canal foramen (figs. 1, 8B, 8C) “transmits a vein communicating with the cavernous sinus and that [sometimes] communicates across the midline with its antimere” (Wible, 2003: 184; see also Wortman, 1902: 440; Gregory, 1910: 223; Archer, 1976b; Aplin, 1990; Sánchez-Villagra, 1998; Sánchez-Villagra and Wible, 2002). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 44; 2009), whereas Sánchez-Villagra and Wible (2002) provided information (partly based on Aplin, 1990) about the presence or absence of the transverse canal foramen in both Recent didelphids and nondidelphid metatherians.

Among the taxa scored for this study, an identifiable transverse canal foramen is absent (state 0) in †Allqokirus (Muizon et al., 2018), †Mayulestes (Muizon, 1998), Caluromys (fig. 8A), Caluromysiops, Planigale (see below), †Lekanoleo, and Tarsipes. The transverse canal foramen has been reported as absent in †Pucadelphys (e.g., by Marshall and Muizon, 1995; Sánchez-Villagra and Wible, 2002), but examination of a large number of recently collected skulls reveals that the transverse canal foramen is clearly present in some specimens (e.g., MHNC 8377; MHNC 8380; Ladevèze et al., 2011), so we scored this taxon as polymorphic (“0+1”). Similarly, a transverse canal foramen is absent in some, but not all, specimens of Myoictis, Gymnobelideus, and Pseudochirops archeri—all of which we also scored as polymorphic. Rougier et al. (1998) coded the transverse canal foramen as absent in Dromiciops, whereas Sánchez-Villagra and Wible (2002: 30) identified this foramen in all but one of the Dromiciops gliroides specimens that they examined. We found the transverse canal foramen to be consistently present in the D. gliroides specimens we examined (e.g., UWBM 78641 [fig. 8B]), so we scored this taxon as corresponding to state 1.

The condition in Dactyopsila and Dactylonax warrants further discussion. Sánchez-Villagra and Wible (2002: 30) implied that the anterior pterygoid foramen present in Dactylopsila (see char. 50) is homologous with the transverse canal foramen, whereas Aplin (1990: 326) stated that Dactylopsila trivirgata has a “very reduced transverse canal foramen” distinct from, and in addition to, the anterior pterygoid foramen. Based on our own observations, a transverse canal foramen that is clearly distinct from the anterior pterygoid foramen appears to be present in some, but not all, adult specimens of D. trivirgata, although it often occurs on only one side of the skull (e.g., in AMNH 157142, 159480, 198710). Therefore, we scored Dactylopsila as polymorphic (“0+1”) for this character. Where present, this foramen is posterior to the pterygoid fossa in Dactylopsila, in a similar position to that seen in Petaurus. In Dactylonax, by contrast, the transverse canal foramen appears to be consistently present but shares a common opening with the anterior pterygoid foramen (fig. 8C).

Although Planigale maculata exhibits a transverse canal foramen (Archer, 1976b: fig. 3A, 3B), we could not identify this foramen in examined specimens of P. ingrami (our generic exemplar). Archer (1976b: 307) stated that “in some Planigale…the venous sinus passes transversely via the anterior edges of the enormous foramina pseudovale [= primary foramen ovale; see table 10 and char. 52] which are so large that they incise the alisphenoid in the normal position of the transverse canal.” Therefore, it is possible that a transverse canal is present in P. ingrami but is externally confluent with the foramen ovale. Despite this, and in the absence of serially sectioned specimens or vascular casts of P. ingrami, we scored Planigale as corresponding to state 0.

Where present, the transverse canal foramen is usually enclosed by the basisphenoid in metatherians (e.g., Marmosa [fig. 1]; Dromiciops [fig. 8B]; Dactylonax [fig. 8C]; Monodelphis [Wible, 2003: 151]). In acrobatids, however, a deep sulcus that houses the transverse canal vein extends laterally from the carotid canal and is floored ventrally either by the posterior part of the pterygoid (in Distoechurus) or a strut of the alisphenoid (in Acrobates; see Aplin, 1990: 326; Sánchez-Villagra and Wible, 2002: 30).

In some metatherians the connection between the left and right transverse canal foramina is roofed dorsally by a continuous sheet of bone that separates the transverse canal from the endocranium, whereas in others the transverse canal foramen opens directly into the endocranial cavity (Sánchez-Villagra and Wible, 2002 [char. 2]; see also Horovitz and Sánchez-Villagra, 2003 [char. 197], and subsequent studies that used modified versions of their matrix). However, we observed an apparently continuous range of intermediate morphologies (sometimes within a single taxon), in which a partial dorsal roof is present but fenestrae of varying sizes open within it. Thus, we were unable to consistently score presence or absence of an intramural transverse canal, and so have not included this character in our analyses.

Ear Region

No part of the skull has contributed more characters to marsupial phylogenetic studies than the ear region. Although little is known with any certainty about the adaptive significance of relevant character transformations, it is reasonable to suppose that auditory specializations are largely responsible for most observed taxonomic differences. However, closely juxtaposed structures—including the jaw joint and various processes for muscle attachment—participate in other (nonauditory) functions and must resist large forces. These mechanically important features, together with several nerves and blood vessels that are also crowded into this part of the skull, may constrain the evolution of auditory phenotypes and account for some of the apparent homoplasy noted by various authors. This region includes four small bones in their entirety (ectotympanic, malleus, incus, and stapes) as well as parts of another four (alisphenoid, exoccipital, petrosal, and squamosal) that help form auditory sinuses, contain the internal organs of hearing and balance, or impinge so closely on auditory structures that they are more conveniently discussed here than elsewhere. Many anatomical features of the ear region are known by multiple synonyms; a nonexhaustive list of these is given in table 10.

Character 52. Extracranial course of mandibular nerve not fully enclosed by bone (0); or extracranial course of mandibular nerve fully enclosed by medial outgrowths of the ossified hypotympanic sinus floor (1) or extracranial course of mandibular nerve traverses a bony canal in the roof of the hypotympanic sinus (2); or extracranial course of mandibular nerve traverses a bony canal in the medial wall of the hypotympanic sinus (3). Following Gaudin et al. (1996) and Voss and Jansa (2003, 2009; see also Wroe, 1997b; Wroe et al., 1998; Muirhead, 2000), we use the general term foramen ovale to refer to all bony enclosures of the mandibular division of the trigeminal nerve, regardless of which bones participate. By contrast, MacIntyre (1967) proposed the term “foramen pseudovale” for cases in which the primary orifice by which the mandibular nerve exits the skull is not fully enclosed by the alisphenoid, and his usage was followed, inter alia, by Archer (1976b; 1981) and Marshall (1977; 1979). We refer to the ventrally directed foramen by which the mandibular nerve exits the braincase as the primary foramen ovale, following most recent works on metatherian cranial anatomy (e.g., Wroe, 1997b, 1999; Wroe et al., 1998; Muirhead, 2000; Voss and Jansa, 2003, 2009; Murray and Megirian, 2006a; Travouillon et al., 2010, 2014b, 2015b; Black et al., 2014c; Gurovich et al., 2014; Butler et al., 2016); however, Gaudin et al. (1996) used the term “true foramen ovale” for this orifice.

TABLE 10

Selected Anatomical Synonyms for Structures of the Ear Regiona

img-AgM1_01.gif

continued

img-z68-2_01.gif

FIG. 9.

Ventral view of left ear region of Marmosa murina (A, AMNH 267368), Phascogale tapoatafa (B, AMNH 160267), and Petaurus breviceps (C, AMNH 154468; note that the pterygoid bone, which when intact extends posteriorly as far as the carotid canal in P. breviceps [char. 47], is missing in this specimen). Alternative states for characters 23, 55, 60, 68, 79, 84, and 87 (see main text for descriptions of these characters and character states) are illustrated as follows: Marmosa 23(1), 55(1), 60(0), 68(0), 79(0), 84(0), 87(0); Phascogale 23(1), 55(2), 60(0), 68(2), 79(3), 84(1), 87(0); Petaurus 23(0), 55(3), 60(2), 68(1[not visible on intact skulls]), 79(4), 84(1[not visible on intact skulls]), 87(1; not visible on intact skulls). Abbreviations: als, alisphenoid; atp, alisphenoid tympanic process; bo, basioccipital; bs, basisphenoid; cc, carotid canal; cf, condyloid foramen; ctp, caudal tympanic process (of pars canalicularis of petrosal); ect, ectotympanic; fgpn, foramen for greater petrosal nerve; fips, foramen for inferior petrosal sinus; gf, glenoid fossa; gpa, glenoid process of alisphenoid; hf, hypoglossal foramen; jf, jugular foramen; jug, jugal; mas, mastoid exposure of pars canalicularis of petrosal); mf, mastoid fenestra; oc, occipital condyle; pes, posterior epitympanic sinus (of squamosal); pfo, primary foramen ovale; pgf, postglenoid foramen; pogp, postglenoid process (of squamosal); pp, paroccipital process (of exoccipital); pro, promontorium (of pars cochlearis of petrosal); pt, pterygoid; ptps, postympanic process (of squamosal); rtp, rostral tympanic process (of pars cochlearis of petrosal); smf, stylomastoid foramen; smn, stylomastoid notch; sq, squamosal; ssf, subsquamosal foramen; tcf, transverse canal foramen. Specimens are not drawn to the same scale.

img-z69-1_01.jpg

In most metatherians in which the primary foramen ovale is clearly externally identifiable, this opening appears to be between the petrosal and the alisphenoid (e.g., in the Recent didelphids Marmosa [fig. 9A], Marmosops [Voss and Jansa, 2003: fig. 8B], and Monodelphis [Voss and Jansa, 2003: fig. 8c; Wible, 2003: fig. 6]), but it is between the petrosal, alisphenoid, and squamosal in Vombatus and Lasiorhinus. In all these taxa, the endocranial exit of the greater petrosal nerve (the “foramen lacerum medium” sensu Marshall and Muizon, 1995; Muizon, 1998; 1999; see Forasiepi, 2009: 37) is confluent with the primary foramen ovale, although the two openings are sometimes partially separated by a postero-laterally directed process of the alisphenoid (fig. 9A; see also right side [anatomical left side] of fig. 6 of Gaudin et al., 1996; Wible, 2003: fig. 6). Forasiepi (2009: 37) proposed the term “spheno-petrosal fissure” for the shared opening by which V3 and the greater petrosal nerve exit the endro-cranium (following Maier, 1987), whereas Gabbert (1998) referred to it as the “piriform fenestra” (see also Murray, 1991; Forasiepi et al., 2019; Wible et al., 2021). In a few metatherians, however, the primary foramen ovale is entirely enclosed by alisphenoid, with the greater petrosal nerve exiting via a separate foramen (see left side [anatomical right side] of fig. 6 of Gaudin et al., 1996). As noted by Voss and Jansa (2003: 28–30, 2009: 39), the location of the primary foramen ovale—between the petrosal and alisphenoid; between the petrosal, alisphenoid, and squamosal; or fully enclosed by the alisphenoid—is often ambiguous or difficult to interpret by external inspection. This difficulty may explain some discrepancies between our observations and those reported by previous authors, which we briefly discuss here.

The primary foramen ovale of Dromiciops appears to be between the alisphenoid and petrosal (based on FMNH 129804 and UWBM 78644), and not entirely within the alisphenoid (as stated by Wroe, 1997: 27), nor is it between the alisphenoid and squamosal (as stated by Rougier et al., 1998: char. 111; Wible et al., 2001: char. 111; and Ladevèze, 2004: char. 30). Marshall and Muizon (1995) and Muizon et al. (1997: fig. 2A) suggested that the squamosal contributes to the posterolateral rim of the primary foramen ovale in †Pucadelphys (see also Muizon et al., 2018; Muizon and Ladevèze, 2020). However, none of the †Pucadelphys specimens we examined provide unambiguous evidence of this (this region is poorly preserved in most available specimens), and in YPFB Pal 6105 (the holotype) a narrow connection between the alisphenoid and petrosal posterolateral to the primary foramen ovale clearly excludes the squamosal from the rim of this foramen (R.M.D.B., personal obs.). Horovitz et al. (2008: 117) stated that “a foramen just anterior to the large alisphenoid bulla…may be the foramen ovale” in a specimen of †Herpetotherium cf. H. fugax (MB.Ma.50672) from the early Oligocene White River Formation in Wyoming. However, we examined their material and additional well-preserved specimens of †Herpetotherium (notably AMNH FM 22304; see Gabbert, 1998) and conclude that the structure referred to by Horovitz et al. (2008) is probably the foramen rotundum. Instead, the primary foramen ovale of †Herpetotherium is medial (rather than anterior) to the alisphenoid tympanic process and appears to lie between the alisphenoid and the petrosal (see Gabbert, 1998: fig. 1).

In several metatherians—including Caluromysiops, Philander (Voss and Jansa, 2009: fig. 48; Wible et al., 2021: fig. 2), Isoodon, many macropodiforms (e.g., Lagostrophus [fig. 52]), and some Sarcophilus individuals (Wroe, 1997: 33–34, figs. 4D, 5C)—the primary foramen ovale is largely or entirely concealed from ventral view in intact crania, due to the presence of an extensive lamina of bone that forms a secondary foramen ovale. In these taxa, the bone(s) that form the primary foramen ovale cannot always be determined by external inspection. We have, therefore, followed Voss and Jansa (2003, 2009) in not scoring the bony composition of the primary foramen ovale as a character. Instead, we scored the morphology of the course of the mandibular nerve after it leaves the endocranium.

In many metatherians—such as †Herpetotherium (Gabbert, 1998), some Recent didelphids (e.g., Marmosa [fig. 9A]), caenolestids, Dromiciops, most dasyurids (e.g., Phascogale [fig. 9B]), Myrmecobius, Aepyprymnus, acrobatids, Cercartetus, and most petaurids—the mandibular nerve is usually not fully enclosed by bone after exiting the endocranium via the primary foramen ovale. Although bony outgrowths from the ossified tympanic sinus floor sometimes partially enclose the nerve in some of these taxa, a complete secondary foramen ovale sensu Gaudin et al. (1996) and Wroe (1997b) is usually absent (state 0). In the only known skull of †Yalkaparidon (QMF13008) the alisphenoid tympanic process is damaged on both sides, but it does not appear that a strut or lamina forming a complete secondary foramen ovale was ever present (Beck et al., 2014: 143), so we scored this taxon as corresponding to state 0.

In other metatherians (e.g., many didelphids, Notoryctes, peramelemorphians, thylacinids, and most diprotodontians), bony outgrowths from the medial side of the ossified hypotympanic sinus floor form a complete secondary foramen ovale that encloses the extracranial course of the mandibular nerve after it exits the primary foramen ovale (state 1). Although Voss and Jansa (2003, 2009) identified two distinct patterns of secondary foramen ovale formation in didelphids, and Wroe et al. (2000: chars. 51–53; see also Wroe, 1997) distinguished three patterns of secondary foramen ovale formation in dasyuromorphians, the alternative conditions described by those authors cannot be consistently distinguished among the wider range of metatherian taxa examined in this study. Therefore, we simply scored a secondary foramen ovale as present or absent, regardless of its apparent osteological origin.

In some taxa in which a secondary foramen ovale is present (e.g., Chironectes, Gracilinanus, most peramelemorphians, Thylacinus, some diprotodontians), the enclosure is a relatively narrow strut, and the primary foramen ovale remains clearly visible dorsal to the secondary enclosure (Voss and Jansa, 2003: fig. 8B). However, as already noted, in some taxa (e.g., Caluromysiops, Philander, Isoodon, many macropodiforms, some Sarcophilus individuals), the bone secondarily enclosing the mandibular nerve is more extensive and forms a lamina that conceals the primary foramen ovale from external view. In such cases, the secondary foramen ovale might be confused for the (concealed) primary foramen ovale. However, the primary foramen ovale is not formed by outgrowths of the ossified hypotympanic sinus floor, is immediately adjacent to the anterior pole of the petrosal, opens ventrally, and is usually in approximately the same horizontal plane as the basisphenoid and basioccipital. By contrast, the secondary foramen ovale (if present) is formed by outgrowths of the ossified hypotympanic sinus floor, and, as a result, is more distant from the anterior pole of the petrosal, opens anterolaterally, and is usually clearly ventral to the basisphenoid and basioccipital.

Both †Yarala and †Galadi have been described as lacking a secondary foramen ovale (Muirhead, 2000; Travouillon et al., 2010), but Wroe et al. (2000: char. 52; see also Wroe and Musser, 2001) scored †Yarala as having a complete secondary foramen ovale formed by a “mesial fold in [the] alisphenoid tympanic wing.” In both †Yarala and †Galadi an outgrowth of the alisphenoid tympanic process (labelled as “arch of alisphenoid” in †Yarala by Muirhead, 2000: fig. 3) forms a complete foramen anterolateral to the primary foramen ovale; the small size of this foramen and its relatively lateral position suggests that it probably enclosed only the masseteric branch of the mandibular nerve, in which case it arguably does not represent a secondary foramen ovale sensu stricto. However, in the absence of soft-tissue evidence regarding foraminal contents for the vast majority of taxa included in the current analysis, we scored †Yarala and †Galadi as possessing a secondary foramen ovale (contra Muirhead, 2000; Travouillon et al., 2010, 2013b).

In Phascolarctos, the mandibular nerve takes an unusual course after leaving the endocranial cavity: it passes through (rather than anteromedial to) the hypotympanic sinus, enclosed within a bony canal in the sinus roof, and then emerges onto the external surface of the cranium via an anterolaterally facing foramen on the anterolateral side of the ossified hypotympanic sinus floor (Aplin, 1987, 1990). Damaged cranial material of †Litokoala and †Nimiokoala reveals the presence of a similar canal in the hypotympanic sinus roof (see Louys et al., 2009). Because the mandibular nerve is enclosed within the hypotympanic sinus (rather than by bony outgrowths that extend from the ossified hypotympanic sinus floor) in Phascolarctos, †Litokoala, and †Nimiokoala, we scored this morphology as a distinct alternative condition (state 2). We observed a somewhat similar morphology in the sparassocynin didelphids †Sparassocynus and †Hesperocynus, in which the mandibular nerve passes within a bony tube in the medial wall (not the dorsal roof, as seen in phascolarctids) of the hypotympanic sinus (Beck and Taglioretti, 2020); nevertheless, given the distinctly different (medial) path of the mandibular nerve compared to the (dorsal) phascolarctid condition, we scored †Sparassocynus and †Hesperocynus as corresponding to an additional state (state 3). Both †Pucadelphys and †Mayulestes lack an ossified hypotympanic sinus floor (Muizon, 1994, 1998; Marshall and Muizon, 1995; Muizon et al., 1997), as do †Allqokirus (Muizon et al., 2018) and †Mimoperadectes (see Horovitz et al., 2009: fig. S3), so we scored these four taxa as inapplicable for this character. As defined here, our states do not form a clear morphocline, so we left this character as unordered in all our analyses.

We observed additional taxonomic variation in the extracranial course of the mandibular nerve that we did not score for phylogenetic analysis, but which merits brief discussion. In some taxa, the masseteric nerve, the buccinator (or buccal) nerve, or both nerves, sometimes emerge from foramina that are separate from the secondary foramen ovale—as in some macropodiforms (e.g., †Balbaroo, †Ganawamaya, Lagostrophus, Onychogalea, and Wallabia; see Aplin, 1990: 182–183), some peramelemorphians (e.g., Chaeropus and Peroryctes; see Aplin et al., 2010: fig. 8; Travouillon et al., 2019: 15), and Pseudochirops cupreus. However, presence or absence of these additional foramina often appears to be highly variable within species. In the peramelemorphian Macrotis, a small foramen in the alisphenoid anteromedial to the glenoid fossa suggests that the masseteric nerve exits the skull far lateral to the secondary foramen ovale. A similar foramen is also present in Notoryctes (Archer, 1976b: 270) and likewise appears to transmit the masseteric nerve. However, Acrobates and Distoechurus are apparently unique in having a distinct foramen at the anteromedial end of the glenoid fossa that opens directly into the endocranium and which transmits the masseteric nerve, a morphology that we have scored as an additional character (see char. 53).

Character 53. Single endocranial exit for mandibular nerve (0); or separate foramen for ramus lateralis present at anteromedial end of the glenoid fossa (1). In most marsupials for which relevant information is available (see, e.g., Aplin, 1990; Tomo et al., 2007), the mandibular nerve branches after exiting the endocranium via the primary foramen ovale (state 0). In taxa with a secondary foramen ovale, the nerve either branches after exiting the secondary foramen ovale (as in Macropus giganteus; Tomo et al., 2007: fig. 2) or immediately before; in the latter case, additional foramina for individual branches of the mandibular nerve may be present, as described above (see char. 52).

By contrast, Aplin (1990: 315) examined sectioned and stained specimens and discovered that the mandibular nerve branches endocrani-ally in the acrobatid Acrobates, with the “ramus lateralis” (= the common trunk of the masseteric and posterior temporal nerves) arising “direct[ly] from the mandibular lobe of the trigeminal ganglion….” This nerve then exits the endocranium via a separate foramen at the anteromedial end of the glenoid fossa—between the alisphenoid and squamosal—rather than via the foramen ovale (state 1). An identical foramen is present in our other acrobatid terminal, Distoechurus, which has also been scored as state 1. A superficially similar foramen that appears to transmit the masseteric nerve is also identifiable within the alisphenoid, anteromedial to the glenoid fossa, in Notoryctes and the peramelemorphian Macrotis (see char. 52). However, careful examination of damaged cranial material of both taxa indicates that the masseteric nerve originates after the mandibular nerve has exited the endocranium, unlike the acrobatid condition (which appears to be unique among the metatherians we surveyed). Direct evidence regarding the branching pattern of the mandibular nerve is, of course, unavailable for fossil metatherians, as it is for most Recent taxa. Therefore, this character has been scored purely on the basis of the presence or absence of a separate foramen (in addition to the foramen ovale) anteromedial to the glenoid fossa—either within or immediately adjacent to the alisphenoid-squamosal suture—that opens directly into the endocranium. Such a foramen is present only in Acrobates and Distoechurus.

Character 54. Hypotympanic sinus roof formed by alisphenoid and petrosal only (0); or by alisphenoid, squamosal, and petrosal (1); or by squamosal and petrosal only (2). A distinct concavity in the ventral surface(s) of one or more bones anterior to the epitympanic recess defines the roof of the hypotympanic sinus (within the middle ear; van der Klaauw, 1931: 19). It has been reported that no such distinct concavity exists in †Pucadelphys, and hence that a true hypotympanic sinus was absent (e.g., Muizon, 1994, 1998; Marshall and Muizon, 1995; Muizon et al., 1997). However, we observed a shallow concavity that appears to correspond to the roof of the hypotympanic sinus in several recently prepared specimens (MHNC 8376, 8377, and 8380; Ladevèze et al., 2011: fig. 1). What we interpret to be a hypotympanic sinus roof is therefore present in all metatherians included in the current analysis for which specimens exposing this region were available.

The roof of the hypotympanic sinus is variously formed in metatherians. Most nondiprotodontians have a sinus roof formed by the alisphenoid with a small (often negligible) petrosal component (state 0), as in most didelphids, Caenolestes, Dromiciops, †Yalkaparidon (see Beck et al., 2014: fig. 8), dasyuromorphians (e.g., Mutpuracinus; Murray and Megirian, 2006a: fig. 8), and most peramelemorphians. However, the squamosal sends a medial process into the hypotympanic sinus of †Mayulestes (Muizon, 1994, 1998; Muizon et al., 1997), †Pucadelphys (Muizon, 1994, 1998; Muizon et al., 1997), †Allqokirus (Muizon et al., 2018), and †Mimoperadectes (see Horovitz et al., 2009: fig. S3A) (state 1). The hypotympanic sinus roof also has both alisphenoid and squamosal components in Lutreolina, Philander, Rhyncholestes, and several diprotodontians (Aplin, 1987; 1990; Gillespie, 1997; Louys et al., 2009), including †Lekanoleo, Phascolarctos, †Nimiokoala, †Litokoala, †Bulungamaya, Caloprymnus, Potorous, Burramys, and Cercartetus. Both the alisphenoid and squamosal contribute to the hypotympanic sinus roof in some specimens of Echymipera, Microperoryctes, and Peroryctes, but the squamosal is excluded from this region in others, so we scored these taxa as polymorphic (“0+1”). Only the squamosal and petrosal roof the hypotympanic sinus of most diprotodontians (Aplin, 1987; 1990; Gillespie, 1997; Black, 2008; Louys et al., 2009; Black and Hand, 2010), including petaurids, pseudocheirids, phalangerids, diprotodontids, vombatids, wynyardiids, †Thylacoleo, and †Ilaria (state 2). We were unable to determine the composition of the hypotympanic sinus roof of Notoryctes, Acrobates, and Distoechurus, which we scored as unknown (“?”). As defined, the states comprising this character represent a plausible morphocline, so this character was ordered in all our analyses.

Although the presence/absence of a medial process of the squamosal and the composition of the hypotympanic sinus were coded as separate characters by Rougier et al. (1998: chars. 140, 141; see also Wible et al., 2001), these are clearly nonindependent aspects of auditory morphology that seem better treated as part of the same transformation series. Springer and Woodburne (1989: char. 1) and Springer et al. (1997: char. 61) coded the medial wall of the “mandibular” (glenoid) fossa as formed by either the alisphenoid or the squamosal, but the medial wall of the mandibular fossa is typically formed by the squamosal whenever the squamosal contributes to the hypotympanic sinus roof, and so these would also appear to be nonindependent characters.

Character 55. Hypotympanic sinus floor unossified (0); or hypotympanic sinus floor ossified but does not contact rostral tympanic process of petrosal (1); or hypotympanic sinus floor ossified and contacts rostral tympanic process of petrosal (2); or hypotympanic sinus floor ossified and extends posteriorly across the petrosal to contact the exoccipital (3). The hypotympanic sinus of †Allqokirus, †Mayulestes, and †Pucadelphys is not enclosed anteriorly or ventrally by any kind of ossified tympanic process (state 0; Muizon, 1994, 1998; Marshall and Muizon, 1995; Muizon et al., 1997, 2018), although an unossified fibrous membrane of the tympanic cavity sensu MacPhee (1981) was presumably present. Horovitz et al. (2009: 3) reported that a “poorly developed tympanic wing of the alisphenoid” is present in †Mimoperadectes (see also Muizon et al. 2018: 402), but we examined their material (USNM 482355) and did not find any trace of an alisphenoid tympanic process. We therefore also scored †Mimoperadectes as state 0 (see also Horovitz et al., 2009: fig. S3A).

By contrast, the hypotympanic sinus of all other metatherian taxa examined here is at least partially floored by bone. We assume here that, where present, the bony floor of the hypotympanic sinus exhibits primary homology (sensu de Pinna, 1991) across Metatheria, regardless of exactly which bones form it. This assumption seems justified by reports that, in specimens of Phalanger orientalis from the Solomon Islands, the hypotympanic sinus floor is alternatively composed of the alisphenoid only, of alisphenoid and squamosal components, or of the squamosal only (Aplin, 1990; Norris, 1993). These morphologies appear otherwise identical; indeed, a number of examples are known in which the hypotympanic sinus floor is entirely alisphenoid on one side and entirely squamosal on the other (Aplin, 1990; Norris, 1993). Similarly, although Aplin (1990) identified the the bony hypotympanic sinus floor of acrobatids (e.g., Distoechurus [fig. 48]) as being formed by the petrosal and an “entotympanic-like” intramembraneous ossification, it otherwise resembles the condition in other diprotodontians, in which the sinus floor is formed by the alisphenoid or the squamosal (figs. 4447, 4954), so we assume that these structures are homologous regardless of bony composition (see also Aplin, 1990: 344–345)7.

When present, the ossified hypotympanic sinus floor exhibits striking taxonomic variation in size and in its contact with adjacent bones. The ossified hypotympanic sinus floor does not extend far enough posteriorly to contact the rostral tympanic process of the petrosal (state 1) in most Recent didelphids (e.g., Marmosa [fig. 9A], Didelphis [Wible, 1990: fig. 1A], Monodelphis [Wible, 2003], Philander (Wible et al., 2021: figs. 2, 4), †Sparassocynus, caenolestids, thylacinids, most peramelemorphians, diprotodontids, wynyardiids, phascolarctids, †Lekanoleo, and †Wakaleo, although the hypotympanic sinus floor itself varies markedly in size and shape among these taxa (it is particularly large in †Sparassocynus, Phascolarctos, the dasyurids Dasycercus and Dasyuroides, and the peramelemorphians Isoodon and Macrotis). By contrast, the ossified hypotympanic sinus floor clearly contacts the rostral tympanic process of the petrosal (state 2) in dasyurids (e.g., Phascogale; fig. 9B), Chaeropus, microbiotheriids, burramyids, Tarsipes, Dactylonax, and in some specimens of Caluromys (e.g., AMNH 267002), Dactylopsila, Pseudochirops, and Ailurops. The ossified hypotympanic sinus floor of Myrmecobius resembles that of dasyurids, except that it extends somewhat farther posteriorly, contacting the caudal tympanic process of the petrosal, as well as the rostral tympanic process (Archer, 1976b: 244; Murray and Megirian, 2006a: fig. 16H), a condition we subsume in state 2. The ossified hypotympanic sinus floor is still larger, extending posteriorly across the petrosal to contact the base of the paroccipital process of the exoccipital (state 3) in macropodiforms and most petaurids (e.g., Petaurus; fig. 9C), phalangerids, and pseudocheirids (as previously described by Springer and Woodburne, 1989: char. 3).

In several taxa—mostly fossil forms represented by damaged cranial material—an ossified hypotympanic sinus floor is clearly present, but its posterior extent cannot be determined based on available specimens. Such taxa have been scored using ambiguity (“either/or”) coding, with plausible alternative character states represented. For example, we scored †Yalkaparidon, †Yarala, †Galadi, and †Litokoala (in which the hypotympanic sinus floor is ossified but clearly did not contact the exoccipital) as “1/2.” Because of sutural fusion, the precise extent of the ossified hypotympanic sinus floor is also unclear in Notoryctes, but it extends posteriorly either as far as the rostral tympanic process of the petrosal, or further posteriorly to contact the exoccipital, so we scored this taxon as “2/3.” Because these conditions appear to represent a successive series of tympanic expansions, we treated this character as ordered (additive) in all our analyses.

Character 56. Ossified hypotympanic sinus floor formed at least partially by alisphenoid (0); or by squamosal only (1); or by petrosal and an “entotympanic-like” ossification (2). There is considerable taxonomic variation in the composition of the ossified hypotympanic sinus floor within Metatheria, where such a structure is present. In most nondiprotodontian metatherians, it appears to be formed exclusively by a tympanic process—referred to as a “wing” by some authors (e.g., Archer, 1975, 1976; Cooper et al., 2000; see table 10), although MacPhee (1981: 26) restricted “wing” to components of the auditory roof only—of the alisphenoid. This morphology is seen in didelphids (e.g., Marmosa: fig. 9A; Philander [Wible et al., 2021: fig. 4), caenolestids, Dromiciops, †Yalkaparidon (Beck et al., 2014), peramelemorphians. and dasyuromorphians (e.g., Phascogale; fig. 9B). In many diprotodontians, by contrast, damaged or sectioned skulls reveal the presence of a squamosal contribution to the floor of the hypotympanic sinus, in addition to an alisphenoid tympanic process (Aplin, 1990; Beck et al., 2008a: char. 244; Sánchez-Villagra and Forasiepi, 2017: 26, fig. 23). However, this squamosal tympanic process is often largely covered by the alisphenoid tympanic process, so its presence is not readily apparent based on external examination of intact skulls (Murray et al., 1987; Sánchez-Villagra, 1998: 184–185; Sánchez-Villagra and Forasiepi, 2017: 26, fig. 23). Indeed, in some diprotodontians, the ossified hypotympanic sinus floor is largely or entirely bilaminar, comprising an internal squamosal tympanic process and an external alisphenoid tympanic process—as in †Wakaleo (Murray et al., 1987), †Namilamadeta, †Muramura, and †Ngapakaldia (based on SAM P13851, the holotype of †N. tedfordi; contra Murray et al., 2000: 98). Because it is difficult to verify the presence and extent of the squamosal tympanic process in taxa that also have an alisphenoid tympanic process (unless broken skulls are available), we scored all taxa in which the alisphenoid contributes to the ossified hypotympanic sinus floor as corresponding to state 0, regardless of whether the squamosal also contributes.

A qualitatively distinct morphology is seen in Vombatus, Lasiorhinus, †Warendja, and †Thylacoleo. In these taxa the ossified hypotympanic floor is formed exclusively by a tympanic process (again, referred to as a “wing” by some authors [e.g., Archer, 1978; Murray et al., 1987; Springer and Woodburne, 1989; see table 10], contra MacPhee, 1981: 26) of the squamosal, without any contribution by the alisphenoid (state 1; Aplin, 1987, 1990; Murray et al., 1987; Murray, 1998). A third type of middle ear enclosure is seen in acrobatids, but the osseous composition of the hypotympanic sinus floor of adult acrobatids is difficult to determine due to seamless fusion of the squamosal, ectotympanic, and petrosal. However, developmental studies by Aplin (1990) suggest that the acrobatid hypotympanic sinus floor is formed by three distinct elements that correspond (from posterior to anterior) to the caudal tympanic process of the petrosal, the rostral tympanic process of the petrosal, and an “entotympanic-like” intramembraneous ossification (state 2). We scored Notoryctes as unknown (“?”), because fusion of the basicranial sutures obscures the homology of bones contributing to the hypotympanic sinus floor (Aplin, 1990: 370). By contrast, both Archer (1976b) and Ladevèze et al. (2008) considered the hypotympanic sinus to be formed by the alisphenoid in this taxon. We scored †Pucadelphys, †Allqokirus, †Mayulestes, and †Mimoperadectes as inapplicable (“-”) for this character because they lack an ossified hypotympanic sinus floor (see char. 55).

As defined, the states of this character do not represent a plausible morphocline, so we treated it as unordered in all our analyses.

Character 57. Ectotympanic laterally exposed (0); or concealed from lateral view by ossified hypotympanic sinus floor (1). In lateral view the ectoympanic is an externally obvious element of the ear region in most metatherians (state 0; e.g., Didelphis [Wible, 1990: fig. 1], Monodelphis [Wible, 2003], Philander [Wible et al., 2021: fig. 4], Phascogale [fig. 10A], and Petaurus [fig. 10B]), but the ectotympanic is concealed by the ossified hypotympanic sinus floor in Dromiciops (state 1; Segall, 1969b; Springer and Woodburne, 1989: char. 2). Aplin (1990: 370) stated that “the ectotympanic in Notoryctes is…phaneric or ‘intrabullar’ in location…and is hidden from external view by other, major elements of the tympanic floor,” implying that the ectotympanic of Notoryctes is concealed in a manner similar to that seen in Dromiciops. Our own examination of N. typhlops specimens in which the ossified hypotympanic sinus floor is broken or damaged (e.g., AMNH 16717 and 198651) confirms Aplin’s (1990: 370) observation, so we also scored Notoryctes as corresponding to state 1. The ectotympanic of some dasyurids (e.g., Dasykaluta and Pseudantechinus; fig. 42) is largely concealed by the ossified hypotympanic sinus floor in ventral view, with only the ends of the anterior and posterior limbs visible, but (unlike the condition in Dromiciops and Notoryctes) it remains clearly exposed in lateral view, so we scored these taxa as state 0.

Character 58. Lateral margin of ectotympanic with posterodorsal incisura (0); or ectotympanic forms a closed tube that completely encircles the ear canal (1). The ectotympanic of most metatherians (e.g., Didelphis [Wible, 1990: fig. 1], Monodelphis [Wible, 2003], Philander [Wible et al., 2021: fig. 4], Phascogale [fig. 10A], and Petaurus [fig. 10B]) does not completely encircle the ear canal, leaving a more or less prominent posterodorsal opening (the incisura tympanica; Segall, 1969a) between the anterior and posterior limbs (state 0); such ectotympanics were referred to as U-shaped by Wroe et al. (1998). Although the incisura exhibits obvious taxonomic variation in size and position among metatherians—tending to be widely open and more posteriorly oriented in taxa with mediolaterally unexpanded (ringlike) ectotympanics, but narrower and more dorsally oriented in taxa with mediolaterally expanded (tubelike) ectotympanics—we were unable to code observed differences in relative closure or location as discrete states due to an almost continuous range of intermediate conditions (contra Springer et al., 1997: chars. 74, 75; Horovitz and Sánchez-Villagra, 2003: chars. 192, 193).

FIG. 10.

Anteroventrolateral views of left ear region of Phascogale tapoatafa (A, AMNH 160267) and Petaurus breviceps (B, AMNH 154468) with reconstructed course of postglenoid vein shown by heavy arrow in each panel. Alternative states for characters 22, 23, 59, 60, 77, 79 and 87 (see main text for descriptions of these characters and character states) are illustrated as follows: Phascogale 22(0), 23(1), 59(1), 60(0), 77(0), 79(3), 87(0); Petaurus 22(1), 23(0), 59(2), 60(2), 77(1), 79(4), 87(1[not visible on intact skulls]). Abbreviations: als, alisphenoid; atp, alisphenoid tympanic process; cc, carotid canal; eam, external auditory meatus; ect, ectotympanic; gf, glenoid fossa; jug, jugal; mas, mastoid exposure of petrosal pars canalicularis (pneumatized in Petaurus); pes, posterior epitympanic sinus (not externally visible in Petaurus); pogp, postglenoid process (of squamosal); prgp, preglenoid process (of squamosal); ptps, posttympanic process of squamosal; rtp + ctp, rostral and caudal tympanic processes of petrosal (fused to form petrosal plate); sq, squamosal (pneumatized in Petaurus); smf, stylomastoid foramen; ssf, subsquamosal fenestra (out of view, between squamosal and parietal, in Petaurus); tcf; transverse canal foramen; zps, zygomatic process of squamosal. Specimens are not drawn to the same scale.

img-z77-1_01.jpg

In a few diprotodontians, however, the ectotympanic forms a fully enclosed tube that completely encircles the ear canal (state 1), as in Tarsipes (Aplin, 1990: 290), Cercartetus (contra Wroe et al., 1998: 743), Vombatus, Lasiorhinus, Macropus, Notamacropus, Osphranter, Petrogale, and Wallabia. Although the ectotympanic forms a complete tube in a few specimens of Dendrolagus lumholtzi, it is incomplete in most individuals that we examined, so we scored Dendrolagus as state 0 based on the modal condition. In some diprotodontians with mediolaterally expanded (tubelike) ectotympanics, the anterior and posterior limbs do not contact each other at the lateral edge of the ectotympanic but are in contact further medially—within the ear canal—as in Hemibelideus (e.g., AMNH 65510) and Petauroides (AMNH 65378); thus, the lateral margin of the ectotympanic has an incisura tympanica whereas the medial margin does not. Determining whether there is medial contact between the anterior and posterior limbs usually requires broken skulls or CT-scans, so we scored this character based on contact between the externally obvious lateral edges of the anterior and posterior limbs. In consequence, Hemibelideus and Petauroides correspond to state 0.

Character 59. Anterior limb of ectotympanic suspended from anterior process of the malleus, not contacting squamosal or petrosal dorsally (0); or loosely attached to the petrosal or squamosal behind postglenoid process (1); or attached firmly to postglenoid process of squamosal (2); or tightly articulated with suspensory process of petrosal (3). The ectotympanic exhibits a wide range of taxonomic variation in its attachments to adjacent bones of the marsupial skull. As described and illustrated by Voss and Jansa (2003: char. 46, fig. 9; 2009), the anterior limb of the ectotympanic is only indirectly attached to the skull via the anterior process of the malleus (state 0) in some didelphids (e.g., Caluromys, Didelphis, and Philander [see also Wible et al., 2021: fig. 4]). However, in most metatherians—including other didelphids, caenolestids, dasyuromorphians (e.g., Phascogale [fig. 10A]), some peramelemorphians, and Dromiciops—the anterior limb of the ectotympanic is loosely attached to either the petrosal or to the squamosal behind the postglenoid process, from which it is usually separated by a distinct gap (state 1). This condition likewise characterizes some diprotodontians, including Tarsipes, Phascolarctos, †Thylacoleo, †Wakaleo, †Neohelos, and many macropodiforms (Aplin, 1990). In most diprotodontians, however, the ectotympanic is firmly sutured to or seamlessly fused with the postglenoid process of the squamosal (state 2; see Springer and Woodburne, 1989; Aplin, 1990; Wroe et al., 1998: 743), as in petaurids (e.g., Petaurus [fig. 10B]), pseudocheirids, burramyids, phalangerids, Recent vombatids, and a few macropodiforms (e.g., Caloprymnus, Dendrolagus, Petrogale, †Balbaroo, †Ekaltadeta, and †Ganawamaya). The ectotympanic is similarly fused to the postglenoid process in Notoryctes. We are unable to meaningfully distinguish the ectotympanic attachment in Burramys—which Wroe et al. (1998: 743) described as showing “less intimate association with the alisphenoid tympanic wing or the postglenoid process of the squamosal”—from that observed in Cercartetus, petaurids, pseudocheirids, or phalangerids (the postglenoid process and ectotympanic of which are said to be fused; Springer and Woodburne, 1989; Marshall et al., 1990: fig. 2, node 71; Wroe et al., 1998: 743).

A qualitatively different morphology is seen in some peramelemorphians (Isoodon, Macrotis, and Perameles) in which the anterior limb of the ectotympanic articulates with a distinct suspensory process of the petrosal (state 3; Archer, 1976b: 310). Taxa in which the ectotympanic is missing in available specimens (as for many fossil taxa in this analysis) have been scored as unknown (“?”) for this character, which (in the absence of any obvious morphological sequence among these conditions) we treated as unordered (nonadditive) in all our phylogenetic analyses.

Character 60. Posterior limb of ectotympanic does not contact pars canalicularis of the petrosal or posttympanic process of squamosal (0); or in contact with but suturally distinct from pars canalicularis of the petrosal and/or posttympanic process of the squamosal (1); or seamlessly fused with pars canalicularis of the petrosal and/or posttympanic process of the squamosal (2). In didelphids (e.g., Didelphis [Wible, 1990: fig. 1], Monodelphis [Wible, 2003], Philander [Wible et al., 2021: fig. 4], Marmosa [fig. 9A]), caenolestids, peramelemorphians, Thylacinus, and Myrmecobius (see Archer, 1976), the posterior limb of the ectotympanic does not directly contact any adjacent bony structure except (sometimes) the rostral tympanic process of the petrosal (state 0). In Recent dasyurids (e.g., Phascogale [figs. 9B, 10A]) the posterior limb of the ectotympanic contacts the petrosal plate (the fused rostral and caudal tympanic processes of the petrosal; see char. 68) anterior to the stylomastoid foramen. In adult dasyurids there is no suture between the rostral tympanic process (an outgrowth of the pars cochlearis) and the caudal tympanic process (an outgrowth of the pars canalicularis; Wible, 1990, 2003), but examination of juveniles indicates that the boundary between these processes coincides with the stylomastoid foramen (the caudal tympanic process is posterior to the foramen). Therefore, the ectotympanic does not contact the pars canalicularis in dasyurids, which we also scored as state 0.

By contrast, in some diprotodontians (e.g., Phascolarctos, †Neohelos, †Wakaleo, †Thylacoleo, vombatids, and most macropodiforms) the posterior limb of the ectotympanic clearly contacts the pars canalicularis of the petrosal, the posttympanic process of the squamosal, or both with a suture clearly marking the point of contact (Murray et al., 1987; Aplin, 1990; Wroe et al., 1998: 743). In other diprotodontians—namely phalangerids, burramyids, petaurids (e.g., Petaurus; figs. 9C, 10B), pseudocheirids, acrobatids, Tarsipes, and †Ekaltadeta—and also in Dromiciops the posterior limb of the ectotympanic is seamlessly fused to the pars canalicularis of the petrosal and/or to the posttympanic process of the squamosal (Segall, 1969b; Murray et al., 1987; Springer and Woodburne, 1989; Aplin, 1990; Marshall et al., 1990: fig. 2, node 71; Wroe et al., 1998: 743). In Notoryctes the posterior limb of the ectotympanic appears to contact the pars canalicularis without fusion in some specimens (e.g., AMNH 16717), but these structures appear fused in others (e.g., AMNH 198651), so we scored this taxon as polymorphic (“1+2”). Available specimens of many fossil taxa lack ectotympanics, but if the ectotympanic is missing and the pars canalicularis of the petrosal and the posttympanic process of the squamosal are intact, we considered this sufficient evidence that the posterior limb of the ectotympanic could not have been fused to either of these structures, and we scored such taxa using ambiguity coding as “0/1.”

Because these states form a plausible morphocline (reflecting increasingly intimate contact between the posterior limb of the ectotympanic and adjacent bones), we ordered this character in all our analyses.

Character 61. Ectotympanic and pars cochlearis of petrosal are separate bones distinguishable by gaps or open sutures (0); or ectotympanic and pars cochlearis seamlessly coossified (1). The ectotympanic and the pars cochlearis of the petrosal are separated by gaps (as in Marmosa [fig. 9A]) or open sutures (as in Phascogale [fig. 9B]) in most metatherians (state 0), but the ectotympanic is seamlessly fused to the acrobatid bulla, a structure that is partly formed by the rostral tympanic process of the pars cochlearis (state 1; see Aplin, 1990: 292). In Tarsipes, the posterior part of the ectotympanic is seamlessly fused to the petrosal plate, which appears to represent conjoined rostral and caudal tympanic processes of the petrosal (originating from the pars cochlearis and pars canalicularis respectively; Aplin, 1990: 290; see char. 68); hence, we also scored this taxon as corresponding to state 1. The ectotympanic might be fused to the rostral tympanic process in Notoryctes (e.g., AMNH 15015; see also Archer, 1976b: 269), but we are not certain which bone(s) form(s) the hypotympanic sinus floor in this taxon, so we scored it as unknown (“?”). We scored fossil taxa in which the ectotympanic is missing, but the petrosal appears undamaged as “0” because loss of the ectotympanic without damage to the petrosal indicates that these elements were not coossified.

Character 62. Ear canal not occluded by ossifications (0); or ear canal largely occluded by a bony disk (1). The external auditory meatus is not occluded in most metatherians (state 0), but this aperture is nearly filled by a bony disk that is attached by a stalk to the ventral wall of the ear canal in Acrobates and Distoechurus (state 1; see Segall, 1971: fig. 5B; Aitkin and Nelson, 1989: fig. 1B-C; Aplin, 1990: plate 5.10c). This extraordinary structure is reported to attenuate all but a narrow range of low-frequency sounds (Aitkin and Nelson, 1989; Aplin, 1990). According to Aplin (1990: 308; see also Aitkin and Nelson, 1989), “detailed calculations based on the physical parameters of the disk and associated tympanic membrane in Acrobates pygmaeus indicate that the cut-off frequency above which transmission is impaired is probably around 4 kHz. Interestingly enough, this is about the frequency generated by the wing beats of owls (Webster and Webster, 1975); coincidentally, owls probably represent the major predator of Acrobates pygmaeus. Comparable calculations have not yet been made for Distoechurus pennatus.” Unfortunately, a fuller exposition of this hypothesis remains to be published (see also Ward and Woodside, 2008; Ashwell, 2010: 184; Harris, 2015: 35). More recently, Archer et al. (2019) proposed that the presence of the disk helps create Helmholtz resonance (which can amplify a particular frequency; Griffiths, 1978) but, again, no detailed explanation of the underlying basis for this conclusion nor for its potential biological significance has yet been published.

Archer (1984) inferred that this “obstructing process” is ectotympanic in origin, presumably based on its attachment to the ventral surface of the ear canal, but the ectotympanic is fused with the petrosal bulla in even the youngest acrobatids we examined, with no trace of sutures that might confirm or refute this conjecture (see char. 61). Despite extensive ontogenetic studies, Aplin (1990: 293–294) similarly could not distinguish whether the occluding disk is ectotympanic in origin or represents an independent ossification.

Character 63. Malleo-incudal and stapedio-incudal articulations unfused in adult specimens, with clearly identifiable sutures (0); or fused, without sutures (1). In most metatherians in which we were able to examine the ossicles, the articulations between the malleus, incus, and stapes are unfused (state 0; see also Doran, 1878; Segall, 1969b; 1969a; 1970; 1971; Sánchez-Villagra et al., 2002; Schmelzle et al., 2005; Czerny, 2015). However, Aplin (1990: 301) reported that both the malleo-incudal and stapedio-incudal articulations are fused in Acrobates and Distoechurus (state 1), with the common body of the malleo-incudal joint perforated by the chorda tympani nerve.

Character 64. Anterior process of the malleus well developed (0); or weakly developed and spine-like (1); or entirely absent (2). The anterior process of the metatherian malleus is usually well developed (state 0). Segall (1969b: 494) stated that the anterior process “is very small and simplified in Dromiciops,” but an elongate process was clearly identifiable in all of the D. gliroides specimens we examined; additionally, a well-developed anterior process is labelled in figure 3 of Segall (1969b), whereas it appears to have broken off the malleus illustrated in his figure 4. The anterior process of the malleus of Dromiciops is unusual, however, in lacking a well-developed lamina (Voss and Jansa, 2009: 44); nevertheless, we scored Dromiciops as corresponding to state 0 because the process is still elongate. Aplin (1990) observed that the anterior process of the malleus is extremely weakly developed in Acrobates and Distoechurus, “reduced to a delicate spine” (Aplin, 1990: 301, fig. 5.10d), an apparently distinct condition (state 1). The anterior process of the malleus is similarly vestigial in Tarsipes (Parker, 1890), which we also scored as state 1. The anterior process is entirely absent in Notoryctes (see Segall, 1970: fig. 3; Mason, 2001: fig. 11D; Czerny, 2015: abb. 19), another distinctive malleolar condition (state 2). As defined, these states comprise a plausible morphocline reflecting gradual reduction in the size of the anterior process of the malleus, so we ordered this character in all our analyses.

We observed additional variation in the morphology of the anterior process of the malleus (as did Doran, 1878; Segall, 1969b; 1969a; 1970; 1971; Sánchez-Villagra et al., 2002; Schmelzle et al., 2005) that we did not score for phylogenetic analysis, but which merits brief discussion. The anterior process is notably elongate (over twice as long as the manubrium) in some diprotodontians, such as Vombatus and Macropus (Doran, 1878), but a continuous range of intermediate lengths among other metatherians precludes defining additional character states for these outliers. In a number of macropodiforms—for example, Dorcopsulus (e.g., AMNH 151854)—the tip of the anterior process is visible on the external surface of the auditory bulla, where it protrudes between the ectotympanic and the alisphenoid tympanic process (Aplin, 1990: 158). The tip of the anterior process is also visible externally in Vombatus and Lasiorhinus, projecting ventrally from the anteromedial border of the ectotympanic within a distinct sulcus on the squamosal tympanic wing; Aplin (1990: 219) reported that the chorda tympani exits the middle ear via this sulcus in Vombatus. This condition is also seen in some didelphids, including some species of Monodelphis (e.g., M. brevicaudata; Wible, 2003: fig. 6; Pavan and Voss, 2016: fig. 10A).

Character 65. Stapes triangular, perforated by a large obturator foramen (0); or stapes columelliform and microperforate or imperforate (1). The morphological diversity of the marsupial stapes was reviewed by Gaudin et al. (1996; see also Czerny, 2015; Gaillard et al., 2021), whose character-state definitions we followed for scoring foraminal size: the maximum width of a large obturator (stapedial or intercrural) foramen—characteristic of triangular or bicrurate stapedes—is equal to or greater than the width of a surrounding crus (Novacek and Wyss, 1986: fig. 5D–G), whereas the maximum width of the opening in a microperforate stapes is less than the width of a surrounding crus (Novacek and Wyss, 1986: fig. 5C, H–J). However, unlike Gaudin et al. (1996), we do not distinguish between the microperforate and imperforate conditions, because these are often very difficult to tell apart (see, e.g., comments by Segall [1970: 177] and Gaudin et al. [1996: 41] regarding Notoryctes and Thylacinus); both morphologies are subsumed by state 1 here. We found other aspects of stapedial morphology incorporated by Gaudin et al. (1996: 38) into their scoring scheme, as well as other features that have been scored in previous analyses—for example, the relative sizes of the openings of the obturator foramen on either side of the stapes (Schmelzle et al., 2005: char. 12), and the stapedial ratio (e.g., Springer et al., 1997: char. 79; Rougier et al., 1998, char. 127; Horovitz and Sánchez-Villagra, 2003: char. 211; Schmelzle et al., 2005: char. 11; Gaillard et al., 2021: char. 3)—to be too subjective or too difficult to score consistently due to the presence of intermediate morphologies.

A large obturator foramen is present (state 0) in the stapedes of most Recent didelphids (e.g., Caluromys [Czerny, 2015: abb. 31–33]; Philander [Czerny, 2015: abb. 34–45]), Dromiciops (Gaillard et al., 2021: fig. 4), and most diprotodontians (e.g., Trichosurus [Czerny, 2015: abb. 38]; Bettongia [Czerny, 2015: abb. 40]). Schmelzle et al. (2005: table 1) coded Didelphis as possessing a collumeliform stapes, but most specimens examined by us exhibit a large obturator foramen (see also Voss and Jansa, 2009: char. 81; Gaillard et al., 2021: fig. 5). In Dactylonax (based on AMNH 221601), there is asymmetry in the size of the obturator foramen: the opening is large on one side of the stapes but much smaller on the other side of the same bone (Schmelzle et al., 2005, reported that connective tissue covers most of the obturator foramen in Dactylonax, and also in Dendrolagus). However, the overall morphology of the stapes of Dactylonax most closely resembles state 0.

By contrast, the stapes is columelliform and either imperforate or microperforate on both sides of the same bone in a few didelphids (Caluromysiops, Lestodelphys, and some specimens of Tlacuatzin; Voss and Jansa, 2009), all Recent caenolestids (e.g., Caenolestes; Gaillard et al., 2021), Notoryctes (Czerny, 2015: abb. 36), all peramelemorphians (e.g., Echymipera [Czerny, 2015: abb. 37]) and dasyuromorphians for which we were able to examine the ossicles (see also Gaillard et al., 2021), and a few diprotodontians (Vombatus, Lasiorhinus, Phascolarctos, Burramys, and Dactylopsila).

Character 66. Stapedial footplate concave (0); or flat or weakly convex (1); or strongly convex (2). Most mammals, including the majority of metatherians, have a more or less flat stapedial footplate (state 1; see Doran, 1878; Czerny, 2015; Gaillard et al., 2021), but the stapes of some diprotodontians are bullate (Wilkins et al., 1999), with a strongly convex footplate that protrudes into the vestibule of the middle ear (state 2; see Sánchez-Villagra and Nummela, 2001: fig. 1; Schmelzle et al., 2005: fig. 6; Czerny, 2015: abb. 38). This obviously derived morphology has previously been reported for Acrobates and phalangerids (Hyrtl, 1845; Segall, 1971; Sánchez-Villagra and Nummela, 2001; Schmelzle et al., 2005; Czerny, 2015)8. Our observations confirm that the stapes is strongly bullate in acrobatids and Recent phalangerids (the stapedial morphology of the fossil phalangerids †Onirocuscus and “Trichosurus” †dicksoni is unknown), but not in other examined metatherians. Although Schmelzle et al. (2005: char. 13) scored taxa with a weakly convex stapedial footplate (e.g., Petauroides) as exhibiting the same character state as those with a strongly bullate footplate, we consider these morphologies to be qualitatively distinct: in the weakly convex condition, the footplate protrudes only slightly, in contrast to the strongly convex condition, in which the footplate is enormously expanded and almost hemispherical (Sánchez-Villagra and Nummela, 2001: fig. 1; Schmelzle et al., 2005: fig. 6; Czerny, 2015: abb. 38). For this reason, we scored taxa with weakly convex stapedial footplates (e.g., pseudocheirids, petaurids, and Cercartetus) as corresponding to state 1.

Schmelze et al. (2005) observed concave stapedial footplates (state 0) in all the dasyurid taxa included in their study (e.g., Dasycercus byrnei, Dasyurus viverrinus [Schmelzle et al., 2005: fig. 5]), and the footplate also appeared distinctly concave in all dasyurid stapedes that we examined. Direct inspection of the stapes of one specimen of Myrmecobius (BMNH 1843.8.12.46) and CT-scans of another (AMNH 155328) indicate that the footplate is also concave in this taxon, but we were unable to score this character for Thylacinus or for any fossil dasyuromorphians. Among nondasyuromorphians, we observed a distinctly concave stapedial footplate only in Isoodon and Potorous (the latter based on AMNH 65330). For a number of taxa (e.g., Lasiorhinus, Chaeropus, Rhyncholestes) the stapedial footplate was sufficiently exposed in available specimens to determine that it is clearly not bullate but not to distinguish between the flat and concave morphologies; we scored such taxa using ambiguity coding (as “0/1”) Because these states represent a plausible morphocline, we ordered this character in all of our analyses.

Character 67. Rostral tympanic process of petrosal absent or indistinct (0); or present and well developed (1). A distinct rostral tympanic process (sensu MacPhee, 1981) is present on the promontorium of the petrosal of most of the metatherian taxa included in the current analysis (e.g., Marmosa; fig. 9A; state 1). However, a distinct rostral tympanic process is absent (state 0) in †Pucadelphys (Marshall and Muizon, 1995: fig. 19; Ladevèze and Muizon, 2007: text-fig. 1), †Badjcinus (Muirhead and Wroe, 1998: figs. 34), vombatids, †Neohelos, †Namilamadeta, Phalanger, and some specimens of †Nimbadon and †Onirocuscus.

In Muizon's (1998) detailed description of the holotype and only known skull of †Mayulestes (MHNC 1249) he discussed the presence of a small tubercle on the promontorium, just ventral to the fenestra cochleae. Based on this, Wroe et al. (2000: char. 65; see also Wroe and Musser; 2001), Sánchez-Villagra and Wible (2002: char. 7), Horovitz and Sánchez-Villagra (2003: char. 215), and several subsequent studies that have employed versions of Horovitz and Sánchez-Villagra's (2003) matrix (e.g., Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Beck, 2012) have all coded the rostral tympanic process as present in †Mayulestes. By contrast, Ladevèze and Muizon (2007: char. 133; 2010: char. 17) scored †Mayulestes as lacking a rostral tympanic process, and commented that “the presence of a low and tiny ridge or tubercle, anterolateral to the fenestra vestibuli is not regarded as equivalent to the development of a tympanic process, since it is not a process, i.e., a raised shelf of bone” (Ladevèze and Muizon, 2007: 1150). Rougier et al. (1998: char. 130; see also Wible et al., 2001) employed a different coding convention, distinguishing between a rostral tympanic process that is “absent or [a] low ridge” (state 0) and one that is a “tall ridge, occasionally contacting ectotympanic” (state 1), with †Mayulestes scored as state 0. We examined MHNC 1249 and confirmed that a small tubercle is indeed present on the promontorium (see also Muizon, 1994: fig. 2A, 1998: fig. 8B; Muizon et al., 1997: fig. 2C; 2018: supplementary data), in the same position occupied by the rostral tympanic process of didelphids. However, this minute protuberance is so weakly developed that it is not comparable to the macroscopically distinct structure coded as state 1 in our matrix; we therefore scored †Mayulestes as state 0. †Allqokirus appears to exhibit a similar morphology to †Mayulestes (see Muizon et al., 2018) and so has also been scored as state 0.

Following Sánchez-Villagra (1998), Sánchez-Villagra and Wible (2002: fig. 11), and Sánchez-Villagra and Forasiepi (2017), we interpret the bulla of Dromiciops as principally formed by (1) an alisphenoid tympanic process and (2) fused rostral and caudal tympanic processes of the petrosal (with smaller contributions by the ectotympanic, basioccipital, and exoccipital; see also Giannini et al., 2004), so we scored the rostral tympanic process as present in this taxon. By contrast, Segall (1969b) and Hershkovitz (1992b; 1999) interpreted the structure we identify as the rostral tympanic process of Dromiciops as an “entotympanic” (i.e., an independent ossification within the tympanic floor). Unequivocal entotympanics do not appear to occur in metatherians (Archer, 1976b; MacPhee, 1979; Aplin, 1990; Sánchez-Villagra, 1998; Sánchez-Villagra and Forasiepi, 2017; but see Norris, 1993), with the possible exception of acrobatids (see char. 56; Aplin, 1990).

In Thylacinus, a distinct crest extends along the medial edge of the promontorium, immediately adjacent to the contact between the petrosal and the basioccipital; we agree with Archer (1976b) and Muirhead and Wroe (1998) that this crest is homologous with the rostral tympanic process of the petrosal. By contrast, Murray and Megirian (2006a) interpreted this crest as a nonhomologous structure—a “ventromedial crest of the petrosal”—that they alleged is present in “derived Thylacinidae” (†Nimbacinus, †Badjcinus, and Thylacinus; Murray and Megirian, 2006a: fig. 16F, G, I), “derived Dasyuridae” (Antechinus and Dasyurus; Murray and Megirian, 2006: fig. 16C, D), and Myrmecobius (Murray and Megirian, 2006a: fig. 16H). They further argued that the apparent rostral tympanic process of the petrosal in Myrmecobius is in fact a “neomorphic structure developed from the ventromedial crest” (Murray and Megirian, 2006a: 258). Based on our own observations of these taxa, we consider the “ventromedial crest of the petrosal” sensu Murray and Megirian (2006a) to be simply the ventromedial edge of the rostral tympanic process, rather than a distinctly different structure. Thus, a rostral tympanic process of the petrosal appears to be present in all dasyuromorphians that we examined (including Thylacinus and Myrmecobius).

The rostral tympanic process is clearly present and very large in acrobatids and Tarsipes, in which it is seamlessly fused with the caudal tympanic process, forming a petrosal plate (see char. 68; Aplin, 1990; Sánchez-Villagra and Forasiepi, 2017: 27). Sánchez-Villagra and Wible (2002: 34) stated that the rostral tympanic process of the petrosal is “lacking entirely” in Vombatus and Phascolarctos (see also Wible, 1990: 198). We agree that a rostral tympanic process is absent in Vombatus (and also Lasiorhinus), but comparisons between the petrosal of Phascolarctos and those of †Litokoala and †Nimiokoala—both of which possess unambiguous rostral tympanic processes (Louys et al., 2009: figs. 4, 9)—con-vinces us that the rugose, crestlike structure along the ventromedial border of the promontorium of Phascolarctos corresponds to the rostral tympanic process. Aplin (1987: fig. 4) referred to this structure as the “medial crest of the pars cochlearis of petrosal” and stated that it represents the zone of attachment of the fibrous membrane of the tympanic cavity (Aplin, 1987: 380; 1990: 221), further supporting its identification as the rostral tympanic process (MacPhee [1981: 16] noted that tympanic processes typically originate “adjacent to the intratympanic surface of the fibrous membrane”). Interestingly, Aplin (1990: 221) found that the fibrous membrane of the tympanic cavity does not attach to the petrosal in Vombatus, but rather to the lateral rim of the basioccipital, which may explain why this taxon lacks any obvious homolog of the rostral tympanic process.

In all macropodiform specimens for which we were able to examine the ventral surface of the pars cochlearis, we observed a distinct promontorial process that contacts the ectotympanic (when the latter element is also present). Aplin (1990: 148) referred to this structure as the “ectotympanic process of the petrosal,” but we tentatively identify it as homologous with the rostral tympanic process (which contacts the posterior limb of the ectotympanic in many didelphids [Wible, 2003; Voss and Jansa, 2009: 42] and dasyuromorphians [Archer, 1976b]). We observed a similar raised process where the ectotympanic contacts the promontorium in petaurids, pseudocheirids, and most phalangerids—with the exceptions of Phalanger (in which no raised process was identifiable in the specimens we examined), and †Onirocuscus (in which it is only variably present [scored as “0+1”]). By contrast, Norris (1994) and Crosby and Norris (2003) identified a rostral tympanic (“ectotympanic”) process on isolated petrosals of Phalanger (e.g., Norris, 1994: fig. 5B, C); we suspect that this feature may show polymorphism in these and other phalangerids, given sufficiently large sample sizes. It should be noted that we identify the rostral tympanic process of phalangerids (where present) as the discrete structure immediately anteromedial to the fenestra cochleae (the “ectotympanic process” of Norris, 1994), and we do not consider the crest extending anteriorly from the promontorium (or “cupula cochleae”) to represent a continuation of this process, contra Norris (1994) and Crosby and Norris (2003; see also comments by Beck et al., 2008a: char. 221).

In some taxa the rostral tympanic process appears as a more or less conical or knoblike structure located anteroventral to the fenestra cochleae—as in some didelphids (e.g., Marmosa: fig. 9A; Philander [Wible et al., 2021: fig. 5A–C), caenolestids (e.g., Caenolestes; Sánchez-Villagra and Wible, 2002: fig. 4a), and some peramelemorphians. In others, however, it forms an elongate crest that extends the entire length of the promontorium to the anterior pole—as in Caluromys (see Sánchez-Villagra and Wible, 2002: fig. 3f), †Acdestis (Goin et al., 2003: text-fig. 3), Notoryctes (Ladevèze et al., 2008: fig. 4), †Yalkaparidon (Beck, 2009: fig. 7; Beck et al., 2014), some peramelemorphians (e.g., †Yarala; Muirhead, 2000: figs. 1, 3), and dasyurids (e.g., Antechinus; Sánchez-Villagra and Wible, 2002: fig. 4c). In a number of these taxa, the rostral tympanic process partially encloses a prominent sinus (the “periotic hypotympanic sinus” of Archer, 1976b), most obviously in dasyurids and the peramelemorphians Isoodon and Macrotis (Archer, 1976b; Muirhead, 1994; Wroe, 1997b; Muirhead, 2000; Travouillon et al., 2010: char. 47).

Several previous studies have distinguished between these conical or knoblike and elongate and crestlike morphologies of the rostral tympanic process of the petrosal as described above (e.g., Sánchez-Villagra and Wible, 2002: char. 7; Horovitz and Sánchez-Villagra, 2003: char. 215; Ladevèze and Muizon, 2007: char. 134). However, within the larger taxonomic context of the current study, we observed an apparently continuous range of intermediate morphologies, particularly in diprotodontian groups such as pseudocheirids. It is also noteworthy that Wible (2003: 157) described a low ridge that extends anteromedially from the main “finger-like” rostral tympanic process for the entire length of the promontorium to the basioccipital in the didelphid Monodelphis, whereas Sánchez-Villagra and Wible (2002: char. 7) and Horovitz and Sánchez-Villagra (2003: char. 215) both scored Monodelphis as having a process that “does not reach the anterior pole of the promontorium.” This apparent discrepancy emphasises the difficulty in consistently distinguishing between “conical” and “crestlike” morphologies of the rostral tympanic process. The presence of intermediate morphologies also prevents us from scoring presence or absence of a sinus formed by the rostral tympanic process (contra Travouillon et al., 2010: char. 47). Instead, we simply distinguish between presence and absence of the rostral tympanic process here.

Character 68. Caudal tympanic process of petrosal separated from pars cochlearis by a distinct gap (0); or caudal tympanic process of petrosal contacts but is not fused with pars cochlearis (1); or caudal and rostral tympanic processes of petrosal seamlessly fused, forming a petrosal plate (2). The caudal tympanic process of the petrosal is separated from the pars cochlearis by a distinct gap in most didelphids (e.g., Marmosa [fig. 9A]; Didelphis [Wible, 1990]; Monodelphis [Wible, 2003]; Philander [Wible et al., 2021: fig. 5A-C]), caenolestids (Ladevèze, 2005: fig. 11), †Yalkaparidon (Beck, 2009: fig. 7; Beck et al., 2014), peramelemorphians (contra Voss and Jansa, 2009: 42), thylacinids, thylacoleonids, vombatids, Phascolarctos, †Litokoala, †Muramura, and most diprotodontids. By contrast, the caudal tympanic process either contacts the rostral tympanic process or it contacts the promontorium posterodorsal to the rostral tympanic process in many other taxa. As described by Voss and Jansa (2009: char. 76), the rostral and caudal tympanic processes of the petrosal are in contact in some didelphids (e.g., Caluromysiops, Lestodelphys, and also †Sparassocynus; see Beck and Taglioretti, 2020), but in others the caudal tympanic process contacts the promontorium dorsal to the rostral tympanic process—as in Hyladelphys (e.g., AMNH 267339). Contra Voss and Jansa (2009: char. 76), we do not distinguish between these two morphologies, both of which are subsumed in our state 1. This revised description has the advantage of enabling meaningful scoring of taxa in which the rostral tympanic process of the petrosal is absent, such as our outgroup taxa †Mayulestes and †Allqokirus, and also vombatids (see char. 67).

In most taxa in which the caudal tympanic process contacts the pars cochlearis, the point of contact is unfused, as in †Sparassocynus, Caluromysiops, Lestodelphys, Thylamys, macropodiforms, phalangerids, burramyids, petaurids, and pseudocheirids (contact between the rostral and caudal tympanic processes of the petrosal can be determined for most diprotodontian taxa only by examining isolated petrosals or crania with broken auditory bullae).

Only one specimen of †Herpetotherium that we examined (AMNH FM 22304) preserves an intact caudal tympanic process; comparisons between this structure (on the left side of AMNH FM 22304) and the left petrosal of MB.Ma.50672 illustrated by Sánchez-Villagra et al. (2007: fig. 1i) and Horovitz et al. (2008: plate 3) suggest that the caudal tympanic process is damaged in the latter specimen. On the left side of AMNH FM 22304, the caudal tympanic process of the petrosal is clearly in contact with, but nevertheless suturally distinct from, the rostral tympanic process; however, this morphology may be the result of taphonomic crushing of this specimen, and hence we score †Herpetotherium as “0/1” here.

In adult specimens of Notoryctes (see Ladevèze et al., 2008: fig. 4), Tarsipes, acrobatids, and dasyurids (e.g., Phascogale; fig. 9B), no suture between the rostral and caudal tympanic processes is identifiable; instead, these processes are seamlessly fused to form a petrosal plate sensu Klaauw (1931), a morphology we score as a distinct state. Giannini et al. (2004) correctly observed that a sutureless petrosal plate is present in juvenile specimens of Dromiciops, but they implied that a suture develops within the petrosal plate such that distinct rostral and caudal tympanic processes can be identified in adult individuals. However, the apparent suture identified by these authors is, in fact, simply an abrupt change of curvature of the external bullar surface that coincides with an internal septum; the absence of any osteological discontinuity within the petrosal plate can be confirmed by microscopic comparison of the sutureless septal junction with the actual suture between the petrosal plate and the alisphenoid tympanic process. We therefore scored Dromiciops (and also †Microbiotherium; Segall, 1969b: fig. 5) as corresponding to state 2 (see also Sánchez-Villagra, 1998; Sánchez-Villagra and Wible, 2002; Sánchez-Villagra and Forasiepi, 2017).

Character 69. Prootic canal foramen on tympanic face of petrosal present (0); or absent (1). Presence of the prootic canal, through which the prootic sinus communicates with the lateral head vein in the middle ear, is thought to be a mammalian plesiomorphy (Wible, 1990; Wible and Hopson, 1995; Wible et al., 2001, 2009; Ekdale et al., 2004; Rougier and Wible, 2006). In metatherians that retain an unambiguous prootic canal (state 0), it is short, roughly horizontal, and fully enclosed by the petrosal, connecting a foramen within the sulcus for the prootic sinus on the squamosal side of the petrosal to a foramen adjacent to the secondary facial foramen on the tympanic side (Wible, 1990; 2003; Ladevèze, 2005; Wible et al., 2021).

Sánchez-Villagra and Wible (2002: chars. 9, 10) scored presence or absence of a foramen for the prootic canal on the squamosal face of the petrosal and presence or absence of a foramen on the tympanic face as two separate characters. However, we scored the presence or absence of a foramen only on the tympanic face of the petrosal because: (1) the two characters scored by Sánchez-Villagra and Wible (2002) are at least partially dependent (assuming that the prootic canal usually transmits a vessel), and (2) scoring the presence or absence of the prootic canal on the squamosal side of the petrosal usually requires access to isolated petrosals. By contrast, it is relatively easy to score presence or absence of a foramen on the tympanic side of the petrosal of taxa in which the middle ear is incompletely enclosed by bone (such as many didelphids). For taxa in which the auditory bulla is more extensive (such as most diprotodontians), however, isolated petrosals are still required to score this character.

A foramen for the prootic canal on the tympanic face of the petrosal is clearly present in †Pucadelphys (Ladevèze and Muizon, 2007: text-fig. 1), †Mayulestes (contra Muizon, 1998; see Muizon et al., 2018: 412), †Allqokirus (Muizon et al., 2018), all Recent didelphids that could be scored for this character (e.g., Didelphis [Wible, 1990: fig. 4] and Monodelphis [Wible, 2003: fig. 7]), Caenolestes (Ladevèze, 2005), †Yalkaparidon (Beck et al., 2014: fig. 9), some peramelemorphians (e.g., Perameles [Sánchez-Villagra and Forasiepi, 2017: fig. 12] and Peroryctes), some dasyurids (e.g., Antechinus, Phascolosorex), the few macropodiforms that could be assessed for this character (e.g., Notamacropus [Sánchez-Villagra and Forasiepi, 2017: fig. 14b]), Pseudochirops cupreus, Ailurops (see Norris, 1994), †Nimbadon, †Lekanoleo, and some specimens of Thylacinus and Petaurus.

Isolated petrosals of Phascolarctos exhibit an enclosed canal that closely resembles the prootic canal in both location and morphology, with a foramen present on both the tympanic and squamosal faces of the petrosal. However, based on sectioned specimens, Aplin (1990: 233) identified the occupant of this canal in Phascolarctos to be a cutaneous branch of the auricular ramus of the vagus nerve, rather than a venous connection. Although this observation raises obvious questions concerning the homology of petrosal canals in the absence of direct observation of their contents, we scored a prootic canal as present in Phascolarctos, given its structural and topographical similarities to the definitive prootic canal of, for example, didelphids.

A foramen that may be the tympanic foramen for the prootic canal is located between the squamosal and petrosal in †Sparassocynus (based on MLP-88-V-25-2), suggesting that at least part of this canal may lie between these two bones (rather than fully enclosed by the petrosal) in this taxon. If so, this morphology would be unique among all known metatherians. The position of this foramen in †Sparassocynus is also unusual because it is lateral to the fossa incudis rather than medial to the epitympanic recess (as in, for example, Monodelphis; Wible, 2003: fig. 7D). Because we are not certain of the identity of this foramen, we scored †Sparassocynus as unknown (“?”).

A tympanic foramen for the prootic canal appears to be absent (state 1) in Notoryctes (Ladevèze et al., 2008), some peramelemorphians (e.g., Chaeropus, Isoodon, Macrotis), some dasyurids (e.g., Dasyurus, Sminthopsis), †Namilamadeta, Lasiorhinus, Strigocuscus celebensis, most pseudocheirids that could be scored for this trait (i.e., Petauroides, Pseudocheirus, Pseudochiropsarcheri), and Dactylopsila.

Overall, our observations are in close agreement with those of Sánchez-Villagra and Wible (2002), but a few discrepancies are noteworthy. In particular, Sánchez-Villagra and Wible (2002: 35, table 2) coded a foramen for the prootic canal on the tympanic surface of the petrosal as absent in †Mayulestes based on Muizon's (1998) description of the holotype and only known skull of this taxon (MHNC 1249). However, our own examination of MHNC 1249 revealed what appears to be a foramen for the prootic canal dorsomedial to the epitympanic recess, in a similar position to that seen in Didelphis (Wible, 1990: fig. 4) and Monodelphis (Wible, 2003: fig. 7), with the epitympanic recess forming its ventral floor, a finding apparently confirmed by Muizon et al. (2018: 412). Similarly, whereas Sánchez-Villagra and Wible (2002) scored this foramen as absent in Echymipera kalubu and Perameles gunnii, we identified a foramen on the tympanic surface of the petrosal in specimens of both these taxa (e.g., E. kalubu specimens AMNH 190982 and 198708; P. gunnii specimens MVZ 127060 and 127072); Sánchez-Villagra and Forasiepi (2017: fig. 12) subsequently identified the lateral head vein within the prootic canal in a Perameles specimen that was not identified to species. Sánchez-Villagra and Wible (2002) also scored this foramen as absent in the macropodoids Aepyprymnus rufescens and Setonix brachyurus, but it is present in specimens that we examined of both species (e.g., A. rufescens specimen AMNH 160121; S. brachyurus specimens AMNH 160044 and 160258). Sánchez-Villagra and Wible (2002) and Ladevèze (2005) both reported the prootic canal as absent in adult Dromiciops specimens, but it is present in a sectioned late juvenile (Sánchez-Villagra and Wible, 2002: fig. 9); we have not examined isolated petrosals of Dromiciops, and prefer to score it as unknown (“?”).

Character 70. Tensor tympani muscle unenclosed ventrally (0); or enclosed ventrally by a bridge of bone derived from petrosal (1). In most metatherians, the tensor tympani muscle, which originates from the anterolateral face of the promontorium, is not enclosed ventrally (state 0). However, Aplin (1990: 302) identified a “narrow bridge of bone derived from the petrosal” that ventrally encloses the tensor tympani of Tarsipes; he also suggested that the “small oval nodule of bone” that Parker (1890: 81) observed in the tensor tympani tendon of Tarsipes was, in fact, a remnant of this bridge. Examination of the left ear region of T. rostratus (e.g., AMNH 119716, the bulla of which has been removed) confirms Aplin's (1990) observations: the ventral bridge of bone can be identified anterior to the stapes, anteromedial to the epitympanic recess, and medial to a dorsally extensive accessory airspace within the petrosal. Dried remnants of the tensor tympani muscle can be seen emerging dorsal to the margins of the bridge in AMNH 119716. Sánchez-Villagra and Wible (2002: fig. 10) illustrated the ear region of AMNH 119716 (probably the left side, although the figure is reversed as if it were the right side), but the structures they identified as the “hiatus fallopii (hF)” and “secondary facial foramen (sff)” in their figure actually represent the anteromedial and posterolateral margins of the bony bridge respectively (Tarsipes entirely lacks a secondary facial foramen; see char. 72). Ventral enclosure of the tensor tympani by bone (state 1) is an autapomorphy of Tarsipes in the current analysis.

Character 71. Cavum supracochleare roofed by petrosal (0); or not roofed by petrosal (1). In most metatherians, the cavum supracochleare is completely roofed by the petrosal (state 0; e.g., in Didelphis; Wible, 1990: fig. 4D). However, Wible (1990: 195) noted that the petrosal roof is incomplete—and, hence, that the cavum supracochleare and cavum epiptericum are connected by a fenestra semilunaris sensu Rougier et al. (1992)—in caenolestids and “some marmosine didelphids.” In taxa in which the cavum supracochleare lacks a complete roof, a prominent depression that houses the geniculate ganglion is visible in isolated petrosals (e.g., Ladevèze, 2004: fig. 2; 2005: fig. 11).

Of the taxa examined in this study for which isolated petrosals were available, the cavum supracochleare lacks a complete roof only in Caenolestes (see Ladevèze, 2005: fig. 11; Wible, 1990: 195; we were unable to score other Recent canolestids or fossil paucituberculatans), Hyladelphys, Marmosa (see Wible, 1990: 195), and †Thylacoleo. Although not included in this analysis, some isolated “Type II” metatherian petrosals from the Eocene Itaborai fossil fauna of Brazil also lack a complete roof (Ladevèze, 2004); Beck (2017a) argued that these petrosals probably belong to the polydolopimorphian Epidolops ameghinoi (but see criticism of this by Muizon et al., 2018: 422–423).

Character 72. Facial nerve exits the substance of the petrosal within the middle ear (0); or is enclosed within a bony tube formed by the petrosal until it exits the skull via the stylomastoid foramen (1). In most metatherians, the facial nerve exits the substance of the petrosal within the middle ear and extends posteriorly within an open facial sulcus of the petrosal before leaving the middle ear via the stylomastoid notch or foramen (state 0). Usually, the facial nerve exits the petrosal via the secondary facial foramen, while the greater petrosal nerve exits via a separate foramen (the hiatus fallopii), as in, for example, Didelphis (Wible, 1990: fig. 4C), Monodelphis (Wible, 2003: fig. 8A), and Philander (Wible et al., 2021). We observed minor taxonomic differences in the relative positions of the hiatus fallopii and secondary facial foramen but, contra previous studies (e.g., Sánchez-Villagra and Wible, 2002: char. 12; Horovitz and Sánchez-Villagra, 2003: char. 218), we were unable to partition this variation into unambiguously distinct states (see also Wible [2003: 174] regarding polymorphism in the position of the hiatus fallopii in Monodelphis).

In some specimens of Lasiorhinus and Vombatus, the floor of the cavum supracochleare (the petrosal chamber that houses the geniculate ganglion, from which arise the facial and greater petrosal nerves) is unossified (as previously noted by Wible, 1990: 188; Sánchez-Villagra and Wible, 2002: 35). In such specimens, the hiatus fallopii and secondary facial foramen are not present as separate openings; instead, a single foramen (the primary facial foramen) opens into the pars cochlearis. However, we observed considerable polymorphism in the degree of ossification of the floor of the cavum supracochleare in Lasiorhinus and Vombatus (Wible, [2010: 11–12] discussed similar polymorphism of this feature in the xenarthran Dasypus novemcinctus), and so we scored both of these taxa as state 0.

A qualitatively distinct morphology is seen in acrobatids and Tarsipes, the facial nerve of which is fully enclosed by a bony petrosal tube that extends all the way to the stylomastoid foramen; hence, there is no exit for the facial nerve anywhere within the middle ear space (state 1; see also Aplin, 1990: 304).

Character 73. Subarcuate fossa deeply excavated (0); or a shallow depression (1). The subarcuate (or floccular) fossa of the petrosal, which houses the petrosal lobule of the paraflocculus of the cerebellum, is a deep cavity (state 0) in most metatherians (e.g., Archer, 1976b; Wible, 1990, 2003; Rougier et al., 1998; Sánchez-Villagra, 2002; Ladevèze, 2004; 2005; Ladevèze and Muizon, 2007, 2010; Sánchez-Villagra et al., 2007; Horovitz et al., 2008, 2009; Muizon et al., 2018). Minor taxonomic variation in subarcuate fossa size and shape has sometimes been partitioned into discrete traits in previous phylogenetic studies (e.g., Ladevèze, 2004: char. 2), but we are unable to distinguish these consistently. However, the subarcuate fossa is just a shallow depression on the cerebellar side of the petrosal in Vombatus and Lasiorhinus, a qualitatively distinct condition (state 1; see Sánchez-Villagra, 2002). The subarcuate fossa is similarly shallow in †Neohelos (Murray et al., 2000a), †Ngapakaldia, †Nimbadon (Black and Hand, 2010: fig. 5B), †Muramura, †Namilamadeta, and some specimens of †Thylacoleo. Although the subarcuate fossa of Sarcophilus is somewhat shallower than that of most other metatherians (Archer, 1976b: 259), it is not comparable to the vombatid condition (Sánchez-Villagra, 2002), so we scored Sarcophilus as state 0.

Character 74. Glenoid fossa essentially flat or slightly concave, sometimes with a raised articular eminence (0); or convex and mediolaterally very broad (1). In most metatherians, the glenoid fossa is flat or slightly concave (state 0), but in most diprotodontians (e.g., phalangerids, pseudocheirids, macropodiforms, diprotodontids, phascolarctids) the glenoid region includes a raised articular eminence anteriorly and a shallow, transversely oriented mandibular fossa posteriorly (Aplin, 1987; 1990). This widespread diprotodontian morphology appears to reflect the spatial displacement of the mandibular condyle for different dental functions: the condyle rests within the mandibular fossa during molar occlusion, but it slides forward onto the articular eminence to bring the upper and lower incisors into occlusion (Aplin, 1987; 1990). Although Aplin (1987: 387) suggested that this feature may be a “key innovation in the origin of Diprotodontia as a whole,” we observed continuous variation in glenoid morphology among most of the diprototodontian taxa we examined (see also comments by Aplin, 1990: 310), so we did not code this trait as a separate character state.

By contrast, the glenoid morphology of Recent vombatids (e.g., Vombatus: fig. 45) is qualitatively distinct (Aplin, 1987; 1990; Murray, 1998), being convex and narrow anteroposteriorly (with no evidence of a mandibular fossa), but very broad mediolaterally (state 1). Lasiorhinus and Vombatus are also unusual in entirely lacking a postglenoid process (see char. 75).

Character 75. Postglenoid process of squamosal well developed (0); or weakly developed or absent (1). The glenoid fossa of most metatherians is buttressed posteriorly by a well-developed postglenoid process of the squamosal (state 0; e.g., in Phascogale [fig. 10A] and Petaurus [fig. 10B]). By contrast, this process is entirely absent in Hypsiprymnodon moschatus (fig. 52), Potorous (fig. 54), Tarsipes (fig. 49), Vombatus (fig. 45), and Lasiorhinus (state 1). Although still identifiable as a distinct structure, the postglenoid process is very weakly developed in †Yalkaparidon (Beck, 2009: fig. 7A; Beck et al., 2014), so we also scored this taxon as state 1.

Character 76. Postglenoid foramen present (0); or absent (1). A postglenoid foramen (which transmits the postglenoid vein) is consistently present in most metatherians (state 0; e.g., Didelphis [Wible, 1990: fig. 1], Monodelphis [Wible, 2003: fig. 6], Philander [Wible et al., 2021: figs. 2, 4], Phascogale [fig. 10A], and Petaurus [fig. 10B]), although its position is taxonomically variable (see char. 77). Aplin (1990: 240) stated that vombatids “lack a conventional postglenoid foramen [but] a variable number of foramina usually perforate the posterior surface of the condylar process/anterior wall of the squamosal epitympanic sinus” and reported that these foramina “typically convey a small branch of the postglenoid artery, and sometimes, a small satellite vein.” We found that a bristle inserted into the largest of the multiple subsquamosal foramina that are present in Vombatus and Lasiorhinus (see chars. 81, 82) exits via a distinct foramen within the zygomatic epitympanic sinus of the squamosal, so we interpret the latter foramen as homologous with the postglenoid foramen. We observed a similar foramen within the zygomatic epitympanic sinus in †Warendja. Therefore, Vombatus, Lasiorhinus, and †Warendja have all been scored as state 0.

By contrast, †Neohelos and †Nimbadon both appear to lack an identifiable postglenoid foramen (state 1). Black and Hand (2010: 37) referred to a “deep, recessed pocket” in the posterior wall of the postglenoid process of the squamosal in some †Nimbadon specimens (e.g., QM F53642) as a “postglenoid foramen,” but this structure is a blind sinus (possibly homologous with the laterally open zygomatic epitympanic sinus seen in vombatids; see chars. 85, 86), rather than a foramen.

Character 77. Postglenoid vein exits skull via the postglenoid foramen above the ear region, posterior or posteromedial to the glenoid fossa and the postglenoid process (if present) (0); or postgenoid vein emerges from the postglenoid foramen in the posteromedial corner of the glenoid fossa, medial or anteromedial to the postglenoid process (if present) (1). In most metatherians in which the postglenoid foramen is present (see char. 76), this foramen opens above the ear region and behind the postglenoid process (state 0), as in, for example, Phascogale (fig. 10A), Didelphis (Wible, 1990: fig. 1), Philander (Wible et al., 2021: figs. 2, 4), and Monodelphis (Wible, 2003: fig. 6). Among these taxa, there is minor variation in the position of this foramen that we have not scored here: for example, it is located in a more medial position (but still slightly posterior to the glenoid fossa and postglenoid process) in Thylacinus than in most other dasyuromorphians (Murray and Megirian, 2006a: appendix 1, fig. 1C; Warburton et al., 2019: fig. 7).

A qualitatively distinct state, however, is observed in most examined diprotodontians: the postglenoid vein passes anteroventrally in a bony canal that is primarily formed by the squamosal (although an ectotympanic component is often also present; Wroe et al., 1998: 744) and emerges medial (or anteromedial) to the postglenoid process, in the posteromedial corner of the glenoid fossa (state 1; e.g., Petaurus, fig. 10B). Springer and Woodburne (1989: table 2, char. 11) described the latter condition as “sheathing of the ventral postglenoid foramen” and reported that it does not occur in phascolarctids or macropodoids (see also Marshall et al., 1990: fig. 2, node 71). Our observations, by contrast, are more consistent with those of Wroe et al. (1998) who reported enclosure (“sheathing”) of the postglenoid vein (or foramen) in macropodiforms, and the same morphology is clearly expressed in the phascolarctids Phascolarctos, †Litokoala, and †Nimiokoala (Louys et al., 2009). Although Hypsiprymnodon moschatus and Potorous both lack a postglenoid process of the squamosal (see char. 75), the enclosure of the postglenoid vein in these taxa and its opening in the posteromedial corner of the glenoid fossa more closely resembles the widespread diprotodontian condition than it does the morphology seen in other metatherians; we therefore scored these taxa as conforming to state 1.

However, this condition is not universal among diprotodontians. In Tarsipes, the postglenoid foramen is at the lateral end of the deep groove posterior to the glenoid fossa (Aplin, 1990: 311); this morphology most closely corresponds to state 0. In Vombatus, Lasiorhinus, and †Warendja, a foramen that appears to be homologous to the postglenoid foramen opens within the zygomatic epitympanic sinus of the squamosal (see char. 76); although these taxa all lack a postglenoid process of the squamosal (see char. 75), this foramen clearly does not open within the glenoid fossa, but rather posteromedial to it, so we also score them as state 0 (contra Springer and Woodburne, 1989: table 2, char. 11, who scored “sheathing of ventral postglenoid foramen” as present in Vombatidae). Similarly, the postglenoid foramen of †Silvabestius, †Muramura, †Namilamadeta, and thylacoleonids opens within the zygomatic epitympanic sinus of the squamosal, posteromedial to the glenoid fossa and postglenoid process, rather than within the glenoid fossa (see comments by Aplin, 1987: 383); because this morphology resembles the vombatid condition, we scored these taxa as state 0.

The morphology of this region in Notoryctes is difficult to interpret, in part because the postglenoid process of the squamosal appears to be fused to the pars canalicularis of the petrosal (with the external auditory meatus below, rather than behind, the postglenoid process; see char. 78). However, the postglenoid vein does appear to emerge medial to the postglenoid process in Notoryctes, in the posteromedial corner of the glenoid fossa. Nevertheless, CT-scans indicate that, unlike the condition in most diprotodontians, it is not enclosed proximally within an elongate bony canal (Beck, in prep.); despite this difference, we tentatively scored Notoryctes as state 1. This character is inapplicable (“-”) for †Neohelos and †Nimbadon, both of which lack a distinct postglenoid foramen (see char. 76).

Character 78. External auditory meatus dorsal to the glenoid fossa (0); posterior to the glenoid fossa (1); or ventral to the glenoid fossa (2). In most metatherians, the external auditory meatus is posterior to the glenoid fossa, behind the postglenoid process of the squamosal (if present) and anterior to the posttympanic process of the squamosal and the mastoid exposure of the petrosal (state 1; e.g., in Phascogale [fig. 10A] and Petaurus [fig. 10B]). However, in Dactylopsila and Dactylonax, the postglenoid process is in direct contact with the mastoid, and the external auditory meatus is located dorsal (rather than posterior) to the glenoid fossa (state 0). As a result, the external auditory meatus appears to lie within the posterior root of the zygomatic arch (Aplin, 1990: plate 5.2a; Beck, 2009: fig. 3). Interestingly, Aplin (1990: 311) reported that Dactylopsila juveniles resemble other metatherians in the position of the external auditory meatus, suggesting that the unique adult morphology develops relatively late in ontogeny. The position of the external auditory meatus is also unusual in Notoryctes, directly ventral to the glenoid fossa (fig. 36); this morphology has been scored as an additional state (state 2). Because the states comprising this character represent a plausible morphocline, reflecting a progressive shift in the position of the external auditory meatus along a dorsoventral axis, we ordered them in all our analyses.

Character 79. Facial nerve exits middle ear via a stylomastoid notch (0); or via a stylomastoid foramen formed by the squamosal and caudal tympanic process of the petrosal (1); or via a stylomastoid foramen formed by the caudal tympanic process of the petrosal only (2); or via a stylomastoid foramen formed by the rostral and caudal tympanic processes of the petrosal (3); or via a stylomastoid foramen formed by the ectotympanic, posttympanic process of the squamosal, and pars canalicularis of the petrosal (4); or via a stylomastoid foramen formed by the ectotympanic and pars canalicularis of the petrosal (5). The facial nerve exits the middle ear via a notch (the stylomastoid notch) in the petrosal, posterior to the tympanohyal, in most Recent didelphids (e.g., Marmosa [fig. 9A]), Caenolestes, Rhyncholestes, all examined peramelemorphians, thylacinids, Myrmecobius, †Barinya, †Thylacoleo, †Wakaleo, Phascolarctos, and some specimens of Vombatus (state 0). In other marsupials for which specimens with an intact basicranium were available, the exit of the facial nerve is entirely enclosed by bone, forming a complete stylomastoid foramen. However, the position and bony composition of this foramen vary markedly within Marsupialia, and we consider that the five different types of stylomastoid foramen we identify here fail the test of primary homology (sensu de Pinna, 1991). We therefore score the different foraminal types as alternative states within a single character that also includes absence of a stylomastoid foramen (state 0).

In Caluromysiops (see comments by Gabbert, 1998: 7) and †Sparassocynus, the stylomastoid foramen is formed by the squamosal and caudal tympanic process of the petrosal (state 1), whereas in Lestoros (based on BMNH 1922.1.1.129) this foramen appears to be entirely within the caudal tympanic process of the petrosal (state 2). In Recent dasyurids (e.g., Phascogale [figs. 9B, 10A]), the stylomastoid foramen is within the petrosal plate formed by the rostral and caudal tympanic processes of the petrosal, which are seamlessly fused in adult specimens; however, juvenile specimens, in which the rostral and caudal tympanic processes are not fused, indicate that the stylomastoid foramen probably lies on the boundary between them (state 3; Archer, 1976b: 255; see also comments by Wroe, 1999: 513). The stylomastoid foramen of Recent dasyurids is sometimes also bordered anterodorsally by the squamosal (Archer, 1976b; Wroe, 1999; Wroe et al., 2000: char. 54). Another form of enclosure is seen in Dromiciops, in which the stylomastoid foramen is between the posterior limb of the ectotympanic and the posttympanic process of the squamosal (which are fused; see char. 60) anteriorly and the caudal tympanic process of the petrosal posteriorly (state 4). In phalangeridan diprotodontians (e.g., Petaurus [figs. 9C, 10B]), the stylomastoid foramen is between the squamosal, the mastoid exposure of the petrosal, and the ectotympanic (Aplin, 1990; Wroe et al., 1998); because these are the same bones that enclose the stylomastoid foramen in Dromiciops, we also scored phalangeridan diprotodontians as state 4. Isolated petrosals reveal the presence of an additional, internal foramen for the facial nerve formed by the caudal tympanic process in macropodoids (Aplin, 1990: 148, plate 3.8e), whereas this internal foramen is absent in other phalangeridans. Nevertheless, lacking isolated petrosals for most taxa, we scored this character based on the external aspect of the stylomastoid orifice only.

In †Neohelos (based on NTMP 87108-1), Lasiorhinus, and some specimens of Vombatus (e.g., AMNH 66197), the stylomastoid foramen is formed by the ectotympanic and the pars canalicularis of the petrosal (state 5). Presence or absence of a stylomastoid foramen appears to be age related in Vombatus because it is more commonly present in older and larger individuals (reflecting increasing ossification of the ear region with advancing age).

It is unclear exactly which bones enclose the stylomastoid foramen in Notoryctes, but the morphology of this region most closely corresponds to state 4 based on the position of the foramen and the fact that the ectotympanic appears to contribute to its ventral border (e.g., in AMNH 15015; see also Archer, 1976b: 313). Thus, we tentatively scored Notoryctes as state 4.

Character 80. Process of the exoccipital does not extend as far as the exit of the facial nerve (stylomastoid notch or foramen) (0); or extends laterally or anterolaterally as far as the stylomastoid notch or foramen (1). In most metatherians, the exoccipital ends laterally at a suture with the pars canalicularis of the petrosal immediately lateral to the paroccipital process (state 0), as in Marmosa, Phascogale, and Petaurus (fig. 9). In acrobatids, however, a process of the exoccipital extends laterally as far as the stylomastoid foramen, covering the posterior part of the auditory bulla in ventral view (state 1), as in Distoechurus (fig. 48; see also Aplin and Archer [1987: lviii]; Aplin [1990: 332]). A somewhat similar morphology is seen in †Thylacoleo, in which the exoccipital is sufficiently laterally extensive to contribute to the external opening of the stylomastoid notch, so we also scored this taxon as state 1.

Character 81. Subsquamosal foramen present (0); or absent (1). In many metatherians, the squamosal is pierced immediately dorsal to the external auditory meatus by a prominent foramen (state 0), as in Marmosa (fig. 2), Caenolestes (Osgood, 1921: plate XX fig. 2), Phascogale (fig. 10A), and Thylacinus (Murray and Megirian, 2006a: appendix 1, fig. 1). This foramen connects to the postglenoid foramen through the squamosal and has been shown to transmit a temporal branch of the postglenoid artery and an accompanying vein in didelphids (Wible, 2003) and dasyurids (Archer, 1976b). It was named the subsquamosal foramen by Cope (1880), and this terminology has been followed by most subsequent authors (e.g., Gregory, 1910; Archer, 1976b; Voss and Jansa, 2009; Pavan and Voss, 2016). By contrast, Wible (2003) referred to this orifice as the “suprameatal foramen” and used “subsquamosal foramen” for one or more small foramina that are variably present within the squamosal dorsal to the suprameatal bridge (the horizontal squamosal crest that extends posteriorly from the dorsal margin of the zygomatic process to the occiput) in Monodelphis. We follow Cope's (1880) terminology here to maintain consistency with most of the metatherian literature.

In some diprotodontians there is no foramen immediately above the external auditory meatus, but a smaller foramen occurs farther dorsally, often close to (or even within) the squamosal suture with the parietal (see char. 83). This foramen can be seen in dorsal view in (for example) some specimens of Phalanger (in which it is enclosed by the squamosal [fig. 47]; Springer and Woodburne, 1989: fig. 2B) and Petaurus (in which it is contained within the squamosal-parietal suture [fig. 50]). Despite its position—dorsal to the suprameatal bridge—this foramen appears to be homologous with the subsquamosal foramen: a bristle introduced into this foramen emerges from the postglenoid foramen. Multiple foramina piercing the dorsal surface of the squamosal in Vombatus (fig. 45) and Lasiorhinus also appear to be homologous with the subsquamosal foramen because they are connected internally to the foramen within the zygomatic epitympanic sinus that we identify as the homologue of the postglenoid foramen (see char. 76); additionally, Aplin (1990: 240) reported that they transmit branches of the postglenoid artery. CT-scans demonstrate that the small foramen posterior to the glenoid fossa and dorsal to the external auditory meatus in Notoryctes connects to the postglenoid foramen, and therefore can also be identified as the subsquamosal foramen (Beck, in prep.).

By contrast, we observed complete absence of the subsquamosal foramen (state 1) in Hemibelideus, Petauroides, Pseudochirops cupreus (but not P. archeri), Acrobates, Distoechurus (fig. 48), and Tarsipes (fig. 49), none of which exhibits any trace of a squamosal perforation above the ear region. The subsquamosal foramen is also only variably present in Phalanger (often being present on only one side of an individual), so we scored this taxon polymorphic (“0+1”).

Character 82. Subsquamosal foramen single (0); or multiple (1). In most metatherians in which the subsquamosal foramen is present (see char. 81), it is marked by a single opening (state 0). However, as discussed above, there are multiple apparently homologous perforations of the squamosal (state 1) in Vombatus (fig. 45) and Lasiorhinus. Two or three subsquamosal foramina are present in †Namilamadeta, which we also scored as state 1. Pledge (1992: 111) reported that the squamosal of †Warendja is “not perforated by foramina as in modern wombats,” based on a partial skull of †W. wakefieldi (SAM P21405), implying that only a single subsquamosal foramen is present. However, we consider the squamosal of SAM P21405 to be too incomplete to determine this with any confidence, and Brewer (2007) described a cranial fragment of †W. wakefieldi (AM F129857) that preserves multiple subsquamosal foramina; although we have not examined the latter specimen, we scored †Warendja as state 1 here.

Character 83. Subsquamosal foramen entirely within squamosal (0); or within squamosal-parietal suture (1). In most metatherians that retain a subsquamosal foramen (see char. 81), this foramen opens entirely within the squamosal (state 0). In the pseudocheirids Pseudocheirus and Pseudochirulus, however, it opens within the squamosal-parietal suture (state 1). The subsquamosal foramen is also within the squamosal-parietal suture in some specimens of Petaurus (fig. 50), so we scored this taxon as polymorphic (“0+1”) for this character.

Character 84. Posterior squamosal epitympanic sinus absent or indistinct, without any lateral enclosure (0); or distinct and at least partially enclosed laterally (1). Epitympanic sinuses are pneumatic (air-filled) cavities that communicate with the epitympanic recess dorsal to the tympanic membrane (van der Klaauw, 1931; Beck and Taglioretti, 2020: 5). The presence of a distinct pneumatic cavity within the squamosal posterolateral to the epitympanic recess was proposed as a synapomorphy of Australian marsupials to the exclusion of South American forms by Marshall et al. (1990), so the distribution of this feature within Metatheria warrants careful assessment. Archer (1976a) simply referred to this cavity as the squamosal epitympanic sinus, but we follow Wroe et al. (1998) in distinguishing a posterior epitympanic sinus (which is posterolateral to the epitympanic recess, as in, e.g., Phascogale [figs. 9B, 10A) from a zygomatic epitympanic sinus (which is anterior to the epitympanic recess and invades the zygomatic process of the squamosal; char. 85). In addition to these two sinuses, Wroe et al. (1998: 743) also referred to a third, the dorsal epitympanic sinus present in some diprotodontians (e.g., Burramys). However, we were unable to consistently determine the presence or absence of a dorsal epitympanic sinus among our terminals.

A posterior squamosal epitympanic sinus is absent or indistinct (state 0) in †Pucadelphys (Marshall and Muizon, 1995), †Allqokirus (Muizon et al., 2018), most Recent didelphids (e.g., Marmosa [fig. 9A]; Didelphis [Wible, 1990: fig. 1]; Monodelphis [Wible, 2003]; Philander [Wible et al., 2021: fig. 4]), Recent caenolestids, Dromiciops, †Yalkaparidon (Archer et al., 1988; Beck et al., 2014), †Badjcinus (Muirhead and Wroe, 1998), wynyardiids (Aplin, 1987; Pledge, 2005), and some macropodiforms (e.g., Dorcopsis, Setonix, Thylogale, and †Ganawamaya). A shallow depression that is ventral or posteroventral to the subsquamosal foramen in †Galadi, Echymipera, Peroryctes, and some didelphids (e.g., Lestodelphys [Voss and Jansa, 2009: fig. 13B]) might be homologous with the posterior epitympanic sinus (see comments by Archer, 1976b: 304; 1982a: 461; Beck and Taglioretti, 2020: 19), but these indistinct concavities are difficult to distinguish from absence, so we scored them all as state 0.

By our definition, a distinct posterior squamosal epitympanic sinus (state 1) is at least partially enclosed laterally by bone; such sinuses occur in †Mimoperadectes (Horovitz et al., 2009), Caluromysiops (Reig et al., 1987: fig. 44C; see also Wible, 1990: 200, who implied that a “suprameatal fossa” is present in this taxon), †Sparassocynus (Beck and Taglioretti, 2020), all examined dasyuromorphians except †Badjcinus (e.g., Phascogale [figs. 9B, 10A]), some peramelemorphians (e.g., Chaeropus [fig. 37]), and most diprotodontians, with the exception of wynyardiids, some macropodiforms (discussed above), and some specimens of †Neohelos. In many diprotodontians (e.g., Petaurus [fig. 10B]), this sinus is fully enclosed by the squamosal such that it is not externally visible (Springer and Woodburne, 1989; Wroe et al., 1998); therefore, we determined sinus morphology in these taxa by examination of broken or damaged skulls. Although Notoryctes lacks external evidence of a posterior squamosal epitympanic sinus, broken skulls reveal the presence of what appear to be auditory sinuses dorsolateral and posterolateral to the epitympanic recess (see also Archer, 1976b: 269); fusion of cranial bones in this region, however, makes it impossible to determine conclusively whether or not these auditory sinuses are within the squamosal, so we scored Notoryctes as unknown (“?”).

Examination of incompletely cleaned crania suggests that, where present, this sinus is covered by the pars flaccida of the tympanic membrane in dasyuromorphians and peramelemorphians (van der Klaauw, 1931; Archer, 1976b; Marshall, 1977; Aplin, 1990). The apparently homologous shallow depression seen in some Recent didelphids also appears to be covered by the pars flaccida, as in Chironectes (e.g., AMNH 212909) and Lutreolina (e.g., BMNH 1939.4221]). However, this does not appear to be the case in diprotodontians (Aplin, 1990: 367). Thus, the homology of the posterior epitympanic sinus in all the taxa scored as possessing this trait is open to reasonable doubt.

Character 85. Zygomatic epitympanic sinus absent (0); or present (1). Most metatherians have, at most, a single squamosal epitympanic sinus, namely the posterior sinus described above (see char. 84). However, an additional epitympanic sinus that invades the squamosal zygomatic process is present (state 1) in all examined diprotodontians for which suitable specimens were available, with the notable exception of Tarsipes. We follow Wroe et al. (1998) in referring to the latter cavity as the zygomatic epitympanic sinus. There is considerable variation in zygomatic epitympanic sinus morphology within Diprotodontia, which we coded as an additional character below (char. 86).

Character 86. Zygomatic epitympanic sinus fully enclosed by squamosal without external opening (0); or enclosed laterally by squamosal but with prominent external opening on posterior surface of squamosal zygomatic process (1); or shallow and largely open laterally (2). The zygomatic epitympanic sinus is fully enclosed by the squamosal and lacks an obvious external opening (state 1) in acrobatids, phalangerids, pseudocheirids, petaurids (e.g., Petaurus [fig. 10B]), and Cercartetus; broken or damaged skulls reveal that the zygomatic epitympanic sinus of these taxa is extensive and usually filled with cancellous bone. By contrast, a prominent external opening is visible on the posterior surface of the squamosal zygomatic process (state 1) in Burramys, †Wakaleo, phascolarctids (see Murray et al., 1987. and Louys et al., 2009: figs. 3, 4, who referred to this orifice as the “postglenoid cavity”), and most macropodiforms (see Springer and Woodburne, 1989 [fig. 2A], where it is the more anterior of the two “openings in squamosal,” and Wells and Tedford, 1995 [fig. 9D], where it is labelled as the “postzygomatic foramen”). In Petrogale, the region of the squamosal posterodorsal to the external auditory meatus is strongly pneumatized and contacts the posterior part of the zygomatic process of the squamosal (Prideaux and Warburton, 2010: char. 12). As a result, there is no large, posteriorly facing external opening leading into the zygomatic epitympanic sinus in Petrogale; at most, there is a tiny, laterally oriented, slitlike opening within the point of contact between the two squamosal processes. Thus, we scored Petrogale as state 0. Although we agree with Prideaux and Warburton (2010: char. 12) that the region of the squamosal posterodorsal to the external auditory meatus is also inflated and contacts the posterior part of the zygomatic process of the squamosal in Dendrolagus and †Bohra, a large, posterior-facing opening into the zygomatic epitympanic sinus remains visible in both of these taxa, which we scored as state 1.

A third distinct morphology is seen in Vombatus (fig. 45), Lasiorhinus, and †Warendja (Aplin, 1987: 381; 1990; Pledge, 1992), in which the zygomatic epitympanic sinus does not deeply invade the zygomatic process of the squamosal but is largely open laterally (state 2). We observed a similarly shallow, laterally open zygomatic epitympanic sinus in †Muramura, †Namilamadeta, †Ngapakaldia, and †Lekanoleo.

In †Neohelos (Murray et al., 2000a: fig. 40), †Nimbadon (Black and Hand, 2010: figs. 4, 17, 18), †Silvabestius (Black, 2008: fig. 5.2A), and †Thylacoleo, an enclosed epitympanic sinus extends deep into the squamosal zygomatic process and communicates with the external surface of the skull via a large “epitympanic fenestra” (sensu Murray, 1992); this morphology most closely resembles state 1. In all four of these taxa, however, another shallow concavity—the “postglenoid sinus” of Black (2008: fig. 5.2A) and Black and Hand (2010: fig. 18A)—also invades the posterior face of the postglenoid process, a trait similar to that seen in taxa that we scored as state 2. The apparent presence of two zygomatic epitympanic sinuses in these taxa raises obvious questions about the homology of such structures among diprotodontians, but in the current study we scored †Neohelos, †Nimbadon, †Silvabestius, and †Thylacoleo using ambiguity coding (as “1/2”).

This character was scored as inapplicable (“-”) for all taxa lacking a distinct zygomatic epitympanic sinus (i.e., all taxa scored as state 0 for char. 85). Because the states of this character form a plausible morphocline reflecting the degree of enclosure of the zygomatic epitympanic sinus, we treated this character as ordered in all our analyses.

Character 87. Squamosal and pars canalicularis of the petrosal not extensively pneumatized posterior and dorsal to external auditory meatus (0); or extensively pneumatized and cancellous (1). In most metatherians (e.g., Marmosa [fig. 1]), the squamosal and pars canalicularis of the petrosal are not greatly pneumatized posterior and dorsal to the external auditory meatus (state 0), apart from the presence of a posterior epitympanic sinus of the squamosal in some taxa (char. 84). By contrast, the squamosal and pars canalicularis are both extensively pneumatized and are composed of cancellous bone in this region (state 1) in all examined petaurids (e.g., Petaurus [fig. 10B]), pseudocheirids, acrobatids, phalangerids, and Petrogale (Aplin, 1987: 384; Aplin, 1990). Notably, this region is not strongly pneumatized in Tarsipes, Burramys, or Cercartetus (Aplin, 1987: 384; Aplin, 1990). Our observations with regard to this feature largely agree with those of Springer and Woodburne (1989: char. 14), except that this region also appears to be extensively pneumatized in Ailurops (e.g., in AMNH 152884), albeit to a slightly lesser extent than in other phalangerids (Flannery et al., 1987: 482); thus, we scored Ailurops as state 1.

Character 88. Squamosal and the pars canalicularis of the petrosal unfused (0); or seamlessly fused, with no evidence of a suture between the two bones even in juveniles (1). In most metatherians, the squamosal and the pars canalicularis of the petrosal are clearly distinguishable throughout ontogeny, with an obvious suture between the two bones (state 0). In phalangerids, the suture between the mastoid exposure of the pars canalicularis of the petrosal (which is restricted to a small ventral strip on the occiput; fig. 12C) and the squamosal is sometimes difficult to identify in adult individuals but is obvious in juvenile and subadult specimens (see char. 90), so we scored phalangerids as state 0. By contrast, the squamosal and the pars canalicularis of Acrobates and Distoechurus are seamlessly fused (state 1), even in the youngest juvenile specimens we examined (contra Voss and Jansa, 2009: 36); indeed, Aplin (1990: 291) specifically compared the acrobatid morphology to the human temporal bone.

Occipital Region

The occipital region includes the basi-, exo-, and supraoccipital bones as well as the mastoid exposure of the pars canalicularis of the petrosal. Broadly speaking, this part of the skull provides surfaces for the cervical articulation, for the insertion of nuchal muscles, and for the origin of the digastric and various hyoid muscles. Occipital bones are also pierced by foramina that transmit cranial nerves and blood vessels, and they form a protective ring around the medula oblongata. Characters of the occipital region have not played a major historical role in debates about metatherian phylogeny, but there is more anatomical variation here than might be expected from this traditional neglect. A nonexhaustive list of synonyms for anatomical features in this region is given in table 11.

Character 89. Lambdoid sesamoids absent (0); or present (1). Sesamoids for the insertion of the dorsal neck musculature are absent (state 0) in most metatherians, but they are present (state 1) in all examined peramelemorphians (including †Yarala and †Galadi; Muirhead, 2000; Travouillon et al., 2010) as bilaterally paired, suturally distinct ossifications on the lambdoid crest between the parietals and the supraoccipital (e.g., in Echymipera [fig. 11]). These sesamoids are markedly smaller, but still identifiable, in Chaeropus (fig. 37) than in other peramelemorphians we examined (e.g., Macrotis [fig. 38] and Perameles [fig. 39]). Although often illustrated by authors (e.g., Freedman, 1967), these distinctive structures were first described explicitly by Filan (1990), who hypothesized their role as force-multipliers for M. rectus capitis dorsalis and other head extensors. Although bilateral thickenings of the lambdoid crest are present in several other marsupials (for example, many dasyurids), they lack obvious, suturally distinct sesamoids; therefore, we scored such taxa as state 0.

Character 90. Mastoid exposure of petrosal extensively exposed on occiput, contacting supraoccipital on each side (0); or restricted to ventralmost part of occiput, not contacting supraoccipital (1). In most metatherians the pars canalicularis of the petrosal is extensively exposed on the occiput, where it is commonly known as the mastoid (or as the mastoid exposure of the petrosal). In this taxonomically widespread morphology the mastoid extends dorsally to contact the supraoccipital (state 0), as in †Mayulestes (Muizon, 1998: fig. 9), didelphids (e.g., Lestodelphys [fig. 12A] and Metachirus [fig. 12B]), Dromiciops (Giannini et al., 2004: fig. 3), and most diprotodontians (e.g., Macropus; Wells and Tedford, 1995: fig. 9). Although Wroe et al. (2000: char. 68) and Wroe and Musser (2001: char. 68) scored †Pucadelphys as lacking a mastoid contribution to the occiput, a large mastoid exposure is, in fact, clearly present on the occiput of this taxon (Marshall and Muizon, 1995; personal obs.). The bones comprising the occiput are fused in adult specimens of Notoryctes, but CT-scans of a fluid-preserved juvenile (NMV C11082) indicate the presence of a large mastoid exposure on the occiput of this taxon as well (Beck, in prep.).

TABLE 11

Selected Anatomical Synonyms for Structures of the Occipital Regiona

img-AHAT_01.gif

By contrast, in all examined phalangerids (e.g., Trichosurus [fig. 12C]) the mastoid exposure is reduced to a narrow strip on the ventrolateral aspect of the occiput (Flannery et al., 1987), where it does not contact the supraoccipital (state 1). Instead, phalangerids exhibit squamosal-exoccipital contact. The transverse suture between the squamosal and mastoid is hard to see on adult skulls of some phalangerids, with the result that the mastoid is sometimes incorrectly illustrated as extending between the squamosal and the exoccipital (e.g., as by Springer and Woodburne, 1989: fig. 2B). However, the squamosal-mastoid suture is ontogenetically persistent in other taxa, notably Trichosurus (fig. 12C). The mastoid exposure is slightly more extensive dorsally in Ailurops than it is in other phalangerids (Flannery et al., 1987: fig. 4D), but it does not contact the supraoccipital in any examined specimen, so we scored this taxon as corresponding to state 1.

As noted above (char. 88), the pars canalicularis of the petrosal is seamlessly fused to the squamosal in Acrobates and Distoechurus, so the relative contribution of the mastoid to the occiput is unclear in these taxa, both of which we scored as unknown (“?”).

Character 91. Mastoid fenestra absent (0); or present dorsal and medial to the mastoid exposure of petrosal (1). In most metatherians the occiput is imperforate (state 0), with the obvious exception of the foramen magnum (e.g., in Metachirus [fig. 12B] and Trichosurus [fig. 12C]). Although a foramen is sometimes present within the mastoid, this trait is often polymorphic, and we did not code it as a character. However, a qualitatively distinct condition is seen in Recent caenolestids (but not †Acdestis) and some didelphids (e.g., Marmosa [fig. 6B]; Lestodelphys [fig. 12A]), which exhibit a large unossified vacuity between the dorsal margin of the mastoid and the supraoccipital (state 1). Osgood (1921: 150–151) referred to this vacuity as a “large persistent mastoid foramen,” but we prefer the term mastoid fenestra, based on the previously explained distinction between foramina and fenestrae (see Palate, above). The same name has been applied to topologically similar vacuities (albeit entirely within the mastoid) in rodents (e.g., by Carleton and Musser, 1989). Most macropodiforms (e.g., Macropus [Wells and Tedford, 1995: fig. 9D]) have a foramen within the mastoid but not a mastoid fenestra, so we scored them as state 0. The sole exception is Setonix, in which a large unossified vacuity is present dorsal to the mastoid (e.g., in AMNH 196398, 196399); therefore, we scored Setonix as state 1.

FIG. 11.

Occiput of Echymipera kalubu (AMNH 192168) illustrating the mature peramelemorphian morphology (a conspecific juvenile is shown in fig. 6C). States for characters 25 and 89 (see main text for descriptions of these characters and character states) are illustrated as follows: 25(1), 89(1). Abbreviations: exo, exoccipital; ju, jugal; ls, lambdoid sesamoid; mas, mastoid; par, parietal; so, supraocciptal; sq, squamosal.

img-z98-1_01.jpg

Character 92. Two hypoglossal foramina (0); or only one hypoglossal foramen opening on the extracranial surface (1). Most metatherians have two hypoglossal foramina opening on the extracranial surface (e.g., Didelphis [Wible, 2003: fig. 1]; Monodelphis [Wible, 2003: fig. 6]; Philander [Wible et al., 2021]), which Forasiepi et al. (2019) referred to as rostral and caudal hypoglossal foramina and which transmit branches of the hypoglossal nerve and accompanying arteries and veins (state 0). Marshall and Muizon (1995: fig. 19) identified three hypoglossal foramina (but called them “condyloid foramina”) in a specimen of †Pucadelphys (YPFB Pal 6110), but our examination of YPFB Pal 6110 suggests that only the medial two foramina would have transmitted branches of the hypoglossal nerve in life; the lateralmost of the three foramina appears to be a venous foramen for the condyloid canal, and we refer to it as a condyloid foramen here (see also Forasiepi et al., 2019: 45–47). The presence of one or more condyloid foramina lateral to the hypoglossal foramen or foramina and the paroccipital process is common in metatherians (e.g., Marmosa [fig. 9A]; Petaurus [fig. 9C]; Didelphis [Wible, 1990: fig. 1]; Monodelphis [Wible, 2003: 171]). Based on the specimens examined for this study, †Pucadelphys appears to consistently exhibit two hypoglossal foramina, and so has been scored as state 0. Three hypoglossal foramina are occasionally present in Phascolarctos (e.g., the left side of AM M6805), but we have not scored such rare variants. However, both Tarsipes and Notoryctes consistently have only a single hypoglossal foramen opening on the extracranial surface (state 1; contra Sánchez-Villagra and Forasiepi [2017: 18]). Thylacinus is polymorphic (“0+1”) for this character.

Our observations conflict with those of Horovitz and Sánchez-Villagra (2003: char. 205), who scored Caenolestes, Cercartetus, Dasyuroides, Echymipera, and Vombatus as having only a single hypoglossal foramen, and Didelphis and Thylogale as polymorphic. Based on the specimens examined for this study, all these taxa consistently have two hypoglossal foramina opening on the extracranial surface.

FIG. 12.

Occipital views of Lestodelphys halli (A, UWZM 22422), Metachirus nudicaudatus (B, AMNH 267009), and Trichosurus vulpecula (C, AMNH 65557). Alternative states of characters 32, 90, 91, 93, and 95 (see main text for descriptions of these characters and character states) are illustrated as follows: Lestodelphys: 32(0), 90(0), 91(1), 93(1), 95(0); Metachirus: 32(1[but note that the modal condition for M. nudicaudatus is 0]), 90(0), 91(0), 93(2), 95(1); Trichosurus vulpecula: 32(-), 90(1), 91(0), 93(2), 95(0). Abbreviations: exo, exoccipital (yellow, coossified with basioccipital); fm, foramen magnum; mas, mastoid exposure of pars canalicularis of petrosal (blue, coossified with pars cochlearis); mf, mastoid fenestra; par, parietal (purple); pp, paroccipital process (of exoccipital); so, supraoccipital; sq, squamosal (orange). Specimens are not drawn to the same scale.

img-z99-1_01.jpg

Character 93. Paroccipital process of exoccipital absent (0); or small and broadly adnate to the petrosal (1); or a large erect process usually directed ventrally (2). There is some debate in the mammalian literature regarding the use of the term “paroccipital process.” Rougier et al. (1998: char. 131; following Wible and Hopson, 1993) and several subsequent works (e.g., Wible and Gaudin, 2004; Wible et al., 2004, 2009; Giannini et al., 2006; Wible, 2008; Wible and Spaulding, 2013) have applied this term to a process of the petrosal from which the sternocleidomastoid muscle originates, a structure that we (and most other morphologists) refer to as the mastoid process. Instead, we follow traditional usage (e.g., Coues, 1872; Muizon et al., 2015) by employing paroccipital process for the bony site of attachment of the posterior belly of the digastric on the exoccipital; by contrast, Wible and coworkers have referred to this process as the “paracondylar process” (following Evans, 1993).

Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 48). In most didelphids (e.g., Marmosa [fig. 9A], Lestodelphys [fig. 12A]) the paroccipital process is small and adnate to the petrosal, but in others (e.g., Metachirus [fig. 12B]) this process is much larger, erect, and directed ventrally. Among other metatherians, a paroccipital process appears to be entirely absent (state 0) in †Herpetotherium (Gabbert, 1998: 4), Dromiciops (fig. 43), Notoryctes (fig. 36), Tarsipes, and Acrobates. This process is also absent in Microperoryctes (which we likewise scored as state 0), but Microperoryctes has a sesamoid attached to the region of the exoccipital where the paroccipital process occurs in other peramelemorphians (see also Voss and Jansa, 2009: 45). The paroccipital process is absent or indistinct in some specimens of †Pucadelphys (e.g., YPFB Pal 6105), and was said to be absent by Marshall and Muizon (1995); however, a small but distinct process is present in other specimens (e.g., MHNC 8376, MHNC 8379, MHNC 8381), so we scored †Pucadelphys as polymorphic (“0+1”). Distoechurus is also polymorphic, with the specimen illustrated here (AMNH 227121 [fig. 48]) lacking a distinct process.

The paroccipital process is large and erect in †Thylophorops, †Galadi (Travouillon et al., 2010: fig. 7B), Thylacinus (Murray and Megirian, 2006a: appendix 1, fig. 1), †Badjcinus (Muirhead and Wroe, 1998: fig. 3A), †Nimbacinus (Wroe and Musser, 2001: figs. 1b, 7), †Barinya, and many diprotodontians. The last includes pseudocheirids, Recent macropodids (e.g., Macropus [Wells and Tedford, 1995: fig. 9]), Vombatus (but not Lasiorhinus), diprotodontids (e.g., †Nimbadon [Black and Hand, 2010: fig. 6]), wynyardiids, and most phalangerids (e.g., Trichosurus [fig. 12C]). We scored Macrotis as state 2 based on its large and erect paroccipital process, even though this process remains in contact with the petrosal due to the proportionally enormous size of the rostral tympanic process. Most remaining taxa that we examined have small paroccipital processes (state 1), such as Phascogale (fig. 9B), Petaurus (fig. 9C), and †Sparassocynus (Beck and Taglioretti, 2020).

We observed additional variation in the morphology of this region that we did not code for phylogenetic analysis but that we discuss here for completeness (see also char. 94). In Echymipera, Peroryctes, and Perameles the paroccipital process is capped with a sesamoid that is suturally distinct in juveniles and subadults but is often partially or completely fused with the paroccipital process—and hence unrecognizable as a discrete structure—in adults (Voss and Jansa, 2009: 45). As already noted, a digastric sesamoid is also present in Microperoryctes, but there is no distinct paroccipital process. There does not, however, appear to be a sesamoid capping the paroccipital process in Isoodon. In Macrotis, the paroccipital process and mastoid process are in contact ventrally, enclosing a distinct fenestra or foramen (we cannot tell whether this orifice transmits any soft-tissue structure; fig. 38). The ventral surface of the paroccipital process is distinctly concave in Cercartetus. In Phascolarctos a long, vertically directed notch or groove is present in the base of paroccipital process just medial to the medi-almost part of the pars canalicularis of the petrosal; although we are uncertain as to the functional significance of this feature, it may be connected with the unusual hyo-laryngeal morphology of koalas (Aplin, 1990; Louys et al., 2009).

Character 94. Paroccipital process of exoccipital not greatly pneumatized (0); or greatly pneumatized, enclosing a large sinus (1). In most metatherians, the paroccipital process is not greatly pneumatized (state 0), but in Dasycercus, Dasyuroides, and Dasykaluta it encloses a large pneumatic sinus that is continuous with the middle ear cavity (state 1; see Jones, 1949; Ride, 1964b; Archer, 1976b). This character was scored as inapplicable (“-”) for all taxa lacking a distinct paroccipital process.

Character 95. Dorsal margin of adult foramen magnum formed by exoccipitals and supraoccipital (0); or by exoccipitals only (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 49), who reported that the supraoccipital is excluded from the dorsal margin of the foramen magnum (state 1) in Didelphis, Lutreolina, Metachirus (fig. 12B), and Philander. By contrast, the dorsal margin of the foramen magnum is formed by both the supraoccipital and exoccipitals (state 0) in other Recent didelphids (e.g., Lestodelphys [fig. 12A]) and in most other metatherians (e.g., Trichosurus [fig. 12C]; see Sánchez-Villagra and Wible, 2002: char. 3). Although the supraoccipital is excluded from the foramen magnum in a few older individuals of Isoodon (e.g., AMNH 108890), the modal condition for this taxon is state 0. However, the supraoccipital does not contribute to the dorsal margin of the foramen magnum in †Thylophorops, Dasyurus, Sarcophilus, Thylacinus, or †Barinya, all of which have been scored as state 1. Notamacropus is polymorphic (“0+1”) for this character. In general, our observations match those of Sánchez-Villagra and Wible (2002: char. 3), except that we score Phascolarctos as state 0 here.

As noted by Abdala et al. (2001: 195), Flores et al. (2003: 4), and Giannini et al. (2004: 6), the supraoccipital contributes to the dorsal rim of the foramen magnum in juveniles of Didelphis and Lutreolina but is excluded from the foramen magnum in subadults and adults due to medial growth of the exoccipitals, and the same may apply more generally within Metatheria. Due to the potentially confounding effect of such ontogenetic variation, we scored this character based on subadult and adult specimens only. However, the occipital sutures were fused in all available specimens of some taxa, which we scored as unknown (“?”) because the osseous composition of the dorsal rim of the foramen magnum could not be determined.

Prideaux and Warburton (2010: char. 11) noted that the foramen magnum is markedly larger in some macropodids (e.g., Dendrolagus) than others; although we concur with this observation, variation in foramen magnum size was not amenable to partitioning into discrete states in the context of the taxonomically broader sample of metatherians used here.

Mandible

The mandible is a much simpler structure than the syncranium and provides correspondingly fewer phylogenetic characters. Most mandibular characters are presumably related, directly or indirectly, to the key mechanical role of the dentary as a supporting platform for the lower dentition and as the site of insertion for masticatory muscles, but a few (e.g., the size and position of mental foramina) likely reflect the development and distribution of tactile organs (vibrissae) on the mandibular integument. Terminology for metatherian mandibular traits is generally not problematic.

Character 96. Mandible large and well developed, with distinct coronoid and angular processes (0); or much reduced and splintlike, lacking coronoid and angular processes (1). In most metatherians, the mandible is a robust bone, with prominent coronoid and angular processes (state 0); this morphology is seen even in Myrmecobius, despite its weakly developed dentition (fig. 41). However, the mandible of Tarsipes (fig. 49) is reduced to little more than a splint of bone that lacks distinct coronoid or angular processes (state 1), an obviously autapomorphic condition.

Character 97. Mandibular symphysis unfused (0); or fused (1). Most metatherians have an unfused mandibular (or mental) symphysis (state 0), such that the two hemimandibles are capable of some degree of independent movement. By contrast, the symphysis is completely fused, or “synostosed” sensu Scapino (1981; see also Prideaux, 2004: 214)—with the hemimandibles consequently immobile relative to one another and no symphyseal suture identifiable between them—in adult specimens of Recent vombatids (but not †Warendja; Hope and Wilkinson, 1982) and Phascolarctos (see Archer, 1984c; Springer et al., 1997: char. 48). Because the symphysis is unfused in juveniles of these taxa, we scored this character based only on fully adult specimens. Although we also observed symphyseal fusion in some adult specimens of †Nimbadon, fusion in this taxon occurs only after the permanent dentition is fully erupted and has undergone some wear (in “Stage VI” adults sensu Black et al., 2010), so we scored †Nimbadon as polymorphic (“0+1”) for this character. Our definition of symphyseal fusion corresponds to “state 4” of Scott et al. (2012: 662), who described it as “completely ossified”; however, those authors recorded the symphysis of Sarcophilus as completely ossified, whereas we scored Sarcophilus as state 0 based on the modal morphology (absence of complete fusion) in our examined material.

Character 98. Mandible usually with one mental foramen (0); or usually with two or more mental foramina (1). We distinguish true mental foramina—which transmit branches of the inferior alveolar nerve and accompanying vessels (Wible, 2003: 178), and which are relatively large and usually consistent in number and position among conspecific specimens—from nutrient foramina (which are often also present on the lateral surface of the mandible but are considerably smaller and highly variable in number and position, even between left and right sides of the same individual). When only a single mental foramen is present, it is invariably a large opening in the position of the anterior foramen of individuals or taxa with two or more foramina.

Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 50; 2009: char. 85). Although Voss and Jansa (2009: char. 85) subsequently included an additional state to account for the presence of three mental foramina, which they observed in specimens of Perameles gunnii, we observed considerable polymorphism in our sample of P. gunnii (with roughly equal numbers of individuals with two and three mental foramina). In other metatherians we examined that sometimes have three mental foramina, we always observed multiple individuals with two foramina when large sample sizes were available (e.g., Chaeropus, Isoodon, Peroryctes, Myrmecobius, Dasyurus, Murexia, Phascolosorex, and Sarcophilus). We therefore decided not to include an additional state for presence of three mental foramina, but simply distinguish between presence of a single foramen (state 0) and presence of two or more foramina (state 1).

We confirmed Voss and Jansa's (2003, 2009) observations regarding the number of mental foramina in Recent didelphids, except that one and two mental foramina appear to be equally common in Philander, which we scored as polymorphic (“0+1”). Two or more mental foramina are usually present in other metatherians, including †Pucadelphys (see Marshall and Muizon, 1995: fig. 8A), †Mayulestes (see muizon, 1998: fig. 3A), †Allqokirus (see Muizon et al., 2018: fig. 28A), and †Herpetotherium. However, only a single mental foramen is typically present in petaurids (e.g., Petaurus [fig. 50]), acrobatids (e.g., Distoechurus [fig. 48]), Tarsipes (fig. 49), Cercartetus, some phalangerids (Strigocuscus, Wyulda, †Onirocuscus), most macropodiforms (†Balbaroo and †Ganawamaya are noteworthy exceptions; Kear et al., 2007: fig. 4.3), †Ngapakaldia (see Black, 2010), vombatids (e.g., Vombatus [fig. 45]), wynyardiids, †Thylacoleo, Antechinomys, Planigale, †Yalkaparidon (see Beck et al., 2014), and Dromiciops (fig. 43). We observed polymorphism in this feature in Notoryctes, Ningaui, Bettongia, Spilocuscus, and Petauroides.

Character 99. Masseteric fossa imperforate (0); or perforated by a distinct masseteric foramen (1); or by a masseteric canal (2); or by a large unossified vacuity (3). The masseteric fossa is a prominent concavity for the insertion of the deep masseter on the posterolateral aspect of the mandible (Turnbull, 1970). This fossa is imperforate (with the occasional exception of tiny nutrient foramina) in †Pucadelphys (see Marshall and Muizon, 1995: fig. 8A), †Allqokirus (see Muizon et al., 2018: fig. 28A), †Herpetotherium (see Horovitz et al., 2008: plate 2A), didelphids, microbiotheriids, peramelemorphians, and all dasyuromorphians except Myrmecobius (state 0). Among diprotodontians, the masseteric fossa is consistently imperforate in most phalangerids, the burramyid Cercartetus, and the ilariid †Ilaria. In most other examined nonmacropodiform diprotodontians, the fossa is penetrated by a distinct masseteric foramen (state 1). However, Dactylopsila, Dactylonax, Pseudochirops archeri, Phalanger, Trichosurus, and †Nimbadon (Black and Hand, 2010: 41; Murray et al., 2000: 81) are all polymorphic for the presence or absence of the masseteric foramen (“0+1”). Abbie (1939) showed that, at least in diprotodontians, this foramen transmits a branch of the inferior alveolar artery (“inferior dental artery” of Abbie, 1939) and always communicates internally with the lumen of the mandibular foramen from the opposite (medial) side of the dentary (as illustrated for Vombatus by Owen, 1866: fig. 220). Voss and Jansa (2009: 46) noted that one or more small foramina within the masseteric fossa of Caenolestes also appear to communicate internally with the mandibular foramen. We observed a similar foramen or foramina within the masseteric fossae of Lestoros, †Stilotherium, Myrmecobius, Notoryctes, and †Yalkaparidon, all of which have been scored as state 1. Rhyncholestes and †Palaeothentes are polymorphic (“0+1”) for the presence of this foramen or foramina.

All examined macropodiforms uniquely exhibit a distinctly different morphology, in which the masseteric fossa is penetrated by a masseteric canal sensu Abbie (1939; state 2; figs. 5254). In contrast with the masseteric foramen described above, the macropodiform masseteric canal is much larger, is bounded laterally by the masseteric crest, provides a clear view into the mandibular canal (= “inferior dental canal” of Abbie, 1939), and is the site of insertion for a slip of the deep masseter (Abbie, 1939; Warburton, 2009). Because we could not consistently recognize Abbie's (1939) distinction between a masseteric canal that is only partially confluent with the mandibular canal (supposedly diagnostic of macropodids; but see comments by Woodburne, 1984a: 1071; Archer and Flannery, 1985: 1344; Cooke, 1999: 249; Kear et al., 2007: 1164) and a masseteric canal that is fully confluent with the mandibular canal (apparently seen in Hypsiprymnodon moschatus, potoroids, and most fossil macropodiforms; Woods, 1960; Flannery et al., 1983, 1984; Case, 1984; Archer and Flannery, 1985; Kear et al., 2007: char. 15), we did not code these as separate conditions. Instead, both are combined among the range of morphologies that we assign to state 2. Uniquely, the masseteric fossa of Tarsipes is pierced by a large, unossified vacuity or fenestra (fig. 49), the homology of which is unclear; we therefore scored this as an additional, autapomorphic condition (state 3).

Abbie (1939) argued persuasively that states 0, 1, and 2 form an ordered transformation series (0 ↔ 1 ↔ 2), but Tarsipes (state 3) is difficult to accommodate in this sequence, so we treated this character as unordered in all of our analyses.

Character 100. Retromolar fossa imperforate (0); or pierced by large retrodental canal (1). The retromolar fossa—a shallow concavity between the last mandibular molar and the base of the coronoid process—is essentially imperforate in most metatherians (state 0), although tiny nutrient foramina are often present in this region. However, a distinct foramen is consistently present behind m4 in Caenolestes, Lestoros, †Palaeothentes, and †Stilotherium (state 1). This foramen is the opening for a vertical canal that, in Caenolestes at least, connects to the mandibular canal (Beck, 2017a: fig. 13c, 405). We refer to this vertical canal as the retrodental canal, following Hoffstetter and Villarroel. (1974: 1950) and subsequent authors (e.g., Sánchez-Villagra et al., 2000; Voss and Jansa, 2009; Goin and Abello, 2013). A similar foramen is also seen in some specimens of Rhyncholestes, which has been scored as polymorphic (“0+1”) contra Voss and Jansa (2009), who scored Rhyncholestes as lacking a foramen for the retrodental canal.

Character 101. Condylar process variously formed but articular surface more or less oblong and dorsoventrally flattened (0); or transversely elongate and medially extensive (1). The condylar process of the mandible exhibits modest taxonomic variation within Metatheria, but in most taxa it can be described as mediolaterally directed and irregularly oblong, with a flattened dorsal surface; in dorsal view, the condylar process extends much farther laterally than it does medially, relative to the coronoid process. By contrast, the condyle is transversely elongate, such that it extends as far or farther medial to the coronoid process as it does lateral to this process in the Recent vombatids Vombatus and Lasiorhinus (but not †Warendja; Hope and Wilkinson, 1982), †Neohelos (Murray et al., 2000a: fig. 43B), †Nimbadon (Black and Hand, 2010: fig. 7), and †Muramura (Pledge, 2003: fig. 19.4).

The only known mandibular material of †Silvabestius is from a juvenile (QM F30504), in which the condylar process is neither transversely elongate nor medially extensive (Black and Archer, 1997a: fig. 3). However, in their ontogenetic study of †Nimbadon, Black et al. (2010: 997) noted that the condylar process “becomes longer, wider, more laterally expansive, and more heavily ossified” with increasing age. Allowing for the possibility that similar ontogenetic changes may also have occurred in †Silvabestius, we scored this taxon as unknown (“?”) pending description of adult mandibular specimens.

Although we have scored Lagostrophus as state 0 here, the morphology of its condylar process is distinctive: a nonarticular outgrowth (medial to the coronoid process) projects anteromedially, such that the medial and lateral parts of the process form an obtuse angle in dorsal view (see also comments by Warburton, 2009).

Character 102. Angular process strongly medially inflected (0); or weakly or not inflected (1). The morphology of the metatherian angular process has been reviewed by Sánchez-Villagra and Smith (1997), Sánchez-Villagra and Wible (2002: char. 4), and Voss and Jansa (2003: char. 51; 2009: 46). Most metatherians, including †Pucadelphys (Marshall and Muizon, 1995), †Allqokirus (Muizon et al., 2018), and †Herpetotherium (Horovitz et al., 2008), have a strongly medially inflected mandibular angle (state 0). However, we agree with previous authors that Caluromys, Caluromysiops, Myrmecobius, Phascolarctos, Dactylopsila, and Dactylonax all possess a weakly inflected or uninflected angle (state 1). We coded this character as inapplicable (“-”) for Tarsipes, which does not possess a distinct angular process.

DENTAL CHARACTERS

Didelphids and microbiotheriids have a total of 50 teeth, as did many nonmarsupial metatherians (Kielan-Jaworowska et al., 2004). We refer to this unreduced marsupial complement as comprising five incisors (I1–5), one canine (C1), three premolars (P1–3), and four molars (M1–4) in the upper dentition, and four incisors (i1–4), one canine (c1), three premolars (p1–3), and four molars (m1–4) in the lower dentition, as in the majority of the metatherian literature. However, a number of alternative schemes have been proposed based on different hypotheses of dental homology (table 12). In particular, recent attempts to homologize dental loci across Mammaliaformes have led to novel hypotheses of homology between the teeth of metatherians and those of other mammals (e.g., Averianov et al., 2010; O'Leary et al., 2013).

Most other marsupial taxa scored for this study have fewer than 50 teeth. In some taxa the suppressed dental loci can be identified with confidence, but dental homologies in other taxa are obscure for lack of relevant evidence. As described by multiple authors (e.g., Flower, 1867; Luckett, 1993b; Luo et al., 2004; van Nievelt and Smith, 2005a; 2005b; Forasiepi and Sánchez-Villagra, 2014), dental replacement in marsupials is known to occur only at the posteriormost premolar locus in both the upper and the lower jaw, where a single first-generation tooth (dP3/dp3) is usually replaced by a permanent successor (P3/p3). Further discussions of dental homologies are provided in the relevant character descriptions below.

TABLE 12

Alternative Hypotheses of Marsupial Tooth Homologies

(modified from Voss and Jansa, 2009: table 7)a

img-AM76_01.gif

Upper Incisors

Five upper incisors are present in didelphids, microbiotheriids, and many nonmarsupial metatherians, but other metatherians have four or fewer upper incisors. In order to meaningfully score taxonomic variation in tooth number for phylogenetic analysis, it is necessary to make assumptions about the homologies of these teeth. Although a seemingly agnostic alternative is to code the total number of upper incisors without reference to specific loci (e.g., as by Kirsch and Archer, 1982: char. 1; Springer et al., 1997: char. 1; Rougier et al., 1998: char. 8; Wroe et al., 2000: char. 1; Horovitz and Sánchez-Villagra, 2003: char. 150), such coding may be problematic because it implicitly assumes that homologous teeth were gained or lost to account for similarities in number. If, for example, there is compelling evidence that one taxon with three upper incisors lacks I1 and I2, whereas another with the same total lacks I4 and I5, then it is obviously inappropriate to score both with the same state. Where possible, we prefer to code the presence or absence of specific dental loci rather than simply counting teeth, on the methodological principle that it is better to make explicit rather than implicit assumptions.

Following Thomas (1887a), and in the absence of information to the contrary, we assume that upper teeth are lost from (or added to) the posterior (distal) end of the incisor row (see also Ziegler, 1971; Wible, 2008: 324–326; 2011: 151).9 Although there appears to be no developmental data to support this notion, we observed that I5 is small and occupies a separate alveolus from I2–I4 (the alveoli of which are usually at least partially confluent) in some taxa with five upper incisors (e.g., Glironia [Voss and Jansa, 2009: fig. 37], Isoodon, Peroryctes, and Perameles [fig. 39]) and that the loss of such a tooth would simply enlarge an already existing diastema between the incisors and C1 rather than opening a fresh gap in the dental arcade. Ziegler (1971: 233) also reported the occasional, unilateral presence of a supernumerary upper incisor in the peramelemorphian Echymipera (which normally has only four upper incisors) in the same position that I5 occupies in peramelemorphians that retain this tooth. Conversely, it would seem less difficult to add teeth at one end of the incisor row than to insert them between occupied loci. Because the anteriormost upper incisor is a conspicuously differentiated (styliform or chisellike) tooth in most metatherians (see char. 104), this locus does not appear to be a plausible site for dental additions or subtractions.

Character 103. Five upper incisors present (0); or I5 absent (1); or I4 and I5 absent (2); or I2–I5 absent (3). The fifth upper incisor (I5) is present (state 0) in †Mayulestes, †Allqokirus, †Pucadelphys, †Herpetotherium, didelphids (e.g., Marmosa [fig. 1], Thylamys [fig. 7]), most peramelemorphians (figs. 3739), Dromiciops (fig. 43), and †Microbiotherium. A single upper incisor (assumed to be I5 for the reasons discussed above) has been lost (state 1) in caenolestids (e.g., Caenolestes [fig. 35]), Rhynchomeles, Echymipera, and dasyuromorphians (figs. 4042). Of the other taxa included in this analysis, †Acdestis (Goin et al., 2003), †Palaeothentes (Marshall, 1980: 69; Forasiepi et al., 2014b), Notoryctes (fig. 36), †Yalkaparidon (Archer et al., 1988; Beck et al., 2014), and most diprotodontians have three upper incisors (I4 and I5 assumed to be missing teeth; state 2). Recent vombatids (e.g., Vombatus [fig. 45]) have only a single upper incisor (state 3); the number of incisors in the fossil vombatid †Warendja is unknown based on available material. Following the argument outlined above (that upper incisors are lost sequentially from posterior to anterior), we treated this character as ordered (additive) in all the analyses reported below.

Character 104. First upper incisor styliform or chisellike, conspicuously unlike I2 and I3 (0); or mesiodistally expanded and flat crowned, resembling I2 and I3 (1). In most metatherians, I1 differs from the more posterior incisors (if present) in being styliform or chisellike, somewhat hypsodont, and often slightly procumbent (state 0; see Voss and Jansa, 2009: 46, fig. 18). Although Vombatus (fig. 45) and Lasiorhinus retain only a single upper incisor, this tooth is large and chisellike, so we scored both taxa as state 0. In all Recent peramelemorphians, by contrast, I1 is mesiodistally expanded and flat crowned, such that I1–3 form a single morphologically continuous series (state 1; figs. 3739). In Thylacinus, I1–3 also form a morphologically continuous series (fig. 40), so we also scored this taxon as state 1. We scored Tarsipes as inapplicable (“-”) for this character as its upper incisors are greatly reduced (fig. 49).

A number of previous studies have distinguished between hypsodonty and nonhypsodonty of the first upper incisor (e.g., Takahashi, 1974; Archer, 1976c: char. 2; Reig et al., 1987; Wroe, 1999; Wroe et al., 2000: char. 3; Wroe and Musser, 2001: char. 3), and between “spatulate” and “peglike” upper incisors (e.g., Archer, 1976c: char. 3; Wroe et al., 2000: char. 2; Wroe and Musser, 2001: char. 2). However, we were unable to consistently score either of these characters for our taxon sample and have not included them in the current study.

Character 105. Second and third upper incisors with enamel present (0); or entirely lacking enamel (1). In all but one of the metatherians we examined that retain I2 and I3, these teeth are enamelled (state 0). By contrast, †Yalkaparidon uniquely lacks enamel on its second and third upper incisors (Beck, 2009; Beck et al., 2014), an obvious autapomorphy. This character was scored as inapplicable (“-”) for Vombatus and Lasiorhinus, both of which lack I2 and I3.

Character 106. Occlusal surface of I3 variously formed, but without a well-developed anterolabial crest separated from the main crest by a mesiodistally elongate groove or basin (0); or occlusal surface of I3 bifid, with a well-developed anterolabial crest separated from the main crest by a mesiodistally elongate groove or basin (1). The occlusal morphology of newly erupted, unworn second and third upper incisors varies considerably within Marsupialia. Of particular note is the complex morphology seen in unworn I2 and I3 of many diprotodontians (e.g., Cercartetus [e.g., AMNH 222711], Dactylopsila [e.g., AMNH 101988, 192155, 157140], Dactylonax [e.g., AMNH 221611], Ailurops [e.g., AMNH 108799], and Pseudochirops cupreus [e.g., AMNH 191142, 191155]). Although this variation is striking, most of the variants we observed are difficult to code as discrete character states. The sole exception is the presence or absence of a distinct anterolabial crest on the occlusal surface of I3. This crest is present (state 1) in several macropodids, where it is separated from the main (posterolingual) crest by a distinct, mesiodistally elongate and posteriorly open groove or (as in Dendrolagus) a posteriorly enclosed basin (Prideaux, 2004: 237–239, figs. 16, 17; Prideaux and Warburton, 2010: char. 15). The major posterolingual crest appears to be homologous with the single blade present on I3 of most nondiprotodontian metatherians, whereas the anterolabial crest appears to be a neomorph (Prideaux and Warburton, 2010). The groove can be lost through wear, so we scored this character based on lightly worn specimens (preferably juveniles) where possible. Our observations are in general agreement with those of Prideaux and Warburton (2010), but we scored Lagostrophus as corresponding to state 1 because a distinct anterolabial crest appears to be present in this taxon (e.g., on WAM M6791; see also Prideaux, 2004: figs. 16B, 17A). We scored Tarsipes as inapplicable (“-”) for this character as its upper incisors are greatly reduced (fig. 49).

Character 107. Fourth upper incisor (I4) similar in size to I1–I3 (0); or I4 markedly larger than I1–I3 (1). In most metatherians, I4 (if present) is similar in size to I1–I3. However, in Thylacinus I4 is markedly larger than more anterior teeth (fig. 40). This character was scored as inapplicable (“-”) for all taxa lacking I4, and as unknown (“?”) for all taxa in which I4 is not preserved in available specimens. Wroe and Musser (2001: 499) stated that the alveolus for I4 in †Nimbacinus is the largest of the upper incisor arcade, as it is in Thylacinus. However, our examination of QM F36357, the well-preserved skull of †N. dicksoni described by Wroe and Musser (2001), suggests otherwise: although the left I4 alveolus is, indeed, markedly larger than the adjacent I3 alveolus, this is due to damage to the anterior margin of the former; the right I4 alveolus, which appears to be intact, is not noticeably larger than that for I3 (see Wroe and Musser, 2001: fig. 1B). Because no specimen of †Nimbacinus preserves the crown of I4 itself, we prefer to score this taxon as unknown (“?”). In effect, state 1 is an autapomorphy of Thylacinus in the current analysis.

Upper Canine

The upper canine (C1), which in most therian mammals is a single-rooted unicuspid tooth that occupies the premaxillary-maxillary suture (Kielan-Jaworowska et al., 2004), is absent in some marsupials and exhibits noteworthy morphological variation in others. Remarkably, marsupial canines are sometimes bi- or tricuspid, sometimes double-rooted, and sometimes wholly contained in the maxilla, as briefly discussed by Voss and Jansa (2009: 48–49). Additionally, C1 is sexually dimorphic in shape and root number in certain clades. Few of these unusual traits have been represented in previous phylogenetic datasets.

Character 108. Upper canine (C1) present (0); or absent (1). The upper canine is present in most metatherians, but some diprotodontians usually lack C1, including some macropodids (e.g., Lagostrophus [fig. 53]), †Nimbadon, †Neohelos, and Recent vombatids (e.g., Vombatus [fig. 45]). Other macropodids exhibit morphologies that are intermediate between simple presence and absence of C1. In Onychogalea, for example, C1 is present in juveniles, subadults, and young adults, but it is lost in old adults (e.g., AMNH 197525); we scored this taxon as state 0 because C1 is present throughout most of its ontogeny. By contrast, C1 is present only in the youngest juveniles we examined of Osphranter (e.g., AMNH 154509), Notamacropus (e.g., AMNH 154504), and Macropus (e.g., AMNH 65017); we scored these taxa as state 1 because C1 is absent throughout most of their ontogeny. In yet other macropodids (e.g., Thylogale, Wallabia) what may be a remnant of the C1 alveolus is sometimes identifiable within the premaxillary-maxillary suture, but this cavity is not occupied by a tooth even in young juveniles; we scored such taxa as state 1. Although most specimens of †Nimbadon lack C1, this tooth is present in a few individuals (e.g., QM F50470 [Black and Hand, 2010]); we scored this taxon as state 1 based on the observed modal condition.

Character 109. Upper canine (C1) single-rooted (0); or double-rooted (1). Where present, C1 is single-rooted (state 0) in most metatherians. However, this tooth is consistently double-rooted in Lestoros (state 1; Martin, 2013). Osgood (1924: 171) and Travouillon et al. (2019: 15) stated that C1 is double-rooted in the peramelemorphian Chaeropus, whereas Thomas (1888: 250) reported that this tooth is “sometimes double-rooted” in this taxon. A shallow, vertically directed sulcus or sulcus is sometimes visible on the lateral face of the root of C1 in Chaeropus (Travouillon et al., 2019: figs. 3b, i, 6b), but in none of the specimens we examined did the root itself appear to bifurcate, and the C1 alveolus is apparently single in crania in which this tooth is missing (Travouillon et al., 2019: figs. 6i, j); therefore, we scored this taxon as state 0. Among fossil paucituberculatans, C1 is single-rooted in known specimens of †Palaeothentes (Marshall, 1980; Forasiepi et al., 2014), †Acdestis (Goin et al., 2003), and †Stilotherium (Marshall, 1980 [fig. 4]), all of which we scored as state 0. However, C1 is single-rooted in males but double-rooted in females of Caenolestes (fig. 35; Bublitz, 1987: abb. 8) and Rhyncholestes (Osgood, 1924; Patterson and Gallardo, 1987; Voss and Jansa, 2009), so we scored these taxa as polymorphic (“0+1”). By contrast, Horovitz and Sánchez-Villagra (2003 [char. 171]), Sánchez-Villagra et al. (2007 [char. 156]), and most subsequent studies that have employed versions of their matrices (e.g., Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Abello and Candela, 2010) scored Caenolestes as having a single-rooted C1.

We acknowledge that coding sexual dimorphism as polymorphism conflates sex-specific differences with variability due to other phenomena (e.g., alleles segregating in a population, geographic variation). An alternative approach would be to explicitly recognize sexual dimorphism in canine root number (or crown morphology; see below) as a distinct state within in an ordered transformation series. Unfortunately, reliable indications of sex are seldom available for fossil taxa, so this suboptimal expedient seems preferable to scoring all fossils as missing (“?”). This character was scored as inapplicable (“-”) for taxa lacking C1 (see char. 108).

Character 110. Upper canine (C1) separated from posteriormost upper incisor by a distinct diastema (0); or diastema between C1 and upper incisor row absent (1). In most metatherians, a conspicuous diastema separates C1 from the posteriormost upper incisor, and these teeth have clearly separate alveoli (e.g., in Marmosa [figs. 1, 34]; state 0). By contrast, in acrobatids (e.g., Distoechurus [fig. 48]), Dactylopsila (Aplin, 1990: plate 5.2a), Dactylonax, Pseudochirops cupreus, some phalangerids (e.g., Phalanger [fig. 47]), Bettongia (but not ?Bettongiamoyesi; Flannery and Archer, 1987b), and †Thylacoleo (Anderson, 1929: plate XXI), no diastema is consistently present between C1 and I3 (state 1), the alveoli of which are often confluent. Pseudochirops archeri and Notoryctes are both polymorphic (“0+1”) for this character, which we scored as inapplicable (“-”) for taxa lacking C1 (see char. 108).

Character 111. Upper canine (C1) alveolus occupies premaxillary-maxillary suture (0); or C1 alveolus contained entirely in maxillary bone (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 30; 2009: 48), who misleadingly described taxa exhibiting state 1 as lacking a premaxillary palatal process. Among didelphids, only Caluromys and Caluromysiops have the C1 alveolus entirely enclosed by the maxilla (state 1). Engelman and Croft (2014: 675) stated that the C1 alveolus is within the maxilla in †Sparassocynus, presumably based on the line drawing shown in figure 4 of Reig and Simpson (1972), but in several †Sparassocynus specimens that we examined the alveolus is in the premaxillary-maxillary suture, as it is in most Recent didelphids (Beck and Taglioretti, 2020); thus, this taxon has been scored as state 0 here.

Among other metatherians, the upper canine is typically within the premaxillary-maxillary suture (state 0), but this tooth is entirely within the maxilla in †Pucadelphys, microbiotheriids (e.g., Dromiciops [fig. 43]), Myrmecobius (fig. 41), dasyurids (e.g., Pseudantechinus [fig. 42]), †Yalkaparidon (Beck et al., 2014: fig. 5), and several peramelemorphians (†Galadi, Macrotis, Echymipera, and Isoodon). Lestoros, Chaeropus, Microperoryctes, Perameles, and Rhynchomeles are all polymorphic (“0+1”) for this character. We scored taxa lacking C1 (see char. 108) as inapplicable (“-”).

Character 112. Upper canine simple, without distinct accessory cusps (0); or with distinct posterior accessory cusp only (1); or with distinct anterior and posterior accessory cusps (2); or with distinct anterior accessory cusp only (3). As described by Voss and Jansa (2003: char. 53; 2009: 48), didelphid upper canines are either simple unicuspids (state 0), or they have only a posterior accessory cusp (state 1), or they have both anterior and posterior accessory cusps (state 2). Some previous authors (e.g., Osgood, 1924; Archer, 1976c: char. 5; 1981; Patterson and Gallardo, 1987; Wroe et al., 2000: char. 4) have described variation in this trait using the modifier “premolariform” for canines with accessory cusps and “caniniform” for simple teeth. Most nondidelphid metatherians—including all diprotodontians that retain C1—have simple, unicuspid upper canines that exhibit only minor variation in morphology, so we scored them as state 0. However, a distinct posterior accessory cusp (state 1) occurs in several dasyurids (e.g., Planigale, Pseudantechinus, Phascogale, Sminthopsis), and an anterior accessory cusp is present (state 3) in some peramelemorphians (e.g., Chaeropus [fig. 37]; Travouillon et al., 2019).

The upper canine is sexually dimorphic in Recent caenolestids (Osgood, 1924: 169, 171; Bublitz, 1987; Patterson and Gallardo, 1987: 1; Voss and Jansa, 2009: 48–49). Whereas C1 is consistently unicuspid in the male caenolestids we examined, this tooth has an anterior accessory cusp (state 3) in female Caenolestes and Lestoros (but see Martin, 2013: 610, who reported nonsexually dimorphic polymorphism in the presence or absence of anterior and posterior cusps in Lestoros). In female Rhyncholestes, C1 has both anterior and posterior accessory cusps (state 2). As for canine root number (see char. 109), we scored these traits as polymorphisms, with Caenolestes and Lestoros scored as “0+3” and Rhyncholestes scored as “0+2” (after Voss and Jansa, 2009: char. 94).

Polymorphism in C1 crown morphology is also seen in some dasyurids. Dasyuroides and Ningaui either lack accessory cusps or exhibit a posterior accessory cusp (“0+1”), whereas Antechinomys has either a posterior accessory cusp or both anterior and posterior accessory cusps (“1+2”). Canine morphology appears to be particularly variable in Micromurexia, for which some specimens lack accessory cusps (e.g., AMNH 190909), others have a distinct anterior cusp only (e.g., AMNH 190914), some have both anterior and posterior accessory cusps (e.g., AMNH 190912), and some have only a posterior accessory cusp (e.g., AMNH 190906); we scored this apparently hypervariable taxon as “0+1+2+3.” The peramelemorphian Rhynchomeles is also polymorphic: specimens either lack accessory cusps or have only an anterior accessory cusp (“0+3”). Unlike the pattern of C1 variation we observed among Recent caenolestids, the polymorphic traits seen in Rhynchomeles and some dasyurids do not appear to be sexual dimorphic.

Although Voss and Jansa (2003) regarded state 1 as intermediate to states 0 and 2 among their terminal taxa and treated this character as an ordered transformation series, the observation of a different bicuspid condition (state 3) in some nondidelphids makes their ordering scheme problematic. We therefore treated this character as unordered in all our analyses following Voss and Jansa (2009: char. 94). This character was scored as inapplicable (“-”) for taxa lacking C1 (see char. 108).

Postcanine Dentition

Character 113. Postcanine dentition with identifiable cusps (0); or postcanine teeth are featureless spicules (1). The postcanine teeth of most metatherians exhibit one or more well-defined and presumably functional cusps (state 0), but in Tarsipes the upper and lower postcanine dentition is uniquely reduced to tiny, featureless, apparently nonfunctional spicules of enamel (Parker, 1890; Bensley, 1903; Archer, 1984c; Russell and Renfree, 1989), an obvious autapomorphy (state 1). In addition, none of the postcanine teeth of Tarsipes can be confidently homologized with specific loci in other metatherians (Bensley, 1903; Archer, 1984c; Russell and Renfree, 1989). We therefore scored this taxon as inapplicable (“-”) for characters relating to the presence or absence of particular structures (e.g., cusps and crests) in the postcanine dentition, and unknown (“?”) for characters relating to the presence or absence of specific postcanine loci.

Upper Premolars

Three upper premolars, the maximum number normally seen in any metatherian (Kielan-Jaworowska et al., 2004), are present in †Mayulestes, †Allqokirus, †Pucadelphys, †Herpetotherium, microbiotheriids, most didelphids, paucituberculatans, peramelemorphians, most dasyuromorphians, †Lekanoleo, †Thylacoleo, pseudocheirids, petaurids, Cercartetus, acrobatids, and some phalangerids. By contrast, other marsupials have fewer upper premolars, with the result that tooth-locus homologies in this series are sometimes problematic. In the accounts that follow, we base decisions about premolar homologies on tooth size and morphology, the presence and position of diastemata, the presence or absence of deciduous precursors, supernumerary teeth, and other relevant evidence.

Apart from variation in tooth number, the marsupial upper premolar dentition seems remarkably conservative except among diprotodontians, which exhibit extraordinary modifications of these teeth. Indeed, it is doubtful whether any other mammalian order exhibits such exuberant variation in premolar morphology as that seen among members of Diprotodontia, which include the only living mammals with plagiaulacoid dentitions (sensu Simpson, 1933). Further studies on diprotodontian premolar function would be a welcome addition to the mammalian dental literature.

Character 114. First upper premolar (P1) present (0); or absent (1). Only two upper premolars are consistently present in Notoryctes (fig. 36), Burramys (fig. 46), most phalangerids (i.e., Ailurops, Spilocuscus, Strigocuscus, Trichosurus [the fossil “Trichosurus” †dicksoni could not be scored for this character], Wyulda), juvenile macropodiforms, and some dasyurids (Dasyurus, Dasykaluta, and Sarcophilus). Specimens of Caluromysiops, Dasycercus, and Pseudantechinus exhibit either two or three upper premolars. Phascolarctids (e.g., Phascolarctos [fig. 44]), vombatids (e.g., Vombatus [fig. 45]), diprotodontids, wynyardiids, and †Yalkaparidon have only a single upper premolar. Although some authors have scored number of premolars as a phylogenetic character without reference to tooth homologies (e.g., Archer, 1976c: char. 7; Rougier et al., 1998: char. 1), plausible premolar homologies can be determined for most of our terminal taxa with fewer than three premolars based on a combination of ontogenetic, positional, and other morphological criteria.

Comparison of three-premolared and two-premolared specimens of Caluromysiops clearly indicates that it is P1 that is sometimes absent in this taxon, which we scored as polymorphic (“0+1”). When present, P1 of Caluromysiops is vestigial and single-rooted (see char. 115).

The dasyurids Dasyurus, Dasykaluta, and Sarcophilus consistently have only two premolars (as does Planigale gilesi, although we used three-premolared P. ingrami to represent the genus in this study). Examination of juvenile specimens of Dasyurus and Sarcophilus indicates that neither tooth seems to have a deciduous precursor in these taxa (we were unable to examine juveniles of Dasykaluta). Because P3 is a replacement tooth in all other metatherians with rooted cheekteeth, we agree with the traditional view that the missing tooth in two-premolared dasyurids is P3 (e.g., Thomas, 1887a; Woodward, 1896; Bensley, 1903; Tate, 1947; Bartholomai, 1971a; Archer, 1976c; 1982b; 1984c; Wroe and Mackness, 2000a; 2000b). This inference is additionally supported by the presence of a small but distinct diastema between the posteriormost premolar and M1 in both Dasyurus and Dasykaluta, whereas no diastema is present between P3 and M1 in dasyurids with three premolars. A diastema is absent in Sarcophilus (see Archer, 1984c: figs. 1415; Rose et al., 2017: fig. 2), but this may be due to autapomorphic rostral shortening. An obviously vestigial tooth that may represent dP3 is sometimes present within the diastema between the fully erupted posteriormost premolar and M1 in specimens of Dasyurus (e.g., AM 35911; see also Woodward, 1896).

Although not in itself direct evidence that Dasykaluta, Dasyurus, and Sarcophilus have lost P3, it is also noteworthy that P3 is only variably present in Dasycercus and Pseudantechinus and that this tooth is small and single-rooted in Dasyuroides and in three-premolared specimens of Dasycercus. Similarly, comparison between the two-premolared Planigale gilesi with other Planigale species indicates that the missing upper premolar in the former is P3 (Archer, 1976a: 356–357). By contrast, P1 and P2 are always well developed, double-rooted, premolariform teeth in three-premolared dasyurids. We therefore scored all dasyurids as retaining P1 (state 0). We acknowledge that our interpretation conflicts with the hypothesis of Luckett (1993b; see also Luckett and Woolley, 1996; Luckett et al., 2019) that the missing tooth in Dasyurus is P2; a fresh appraisal of the ontogenetic basis for inferring premolar homologies in this taxon would be welcome.

The posterior macropodiform premolar—a large sectorial tooth that replaces a molariform deciduous precursor—is clearly P3, and so the missing tooth in this group is either P1 or P2. We favor the hypothesis that P1 is the missing tooth in macropodiforms because a prominent diastema is present immediately behind C1 in those macropodiforms that retain this tooth (see char. 108), but not between the anterior and posterior premolars; thus, we scored all of our macropodiform terminal taxa as corresponding to state 1. Although data from taxa unrepresented in our study cannot be taken as evidence for a priori homology, we note that the hypothesis that macropodiforms have lost P1 is consistent with observations of a suppressed tooth bud at this locus in sectioned pouch young of several Recent macropodoid species (Woodward, 1893; Berkovitz, 1966; Kirkpatrick, 1969; 1978).

In Burramys (fig. 46), the more posterior of the two upper premolars is very large, sectorial, and in direct contact with M1; we identify it as P3 (after, e.g., Ride, 1956; Brammall and Archer, 1997). The more anterior upper premolar is tiny, single-rooted, and in direct contact with the base of P3 but separated from C1 by a large diastema; based primarily on the presence of this diastema, we identify the anterior premolar of Burramys as P2 (again, as in, e.g., Ride, 1956; Brammall and Archer, 1997), and therefore score P1 as absent in this taxon.

Two of the thylacoleonids included in our study—†Lekanoleo and †Thylacoleo—retain three premolars; in these taxa P1 is separated from C1 and P2 by flanking diastemata, but P2 is in contact with the hypertrophied, sectorial P3 (Gillespie, 2007). By contrast, the species used to score our †Wakaleo terminal (†W. vanderleuri) has only two premolars (Murray et al., 1987; Gillespie, 2007; Gillespie et al., 2017), the more posterior of which is hypertrophied, sectorial, and in direct contact with M1, and so can be confidently identified as P3 based on its resemblance in morphology and position to the unequivocal P3 of †Lekanoleo and †Thylacoleo. As discussed by Gillespie (2007: 75) and Gillespie et al. (2017: 26–27), the anterior premolar of †Wakaleo is well separated from both P3 and C1 by diastemata, suggesting that it is homologous with P1 of †Lekanoleo and †Thylacoleo, rather than P2 (contra Murray et al., 1987); therefore, we scored all three thylacoleonids as retaining P1.

The single upper premolar of Phascolarctos is a large sectorial tooth that replaces a vestigial milk precursor that may or may not erupt (Thomas, 1887b; Blanshard, 1990), so the missing premolars in this taxon are clearly P1 and P2. A single upper premolar is also present in †Nimiokoala and †Litokoala; this tooth is similar in morphology and position to the P3 of Phascolarctos (i.e., large, sectorial, and in contact with M1 but separated from C1 by a large diastema; Black and Archer, 1997b; Louys et al., 2007, 2009; Black et al., 2014a), so we identify it also as P3 and score all three phascolarctids as lacking P1. †Muramura (see Pledge, 1987c; 2003) and †Namilamadeta (see Rich and Archer, 1979; Pledge, 2005) also have a single large sectorial premolar in contact with M1 but widely separated from C1, suggesting that it is P3, and this inference is further supported by a juvenile specimen of †Namilamadeta (QM F51348; Pledge, 2005: fig. 11D), which indicates that the sectorial premolar of this taxon has a deciduous precursor. We therefore scored both wynyardiids as lacking P1.

In the diprotodontids †Neohelos, †Ngapakaldia, †Nimbadon, and †Silvabestius the single upper premolar is molariform rather than sectorial, but, as in phascolarctids and wynyardiids, this tooth is in direct contact with M1 and there is a large diastema between it and C1 in those taxa that retain the latter tooth (see char. 108); therefore, we scored these diprotodontids as corresponding to state 1. Juveniles of †Nimbadon (see Black and Hand, 2010) and †Ngapakaldia (e.g., SAM P35313, a specimen of †N. bonythoni) demonstrate that the single premolar of both taxa had a deciduous precursor, confirming its identity as P3.

The enigmatic †Yalkaparidon also retains a single upper premolar, which is nearly in contact with M1 but separated from C1 by a large diastema (Beck, 2009; Beck et al., 2014). Although juvenile specimens that might reveal the presence or absence of a deciduous precursor are unknown, we tentatively identified this tooth as P3 (following Beck, 2009; Beck et al., 2014) and scored this taxon as lacking P1 and P2.

Premolar homologies in phalangerids also merit discussion. Phalanger has three premolars, all of which are morphologically distinctive (fig. 47): P1 is large, single-rooted, and caniniform; P2 is a tiny, single-rooted, and occlusally featureless button flanked by diastemata; and P3 is a large sectorial replacement tooth (see, e.g., Flannery et al., 1987; Norris and Musser, 2001: fig. 5). A similar premolar complement was apparently present in †Onirocuscus, but P1 is double-rooted in this taxon, and there is no diastema between P2 and P3 (Crosby, 2007; Case et al., 2008). By contrast, Spilocuscus, Strigocuscus, Trichosurus, Wyulda, and Ailurops each have only two premolars, of which the posterior one (a large, sectorial, replacement tooth closely appressed to M1) is unambiguously identifiable as P3 (Flannery et al., 1987). Following previous authors (Bensley, 1903; Archer, 1984c; Flannery et al., 1987), we hypothesize that the anterior premolar of Trichosurus and Wyulda—a single-rooted, more or less caniniform tooth—is P1 on the basis of its morphology, proximity to the canine, and the presence of a large diastema between it and P3 in juvenile specimens (e.g., AMNH 65547 [Trichosurus vulpecula] and WAM 21625 [Wyulda squamicaudata]). This inference is also consistent with Berkovitz's (1968) report of a nonfunctional enamel organ at the second upper premolar locus in Trichosurus vulpecula pouch young. Therefore, we scored both Trichosurus and Wyulda as retaining P1 (state 0).

Although Spilocuscus maculatus (the generic exemplar used for scoring purposes in this study) has only two premolars (Archer, 1984c: fig. 166; Flannery et al., 1987: fig. 15; Flannery and Calaby, 1987), several congeners retain three premolars in at least some specimens (Flannery and Calaby, 1987; Helgen and Flannery, 2004). In the latter taxa (e.g., S. papuensis, S. rufoniger, S. wilsoni), P2 is very small (when present) and is clearly the polymorphically absent tooth. Based on the size, position, and morphological similarity of the two premolars in S. maculatus with the teeth of congeneric species that sometimes retain three, we identify the missing tooth in S. maculatus as P2.

We examined two juvenile specimens of Ailurops (BMNH 1950.1011 and 1950.1012) in which a distinct diastema separates the erupted anterior premolar from dP3, but no diastema separates the anterior premolar from C1 (with which it shares a partially confluent alveolus). We consider this a plausible indication that P2 has been lost in Ailurops (Flannery et al., 1987). In Strigocuscus, however, the anterior premolar is equally separated from C1 and P3 by diastemata (Flannery et al., 1987: fig. 9; Groves, 1987), rendering its homology uncertain. Although multiple tiny foramina in the diastema between P3 and the anterior premolar of Strigocuscus may represent vestiges of a dental alveolus (in which case it is P2 that has been lost in this taxon), based on the evidence at hand we prefer to code Strigocuscus as unknown (“?”) for this character (and for other characters based on P1 and/or P2; see below).

The single, open-rooted upper premolar of Vombatus (fig. 45), Lasiorhinus, and †Warendja (see Pledge, 1992: fig. 1) lacks any diastema between it and M1, consistent with the traditional identification of this tooth as P3. However, the evidence that this is a replacement tooth rather than a retained dP3 is somewhat less compelling. Thomas (1887b) argued that this tooth is P3 rather than dP3 based, in part, on the observation that the open-rooted premolars of placentals (e.g., Cavia) are known to be replacement teeth (see also Adloff, 1904; Harman and Smith, 1936; Berkovitz, 1972a; Luckett, 1985). Thomas (1887b) also noted that Phascolarctos (the presumed closest extant relative of vombatids) has vestigial milk premolars, which he suggested may represent an intermediate condition between the large and persistent deciduous teeth of other diprotodontians and their apparent absence in vombatids. Brewer et al. (2018: 2–3) also noted that the premolars of Vombatus erupt at about the same time as the replacement premolars of Phascolarctos (Green and Rainbird, 1987; Blanshard, 1990; Triggs, 2009), suggesting that they are also replacement teeth in Vombatus. Nevertheless, because replacement of dP3 by P3 has not (to our knowledge) been directly observed in vombatids, and that dP3 is often a well-developed molariform tooth in other marsupials (including diprotodontids and wynyardiids; see below), it is possible that the large open-rooted tooth in the third premolar position of vombatids is a retained dP3.

The only study of dental development directly bearing on this question seems to be that of Röse (1893), who serially sectioned a single pouch young (1.9 cm body length) of “Phascolomys wombat” (= Vombatus ursinus). Röse (1893) identified dP3 in this specimen, but concluded that this tooth is subsequently replaced by another that develops from the successional lamina (“Ersatzleiste”; Röse, 1893: fig. 3). However, the sectioned specimen examined by Röse (1893) does not, in fact, preserve any evidence of a developing P3 because the successional lamina he illustrated shows no indication of the localized thickening that is characteristic of the earliest visible (= early bud) stage of dental development (Luckett, 1993a). Despite such equivocal evidence for the homology of the vombatid premolar, we treat this tooth as P3, in agreement with most previous studies (see also char. 120), but contra Brewer et al. (2018: 2–3), who treated the identity of this tooth in vombatids as unknown.

Notoryctes typically has two upper premolars (Gadow, 1892; Archer, 1984c: 657), the more posterior of which is well developed and occludes with the paracristid of m1, suggesting that it is P3. However, the anterior upper premolar is of less certain homology. As noted by Archer et al. (2011), P1 is tiny compared with P2 and P3 in the Miocene notoryctid †Naraboryctes philcreaseri, which is evidently closely related and potentially ancestral to Notoryctes (see Beck et al., 2016). Further, in specimens of Notoryctes that appear to retain three upper premolars, the first of these is the smallest (Thomas, 1920: 112), suggesting that it is indeed P1 absent in specimens with only two upper premolars. Based on this collective evidence, we assume here that it is P1 has been lost in Notoryctes.

The upper postcanine dentition of Tarsipes (fig. 49) comprises between three and four undifferentiated spicules of enamel that cannot be homologized to specific dental loci (see char. 113), so we scored this taxon as unknown (“?”) for this and for most other characters of the upper dentition.

Character 115. First upper premolar (P1) conventionally premolariform (0); or single-rooted (1). When present, the first upper premolar of most metatherians is conventionally premolariform, having two subequal roots (one anterior and one posterior) and a more or less laterally compressed crown bearing a single dominant median cusp that may or may not be flanked by much smaller anterior and/or posterior accessory cusps (state 0); in these respects it resembles the widespread and presumably primitive therian condition (Kielan-Jaworowska et al., 2004). By contrast, pseudocheirids (with the exception of some Pseudocheirus specimens, in which P1 is double-rooted; Archer, 1984c: figs. 185–186), phalangerids (except †Onirocuscus and some specimens of Ailurops, in which this tooth is also double-rooted; Flannery et al., 1987; Crosby, 2007), thylacoleonids (Gillespie, 2007), Dactylopsila (Archer, 1984c: fig. 208), Dactylonax, Gymnobelideus (Archer, 1984c: fig. 192), and Caluromysiops have a single-rooted, subconical P1 (state 1). Pseudocheirus, Cercartetus, Ailurops, and Lestoros are polymorphic (“0+1”) for this character, which we scored as inapplicable (“-”) for taxa lacking P1 and as unknown (“?”) for taxa with indeterminate anterior upper premolar homologies (see char. 114).

Character 116. Second upper premolar (P2) present (0); or absent (1). Scoring for this character is based on previously discussed assumptions about upper premolar homologies (char. 114). However, we note that P2 is present in juvenile macropodiforms but is lost in adult specimens of most taxa, in which it is displaced by the eruption of P3 (the only known exception is †Ekaltadeta; see char. 117). Therefore, metatherians that unambiguously lack P2 as adults were scored for this character only if juvenile specimens were available for examination. In the absence of juvenile material, we scored such taxa (namely †Bohra, †Hadronomas, †Rhizosthenurus, ?Bettongiamoyesi, †Balbaroo, †Ganawamaya, Hypsiprymnodonbartholomaii, †Neohelos, †Ngapakaldia, †Warendja, †Muramura, †Nimiokoala, †Litokoala, †Wakeleo, and †Yalkaparidon) as unknown (“?”).

Based on examination of juvenile material, P2 is unequivocally absent in some phalangerids (Ailurops, Spilocuscus, Trichosurus, and Wyulda), †Silvabestius, †Nimbadon, Vombatus (fig. 45), Lasiorhinus, †Namilamadeta, and Phascolarctos (fig. 44). Taxa in which anterior upper premolar homologies are unrecognizable (i.e., Strigocuscus and Tarsipes; see above) were scored as unknown (“?”) for this character.

Character 117. Second upper premolar (P2) ontogenetically persistent (0); or P2 retained briefly and then shed after dP3 is replaced by P3 (1); or P2 and dP3 simultaneously replaced by P3 (2). When present, the metatherian second upper premolar usually persists into adulthood (state 0). In most macropodiforms, however, the erupting third premolar (P3) simultaneously displaces P2 and dP3 (state 2), a remarkable phenomenon admirably described and illustrated by Thomas (1888: figs. 1, 2; see also Flower, 1867: plate XXIX fig. 4; Archer, 1984c: fig. 214). An obviously intermediate condition occurs in Hypsiprymnodon, where P2 is retained for a brief period after dP3 has been replaced by P3 (state 1). In this taxon, both P2 and P3 are present in younger individuals (Ride, 1961: fig. 1b) whereas older specimens have only P3 (fig. 52; see also Woods, 1960; Ride, 1961; Johnson and Strahan, 1982). Uniquely, the fossil macropodiform †Ekaltadeta ima retained P2 into adulthood (Wroe, 1996b; Wroe et al., 1998). Because dental replacement patterns in the macropodiform upper dentition are paralleled by those in the mandibular teeth, we did not code the latter as a separate character.

Interestingly, although a tiny, single-rooted P2 is consistently present in juvenile and subadult specimens of Dactylonax and Dactylopsila, this tooth is sometimes absent from one or both sides of adults, particularly older individuals (e.g., AMNH 194760, a specimen of Dactylonax palpator). However, this occasional, apparently ontogeny-related loss of P2 in Dactylopsila and Dactylonax is unconnected with the eruption of P3, and so we scored both taxa as state 0.

As for the preceding character, taxa that unequivocally lack P2 were scored as inapplicable (“-”), whereas those that lack P2 as adults but for which juvenile material is unavailable (including †Yalkaparidon and many fossil diprotodontians) were scored as unknown (“?”), because it is unclear in these taxa whether P2 is absent throughout ontogeny or, alternatively, present in juveniles but subsequently evicted by P3. Taxa in which P2 cannot be unequivocally identified (see above) were also coded as unknown.

This character was treated as ordered (additive) in all the analyses reported below.

Character 118. Second upper premolar conventionally premolariform (0); or P2 “pseudocheiriform” (1); or P2 sectorial (2); or P2 bicuspid (3); or P2 single-rooted and caniniform or vestigial and occlusally featureless (4). The second upper premolar of most metatherians (when present) is conventionally premolariform (as defined above; see char. 115), as it is in †Mayulestes, †Allqokirus, †Pucadelphys, †Herpetotherium, didelphids, paucituberculatans, Dromiciops, dasyuromorphians, peramelemorphians, acrobatids, Cercartetus, and Petaurus (state 0). By contrast, the unworn P2 of pseudocheirids is a distinctively bulbous tooth about as wide as it is long; the occlusal surface of this tooth usually consists of two subequal labial cusps connected by a sharp crest, and there is a prominent lingual heel that is not present in other taxa (state 1; see Archer, 1984c: figs. 180, 185, 186). As in the widespread therian condition, the pseudocheiriform premolar has just two roots, but the posterior root is substantially larger than the anterior root.

The second upper premolar crown of all macropodiforms except Onychogalea and Macropus consists of a single large shearing crest (state 2), a morphology that has traditionally been described in the literature as “sectorial” (e.g., by Bensley, 1903) or “secant” (e.g., by Tate, 1948). As far as can be determined from the material at hand, macropodiform P2s always have three roots, one anterior and two posterior. Although there is substantial taxonomic variation in the morphology of P2 among macropodiform genera, this variation seems to correspond closely with that observed for P3 (also a sectorial tooth) as described below (see chars. 125–127); therefore, we did not code these simultaneous transformations as separate characters. The crown morphology of P2 in Onychogalea (e.g., AMNH 183443) and Macropus (e.g., AMNH 65017) differs from that of other examined macropodiforms (state 3): in both taxa, two large, laterally compressed cusps (of which the anterior is larger) are arranged anteroposteriorly along the labial edge of the tooth and are not connected by a crest (Thomas, 1888: 17; Tate, 1948: 277–276, 331–332; Bartholomai, 1971b: fig. 1). A prominent posterolingual cusp and somewhat smaller anterolingual cusp are also present in Macropus (Tate, 1948: 331–332; Bartholomai, 1971b: 2–3, fig. 1).

The second upper premolar is single-rooted and caniniform or occlusally featureless (state 4) in Dactylopsila (Archer, 1984c: fig. 208), Dactylonax, Gymnobelideus (Archer, 1984c: fig. 192), Burramys (fig. 46), the phalangerids Phalanger (fig. 47) and †Onirocuscus (see discussion under char. 114), the †thylacoleonids †Lekanoleo and †Thylacoleo (Woods, 1956; Archer, 1984c; Gillespie, 1997, 2007), and Notoryctes (Gadow, 1892: 367).

In the absence of any clear sequence of transformations among these alternative conditions, we treated this character as unordered (nonadditive) in all the analyses reported below. This character was coded as inapplicable (“-”) for taxa in which P2 is unambiguously missing and as unknown (“?”) for taxa in which P2 could not be identified with certainty (see above).

Character 119. Premolariform second upper premolar (P2) distinctly taller than premolariform P3 (0); or premolariform P2 and premolariform P3 subequal in height (1); or premolariform P3 distinctly taller than premolariform P2 (2). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 55; 2009), who observed considerable variation in the relative heights of P2 and P3 within Didelphidae; however, only in Caluromys, Caluromysiops, and Hyladelphys is P2 distinctly taller than P3.

The second and third upper premolars are always present and always premolariform in didelphids (Voss and Jansa, 2003; 2009). However, one or the other of these loci is absent in many other metatherians (see chars. 116, 122) and, when both teeth are present, there is considerable variation in their crown morphology (see chars. 118, 123). To minimize character non-independence (for example, in cases where P2 or P3 is vestigial), we scored this character only for taxa in which P2 and P3 are both present and premolariform (i.e., taxa scored as state 0 for chars. 118, 123). We scored taxa in which one or both teeth are missing or exhibit other (nonpremolariform) crown morphologies as inapplicable (“-”). Among nondidelphid taxa with premolariform teeth at both loci, P3 is taller than P2 in †Pucadelphys, †Allqokirus, †Mayulestes, †Herpetotherium, paucituberculatans (e.g., Caenolestes [fig. 35]), thylacinids (e.g., Thylacinus [fig. 40]), several dasyurids (e.g., Antechinomys, Murexia, Ningaui), most peramelemorphians (e.g., †Galadi, Echymipera, Isoodon), and Petaurus (fig. 50). By contrast, these teeth are subequal in height in some dasyurids (e.g., Antechinus, Pseudantechinus [fig. 42], some specimens of Phascogale) and a few peramelemorphians (Macrotis [fig. 38], some specimens of Microperoryctes and Peroryctes). Finally, premolariform P2 is distinctly taller than premolariform P3 in acrobatids (e.g., Distoechurus [fig. 48]), Myrmecobius (fig. 41), and several dasyurids (Myoictis, Neophascogale, and some specimens of Phascogale). This character was treated as ordered (additive) in all the analyses reported below.

FIG. 13.

Occlusal views of right deciduous third upper premolars (dP3) of Dromiciops gliroides (A, FMNH 127415), Sminthopsis crassicaudata (B, AMNH 196686), and Echymipera kalubu (C, AMNH 221654). Alternative states of character 120 (see main text for description of this character and character states) are illustrated as follows: Dromiciops 120(0); Sminthopsis 120(1); Echymipera: 120(2). Note the obvious wear facet on the postmetacrista of dP3 in S. crassicaudata, indicating that this tooth is occlusally functional.

img-z117-1_01.jpg

Character 120. Deciduous upper premolar (dP3) large and molariform, occluding with both dp3 and m1 (0); or dP3 smaller and occluding only with m1 (1); or dP3 very small, nonoccluding, and structurally simplified, or absent (2). The deciduous upper premolar is a large, molariform tooth that occludes with both the deciduous lower premolar (dp3) and with m1 (state 0) in most Recent didelphids (Voss and Jansa, 2009). We scored †Herpetotherium as state 0 based on Rothecker and Storer's (1996) description of dP3 in the taxon they identified as †Herpetotherium cf. marsupium, but this tooth is currently unknown for many other nonmarsupial metatherians, including †Pucadelphys, †Mayulestes, and †Mimoperadectes. Among marsupials, dP3 is also a large molariform tooth in †Sparassocynus (Beck and Taglioretti, 2020), †Thylophorops, Dromiciops (fig. 13A; see also Hershkovitz, 1999: fig. 30), macropodiforms, phalangerids, †Namilamadeta (Pledge, 2005: 150–152, fig. 11; but note that this author referred to dP3 as “DP2,” an obvious lapsus), †Nimbadon (Black and Hand, 2010), and †Ngapakaldia (based on SAM P35313, a juvenile †N. bonythoni).

Among other marsupials for which we were able to examine juvenile material, dP3 is variously reduced in size, functional relationships, and occlusal complexity. This tooth is small, does not occlude with dp3, but maintains occlusal contact with m1 (state 1) in some dasyurids (e.g., Antechinomys, Murexia, and Sminthopsis [fig. 13B]), Macrotis, and Chaeropus. By contrast, dP3 is very small, does not occlude with any element of the lower dentition, and retains few (if any) distinct occlusal features (state 2) in Hyladelphys (Voss et al., 2001), Caenolestes (Luckett and Hong, 2000), some dasyurids (e.g., Dasyuroides, Myoictis, Planigale), Thylacinus (Flower, 1866; 1867; Luckett et al., 2019), some peramelemorphians (Echymipera [fig. 13C], Isoodon, Microperoryctes, Perameles, Peroryctes, Rhynchomeles), Distoechurus, petaurids, and pseudocheirids. As noted above, Dasyurus lacks P3 but we observed a vestigial tooth that appears to represent dP3 in some specimens (e.g., AM 35911; see also Woodward, 1896), so we also scored Dasyurus as state 2.

In Phascolarctos, dP3 is vestigial and appears to be resorbed early in development (Thomas, 1887b), and so we scored this taxon as state 2. The condition in vombatids is more equivocal. As already discussed (see char. 114), we assume that the tooth immediately anterior to M1 in vombatids is P3 rather than a retained dP3. Röse (1893: fig. 3) illustrated a developing dP3 in Vombatus, but the ultimate fate of this tooth is uncertain. However, the youngest juvenile specimens of Vombatus and Lasiorhinus we examined have P3 already erupting, with no evidence of dP3, which we interpret as indicating that dP3 was either absent or vestigial and lost very early in development. We therefore scored Vombatus and Lasiorhinus as state 2. However, we scored †Warendja as unknown (“?”) due to an absence of juvenile specimens for this taxon.

The dP3 of Myrmecobius—which is retained into adulthood and positioned behind P3 (see char. 121)—varies considerably in morphology between individuals and on opposite sides of the same skull, being a small, essentially featureless unicuspid in some cases (e.g., on the right side of UMZC A6.41/3) but bicuspid (on the left side of UMZC A6.41/6) or tricuspid and distinctly premolariform (on the right side of UMZC A6.41/6) in others. In addition to this variability, the difficulty of meaningful scoring of Myrmecobius for this character is further complicated by the fact that the entire postcanine dentition of this taxon is structurally simplified compared with those of most other metatherians, and there is minimal occlusal contact between the upper and lower postcanines as a whole; therefore, Myrmecobius has been scored as inapplicable (“-”) for this character.

Because these states appear to represent successive stages of reduction and loss, we treated this character as ordered (additive) in all our analyses.

Character 121. Deciduous upper premolar (dP3) absent in adults (0); or retained in adults (1). There has been considerable debate in the literature about whether the large number of postcanine teeth in Myrmecobius (eight or nine depending on the specimen, versus a maximum of seven in all other metatherians included in our analysis; char. 130) is in part due to retention of the deciduous premolars (Winge, 1882; Dependorf, 1898; Tate, 1951; Forasiepi and Sánchez-Villagra, 2014) or whether dP3 and dp3 are replaced by P3 and p3 respectively (Thomas, 1887a; Leche, 1891; Bensley, 1903). Part of this controversy stems from the fact that dP3 and dp3 appear to be more variable in morphology and size in Myrmecobius than in other marsupials (see chars. 120, 157), leading to these teeth sometimes being misidentified as molars.

On the assumption that dP3 develops and erupts before M1 in Myrmecobius (as in other marsupials; Wilson and Hill, 1897; Fosse, 1969; Fosse and Risnes, 1972; Archer, 1974a, 1974b; Berkovitz, 1978; Kirkpatrick, 1978; Luckett and Woolley, 1996; van Nievelt and Smith, 2005a; Black et al., 2010; Luckett et al., 2019), we were able to confidently identify dP3 in very young juvenile specimens of this taxon (e.g., BMNH 1844.9.2.18, 1844.9.2.19 and 1844.1.22). None of the older Myrmecobius specimens that we examined (which included juveniles, subadults, and adults) show any evidence of dP3 being evicted by P3, whereas a tooth that closely resembles the unequivocal dP3 of juveniles is consistently present in adults of this taxon (contra Bensley, 1903). Thus, we conclude that dP3 is retained in Myrmecobius.

In the lower dentition, Thomas (1887a: plate 27 fig. 9) and Bensley (1903: fig. 3) illustrated juvenile BMNH specimens that they interpreted as indicating replacement of dp3 by p3. Neither Thomas (1887a) nor Bensley (1903) gave catalog numbers, but it is clear from our reexamination of BMNH material that the specimen illustrated by Thomas (1887a: plate 27 fig. 9) and Bensley (1903: fig. 3c) is BMNH 44.1.22.21,10 whereas the second specimen illustrated by Bensley (1903: fig. 3a, b) is BMNH 44.9.2.18. Unfortunately, the teeth from the left side of BMNH 44.9.2.18 that Bensley (1903: fig. 3a) identified as dp3 and m1 are now missing from this specimen, but the equivalent teeth remain in place on the right side, and we concur with Bensley's (1903: fig. 3a, b) proposed homologies. As for the second specimen illustrated by Thomas (1887a: plate 27 fig. 9) and Bensley (1903: fig. 3c)—namely BMNH 1844.1.22.21—we believe that both of these authors wrongly identified dp3 as m1 and that the true m1 is the more posterior tooth that Thomas (1887a: plate 27 fig. 9) labelled as m2. As a result, we interpret the so-called sixth lower molar seen in some Myrmecobius specimens (Thomas, 1887a; Thomas, 1888; Bensley, 1903; Archer and Kirsch, 1977; Archer, 1984c) as m5.

Our examination of a large sample of Myrmecobius individuals of various ages suggests that (as in the upper dentition) the sizes of dp3 and of the lower molars vary considerably in this taxon. Although the tooth we identify as dp3 in BMNH 1844.1.22.21 is somewhat larger than the corresponding tooth of BMNH 1844.9.2.18, they share a similar crown morphology (namely three cusps arranged anteroposteriorly), and we are confident that they are homologous. By contrast, m1 of both specimens exhibits a distinctly triangulated trigonid and talonid. In other, older specimens, however, the protoconid and hypoconid are often lost as the result of wear, resulting in an apparently linear tricuspid morphology. We therefore agree with Winge (1882; see also Dependorf, 1898; Tate, 1951; Forasiepi and Sánchez-Villagra, 2014) that Myrmecobius retains both dP3 and dp3 into adulthood (state 1), presumably because elongation of its jaws means that this tooth is no longer evicted by the erupting third premolar, which, instead, erupts anterior to it. Retention of dP3 and dp3 do not appear to vary independently, so we scored presence/absence of these teeth within the context of a single character. In other metatherians known to have dP3/dp3 as juveniles (i.e., those not scored as unknown [“?”] for char. 120), these teeth are replaced by P3/p3 in adulthood (state 0), so state 1 is an autapomorphy of Myrmecobius.

Character 122. Third upper premolar (P3) present (0); or absent (1). As already discussed (see char. 114), P3 is present (state 0) in most metatherians, but this tooth is consistently absent (state 1) in Dasyurus, Dasykaluta, and Sarcophilus (Thomas, 1887a; Woodward, 1896; Bensley, 1903; Tate, 1947; Bartholomai, 1971a; Archer, 1976c, 1982b, 1984c; Wroe and Mackness, 2000a, 2000b). Two other dasyurids, namely Dasycercus (Woolley et al., 2013) and Pseudantechinus (Spencer, 1896; Ride, 1964b; Woolley, 2011), are polymorphic (“0+1”) for this character (the specimen of Pseudantechinus illustrated here [fig. 42] retains this tooth).

Character 123. Third upper premolar (P3) conventionally premolariform (0); or P3 semisectorial (1); or P3 fully sectorial (2); or P3 open-rooted (3); or P3 molariform (4); or P3 reduced and single-rooted (5). After the molars, the third premolar position is arguably the most variable dental locus in terms of occlusal morphology within Metatheria. When present (see char. 122), P3 is conventionally premolariform (state 0) in most metatherians, including †Mayulestes, †Allqokirus, †Pucadelphys, †Mimoperadectes, †Herpetotherium, didelphids, paucituberculatans, microbiotheriids, Notoryctes, most dasyuromorphians (except Dasycercus, Dasyuroides, and some specimens of Pseudantechinus), peramelemorphians, acrobatids, Petaurus, Gymnobelideus, and some specimens of †Yalkaparidon (Beck et al., 2014). By contrast, some diprotodontians exhibit very different P3 morphologies that can be partitioned into qualitatively distinct states.

The pseudocheirid P3 is a bulbous tooth, usually bearing three subequal labial cusps connected by tall, sharp, anteroposteriorly oriented crests; a prominent lingual heel gives this tooth a broad, subtriangular outline in occlusal view (state 1; Archer, 1984c: figs. 180, 185, 186). Because the pseudocheirid condition is neither conventionally premolariform nor unambiguously sectorial, we refer to it as semisectorial. Other diprotodontians have semisectorial P3s in which distinct cusps can be recognized but which are connected by tall, sharp, anteroposteriorly oriented crests; in this respect, they somewhat resemble those of pseudocheirids, although the overall morphology of the teeth varies considerably among taxa. Non-pseudocheirid taxa with semisectorial P3s include Dactylopsila (Archer, 1984c: fig. 208), Dactylonax, the wynardiids †Muramura (Pledge, 1987c; 2003: fig. 19.2) and †Namilamadeta (Rich and Archer, 1979), †Lekanoleo (Gillespie, 1997), †Nimiokoala (Black and Archer, 1997b), †Litokoala (Louys et al., 2007; Pledge, 2010: fig. 5), and Macropus (Bartholomai, 1971b), all of which we scored as state 1.

Many other diprotodontians (including phalangerids, burramyids, Phascolarctos, †Thylacoleo, and all macropodiforms except Macropus) have a fully sectorial P3, the dominant crest of which lacks clearly distinguishable cusps (state 2). In available specimens of the species of †Wakaleo used for scoring purposes here, namely †W. vanderleuri, P3 is too worn to determine whether it exhibits a semisectorial or fully sectorial morphology, so we scored this taxon using ambiguity coding (as “1/2”); however, P3 of †W. schouteni is semisectorial (Gillespie et al., 2017).

In vombatids (including †Warendja; Hope and Wilkinson, 1982; Flannery and Pledge, 1987), the tooth we identify as P3 (see char. 114) is evergrowing and open-rooted (state 3), similar in overall morphology to M1–4, but smaller in occlusal area. The P3 of the four diprotodontid taxa included here (†Nimbadon, †Ngapkaldia, †Silvabestius, and †Neohelos) lacks distinct crests, but each has a well-developed cusp in the protocone position, and a cusp in the hypocone (posterolingual) position is sometimes also present (Stirton, 1967; Stirton et al., 1967a; Hand et al., 1993; Murray et al., 2000; Black, 2010; Black and Hand, 2010; Black et al., 2013); we refer to this morphology as molariform (state 4). In Dasycercus and Dasyuroides, P3 is vestigial and single-rooted (state 5; Archer, 1982b: table 2; 1984c: fig. 7; Woolley et al., 2013). Both †Yalkaparidon (Beck et al., 2014) and Pseudantechinus are polymorphic for this character, because P3 is double-rooted and premolariform in some specimens but reduced and single-rooted in others (“0+5”).

Although the transition from premolariform (state 0) to semisectorial (state 1) to fully sectorial (state 2) represents a plausible morphocline, the other states we code for this character cannot be incorporated into a linear transformation sequence. Therefore, we treated this character as unordered (nonadditive) in all the analyses reported below. We scored this character as inapplicable (“-”) for Dasyurus, Dasykaluta, and Sarcophilus (all of which lack P3; see char. 122), and also for Tarsipes, whose entire postcanine dentition is reduced to vestigial elements (see char. 113), none of which can be homologized with P3.

Character 124. Premolariform P3 with well-developed anterior and posterior cutting edges (0); or only posterior cutting edge well developed (1); or neither anterior nor posterior cutting edges well developed (2). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 56; 2009), who observed that a well-developed anterior cutting edge on P3 (state 0) consistently occurs only in Caluromys, Caluromysiops, Glironia, and Hyladelphys. Among other metatherians with a premolariform P3 (see char. 123), only the posterior cutting edge is well developed (state 1) in most taxa examined here. However, P3 lacks both anterior and posterior cutting edges (state 2) in †Barinya (Wroe, 1999; Archer et al., 2016a: 8) and Myrmecobius. Because the states of this character form a plausible morphocline, they have been ordered in all our analyses. We scored this character as inapplicable (“-”) for taxa in which P3 is nonpremolariform or absent (see chars. 122, 123).

Character 125. Distinct posterolingual cusp on semisectorial or fully sectorial P3 absent (0); or present (1). Among the metatherians included in our study that possess a semisectorial or fully sectorial P3 (i.e., those taxa scored as states 1 or 2 for char. 123), a distinct posterolingual cusp is absent (state 0) in thylacoleonids, a few macropodiforms (e.g., †Ekaltadeta [see Wroe, 1996], Hypsiprymnodonbartholomaii and H. moschatus [Woods, 1960; Ride, 1961; Flannery and Archer, 1987b; Bates, et al., 2014], Potorous, and some specimens of Aepyprymnus), most phalangerids (with the exception of Ailurops), burramyids (fig. 46; Ride, 1956; Archer, 1984b: figs. 145, 154), pseudocheirids (Archer, 1984b: figs. 180, 185, 186), Dactylopsila (Archer, 1984b: fig. 208), and Dactylonax (whose teeth may differ in other details of occlusal structure but are similar in this respect). By contrast, a prominent cusp is consistently present lingual to the posterior terminus of the P3 blade (state 1) in Ailurops, wynyardiids (Rich and Archer, 1979; Pledge, 1987c; 2003; 2005), phascolarctids (this cusp is anteroposteriorly elongate and crestlike in Phascolarctos but more cusplike in †Litokoala and †Nimiokoala; Black and Archer, 1997b; Louys et al., 2007; Pledge, 2010: fig. 5; Black et al., 2014a), and most macropodiforms. Rich and Archer (1979) and Pledge (1987c; 2003; 2005) referred to this cusp as a hypocone in the wynyardiids †Namilamadeta and †Muramura. This character was scored as inapplicable (“-”) for taxa in which P3 is either nonsectorial or absent (see chars. 122, 123).

Character 126. Sectorial P3 smooth sided, without enamel ridges (0); or with weakly developed ridges (1); or with well-developed ridges extending from apex to base of crown (2). The sectorial or semisectorial third upper premolars of Dactyopsila (Archer, 1984b: fig. 208), Dactylonax, thylacoleonids, Cercartetus (Archer, 1984b: fig. 154), Phascolarctos (Black et al., 2014a), and many macropodiforms (e.g., Dendrolagus, Macropus, Notomacropus, Osphranter, Onychogalea, Petrogale, and Caloprymnus) are smooth-sided teeth that may be dimpled or creased but lack distinct ridging (state 0). The unworn sectorial P3s of all examined phalangerids, however, have incipient (or vestigial) ridges that are often distinct near the apex of the tooth but do not extend to the base of the crown (state 1), a morphology also seen in some macropodiforms (e.g., Lagostrophus, Setonix, Thylogale, and Potorous). By contrast, the third upper premolars of Burramys (Ride, 1956; Archer, 1984b: fig. 145) and many other macropodiforms (e.g., Dorcopsis, Dorcopsulus, Lagorchestes, †Hadronomas, Aepyprymnus, Bettongia and ?Bettongiamoyesi, Hypsiprymnodonbartholomaii and H. moschatus, †Balbaroo, †Ganawamaya, and †Ekaltadeta; Woods, 1960; Ride, 1961; Woodburne, 1967: fig. 15; Flannery and Archer, 1987a, 1987b; Wroe, 1996; Cooke, 2000; Prideaux, 2004; Kear et al., 2007; Bates et al., 2014) are distinctly ridged from the dental apex to the base of the crown (state 2). Taxa exhibiting the latter condition are sometimes described in the literature as “plagiaulacoid” (after Simpson, 1933).

The lingual and labial faces of P3 in pseudocheirids are characterized by a single broad “rib” that extends vertically from the large central cusp; because it is unclear whether this rib is homologous or not with the multiple, much narrower ridges seen in phalangerids, Burramys, and some macropodiforms, we scored pseudocheirids as unknown (“?”) for this character. Between two and four broad ribs descend vertically from the major cusps of P3 in †Namilamadeta (Rich and Archer, 1979: fig. 1A; Pledge, 2005: fig. 1), †Muramura (Pledge, 1987c: fig. 2; 2003: fig. 19.2), †Nimiokoala (Black and Archer, 1997b), and †Litokoala (Louys et al., 2007: fig. 1; Pledge, 2010: fig. 5); again, because the homology of these structures with the much narrower ridges seen in other diprotodontians is not obvious, we also scored these taxa as unknown. This character was scored as inapplicable (“-”) for taxa in which P3 is neither semisectorial nor fully sectorial or is absent (see chars. 122, 123).

Character 127. Major crest of semisectorial or fully sectorial P3 oriented posterolingual to anterolabial (0); or major crest of semisectorial or fully sectorial P3 parallel to molar row (1); or major crest of semisectorial or fully sectorial P3 oriented posterolabial to anterolingual (2). The cutting edge of the semisectorial or fully sectorial P3 is oriented obliquely to the molar row, slanting from posterolingual to anterolabial in occlusal view (“flexed buccally” sensu Flannery, 1989; state 0) in most phalangerids that we examined, with the exception of Spilocuscus and some specimens of Ailurops. Flannery et al. (1987: table 2, char. 15) did not score Phalanger as having an obliquely oriented P3, but all the Phalanger material that we examined is indistinguishable from most other phalangerids in this respect (see, e.g., Norris and Musser, 2001: fig. 5). The same posterolingual-to-anterolabial orientation is seen in several macropodiforms, namely Hypsiprymnodon moschatus (fig. 52; Woods, 1960; Ride, 1961) and H.bartholomaii (see Flannery and Archer, 1987c: fig. 1b), †Balbaroo (Cooke, 2000: fig. 2), †Ganawamaya (Kear et al., 2007: figs. 1, 3.1), and †Ekaltadeta (Wroe, 1996b: fig. 1.1; Wroe et al., 1998: fig. 1.3).

By contrast, the major crest of the semi- or fully sectorial P3 is more or less in line with the long axis of the cheektooth row (state 1), in pseudocheirids, burramyids, wynyardiids, phascolarctids, thylacoleonids, most macropodiforms, and Spilocuscus (Flannery et al., 1987: fig. 15). Ailurops is variable for this character, with P3 oriented posterolingual to anterolabial in some specimens (e.g., AMNH 196495; see also Flannery et al., 1987: fig. 2A) but in line with the molar row in others (e.g., AMNH 153361), so we scored it as polymorphic (“0+1”).

A third, qualitatively distinct morphology is seen in Dactylopsila (Archer, 1984c: fig. 208) and Dactylonax: the semisectorial P3 of these taxa is oriented obliquely but in the opposite direction (i.e., posterolabial to anterolingual; state 2) to that seen in phalangerids and some macropodiforms. These differing orientations of sectorial or semisectorial P3 among diprotodontians were well illustrated by Ride (1956: fig. 3); because they form a plausible transformation series, we ordered this character in all of our analyses. The orientation of the occluding mandibular tooth (p3) is an obviously nonindependent trait that we did not code separately for phylogenetic analysis. This character was scored as inapplicable (“-”) for taxa in which P3 is neither semisectorial nor fully sectorial or is absent (see chars. 122, 123).

Upper Molars

Available evidence (see, e.g., Kielan-Jaworowska et al., 2004) supports the hypothesis that the last common therian ancestor was characterized by a tribosphenic molar dentition (sensu Simpson, 1936) in which each upper molar had three well-developed cusps (paracone, metacone, and protocone) forming a triangular array with the protocone at its lingual apex (figs. 14, 15A). The upper molar dentition of nonmarsupial metatherians, didelphids, microbiotheriids, and many dasyuromorphians closely approximates this ancestral morphology, in which the principal cusps are interconnected by crests (cristae) that are named following widely accepted conventions (figs. 14, 15A; see, e.g., Bown and Kraus, 1979: fig. 9-1; Reig et al., 1987: fig. 1; Wroe, 1999: text-fig. 1A; Goin and Candela, 2004: fig. 2A; Kielan-Jaworowska et al., 2004: fig. 11.1; Williamson et al., 2014: fig. 3). Some nonmarsupial metatherians and tribosphenic marsupials also have additional small cusps (known as conules) that occupy the pre- and postprotocristae; of these, the paraconule (on the preprotocrista) is usually absent or vestigial, but the metaconule (on the postprotocrista) is sometimes distinct. In all the taxa properly considered tribosphenic in this report, the postparacrista is connected with the premetacrista, forming the so-called centrocrista (figs. 14, 15A).

FIG. 14.

Occlusal views of left upper and right lower didelphid molars illustrating features of tribosphenic upper molar crown morphology discussed in the text. Abbreviations: acid, anterior cingulid; alci, anterolabial cingulum; crdob, cristid obliqua; ecf, ectoflexus; encd, entoconid; encrd, entocristid; hycd, hypoconid; hycld, hypoconulid; hycld.n., hypoconulid notch; hyfd, hypoflexid; mec, metacone; mecd, metaconid; mecrd, metacristid; pac, paracone; pacd, paraconid; pacrd, paracristid; pcid, posterior cingulid (not present in most didelphids); pomecr, postmetacrista; popacr, postparacrista; poprcr, postprotocrista; prc, protocone; prcd, protoconid; prmecr, premetacrista; prpacr, preparacrista; prprcr, preprotocrista; stB, stylar cusp B; stD, stylar cusp D.

img-z123-1_01.jpg

The upper molars of most tribosphenic metatherians are also characterized by several more or less distinct cusps on the stylar shelf, a (usually broad) platform labial to the paracone and metacone (figs. 14, 15A). Most recent studies have employed some version of Clemens' (1966) terminology for these structures, recognizing five more or less distinct positions: (1) stylar cusp A (stA, sometimes referred to as the parastyle) is anterolabial to the paracone and is often indistinct or soon reduced by wear; (2) stylar cusp B (stB), almost always larger than stA, is more or less labial to the paracone (to which it is often connected via the preparacrista), and appears to be homologous with the stylocone (or cusp “B”) of various Mesozoic mammals (Patterson, 1956; Kielan-Jaworowska et al., 2004); (3) stylar cusp C (stC, sometimes referred to as the mesostyle), when present, is posterolabial to the paracone and anterolabial to the metacone; (4) stylar cusp D (stD) is labial to the metacone; and stylar cusp E (stE, sometimes referred to as the metastyle), often indistinct or absent, is posterolabial to the metacone. Although this strictly topological scheme has the advantage of consistency, other authors have modified it somewhat to maintain assumed homologies between stylar cusps that appear to have shifted position.11

FIG. 15.

Occlusal views of left upper second molars (M2) of Metachirus nudicaudatus (A, AMNH 266453), Caenolestes fuliginosus (B, UMMZ 155575), Echimipera kalubu (C, AMNH 221654), and Notoryctes typhlops (D, AMNH 198651). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 132, 133, 136, 137, 140, 141, 143, and 144 (see main text for description of these characters and character states), as follows: Metachirus 132(1), 133(0), 136(0), 137(1), 140(3), 141(0), 143(0), 144(0); Caenolestes 132(-), 133(1), 136(2), 137(-), 140(-), 141(0), 143(1), 144(0); Echimipera 132(1), 133(0), 136(0), 137(1), 140(4), 141(0), 143(1), 144(0); Notoryctes 132(1), 133(0), 136(0), 137(2), 140(-), 141(0), 143(0), 144(0). Note, however, that some C. fuliginosus specimens retain a paracone and can be scored as 137(1). Abbreviations: alci, anterolabial cingulum; ccr, centrocrista; mec, metacone; mecl, metaconule; nci, neomorphic cingulum; pac, paracone; prc, protocone; stB, stylar cusp B; stD, stylar cusp D. Teeth are not drawn to the same scale.

img-z124-1_01.jpg

There are, in addition to the tribosphenic morphotype just described, several nontribosphenic molar bauplans represented among the marsupial taxa included in this study. Fortunately, intermediate phenotypes link most nontribosphenic designs to their tribosphenic precursors, allowing cusp homologies to be identified in many otherwise problematic cases (Archer, 1976e; Ride, 1993; Abello, 2013). The details by which relevant cusp homologies have been inferred are discussed in the following accounts, but it is useful to give a brief description here of each of the major upper molar variants, illustrating examples and providing our terminology for trenchant occlusal features.

The upper molars of caenolestids (fig. 15B; Abello, 2013: fig. 2B) differ from the tribosphenic morphotype by having hypertrophied stylar elements (stB and stD) as the main labial cusps, the paracone having been lost by fusion with stB (a remnant of the paracone may be identifiable in the unworn teeth of some caenolestid specimens; see char. 136), and the metacone being much reduced in size and partially fused to stD. In addition, the dental outline is quadrangular rather than triangular because an enlarged metaconule occupies the posterolingual (“hypocone”) position. Other caenolestid departures from the tribosphenic plan include loss of the centrocrista (postparacrista + premetacrista)—the result of fusion of the paracone to stB—and the presence, in some taxa, of a neomorphic labial cingulum. This molar bauplan lacks a name, although some of its distinctive features are shared among other paucituberculatans and a few diprotodontians.

Recent peramelemorphian molars do not depart quite so markedly from the tribosphenic design, but two differences are noteworthy (fig. 15C). The first is that the centrocrista is usually interrupted by a gap between the postparacrista and the premetacrista (both of which terminate at or near the labial margin of the tooth), effectively dividing the stylar shelf into separate V-shaped segments. The second is the typically quadrangular occlusal outline that, as in caenolestids, is usually the consequence of a hypertrophied metaconule in the “hypocone” position. Other unusual features of peramelemorphian upper molars are restricted to single teeth (e.g., M1) or to single taxa (e.g., Macrotis) as described below.

The molars of Notoryctes (fig. 15D; Murray and Megirian, 2006b: fig. 6I) and †Yalkaparidon (Archer et al., 1988; Beck, 2009; Beck et al., 2014) are zalambdodont, a distinctive morphotype that is also seen among several groups of placental mammals (e.g., chrysochlorids, tenrecids, and solenodontids; Asher and Sánchez-Villagra, 2005). Zalambdodont molars are characterized by suppression of either the paracone or the metacone and by the hypertrophy of the remaining cusp, which becomes the dominant upper molar feature; there is little or no occlusion between the protocone and the talonid basin of the lower molars, and the talonid itself is usually reduced or entirely absent (Asher and Sánchez-Villagra, 2005). The dominant upper molar cusp in Notoryctes is the metacone, but this taxon also retains a protocone, and the unworn teeth of some specimens retain what may be an anterolabial vestige of the paracone (fig. 15D; Asher and Sánchez-Villagra, 2005; Murray and Megirian, 2006b: fig. 6I; Archer et al., 2011). Based on occlusal relationships, the dominant upper-molar cusp of †Yalkaparidon seems likely to be the metacone (Beck et al., 2014), but this homology remains open to question in the absence of annectant taxa (see char. 137).

FIG. 16.

Occlusal views of left upper second molars (M2) of Phascolarctos cinereus (A, AMNH 173504) and Pseudochirops cupreus (B, AMNH 109633). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 132, 133, 136, 137, 140, 141, 143, and 144 (see main text for description of these characters and character states), as follows: Phascolarctos 132(0), 133(0), 136(0), 137(0), 140(4), 141(0), 143(1), 144(0); Pseudochirops cupreus 132(0), 133(0), 136(0), 137(0), 140(5), 141(0), 143(1), 144(0). Abbreviations: mec, metacone; mecl, metaconule; pac, paracone; pacl, paraconule; prc, protocone; stB, stylar cusp B; stD, stylar cusp D. Teeth are not drawn to the same scale.

img-z126-1_01.jpg

Several molar morphotypes among diprotodontians have traditionally been recognized (e.g., by Bensley, 1903; Archer, 1976e; Archer, 1984c). In the so-called selenodont condition the dominant molar cusps and their associated crests somewhat resemble crescent moons, as they do in the selenodont teeth of various placental clades (e.g., numerous artiodactyl groups; Loomis, 1925; Fortelius, 1985; Prothero and Foss, 2007). Among Recent marsupials, the selenodont condition is exemplified by Phascolarctos (fig. 16A) and pseudocheirids (fig. 16B; see also Archer, 1976e; Archer, 1984c), but this superficial resemblance masks underlying differences in cusp homologies. Whereas the dominant labial cusps of phascolarctids are the paracone and metacone (as in tribosphenic marsupials), the dominant labial and lingual cusps of pseudocheirids appear to be hypertrophied stylar elements (nominally stB and stD), with the paracone and metacone reduced to smaller anterointernal and posterointernal elements, respectively (see discussion under char. 136).

Other diprotodontians—except Tarsipes, whose postcanine dentition is vestigial (fig. 49; char. 113)—have nonselenodont upper molars that have variously been described as bunodont, bunolophodont, or lophodont (fig. 17). These distinctions, which are not always consistently recognized in the literature, refer to the presence or absence of transverse crests or lophs that (when fully developed) connect the principal labial cusps—which we identify as hypertrophied stylar elements in the B and D positions (Ride, 1971; see also Archer, 1976e, 1984c: 683; Tedford et al., 1977; Woodburne and Clemens, 1986a: 96–97, 1986b: 20–21, Tedford and Woodburne, 1987: 416; Woodburne et al., 1987a: 600; Ride, 1993; Crosby and Archer, 2000: 331; Beck et al., 2020)—with their lingual counterparts (protocone and metaconule). Such lophs are altogether absent in some taxa (e.g., Acrobates, Dactylopsila, and †Thylacoleo) that might properly be called bunodont, but they are increasingly distinct in others (e.g., Distoechurus [fig. 17A], Hypsiprymnodon [fig. 17B], and phalangerids [fig. 17C]) that we refer to as bunolophodont (after, e.g., Case, 1984; Flannery et al., 1984; Woodburne, 1984a). In fully lophodont taxa (such as diprotodontids and Recent macropodids; fig. 17D), the cusps are fully incorporated into tall transverse crests and no longer appear as distinct structures. In bunolophodont and fully lophodont forms, the anterior crest (connecting stB with the protocone) is commonly referred to as the protoloph, and the posterior crest (connecting stD with the metaconule) as the metaloph (fig. 17). A median crest that sometimes connects the protoloph and metaloph and that appears to be homologous with the postprotocrista in macropodiforms (e.g., Cooke, 2000; Kear et al., 2007; Prideaux and Warburton, 2010: char. 25; Travouillon et al., 2015a; Butler et al., 2016) has been referred to as the “midlink” by some authors (e.g., Stirton, 1955; Tedford, 1966; Woodburne, 1967; Bartholomai, 1971b; Sanson, 1980). Another that projects posteriorly from the metaloph has been alternatively referred to as the “postlink” (e.g., Flannery, 1983; Flannery and Rich, 1986), the “neometaconule crista” (Travouillon et al., 2014b, 2015a, 2016; Butler et al., 2016), or the “urocrista” (e.g., Prideaux, 2004; see char. 146). Among the taxa we examined, the midlink (formed by the postprotocrista) and postlink/neometaconule crista/urocrista occur only in some macropodiforms (fig. 17D).

Character 128. Molars with closed roots and static crowns (0); or molars with open roots, the crowns growing throughout adulthood (1). Most adult metatherians have molars with closed roots and static crowns (state 0), but vombatids—including †Warendja (see Hope and Wilkinson, 1982; Flannery and Pledge, 1987; Brewer, 2007)—have open-rooted, evergrowing cheekteeth (state 1) corresponding to the “hypselodont” condition of Simpson (1970) and the “euhypsodont” condition of Mones (1982). The molar roots of Isoodon and Macrotis appear to close late in ontogeny (e.g., the molar roots of AMNH 35717, a young adult specimen of Macrotis lagotis, are still partially open), but differ from those of vombatids, which remain widely open even in the oldest examined individuals. Therefore, we scored Isoodon and Macrotis as state 0.

Character 129. M1–4 present (0); or M4 absent (1); or M2–4 absent (2). Most metatherians with clearly differentiated cheek teeth have four upper molars (state 0). By contrast, M4 is absent (state 1) in acrobatids (e.g., Distoechurus [fig. 49]) and †Wakaleo (see Gillespie et al., 2017). We scored Cercartetus as state 0 based on C. caudatus, which retains a small M4, but this tooth is absent in C. concinnus and C. nanus (see Archer, 1984c: figs. 154–156). †Thylacoleo retains only a single upper molar, which, based on its position and morphology in comparison to teeth of other thylacoleonids, represents M1 (state 2). Postcanine homologies are not resolvable in Tarsipes (char. 113), so the precise number of molars cannot be determined for this taxon (Bensley, 1903; Archer, 1984c; Russell and Renfree, 1989), which we scored as unknown (“?”).

All examined taxa with four upper molars also have four lower molars, with the unique exception of Myrmecobius, in which some specimens exhibit a supernumerary fifth lower molar (referred to by some previous authors, e.g., Thomas, 1887a; Thomas, 1888; Bensley, 1903; Archer and Kirsch, 1977; Archer, 1984c as a “sixth” molar; see discussion under char. 121). Similarly, all taxa with three upper molars also have three lower molars. Although †Thylacoleo has one upper molar and two lower molars (Archer and Dawson, 1982; Archer and Rich, 1982; Archer, 1984c: fig. 128; Rauscher, 1987; Gillespie, 1997, 2007; Gillespie et al., 2016), it seems probable that the numbers of upper and lower molars are evolutionarily correlated, and on this assumption we did not score them as separate characters. Although not included here as a terminal, Petrogale (= “Peradorcas”) concinna is unique among known metatherians in exhibiting continuous molar replacement (Thomas, 1904; Tate, 1948; Sanson, 1989; Rodrigues et al., 2017). Recognizing the progressive loss of more posterior molars as steps in an ordered transformation series, we ordered this character in all our analyses.

FIG. 17.

Occlusal views of left upper molars of Distoechurus pennatus (A, AMNH 104058 [M1]), Hypsiprymnodon moschatus (B, AMNH 160120 [M2]), Trichosurus vulpecula (C, AMNH 65547 [M2]), and Lagostrophus fasciatus (D, AMNH 155105 [M2]). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 132, 133, 136, 137, 140, 141, 143, 144, 145, and 146 (see main text for description of these characters and character states), as follows: Distoechurus 132(-), 133(1), 136(2), 137(-), 140(-), 141(0), 143(1), 144(1), 145(0), 146(0); Hypsiprymnodon moschatus 132(-), 133(0), 136(2), 137(-[but note that this can be scored in lightly worn specimens of H. moschatus; see fig. 18A]), (140(-), 141(0), 143(1), 144(1), 145(-), 146(-); Trichosurus vulpecula 132(-), 133(0), 136(2), 137(-), 140(-), 141(0), 143(1), 144(1), 145(0), 146(0); Lagostrophus 132(-), 133(0), 136(2), 137(-), 140(-), 141(0), 143(1), 144(2), 145(1), 146(1). Note that in the illustrated specimen of D. pennatus the neomorphic labial cingulum is restricted to the region between the major labial cusps, but in other specimens it extends along the entire labial margin of the upper molars. Abbreviations: mec, metacone; mecl, metaconule; mel, metaloph; prc, protocone; prl, protoloph; stB, stylar cusp B; stD, stylar cusp D; urcr, urocrista. Teeth are not drawn to the same scale.

img-z128-1_01.jpg

Character 130. P3 erupts after M4 (0); or P3 and M4 erupt simultaneously (1); or P3 erupts before M4 but after M3 (2); or P3 and M3 erupt simultaneously (3); or P3 erupts before M3 (4). The timing of the eruption of P3 relative to the upper molars has been shown to encompass potentially useful phylogenetic information among didelphids (Tribe, 1990; Voss and Jansa, 2003: char. 61; 2009), and additional phylogenetically informative variation occurs among other metatherians. In most examined taxa (including all didelphids), P3 erupts after M3, but P3 can erupt after M4 (state 0), simultaneously with M4 (state 1) or before M4 (state 2; Tribe, 1990; Hershkovitz, 1999; Voss and Jansa, 2003: char. 61; 2009). By contrast, P3 erupts earlier, either simultaneously with (state 3) or before M3 (state 4), in petaurids, pseudocheirids (with the possible exceptions of Petauroides and Pseudochirops archeri, for which we were unable to examine appropriate juvenile or subadult material), burramyids, Acrobates (but not Distoechurus, in which P3 erupts after M3; e.g., in AMNH 191045), vombatids, Phascolarctos, Thylacinus, and Myrmecobius (Flower, 1867).

We are unaware of intact maxillary specimens of juveniles or subadults of †Herpetotherium, which we scored as unknown (“?”). We examined several subadult specimens of †Pucadelphys in which M3 is fully erupted and P3 is incompletely erupted, but none preserve M4 (e.g., MNHC 8265; see Cifelli and Muizon, 1998a: fig. 1B; Muizon et al., 2018: 373–375); on the assumption that M4 was not fully mineralized or was still within its crypt in these subadult individuals, we scored †Pucadelphys as state 2. Cifelli and Muizon (1998a) discussed a supposed subadult specimen of †Mayulestes (MNHC 8267), but this specimen was subsequently referred to †Allqokirus (see Muizon et al., 2018), so the eruption pattern of †Mayulestes is currently unknown; however, based on MNHC 8267 (in which M3 is fully erupted, P3 is nearly fully erupted, and M4 is partially erupted; Cifelli and Muizon, 1998a; Muizon et al., 2018), †Allqokirus can be scored as state 2. Several taxa that could not be assigned a unique character state based on available specimens (e.g., Petauroides, Pseudochirops archeri, Distoechurus, Strigocuscus celebensis, Wyulda, †Ekaltadeta, †Neohelos) were scored using appropriate ambiguity codings. Because these states form a plausible transformation series, we ordered this character in all our phylogenetic analyses.

Character 131. Molar progression leading to eviction of P3 absent (0); or present (1). In most metatherians, there is no major change in the position of the molars within the maxilla and dentary over ontogeny (state 0). However, in some macropodids there is a gradual anterior or mesial movement by the upper and lower molars with age, a phenomenon known as molar progression (Tate, 1948; Ride, 1957; Sanson, 1980, 1989; Lentle and Hume, 2010; Prideaux and Warburton, 2010: char. 22). The degree of molar progression is often slight and can be hard to recognize in the absence of convenient landmarks, but molar progression in several taxa leads to the eviction of P3 and p3 from the upper and lower jaws (state 1). In older individuals of Macropus and Osphranter, the anterior molars are also evicted (see, e.g., Kido et al., 2018: fig. 2d), and Sanson (1980) reported that some very old individuals retain only a single molar as a result of this process.

Character 132. Upper molars without a distinct ectoflexus on any tooth (0); or distinct ectoflexus present on one or more teeth (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 59; 2009), who defined the ectoflexus as a “V-shaped labial indentation of the stylar shelf” (see fig. 14). Caluromys and Caluromysiops are the only Recent didelphids that lack a distinct ectoflexus on all their upper molars (Voss and Jansa, 2009: fig. 21).

Among other metatherians, an ectoflexus is present on at least one molar in †Pucadelphys (Marshall and Muizon, 1995: figs. 6B, 7B; Ladevèze et al., 2011: fig. 4d-f), †Allqokirus (Muizon et al., 2018: figs. 79), †Mayulestes (Muizon, 1998: fig. 2), †Mimoperadectes (Horovitz et al., 2009: fig. 1), †Herpetotherium (Horovitz et al., 2008: plate 1A), †Hesperocynus (Forasiepi et al., 2009), †Sparassocynus (Reig and Simpson, 1972: fig. 8; Beck and Taglioretti, 2020: fig. 3a), †Thylophorops (Reig et al., 1987: fig. 28), †Thylatheridium (Reig, 1952: fig. 11A; Reig et al., 1987: fig. 25A), †Evolestes (Goin et al., 2007b: figs. 2, 3), Dromiciops (Marshall, 1982: fig. 16), †Microbiotherium (Marshall, 1982: fig. 9), †Yalkaparidon (Beck, 2009: fig. 6B; Beck et al., 2014), Notoryctes (Figure 15D; Asher and Sánchez-Villagra, 2005: fig. 6A; Murray and Megirian, 2006b: fig. 6I; Archer et al., 2011: fig. 1), peramelemorphians (but note that the weak ectoflexus seen in unworn molars of Macrotis is lost with slight wear), dasyurids, thylacinids, †Muramura (Pledge, 2005: fig. 19.2), and young individuals of Lasiorhinus (Brewer et al., 2018: fig. 13.1). By contrast, the ectoflexus is absent among other diprotodontians that retain a distinct stylar shelf, namely pseudocheirids and phascolarctids (fig. 16).

We scored this character as inapplicable (“-”) in taxa that lack a distinct stylar shelf, as evidenced by fusion between the paracone and stB and between the metacone and stD on all upper molars (see char. 136); such taxa include all examined paucituberculatans except †Evolestes (Goin et al., 2007b; Abello, 2013) and all examined diprotodontians with the exception of phascolarctids (fig. 16A), pseudocheirids (fig. 16B), †Muramura, and Lasiorhinus. In addition, we scored this character as inapplicable in all taxa in which a neomorphic cingulum is present labial to the stylar region (see char. 133), because this cingulum fills the region normally occupied by the ectoflexus (if present). Finally, we also scored Myrmecobius as inapplicable, because its highly modified upper molar morphology makes it unclear as to whether a true stylar shelf is present (Archer, 1984c: fig. 23), and Tarsipes, because its vestigial postcanine dentition (Parker, 1890; Bensley, 1903; Archer, 1984c; Russell and Renfree, 1989) precludes meaningful scoring of this character (see char. 113).

Character 133. Neomorphic labial cingulum absent (0); or present on M1–3 (1). As discussed by Voss and Jansa (2009: 54), an apparently neomorphic cingulum that extends along the entire labial margin of each upper molar (state 1) is present in Caenolestes (fig. 15B and 18B) and Rhyncholestes. The identification of this ridge as an evolutionary novelty rather than as a remnant of the stylar shelf follows from the interpretation of caenolestid cusp homologies suggested by Osgood (1921) and subsequently accepted by Marshall (1987) and most other recent authors (e.g., Goin and Candela, 1996, 2004; Goin et al., 2007b, 2009a; Abello, 2013; Rincón et al., 2015; Engelman et al., 2016), namely that the major labial cusps in living caenolestids are stylar in origin, rather than representing the paracone and metacone (see char. 136). Among other paucituberculatans, a well-developed cingulum extending along the entire labial margin of the upper molars appears to be absent (state 0) in Lestoros (Martin, 2013), †Evolestes (Goin et al., 2007b: figs. 23), †Acdestis (Engelman and Croft, 2016: fig. 4), and †Pichipilus (Abello, 2007: lámina 6I), but it is identifiable in †Stilotherium (Marshall, 1980: fig. 5). This structure is only variably present in †Palaeothentes, which we scored as polymorphic (“0+1”).

A cingulum is also present labial to stB and stD (the “paracone” and “metacone,” respectively, of previous authors; see char. 136) on the first upper molar of †Lekanoleo (see Gillespie, 1997, and Gillespie et al., 2020, who referred to it as a “stylar basin”) and †Wakaleo (Murray et al., 1987: fig. 8A), both of which we scored as state 1. The occlusal morphology of M1 (the only remaining upper molar) in †Thylacoleo is too heavily modified (Archer and Rich, 1982; Archer, 1984c: fig. 128) to score this character with any confidence, so we scored this taxon as inapplicable (“-”).

A cingulum that extends along the entire labial margin of the upper molars is also present in some specimens of Cercartetus (e.g., AMNH 222711), Acrobates (e.g., AMNH 37185; fig. 18C), and Distoechurus (e.g., AMNH 157169); in other specimens of these three taxa, however, the cingulum does not extend along the entire labial margin of the teeth, but is instead restricted to the region immediately labial to the space between the major labial cusps (stB and stD; see char. 136). Nevertheless, based on the presence of this cingulum, we scored Cercartetus, Acrobates, and Distoechurus as state 1 for this character (see also comments regarding Cercartetus by Turnbull and Schram, 1973: 443).

Character 134. Stylar cusp B (stB) taller than or subequal in height to stD on M2 (0); or stD taller than stB on M2 (1). The number, position, and size of the stylar cusps are commonly cited as characters in metatherian systematics, particularly for fossil taxa known primarily from teeth (e.g., by Wroe et al., 2000; Wroe and Musser, 2001; Goin et al., 2006; Davis, 2007; Ladevèze and Muizon, 2007; 2010; Carneiro and Oliveira, 2017b). However, examination of multiple conspecific individuals of both Recent and fossil taxa often reveals polymorphism in stylar cusp number and morphology (e.g., Bensley, 1906; Korth, 1994; Hershkovitz, 1997: fig. 19; Flores and Abdala, 2001; Hayes, 2005; Martin, 2005; Ladevèze et al., 2011, 2012; Korth, 2018). Despite such variability, the stylar cusp in the B position is consistently taller than or subequal in height to the stylar cusp in the D position (state 0) in most tribosphenic metatherians (Godthelp et al., 1999: 302). This morphology is observed in †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, didelphids, and all peramelemorphians with the exception of †Yarala (Muirhead and Filan, 1995).

By contrast, stD is taller than stB (state 1) in most dasyuromorphians (Wroe, 1997b). This trait is most obvious on M2, where stD is usually greatly enlarged relative to stB (fig. 19A). However, stB and stD are subequal in height on the M2 of some specimens of Thylacinus, Micromurexia, Murexia, Phascolosorex, and Phascomurexia, all of which we scored as polymorphic (“0+1”).12 Stylar cusp D is also taller than stB in †Herpetotherium (Korth, 1994; Horovitz et al., 2008), †Stilotherium (Marshall, 1980: fig. 5), and †Yarala (Muirhead and Filan, 1995). We scored Dromiciops and †Microbiotherium as inapplicable (“-”), because the stylar cusps are not identifiable as discrete structures in these taxa but instead form a continuous, low crest along the labial margin of M2 and M3 (see also Goin et al., 2007b: 600). We scored Myrmecobius as unknown (“?”) because upper molar cusp homologies cannot be identified with confidence in this taxon.

Among metatherians with more derived upper molar morphologies, stB is taller than or subequal in height to stD on M2 in †Evolestes (which retains a clearly tribosphenic bauplan) and in all other paucituberculatans we examined (e.g., Caenolestes; fig. 19B)—in which the major labial cusps are enlarged stB and stD—with the exception of †Stilotherium, in which stD is taller (Marshall, 1980: fig. 5). Although Notoryctes is zalambdodont rather than tribosphenic, the cusps that we identify as stB and stD are still present on the labial margin of the tooth crown (Murray and Megirian, 2006b: fig. 6I), where stB is either slightly taller than or subequal in height to StD.

Small but distinct stylar cusps are identifiable labial to the paracone and metacone of M2 in Phascolarctos and †Litokoala (Black et al., 2014a), with stB slightly taller than or subequal in height to stD (state 0); however, stylar elements are indistinct in †Nimiokoala, which we scored as inapplicable (“-”). Among other diprotodontians, scoring this character is somewhat less straightforward. As discussed in greater detail below (char. 136), the cusps conventionally identified as the paracone and metacone in pseudocheirids appear to correspond to stB and stD, respectively. In all pseudocheirids included in this analysis, stB (the “paracone” of previous authors) is taller than or subequal in height to stD (the “metacone” of previous authors). We also consider the labial ends of the protoloph and metaloph of lophodont and bunolophodont diprotodontians to be homologous with stylar cusps B and D, respectively (see char. 136; see also Ride, 1971; Archer, 1976e, 1984c: 683; Tedford et al., 1977; Woodburne and Clemens, 1986a: 96–97, 1986b: 20–21; Tedford and Woodburne, 1987: 416; Woodburne et al., 1987a: 600; Ride, 1993; Crosby and Archer, 2000: 331; Beck et al., 2020), so this character can be scored for such taxa based on the relative heights of the structures that most previous authors have identified as the paracone and metacone. Stylar cusp B (by our interpretation) is taller than or subequal in height to stD in all of the lophodont and bunolophodont diprotodontians we examined, with the exception of some specimens of Dorcopsulus and Lagorchestes (in which stD is taller than stB); we scored both of the latter taxa as polymorphic (“0+1”).

Character 135. Stylar cusp D (stD) shorter than or subequal in height to the metacone on M2 (0); or stD markedly taller than the metacone on M2 (1). In most metatherians in which stD and the metacone can both be identified as distinct cusps on M2, stD is shorter than or subequal in height to the metacone (state 0). This trait is seen in †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, †Herpetotherium, didelphids, dasyurids (e.g., Pseudantechinus; fig. 19A), thylacinids, peramelemorphians (except Macrotis, which is inapplicable for this character), and †Evolestes. In other paucituberculatans (e.g, Caenolestes; fig. 19B), however, the metacone is markedly shorter than stD (state 1) and is partially fused to the lingual face of that cusp (figs. 18B, 19B; see also Marshall, 1980: fig. 5c; Abello, 2007, 2013; Goin et al., 2009a).

We scored this character as inapplicable (“-”) for most lophodont and bunolophodont diprotodontians, because the metacone is no longer identifiable as a distinct cusp in such taxa (having been incorporated into the metaloph; see char. 136). The exceptions are Hypsiprymnodon moschatus and H.bartholomaii, in both of which a cusp that we consider to be homologous with the metacone—but identified as a “neometaconule” by Flannery and Archer (1987c)—is identifiable lingual to stD (figs. 17B, 18A; see also Ride, 1993). In both H. moschatus and H.bartholomaii, the metacone is subequal in height to stD, so we scored these taxa as “0.” Among other diprotodontians, stD is shorter than or subequal in height to the metacone in Phascolarctos, †Litokoala, †Muramura, †Namilamadeta, and (based on unworn molars of juvenile specimens) Lasiorhinus. By contrast, stD (the “metacone” of previous authors; see char. 136) is clearly taller than the metacone (the “neometaconule” of previous authors) in all examined pseudocheirids. Because stylar cusps are indistinct in †Nimiokoala, we scored this taxon as inapplicable (“-”). We also scored Macrotis as inapplicable because the metacone has been displaced lingually in this taxon (see char. 143) and is no longer in close proximity to StD.

Character 136. Paracone not completely fused with any stylar elements (0); or paracone completely fused with stB on M1 but not on more posterior teeth (1); or paracone completely fused with stB on all upper molars (2); or paracone unfused with stB on M1 and M2 but completely fused with stB on M3 and M4 (3). The paracone is remote from the labial margin on all the upper molars and is not fused with any stylar element (state 0) in most tribosphenic metatherians (e.g., †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, †Herpetotherium, didelphids [figs. 14, 15A], peramelemorphians [fig. 15C], most dasyurids [fig. 19A], and most thylacinids). By contrast, the paracone lies on the labial margin of M1 and appears to have completely fused with stB on that tooth, but not on more posterior molars (state 1), in some dasyurids (including Dasycercus, Dasyuroides, Dasyurus, and Sarcophilus; Archer, 1976c, 1982b, 1984c; Wroe, 1996a, 1998; Wroe and Mackness, 2000a) and in †Badjcinus (Muirhead and Wroe, 1998: 613). The same morphology is seen in some, but not all, specimens of Phascolosorex and Pseudantechinus, which we scored as polymorphic (“0+1”).

A single cusp close to the anterolabial margin of M1 in Dromiciops (see Marshall, 1982: fig. 15) and †Microbiotherium superficially resembles the structure that results from fusion of the paracone with stB in some dasyuromorphians (the taxa we scored as state 1, above). However, in both microbiotheriids the stylar shelf of M1 is narrowly continuous along the entire labial margin of the tooth, whereas the stylar shelf is discontinuous anteriorly in dasyuromorphians with fused paracone + stB. This observation, together with the fact that stB is not identifiable as a clearly distinct structure on M2–4 of Dromiciops and †Microbiotherium, suggests that the anterolabial cusp on M1 in these taxa is nothing more than a labially displaced paracone. Therefore, we scored them both as state 0.

We agree with Osgood (1921: 122–126) and Marshall (1987: 140) that the paracone is fused with stB on all of the upper molars of most paucituberculatans (state 2; see also Goin and Candela, 1996, 2004; Goin et al., 2007b; Abello, 2013; Rincón et al., 2015; Engelman et al., 2016). This conclusion is principally based on the identification of the postero-internal cusp on unworn M1–3 as the metacone (figs. 15B, 18B, 19B), from which it follows that the large posterolabial cusp is stD, and that the enamel ridge along the labial margin of the tooth crown is a neomorph (see char. 133). This hypothesis of cusp homologies is further supported by the presence of a small but distinct paracone on the lingual flank of stB in †Pichipilus (Abello, 2007: lámina 6H, I), †Palaeothentes (Engelman et al., 2016), and several other fossil paucituberculatans not included here, such as †Phonocdromus, †Palaepanorthus, †Pitheculites, and †Perulestes (Goin and Candela, 2004; Abello, 2013). We have also examined a subadult specimen of Caenolestes fuliginosus (BMNH 1954.295) and another of C. caniventer (BMNH 1954.302) in which the paracone is identifiable as a small but distinct structure on M3, where it is closely appressed to the lingual face of stB, and Martin (2013: fig. 1A, 611) reported a similar morphology in Lestoros. An alternative—but to us less convincing—interpretation of caenolestid crown homologies was expounded by Marshall (1980: fig. 5b) and was apparently followed by Springer et al. (1997: chars. 6, 7), Sánchez-Villagra (2001), Horovitz and Sánchez-Villagra (2003: chars. 154, 155), and a number of subsequent studies that used versions of their craniodental matrix (e.g., Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009); however, this was revised by Abello and Candela (2010: 1516; see also Beck, 2012, 2017a; Beck et al., 2014, 2016; Maga and Beck, 2017).

Among the paucituberculatans included in the current analysis, specimens of Lestoros, Rhyncholestes, †Acdestis, †Palaeothentes, and †Stilotherium that we examined appear to exhibit complete fusion between the paracone and stylar cusp B on all upper molars (but see Martin, 2013: fig. 1A, 611; Engelman, et al., 2016), so these taxa have been scored as state 2. Although we observed a vestigial paracone on M3 of two Caenolestes specimens (see above), most individuals lack any trace of the paracone, so we also scored Caenolestes as state 2. However, because †Pichipilus retains a distinct paracone on M1–3 (see Abello, 2007), we scored this taxon as state 0.

Although the major upper molar cusp of †Yalkaparidon is presumably either the paracone or (perhaps more likely) the metacone (Beck et al., 2014), its homology is uncertain in the absence of annectant, dentally more plesiomorphic relatives, so we scored †Yalkiparidon as unknown (“?”). Notoryctes is also zalambdodont, with the major upper molar cusp homologous with the metacone (Archer et al., 2011), but a tiny remnant of the paracone, distinct from stylar cusp B, appears to be present on the premetacrista of some specimens (fig. 15D; Murray and Megirian, 2006b: fig. 6I), and the fossil notoryctid †Naraboryctes retained a paracone that is distinct from stB on M1–4 (Archer et al., 2011). Collectively, these observations suggest that the paracone has not fused with stB in Notoryctes, so we scored this taxon as state 0.

The paracone is clearly distinguishable in phascolarctids, †Muramura (Pledge, 2003: fig. 19.2), and on the unworn molars of Lasiorhinus (in which it is removed from the labial margin of the teeth and connected to stylar cusp B by an obvious preparacrista; Brewer et al., 2018: fig. 13.1). In bunodont, bunolophodont, and lophodont diprotodontians (figs. 17, 18A, 18C), however, the principal anterolabial cusp appears to be stB and the paracone is only weakly distinct (as in unworn molars of Hypsiprymnodon moschatus; fig. 18A; Ride, 1993: fig. 8) or is entirely absent, having largely or completely fused to stB or (equivalently) having been incorporated into the labial terminus of the protoloph (Ride, 1971; Archer, 1976e, 1984c: 683; Tedford et al., 1977; Woodburne and Clemens, 1986a: 96–97, 1986b: 20–21; Tedford and Woodburne, 1987: 416; Woodburne et al., 1987a: 600; Ride, 1993; Crosby and Archer, 2000: 331; Beck et al., 2020), so we scored these taxa as state 2. In †Namilamadeta, the paracone is basally fused with stB but is still clearly identifiable as a discrete structure on M1 and M2, whereas this cusp appears to have completely fused with stylar cusp B on M3 and M4 (Rich and Archer, 1979; Archer, 1984b: 682–683; Pledge, 2005), an apparently autapomorphic trait (state 3).

Living pseudocheirids are selenodont, but Roberts (2008) and Roberts et al. (2008: 843) noted that the upper molars of the oldest known members of the family (the early Oligocene †Pildra antiquus and †P. secundus; Woodburne et al., 1987b) are only weakly selenodont, being dentally similar to Recent (bunolophodont) petaurids and, most strikingly, to the “bunoselenodont” petauroid †Djaludjangi yadjana (see Brammall, 1998 and Roberts, 2008: figs. 4.1.12A, 4.1.12B, 4.1.15). The existence of such structurally intermediate forms supports the conclusions of Roberts (2008) and Roberts et al. (2008: 843) that the fully selenodont molars of Recent pseudocheirids are derived from bunolophodont precursors, in which the principal labial cusps are stB and stD (as discussed above). On this assumption, the antero-internal pseudocheirid cusp previously identified by authors as the paraconule is probably the paracone, and the postero-internal pseudocheirid cusp previously identified as the “neometaconule” is the metacone (see also Black et al., 2012a: 132). Another consequence of this hypothesis is that the structure conventionally identified as the “centrocrista” in pseudocheirids is, in fact, a nonhomologous stylar crest linking stB (the “paracone” of authors) and stD (the “metacone”).13 Therefore, all pseudocheirids included in the current analysis have been scored as state 0.

FIG. 18.

Occlusal views of upper molars of Hypsiprymnodon moschatus (A, AMNH 160120 [right M1]), Caenolestes fuliginosus (B, BMNH 1954.295 [left M2, reversed]), and Acrobates pygmaeus (C, AMNH 37185 [right M2]). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 132, 133, 136, 137, 140, 141, 143, 144, 145, and 146 (see main text for description of these characters and character states), as follows: Hypsiprymnodon moschatus 132(-), 133(0), 136(2), 137(1), 140(-), 141(0), 143(1), 144(1), 145(-), 146(-); Caenolestes 132(-), 133(1), 136(2), 137(-[but note that some Caenolestes specimens retain a paracone and can be scored as 1]), 140(-), 141(0), 143(1), 144(0), 145(-), 146(-); Acrobates 132(-), 133(1), 136(2), 137(-), 140(-), 141(0), 143(1), 144(0), 145(-), 146(-). Abbreviations: mec, metacone; nci, neomorphic cingulum; pac, paracone; stB, stylar cusp B; stD, stylar cusp D. Teeth are not shown to the same scale.

img-z135-1_01.jpg

Although states 0, 1, and 2 form a plausible morphocline, state 3 cannot be incorporated into a linear transformation series, so this character has not been ordered in any of our analyses.

Character 137. Paracone and metacone subequal in height (0); or metacone much taller than paracone (1); or paracone absent or vestigial, only metacone present (2). Among the tribosphenic metatherians we examined, the paracone and metacone are subequal in height (state 0) only in the didelphid Caluromys (Reig et al., 1987: fig. 12B), microbiotheriids (Marshall, 1982: figs. 9, 15; Szalay, 1993: fig. 15.3), and the peramelemorphian Chaeropus (Wright et al., 1991: plate 1A). By contrast, the metacone is clearly taller (state 1) in †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, †Herpetotherium, all didelphids except Caluromys, thylacinids, dasyurids, and all peramelemorphians except Chaeropus. Wroe et al. (2000: char. 8) and Wroe and Musser (2001: char. 8) coded Didelphis as exhibiting state 0, an obvious lapsus. The relative heights of the paracone and metacone vary along the molar row in some taxa, (e.g., †Microbiotherium, in which the metacone is taller than the paracone on M1 and M2 but subequal in height on M3; Marshall, 1982: fig. 9; Szalay, 1993: fig. 15.3); in such cases, we scored this character based on M3 only.

Among the paucituberculatans included in the current analysis, an obvious paracone is present and clearly shorter than the metacone in †Evolestes (Goin et al., 2007b) and †Pichipilus (Abello, 2007). Although the paracone is not identifiable in most specimens of Caenolestes, it is present on M3 of two subadult specimens (BMNH 1954.295 [= C. fuliginosus] and 1954.302 [= C. caniventer]) and is clearly shorter than the metacone. All examined specimens of other paucituberculatans (i.e., Rhyncholestes, Lestoros, †Acdestis, †Palaeothentes, and †Stilotherium) lack an identifiable paracone and have been scored as inapplicable (“-”; but see Martin, 2013: fig. 1A, 611; Engelman, et al., 2016).

FIG. 19.

Upper second molars (M2) of Pseudantechinus macdonnellensis (A, AMNH 196786 [left M2 in labio-occlusal view]) and Caenolestes caniventer (B, BMNH1954.302 [right M2 in lingual view and reversed]). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 134, 135, 140, 141, 143, and 144 (see main text for description of these characters and character states), as follows: Pseudantechinus 134(1), 135(0), 140(3), 141(0), 143(0), 144(0); Caenolestes 134(0), 135(1), 140(-), 141(0), 143(1), 144(0). Abbreviations: mec, metacone; mecl, metaconule; pac, paracone; prc, protocone; stB, stylar cusp B; stD, stylar cusp D. Teeth are not shown to the same scale.

img-z136-1_01.jpg

In most selenodont diprotodontians that retain a paracone and metacone distinct from the stylar cusps, namely phascolarctids, wynyardiids and pseudocheirids (note that in pseudocheirids, we assume that the “paraconule” and “neometaconule” of previous authors [e.g., Archer, 1984b; Woodburne et al., 1987; Roberts, 2008; Roberts et al., 2008] actually represent the paracone and metacone; see char. 136), the paracone and metacone are usually subequal in height, although the paracone is sometimes slightly taller, particularly on M1 and M2. However, Ilaria is an exception, with the metacone distinctly taller than the paracone on M1–3 (Tedford and Woodburne, 1987). A distinct paracone and metacone are also present and subequal in height on the unworn molars of Lasiorhinus (Brewer et al., 2018: fig. 13.1). In most bunodont, bunolophodont, and lophodont diprotodontians the paracone and metacone are not identifiable as distinct cusps due to fusion with stylar elements and/or incorporation into the protoloph and metaloph respectively (see char. 136), so we scored these taxa as inapplicable (“-”). The exceptions are Hypsiprymnodon moschatus and H.bartholomaii, in which the paracone and metacone (identified as the “protoconule” and “neometaconule” respectively by Flannery and Archer, 1987c) are identifiable as distinct cusps (see char. 136; fig. 18A, see also Ride, 1993: fig. 8); in both taxa the paracone appears only slightly taller than the metacone, so we scored both as state 0.

The zalambdodont morphology of Notoryctes resulted from extreme reduction of the paracone (Archer et al., 2011), a distinct condition we coded as an additional state (state 2). This hypothesis is supported by the presence of what may be a tiny remnant of the paracone in some Notoryctes specimens (fig. 15D; Murray and Megirian, 2006b: fig. 6I), and the presence of a greatly reduced (but still identifiable) paracone in the fossil notoryctid (not included here) †Naraboryctes (Archer et al., 2011). Although Beck et al. (2014) concluded that the major cusp of the upper molars of †Yalkaparidon probably represents the metacone based on occlusal relations (following Asher et al., 2007: 7), this identification remains tentative in the absence of taxa exhibiting intermediate morphologies. Indeed, occlusal relations are evidently not an infallible guide to cusp homologies, as they led Asher et al. (2007: 7) to identify the major upper molar cusp of the meridiolestidan “dryolestoid” †Necrolestes as probably representing the metacone, whereas subsequent evidence strongly suggests that it is in fact the paracone (Rougier et al., 2012; O'Meara and Thompson, 2014). Even if the major cusp were shown to be the metacone in †Yalkaparidon, it would still be unclear whether the paracone had been suppressed (as in Notoryctes) or lost via another mechanism (e.g., by fusion with stylar cusp B; see char. 136); we therefore scored †Yalkaparidon as unknown (“?”) for this character.

We score this character as inapplicable (“-”) for Myrmecobius, in which homologs of the paracone and/or metacone cannot be distinguished with certainty (Archer, 1984b: 641–642), and Tarsipes, in which the entire postcanine dentition is reduced to tiny spicules of enamel (see char. 113).

Character 138. M1 preparacrista present (0); or absent or indistinct (1). The preparacrista of conventionally tribosphenic metatherians is a low crest that extends anterolabially or labially from the apex of the paracone to terminate on the stylar shelf at or near stA or stB (figs. 14, 15A, 19A, 20A). By contrast, several authors (e.g., Archer, 1976c; Muirhead, 1994, 2000; Muirhead and Filan, 1995; Travouillon et al., 2010) have remarked on the unusual morphology of the M1 “preparacrista” of peramelemorphians, in which a tall crest passes posterolabially from the paracone to connect with a large stylar cusp at or near the stC position (but usually identified as stB), with which it forms a prominent crescentic or bladelike structure (fig. 20B; see also Turnbull et al., 2003: fig. 18.1). However, in Chaeropus it can be seen that two crests extend from the paracone of M1; this morphology is most obvious in young Chaeropus specimens with unworn molars (e.g., AM PA422; Travouillon et al., 2019: fig. 4A), but can still be observed in some specimens with worn teeth (Travouillon et al., 2019: figs. 7c, 12a). One of these crests extends anterolabially and diminishes rapidly in height, terminating at stA near the anterolabial corner of the tooth (Wright et al., 1991: fig. 3); in height and orientation, this crest closely resembles the preparacrista of conventionally tribosphenic metatherians (such as didelphids). The second crest, however, is much taller and extends posterolabially to a cusp at the stC position (but usually referred to as “stB”; Archer, 1976c; Muirhead, 1994, 2000; Muirhead and Filan, 1995; Travouillon, et al., 2010), resembling the “preparacrista” previously described from M1 of other peramelemorphians. Because two homologous structures cannot (by definition) coexist in the same taxon (Patterson, 1982), we infer that the tall, posterolabially directed crest connecting the paracone to stB on the M1 of peramelemorphians is not the preparacrista, but a neomorph that we here designate the “pseudopreparacrista” (see char. 139). On this hypothesis, we scored the preparacrista as absent (state 1) in all peramelemorphians except Chaeropus (which we scored as “0”).

In Neophascogale (e.g., AMNH 156343 and 109516), there appear to be two separate crests extending from the paracone: one reduces in height as it extends anterolabially from the apex of the paracone, terminating at a stylar cusp (?StA) at the anterolabial corner of the tooth, and we identify it here as the preparacrista; the second, taller crest, extends posterolabially, terminating at StB, and we refer to it as the pseudopreparacrista, based on its overall topological similarity to the crest to which we assign that name in peramelemorphians. The condition seen in Neophascogale may explain Archer's (1982: 429) enigmatic comment that “some of the unusual stylar shelf morphology [of the molars, and particularly M1, of Neophascogale and Phascolosorex] is reminiscent of that found in bandicoots.” However, the condition in the Phascolosorex species we have used for scoring purposes here, P. dorsalis, resembles most other dasyurids in having a preparacrista only, rather than preparacrista and pseudopreparacrista as in Neophascogale.

FIG. 20.

Left first upper molars (M1) of Murexia longicaudata (A, AMNH 101970) and Echymipera kalubu (B, AMNH 221654). General features of upper molar crown morphology discussed in the text are illustrated, as are characters 132, 133, 136, 137, 138, 139, 140, 141, 143, and 144 (see main text for description of these characters and character states), as follows: Murexia 132(1), 133(0), 136(0), 137(1), 138(0), 139(0), 140(3), 141(0), 143(0), 144(0); Echymipera 132(1), 133(0), (136(0), 137(1), 138(1), 139(1), 140(4), 141(0), 143(1), 144(0). Abbreviations: mec, metacone; mecl, metaconule; pac, paracone; prc, protocone; prpacr, preparacrista; psppcr, pseudopreparacrista; stD, stylar cusp D. Teeth are not shown to the same scale.

img-z138-1_01.jpg

We scored dasyuromorphians in which the paracone and stB are consistently fused on M1 (char. 136) as inapplicable (“-”). We also scored this character as inapplicable for Myrmecobius, Notoryctes, †Yalkaparidon, and most examined paucituberculatans, because examined specimens of these taxa lack unambiguously identifiable paracones (a necessary landmark for identifying presence or absence of the paracrista; see char. 136). However, the M1 preparacrista is absent in †Evolestes (Goin et al., 2010: fig. 6.1G-H), and present (albeit very short) in †Pichipilus (Abello, 2007: lámina 6H-I).

Among diprotodontians, the preparacrista is unambiguously present on M1 of pseudocheirids, wynyardiids, and phascolarctids. In these taxa, all of which we scored as state 0, the preparacrista is anterolabially or labially directed. By contrast, in lophodont and bunolophodont diprotodontians a low crest usually extends anteriorly from the labial end of the protoloph toward the anterolabial corner of the tooth. Although this crest has been called the preparacrista by numerous authors (e.g., Flannery and Rich, 1986; Cooke, 1997c; Prideaux, 2004; Kear et al., 2007; Prideaux and Warburton, 2008, 2009; Bates et al., 2014; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Butler et al., 2016, 2018), we believe it to be a remnant of a stylar crest (on the hypothesis that the labial terminus of the protoloph is stB rather than the paracone; see char. 136). Instead, the preparacristae of lophodont and bunolophodont diprotodontians may have been incorporated into the labial portion of the protoloph (see chars. 136, 144). Therefore, we scored these taxa as inapplicable (“-”).

Character 139. M1 pseudopreparacrista absent (0); or present (1). As discussed above (char. 138), the tall crest that extends posterolabially from the paracone to stB (or stC) on M1 of most peramelemorphians (fig. 20B) does not appear to be homologous with the preparacrista but, instead, appears to be a neomorph—the pseudopreparacrista. This crest is present (state 1) on the M1 of all extant and fossil peramelemorphians that we examined, with the exception of Macrotis (in which the preparacrista and pseudopreparacrista are both consistently absent, even on minimally worn specimens; state 0), some specimens of Isoodon (e.g., AMNH 154499), and Microperoryctes (e.g., left side of AMNH 108546), with the latter two taxa scored as polymorphic (“0+1”). As already discussed (see char. 138), a topologically and morphologically similar (although presumably independently derived) crest is seen in the dasyurid Neophascogale (e.g., AMNH 156343 and 109516), which has also been scored as state 1 here.

As for the preceding character, we coded this character as inapplicable (“-”) in Myrmecobius, Notoryctes, †Yalkaparidon, and all examined paucituberculatans except †Evolestes and †Pichipilus (in both of which the pseudopreparacrista is undoubtedly absent; see Goin, et al., 2010: fig. 6.1G-H; Abello, 2007: lámina 6H-I) because the absence of an identifiable homolog of the paracone in these taxa precludes meaningful scoring of this character.

Among diprotodontians, an M1 pseudopreparacrista is unequivocally absent in pseudocheirids, wynyardiids and phascolarctids, all of which we scored as state 0. As explained above in the context of scoring the preparacrista character, the probable incorporation of the paracone into the protoloph of lophodont and bunolophodont diprotodontians precludes meaningful observations from these taxa, which were all scored as inapplicable (“-”).

Character 140. Centrocrista straight, midpoint almost level with the floor of the trigon basin (0); or straight, midpoint high above floor of trigon basin (1); or weakly Λ-shaped, midpoint distinctly elevated above floor of trigon basin (2); or strongly Λ-shaped, midpoint high above floor of trigon basin (3); or centrocrista discontinuous, postparacrista and premetacrista extending to labial margin of tooth and forming a narrowly open embrasure (4); or centrocrista discontinuous, with postparacrista and premetacrista terminating in stylar region, not extending to labial margin of tooth (5). Tribosphenic metatherian molars exhibit considerable variation in the shape of the centrocrista (postparacrista + premetacrista), which is associated with a traditional and sometimes confusing descriptive nomenclature (see discussion by Voss and Jansa, 2003: char. 58; 2009: 51–52). Taxa with a Λ-shaped (labially inflected) centrocrista are traditionally described as “dilambdodont” because the ectoloph (preparacrista + centrocrista + postmetacrista) then resembles two inverted capital Greek lambdas, whereas taxa with a straight centrocrista have often been called “predilambdodont” (e.g., by Reig et al., 1987) or rectilinear.

In some taxa with a straight centrocrista (e.g., †Mimoperadectes, Caluromysiops, and microbiotheriids) the midpoint is almost level with the floor of the trigon basin (state 0; see discussion by Johanson, 1996a), but †Mayulestes (Muizon, 1998: fig. 2; Muizon et al., 2018: fig. 18B), †Allqokirus (Muizon et al., 2018: figs. 7B, 8B, 9, 10A) and Thylacinus (Muizon and Lange-Badré, 1997: fig. 4A) have a straight centrocrista that is elevated above the trigon basin (state 1).

In taxa with a labially inflected centrocrista, the apex of this crest is always raised above the trigon basin, but in some it is only weakly Λ-shaped (state 2), as in †Pucadelphys (Marshall and Muizon, 1995: figs. 6B, 7B; Ladevèze et al., 2011: fig. 4d-f), †Sparassocynus (Reig et al., 1987: fig. 29B; Beck and Taglioretti, 2020: fig. 3a), †Thylophorops (Reig et al., 1987: fig. 28B), many Recent didelphids, †Badjcinus (Muirhead and Wroe, 1998: fig. 2B), and †Nimbacinus (Muirhead and Archer, 1990: fig. 1B).

By contrast, in †Herpetotherium (Fox, 1983: figs. 1A, 2A;; Horovitz et al., 2008: plate 1A; Korth, 2018: fig. 6a), †Thylatheridium (Reig, 1952: fig. 11A; Reig et al., 1987: fig. 25A), many Recent didelphids (e.g., Marmosa [fig. 14]), †Barinya, †Mutpuracinus (Murray and Megirian, 2000: fig. 3), all Recent dasyurids (e.g., Dasyurus [Smits and Evans, 2012: fig. 1A]; Pseudantechinus [fig. 19A]; Murexia [fig. 20A]), and †Yarala (Muirhead and Filan, 1995: figs. 1.2, 2.2), the centrocrista is strongly Λ-shaped, with the postparacrista and premetacrista of unworn molars forming a lingual angle of 90 degrees or less in occlusal view (state 3). Wroe et al. (2000: char. 10) scored Sarcophilus as having a rectilinear centrocrista, but we consider the centocrista of this taxon to be strongly dilambdodont, resembling that of other dasyurids (see, e.g., Sakai and Yamada, 1992: fig. 2), so we scored it as state 3.

Recent peramelemorphian molars are also strongly dilambdodont, but the centrocrista is usually discontinuous (state 4; figs. 15C, 20B), because the postparacrista and premetacrista extend through the stylar region to terminate on the labial margin of the tooth, where they are separated by a gap (Archer, 1976c, 1976e, 1984c; Muirhead, 1994; Muirhead and Filan, 1995; Kear et al., 2016; Warburton and Travouillon, 2016; Travouillon and Phillips, 2018); Muirhead and Filan (1995: 131) described this condition as resulting from the centrocrista “breaching” the ectoloph, whereas other authors have referred to it as an “incomplete” centrocrista (e.g., Travouillon et al., 2010, 2013b, 2014a; Kear et al., 2016). Although such a discontinuous centrocrista is characteristic of most Recent peramelemorphians, we observed a few specimens of Peroryctes in which the centrocrista is strongly dilambdodont but continuous (e.g., AMNH 151958; see also comments by Aplin et al., 2010: 20, fig. 10)14; nevertheless, we scored Peroryctes as state 4 based on the modal condition. In †Galadi, the centrocrista is strongly dilambdodont but continuous on M1 and M2, whereas it is discontinuous on M3 (Travouillon et al., 2010); we therefore scored †Galadi as polymorphic (“3+4”).

Among diprotodontians, the centrocrista of phascolarctids (see, e.g., Archer, 1976e, 1984c: fig. 62; Woodburne et al., 1987c; Black et al., 2014a) and wynyardiids (see, e.g., Pledge, 2003: fig. 19.2) is discontinuous, with the postparacrista and premetacrista extending to the labial margin of the tooth and forming an open embrasure, resembling in this respect the morphology seen in most peramelemorphians; thus, we scored representatives of both taxa as state 4. Based on our hypothesis of pseudocheirid cusp homologies (in which the cusps referred to by most recent authors as the paraconule and “neometaconule” are identified as the paracone and metacone respectively; see char. 136), the centrocrista of pseudocheirids is also discontinuous, because the labial ends of the postparacrista (the “postparaconulecrista” of previous authors) and premetacrista (the “preneometaconulecrista” of previous authors) are separated by a distinct gap (Archer, 1984c: fig. 178A-C; Roberts, 2008; Roberts et al., 2008). However, these crests do not extend to the labial margin of the upper molars, but instead terminate lingual to the major labial cusps (identified here as stB and stD), a morphology we score as state 5.

In many lophodont and bunolophodont diprotodontians (e.g., acrobatids, burramyids, petaurids), an anteroposteriorly directed crest connects the labial ends of the protoloph and metaloph. Although this crest has been homologized with the centrocrista by previous authors (e.g., Springer et al., 1997: char. 8; Horovitz and Sánchez-Villagra, 2003: char. 156), this crest cannot be the centrocrista if the labial termini of the protoloph and metaloph are, in fact, stylar cusps (see chars. 134–136). In fact, what appears to be a vestigial centrocrista is present lingual to the crest connecting stB and stD on the unworn molars of Hypsiprymnodon moschatus (fig. 18A; see also Ride, 1993: 451, fig. 8). Although this vestige appears to be approximately linear in H. moschatus, the postparacrista and premetacrista are weakly developed and do not form an obviously continuous crest, making interpretation difficult. We therefore scored H. moschatus and all other lophodont and bunolophodont diprotodontians as inapplicable (“-”).

In †Evolestes the centrocrista of M2 is discontinuous: the postparacrista terminates at the base of the lingual face of stB, the premetacrista terminates at the base of the lingual face of stD, and the labial ends of these crests are separated by a narrow gap (Goin et al., 2007b; Abello, 2013: fig. 2A). Although this morphology most closely resembles state 5 as defined above, the M3 centrocrista of this taxon is strongly dilambdodont but complete (Goin et al., 2007b). We have therefore scored †Evolestes as polymorphic (“3+5”).

Specimens of most of the other paucituberculatans we examined for this study lack an unambiguous homolog of the centrocrista because the paracone is fused with stB and a distinct postparacrista is not present (see char. 136). Thus, we scored Caenolestes, Lestoros, Rhyncholestes, †Acdestis, †Palaeothentes, and †Stilotherium as inapplicable (“-”) for this character. By contrast, Goin et al. (2007b: char. 31; 2009a: char. 33) scored Caenolestes, Rhyncholestes, and †Stilotherium as having a discontinuous centrocrista. However, the centrocrista is identifiable in †Pichipilus, in which the paracone and metacone are both distinct on M1–3 (Abello, 2007; 2013: fig. 2F, I); the centrocrista is discontinuous in this taxon, with the postparacrista and premetacrista terminating against the enlarged stylar cusps (as on M2 of †Evolestes), so we scored it as state 5.

This character was also scored as inapplicable for Tarsipes, Notoryctes, †Yalkaparidon, and Myrmecobius, all of which lack an unambiguously identifiable centrocrista. Because the states we define do not form an obvious morphocline, we did not order this character in any of our analyses. As discussed by Beck et al. (2008b), the shape of the centrocrista can change due to wear by the hypoconid (with the centrocrista becoming increasingly strongly inflected in more heavily worn molars), so we scored this character based on specimens with unworn or lightly worn teeth (see also comments by Travouillon and Phillips, 2018: 241, on the impact of wear on centrocrista shape in peramelemorphians).

Character 141. Protocone present (0); or absent (1). Among the metatherians included in this study, the zalambdodont taxon †Yalkaparidon uniquely lacks the protocone and other related structures—preprotocrista and postprotocrista—lingual to the metacone (Archer et al., 1988; Beck, 2009; Beck et al., 2014). Although Notoryctes is also zalambdodont, it resembles nonzalambdodont metatherians in retaining a distinct protocone (fig. 15D; Asher and Sánchez-Villagra, 2005: fig. 6A; Murray and Megirian, 2006b: fig. 6I; Archer et al., 2011: fig. 1).

Character 142. Preprotocrista and anterolabial cingulum joined to form a continuous shelf along anterior margin of M3 (0); or preprotocrista and anterolabial cingulum separate on M3, not forming a continuous shelf (1). The preprotocrista of many metatherians passes anterolabially around the base of the paracone and merges with the anterolabial cingulum (or paracingulum; see Van Valen, 1966: fig 1A; Marshall, 1987: fig. 2; Reig et al., 1987: fig. 1), forming a continuous shelf along the anterior margin of one or more upper molars (Voss and Jansa, 2003: fig. 13, left); such taxa include †Mimoperadectes (Horovitz et al., 2009: fig. 1), †Allqokirus (Muizon et al., 2018: figs. 7B, 8B, 10A), †Herpetotherium (Horovitz et al., 2008: plate 1A), many Recent didelphids (e.g., Caluromys, Glironia, Hyladelphys, Marmosa; Voss and Jansa, 2003; 2009), microbiotheriids (Marshall, 1982: fig. 9 and 16), †Galadi (Travouillon et al., 2010), Echymipera, Isoodon, Macrotis, many Recent dasyurids (e.g., Antechinus, Murexia, Phascogale, Planigale), phascolarctids, wynyardiids, and pseudocheirids. By contrast, the preprotocrista terminates at or near the base of the paracone, without contacting the anterolabial cingulum (Voss and Jansa, 2003: fig. 13, right), in other Recent didelphids (e.g., Didelphis, Chacodelphys, Chironectes, Lestodelphys, Lutreolina, Metachirus, Monodelphis, Philander, Thylamys; Voss and Jansa, 2003, 2009), fossil didelphids (e.g., †Sparassocynus, †Thylophorops, †Thylatheridium), and many nondidelphimorphian metatherians, including †Pucadelphys (Marshall and Muizon, 1995: figs. 6B, 7B; Ladevèze et al., 2011: fig. 4d–f), †Mayulestes (Muizon, 1998: fig. 2), †evolestes (goin et al., 2007b: text figs. 2, 3), Notoryctes (fig. 15D; Asher and Sánchez-Villagra, 2005: fig. 6A; Murray and Megirian, 2006b: fig. 6I; Archer et al., 2011: fig. 1), †Yarala (Muirhead and Filan, 1995: figs. 1.2, 2.2), Rhynchomeles, Peroryctes (Aplin et al., 2010: fig. 10C, D), Microperoryctes, and some dasyuromorphians (e.g., thylacinids, †Barinya, and several Recent dasyurids).

We scored this character as inapplicable (“-”) in all taxa in which an anterolabial cingulum is absent on M3 or cannot be certainly identified as such: this includes most paucituberculatans (except †Evolestes; see above), †Yalkaparidon, Notoryctes, Perameles, Chaeropus, Myrmecobius, and all diprotodontians except †Ilaria, phascolarctids, and wynyardiids.

Character 143. Upper molar posterolingual cusp absent (0); or posterolingual cusp formed by metaconule (1), or posterolingual cusp is the metacone (2). A distinct posterolingual cusp is consistently absent from the upper molars of †Mayulestes (Muizon, 1998: fig. 2; Muizon et al., 2018: fig. 10B), †Allqokirus (Muizon et al., 2018: figs. 7B, 8B, 9, 10A), †Pucadelphys (Marshall and Muizon, 1995: figs. 6B, 7B; Ladevèze et al., 2011: fig. 4d–f), †Mimoperadectes (Horovitz et al., 2009: fig. 1), †Herpetotherium (Horovitz et al., 2008: plate 1A), didelphids, most dasyuromorphians, microbiotheriids, Notoryctes (fig. 15D; Asher and Sánchez-Villagra, 2005: fig. 6A; Murray and Megirian, 2006b: fig. 6I; Archer et al., 2011: fig. 1), and †Yarala (Muirhead and Filan, 1995: figs. 1.2, 2.2), all of which have triangular or subtriangular teeth (state 0).

By contrast, a distinct posterolingual cusp (or “hypocone”) is present in many other taxa (figs. 15B, 15C, 16, 17, 18, 20B), giving their upper molars a more quadrilateral occlusal outline (see, e.g., Springer et al., 1997: char. 5; Horovitz and Sánchez-Villagra, 2003: char. 153). The homology of this structure has been disputed by several authors (e.g., Tedford and Woodburne, 1987; Sánchez-Villagra and Kay, 1996; Tedford and Woodburne, 1998). In peramelemorphians (with the notable exception of Macrotis), and several diprotodontian groups (e.g., acrobatids, burramyids, and petaurids), this cusp occupies the postprotocrista, and we agree with Tedford and Woodburne (1987; 1998) that it probably corresponds to the metaconule of the tribosphenic bauplan; for this reason, Beck et al. (2008b) referred to it as a “metaconular hypocone.” However, other authors (e.g., Ladevèze et al., 2010; Gheerbrant et al., 2016) have used the term “pseudohypocone” for this cusp to distinguish it from a “true” hypocone (= “cingular hypocone” sensu Beck et al., 2008b), which is a neomorphic outgrowth of the talon or postprotocingulum. A virtually continuous array of intermediate forms (e.g., the macropodiform †Jackmahoneya; Ride, 1993) plausibly establishes the metaconular origin of the posterolingual cusp (state 1) in other diprotodontians. Among thylacoleonids, the upper molars of †Lekanoleo are relatively quadrilateral in outline, due to the presence of a well-developed, posterolingual metaconule (state 1; see Gillespie, 1997: figs. 2, 3; Gillespie et al., 2020: fig. 4), but this cusp is less well developed and more labially positioned in †Wakaleo, resulting in a more triangular outline (state 0; see Murray et al., 1987: fig. 8); the sole upper molar (M1) of †Thylacoleo, meanwhile is highly modified, with uncertain cusp homologies (Archer and Rich, 1982; Archer, 1984c: fig. 128), and so we score it as unknown (“?”) here.

By contrast, the posterolingual cusp of Macrotis appears to be the metacone (state 2), the large posterolabial cusp in that taxon corresponding to a hypertrophied stylar element (usually interpreted as stylar cusp D; Bensley, 1903: fig. 52; Archer and Kirsch, 1977; Archer, 1984c). Presence or absence of an enlarged posterolingual cusp appears to be constant within most of our terminal taxa, with the notable exception of Microperoyctes ornata, in some specimens of which the metaconule is large (giving the upper molars a somewhat quadrilateral appearance; e.g., AMNH 108546), whereas in others the metaconule is distinctly smaller (and the upper molars correspondingly more triangular in occlusal outline; e.g., AMNH 190939). We therefore scored Microperoryctes as polymorphic (“0+1”).

The large posterolingual cusp on M1–3 of paucituberculatans has been interpreted as the metaconule by most recent authors (e.g., Marshall, 1987; Goin and Candela, 2004; Abello, 2007, 2013; Goin et al., 2007b; Rincón et al., 2015; Engelman et al., 2016), but as a “true” hypocone (“cingular hypocone” sensu Beck et al., 2008b) by Hunter and Jernvall (1995: table 1). The reasoning behind the latter authors' interpretation was not made explicit, but it is seemingly supported by the existence of a crest on the unworn molars of Caenolestes (figs. 15B, 18B, 19B) and Rhyncholestes that extends labially from a point on the lingual margin of the tooth about midway between the protocone and the posterolingual cusp and terminates at the base of the metacone (which is partially fused to the lingual face of stylar cusp D). If this crest is assumed to be the postprotocrista, then the posterolingual cusp in these taxa would appear not to be the metaconule, which (when present in tribosphenic metatherians) always lies on the postprotocrista (Tedford and Woodburne, 1998), not posterolingual to it. However, evidence from fossil taxa (e.g., †Evolestes, †Palaeothentes, †Palaepanorthus, †Perulestes, †Pitheculites, †Parabderites, †Acdestis) that provide a graded series of morphological intermediates from tribosphenic molars to the nontribosphenic teeth of Recent caenolestids suggests that the crest in question is, in fact, the premetaconulecrista or “pre-metaconular crest” (Goin and Candela, 2004; Abello, 2007, 2013; Goin et al., 2007b; Rincón et al., 2015). We therefore scored the posterolingual cusp of Recent caenolestids and other paucituberculatans as representing the metaconule, in agreement with most recent authors (Goin and Candela, 2004; Abello, 2007, 2013; Goin et al., 2007b; Rincón et al., 2015; Engelman et al., 2016).

In Notoryctes (fig. 15D; Asher and Sánchez-Villagra, 2005: fig. 6A; Murray and Megirian, 2006b: fig. 6I; see Archer et al., 2011: fig. 1), both the protocone and postprotocrista are identifiable and there is no enlarged metaconule on the postprotocrista, so we scored this taxon as state 0. However, †Yalkaparidon (in which the entire region of the tooth crown lingual to the metacone is absent; see char. 141), Tarsipes (in which the molars are so reduced that individual cusps can no longer be identified; char. 113), and Myrmecobius (whose teeth are too derived for this character to be scored confidently) were all scored as inapplicable (“-”). In the absence of any clear sequence of transformations among these alternative conditions, we treated this character as unordered (nonadditive) in all the phylogenetic analyses reported below.

Character 144. Principal labial and lingual cusps of upper molars not connected by transverse lophs (0); or connected by incipient lophs (1); or connected by well-developed lophs (2). The principal labial and lingual cusps of tribosphenic metatherian molars are not connected by transverse lophs (state 0), nor are transverse lophs present in such nontribosphenic taxa as paucituberculatans, peramelemorphians, phascolarctids, pseudocheirids, and Acrobates. By contrast, in fully lophodont (or “bilophodont”) diprotodontians, each tooth is provided with two tall transverse crests (state 2) that are relatively constant in height throughout their length. On lophodont upper molars, the more anterior protoloph extends labially from the protocone to terminate labially at what we identify as stylar cusp B, whereas the metaloph connects the metaconule (or “hypocone”; see char. 143) to what we interpret as stylar cusp D (fig. 17D). On lophodont lower molars, the protolophid connects the protoconid with the metaconid, and the hypolophid connects the hypoconid with the entoconid. Among the taxa included in this study, only †diprotodontids, Recent macropodids and some fossil macropodiforms (e.g., †Hadronomas, †Rhizosthenurus, †Balbaroo, †Bulungamaya, †Ganguroo, and †Ganawamaya) are fully lophodont.

An apparently annectant condition (state 1) is exhibited by many other diprotodontian taxa that are sometimes referred to as “incipiently lophoid” (by Bensley, 1903: 143) or “bunolophodont” (e.g., by Case, 1984; Flannery et al., 1984; Woodburne, 1984a; Springer et al., 1997: char. 42; Kear and Cooke, 2001; Horovitz and Sánchez-Villagra, 2003: char. 173; Prideaux, 2004; Kear et al., 2007; Prideaux and Warburton, 2010: char. 23). In this morphotype the protoloph and metaloph are less distinct and lower (particularly toward their midpoints) than in fully lophodont taxa, and the cusps at the ends of each loph are identifiable as distinct structures rather than seamlessly incorporated into the lophs (figs. 17A–C, 18A). Additionally, remnants of the paracone and metacone can sometimes be distinguished on the unworn teeth of some bunolophodont taxa (chars. 134–136; figs. 17B, 18A).

Within Diprotodontia, we observed an almost continuous range of intermediate morphologies between the nonlophodont condition (state 0) and bunolophodonty (state 1). We scored taxa in which lophs are identifiable on unworn upper molars as state 1, even if such lophs were only weakly developed; only taxa lacking any trace of lophodonty were scored as state 0. Under this scoring criterion, bunolophodont taxa (scored as state 1) include phalangerids, Petaurus, Gymnobelideus, burramyids, Distoechurus, several nonmacropodid macropodiforms (Aepyprymnus, Bettongia and ?Bettongiamoyesi, Caloprymnus, Potorous, Hypsiprymnodon moschatus [figs. 17B, 18A], H.bartholomaii, and †Ekaltadeta), wynyardiids, and Recent vombatids (the molars of †Warendja are too worn to score for occlusal characters). In fact, with the exception of pseudocheirids and phascolarctids (both of which are selenodont), most diprotodontians included in this study are either incipiently or fully lophodont. Strikingly, the upper molars of †Muramura and †Namilamadeta are both selenodont and incipiently lophodont (Rich and Archer, 1979; Pledge, 1987c; 2003; 2005; Beck et al., 2020), as is another fossil taxon not included here, the petauroid †Djaludjangi (Brammall, 1998).

Noteworthy nonlophodont exceptions among diprodotontians include Acrobates (fig. 18C), Dactylopsila, Dactylonax, and †Thylacoleo (the molars of †Lekanoleo, however, exhibit weakly developed lophs; Gillespie, 1997: fig. 2; Gillespie et al., 2020: fig. 4); all these bunodont taxa have been scored as state 0. †Ilaria lacks any trace of lophs on its (strongly selenodont) upper molars, and so has also been scored as state 0, but we note that its lower molars have weakly developed transverse lophids (Tedford and Woodburne, 1987). We scored †Yalkaparidon (in which the entire region of the crown lingual to the metacone is absent; char. 141) and Tarsipes (in which the upper molars are spicules lacking any identifiable structures; char. 113) as inapplicable (“-”).

Character 145. Protoloph and metaloph not connected by a median enamel crest (0); or protoloph and metaloph connected by a midlink (1). The protoloph and metaloph are entirely separate structures in most incipiently and fully lophodont diprotodontians (e.g., Trichosurus [fig. 17C]), but a median enamel crest—the so-called midlink (e.g., Stirton, 1955; Tedford, 1966; Woodburne, 1967; Bartholomai, 1971b; Sanson, 1980)—connects the protoloph and metaloph in all examined Recent macropodids (e.g., Lagostrophus [fig. 17D]) and several fossil macropodiforms (e.g., †Hadronomas, †Rhizosthenurus, †Balbaroo, †Ganawamaya). Within Macropodiformes, this crest appears to be formed at least partially by the postprotocrista: in unworn specimens of at least some taxa (e.g., Lagostrophus [fig. 17D]), the anterior part of the midlink extends from the apex of the protocone. The postprotocrista is also present, extending from the protocone apex at the anterolingual margin of the tooth, in macropodiforms that lack a midlink (Woodburne, 1967; Cooke, 1999; 2000; Prideaux and Warburton, 2010: char. 25, fig. 5). By contrast, the midlink of palorchestid diprotodontoids (not included in this analysis) appears to have a different evolutionary origin (Murray, 1990; Black, 2006).

This character was scored as inapplicable (“-”) for nonlophodont taxa and for incipiently lophodont taxa (see char. 144) in which the lophs are so weakly developed that it cannot be meaningfully scored (e.g., Hypsiprymnodon moschatus [figs. 17B, 18A]).

Character 146. Metaloph without a posteriorly projecting median crest (0); or urocrista (“postlink”) present (1). A distinct enamel crest projecting posteriorly from the middle of the metaloph is observed in some macropodiforms (e.g., Lagostrophus [fig. 17D]). This crest has been called the postlink by some authors (Flannery, 1983: fig. 4; Flannery and Rich, 1986), but other researchers have applied that term to a different occlusal structure, namely the postmetaconulecrista or posthypocrista (e.g., Tedford, 1966: fig. 3; Woodburne, 1967); we therefore follow Prideaux (2004) in referring to the crest in question as the urocrista.15 Travouillon et al. (2014b; 2015a; 2016) and Butler et al. (2016), however, called this crest the neometaconule crista.

Among the Recent macropodoids included in this study, only Lagostrophus exhibits a distinct urocrista (fig. 17D). Contra Prideaux (2004: 209), we did not observe this crest in any examined specimens of Dorcopsulus, Dendrolagus, Thylogale, or Wallabia. However, an obvious urocrista is present in some specimens of †Ganguroo (see Travouillon et al., 2014b; Cooke et al., 2015) and †Rhizosthenurus (Kirkham, 2004: plate 9), both of which have been scored as polymorphic (“0+1”). This character was scored as inapplicable (“-”) for nonlophodont taxa and for incipiently lophodont taxa (see char. 144) in which the lophs are so weakly developed that it cannot be meaningfully scored (e.g., Hypsiprymnodon moschatus [figs. 17B, 18A]).

Lower Incisors

Didelphids, microbiotheriids, and some nonmarsupial metatherians have four lower incisors, but many other living and extinct metatherians have three or fewer lower incisors, and the identity of the missing tooth (or teeth) in clades with fewer than four lower incisors has vexed researchers for many years. Although embryological studies (cited below) have sometimes been cited to support alternative theories about lower incisor homologies, we expect that only the discovery of fossil intermediates between the unreduced dentitions exemplified by didelphids and the derived morphologies of other groups is likely to convincingly resolve these issues. In addition to taxonomic variation in number, metatherian lower incisors exhibit striking differences in morphology, most notably by the presence of an enlarged gliriform tooth of notoriously uncertain homology at the first erupted mandibular locus in paucituberculatans and diprotodontians.

Character 147. Anteriormost lower incisor small and short crowned (0); or large, long crowned, and conspicuously procumbent (1). Most metatherians have a small, short-crowned incisor at the anteriormost (mesialmost) mandibular dental locus (state 0). By contrast, the anteriormost lower incisor of paucituberculatans (e.g., Caenolestes [fig. 35]) and diprotodontians (figs. 4454) is a large, long-crowned, procumbent tooth (state 1) that is sometimes described as “gliriform” by analogy with the corresponding mandibular teeth of rodents and lagomorphs. Many previous students of marsupial dental morphology (e.g., Luckett, 1994: char. 12; Springer et al., 1997: char. 33; Sánchez-Villagra, 2001: char. 33; Horovitz and Sánchez-Villagra, 2003: char. 167) have regarded these teeth as nonhomologous in paucituberculatans and diprotodontians (but see Beck et al., 2014; Beck, 2017a: char. 163; Maga and Beck, 2017: char. 163). This is based on evidence from serially sectioned developmental series of a few macropodids and phalangerids—(which have been used to conclude that the gliriform tooth in diprotodontians is i2, i3, or i4; e.g., Woodward, 1893; Dependorf, 1898; Hopewell-Smith and Marett Tims, 1911; Bolk, 1929; Engelhardt, 1932; Ride, 1962; Berkovitz, 1972b; Kirkpatrick, 1978)—and tooth-counts from fossil paucituberculatans (which have been used to argue that the gliriform tooth in Recent caenolestids is i1 or i2; Sinclair, 1906; Ride, 1962; Marshall, 1980). However, coding the gliriform lower incisor of unsectioned diprotodontians as i2 (for example) and that of Recent caenolestids as i1 cannot be justified unless it is assumed a priori that Diprotodontia and Paucituberculata are separate clades. Analyses of character data based on such a priori assumptions—which are seldom stated explicitly—cannot then be validly interpreted as supporting either diprotodontian or paucituberculatan monophyly (contra Luckett, 1994; Springer et al., 1997; Sánchez-Villagra, 2001; Horovitz and Sánchez-Villagra, 2003). In addition, using this approach it is unclear how to score taxa that have a gliriform lower incisor but do not appear to be members of either Paucituberculata or Diprotodontia (as in the case of †Yalkaparidon; Archer et al., 1988; Beck et al., 2014). Rather than perpetuate such circular inference, we prefer to code the presence/absence of a gliriform tooth at the anteriormost mandibular position (i.e., an assumption of primary homology sensu de Pinna, 1991), an option that permits global parsimony (or other phylogenetic optimality criterion, such as maximum likelihood) to distinguish between the alternative interpretations of the diprotodontian, paucituberculatan, and yalkaparidontian conditions as homologous or homoplastic (see also Beck et al., 2014; Beck, 2017a: char. 163; Maga and Beck, 2017: char. 163).

Although it may seem perverse to reject evidence that so many other morphologists have considered sufficient to code these teeth as nonhomologous, we note that there is still no consensus about their hypothesized positional identity (Ride, 1962; Marshall, 1980; Murray et al., 1987; Luckett, 1994; Springer et al., 1997; Luckett and Hong, 2000; Sánchez-Villagra, 2001; Horovitz and Sánchez-Villagra, 2003). In fact, a careful review of available evidence suggests that the gliriform incisors of paucituberculatans and diprotodontians plausibly occupy the same dental locus—which we refer to as i1 (see Materials and Methods and table 12), but which several other researchers have referred to as i2 (following Hershkovitz, 1982). In several fossil paucituberculatans (see, e.g., Sinclair, 1906; Hershkovitz, 1982; Abello, 2007; 2013; Engelman et al., 2016) the alveolus immediately behind the gliriform tooth is “staggered” (sensu Hershkovitz, 1982; 1995; see char. 150), and hence presumably homologous with the staggered alveolus seen in didelphids (i2); if so, then the gliriform paucituberculatan tooth occupies the same locus as the anteriormost lower incisor of didelphids (i1). The i2 alveolus is also staggered in dasyurids, thylacinids, and peramelemorphians (see char. 149), suggesting that the missing lower incisor (relative to the four seen in didelphids) in these taxa has been lost from the rear of the incisor series, in which case they retain i1–3 (see char. 148). Woodward (1896) reported an unerupting incisor anterior to the erupting i1 in sectioned pouch young of the dasyurid Dasyurus, as did Wilson and Hill (1897) in young specimens of Perameles. Additionally, Kirkpatrick (1978) examined dental development in Macropus giganteus and found evidence of an unerupting deciduous and successor incisor at a locus anterior to the gliriform tooth, which itself has an unerupting deciduous predecessor.16 As noted by Kirkpatrick (1978), if the anterior, nonerupting lower incisor locus in Macropus is homologous with the nonerupting locus reported in Dasyurus and Perameles by Woodward (1896) and Wilson and Hill (1897) respectively, then the gliriform tooth of the former is plausibly homologous with the anteriormost erupting lower incisor of dasyurids, thylacinids, and peramelemorphians, which in turn appears to be (for reasons discussed above) homologous with the anteriormost erupting lower incisor of paucituberculatans and didelphids (the tooth we refer to as i1). Ultimately, however, confident resolution of this question will probably require further ontogenetic studies of living taxa (most informative in this regard would be caenolestid pouch young, which have so far proven impossible to collect) using more modern techniques (such as Diffus-ible Iodine-based Contrast-Enhanced CT; Nasrullah et al., 2018), and discovery of fossil taxa that unambiguously demonstrate which locus (or loci) has been lost in particular clades.

FIG 21.

Lateral views of anterior mandible of Perameles gunnii (A, MVZ 127060 [left mandible]) and Dromiciops gliroides (B, FMNH 127445 [right mandible]). Alternative states of characters 148, 149, and 150 (see main text for descriptions of these characters and character states) are illustrated as follows: Perameles 148(1; note that the anteriormost visible incisor is the i1 of the right mandible), 149(0), 150(1); Dromiciops 148(0), 149(1), 150(0). Abbreviations: i1, first lower incisor; i2, second lower incisor; i3, third lower incisor; i4, fourth lower incisor. Specimens are not shown to the same scale.

img-z147-1_01.jpg

Character 148. Four lower incisors present (0); or i4 missing (1); or i3 and i4 missing (2); or i2–4 missing (3). Four lower incisors, the complete metatherian complement, are present in †Mayulestes, †Allqokirus, †Pucadelphys, †Herpetotherium, and didelphids (e.g., Marmosa [fig. 34]). In all these taxa, the numerical second incisor (i2) is unambiguously recognizable by its staggered alveolus (see char. 149). Both Dromiciops (figs. 21B, 43) and †Microbiotherium also have four lower incisors, although the alveolus of i2 is not staggered in either taxon (see char. 149; Hershkovitz, 1982, 1995, 1999).

In most metatherians with three lower incisors (dasyuromorphians and peramelemorphians; see fig. 21A) the identity of the middle tooth as i2 is suggested by its staggered alveolus (see char. 149). This, together with the assumption that reductions in incisor number result from losses of the posteriormost teeth in ancestral dentitions (see char. 103), leads us to identify the missing element in these taxa as i4 (state 1). The presence of a distinct posterior lobe on the numerical third incisor of peramelemorphians and some dasyuromorphians (char. 150) may be further evidence that i4 was lost if, as Muirhead (1994: 60) suggested, this lobe evolved as a functional replacement.

Myrmecobius also has only three lower incisors (fig. 41), but none of these teeth are staggered. Thomas (1887b: 311) noted two BMNH specimens of Myrmecobius that have four rather than three lower incisors unilaterally and concluded, based on this material, that the missing tooth is i4. However, it should be noted that Myrmecobius exhibits numerous other dental anomalies, such as retention of dP3 and dp3 into adulthood (see char. 121), presence of a supernumerary lower fifth molar in many specimens, and the occasional presence of a supernumerary tooth (usually a largely featureless unicuspid) between P3 and dP3 (as in BMNH 1906.8.1.357) or between p3 and dp3 (as in BMNH 1906.8.1.361). Nevertheless, given our general assumption that incisors are lost from the posterior end of the series, and in the absence of compelling evidence for alternative interpretations, we also assume that Myrmecobius has lost i4.

Scoring this character is more difficult in metatherian taxa with problematic lower antemolar homologies, such as paucituberculatans, most diprotodontians, †Yalkaparidon, and Notoryctes. In paucituberculatan terminals that exhibit an unambiguously identifiable double-rooted p2, and in which the anterior part of the dentary is intact in at least one specimen—namely Caenolestes (fig. 35), Rhyncholestes, Lestoros, and †Stilotherium—there are typically four unicuspids between i1 and p2 (Marshall, 1980; Abello, 2007, 2013; Martin, 2007, 2013). Assuming that the ancestral metatherian lower antemolar formula is i1–4, c1, p1–3, a minimum of two and a maximum of three of these unicuspids could be incisors. Hence, the total lower incisor count of these taxa, including the gliriform i1 (see char. 147), is either three (i1–3) or four (i1–4); therefore, we scored these taxa using ambiguity coding (“0/1”) to account for this uncertainty.

Diprotodontians can have anywhere from zero to three unicuspids between i1 and p2 (in the case of macropodiforms) or p3 (in all others). In those that consistently lack uncuspids in this interval—namely diprotodontids, phascolarctids (e.g., Phascolarctos [fig. 44]), vombatids (e.g., Vombatus [fig. 45]), wynyardiids, and most macropodiforms (e.g., Potorous [fig. 54])—three incisors (i2–4) are obviously absent (state 3). Tarsipes also lacks any teeth between i1 and its undifferentiated lower postcanine dentition (see fig. 49), so we also scored it as state 3. In diprotodontians that consistently have only a single tooth in this interval—namely †Wakaleo (Megirian, 1986; Murray, et al., 1987; Murray and Megirian, 1990; Gillespie, 2007; Yates, 2015b; Gillespie et al., 2017), Pseudochirops archeri (see Archer, 1984c: fig. 186), P. cupreus (fig. 51), Pseudochirulus, †Balbaroo (Cooke, 2000), ?Bettongiamoyesi (see Flannery and Archer, 1987b), †Bulungamaya (Travouillon et al., 2014b), †Ganguroo (Cooke, 1997b; Travouillon et al., 2014b; Cooke et al., 2015), †Ganawamaya (Kear et al., 2007), †Ekaltadeta (Wroe, 1996b), and Hypsiprymnodon moschatus (fig. 52)—at least two (i3–4) and potentially three (i2–4) of the lower incisors have been lost, so we scored these taxa as “2/3.” Diprotodontians with two unicuspids between i1 and p3 (†Thylacoleo, Petropseudes, acrobatids [fig. 48], and some phalangerids) must have lost at least one and potentially two or three lower incisors; we scored such taxa using ambiguity coding (as “1/2/3”). Finally, taxa with three unicuspids between i1 and p3 (e.g., †Lekanoleo [Gillespie et al., 2020: fig. 5]; petaurids, burramyids [fig. 46], some pseudocheirids, and some phalangerids [fig. 47]) could feasibly retain i2–4 or, alternatively, might have lost between one and three of these teeth; these alternatives are not distinguishable based on available evidence, so we scored such taxa as unknown (“?”).

The number of unicuspids present between i1 and m1 varies between one and four among specimens of †Yalkaparidon coheni, so we scored this taxon as unknown (by contrast, †Y. jonesi lacks any teeth between i1 and m1; Archer et al., 1988; Beck et al., 2014). Because Notoryctes has four or five teeth anterior to p3, it is possible that this taxon retains four lower incisors; however, based on our own observations and those of previous authors (e.g., Gadow, 1892; Archer, 1984c; Archer et al., 2011), Notoryctes appears to have two or three lower incisors, and has therefore been scored using ambiguity coding as “1/2.”

This character was ordered (0 ↔1 ↔2 ↔ 3) based on the hypothesized pattern of sequential loss.

Character 149. Second lower incisor (i2) with “staggered” alveolus (0); or i2 alveolus like those of i1 and i3 (1). The alveolus of i2 is “staggered”—displaced lingually and dorsally relative to the alveoli of adjacent incisors (i1 and i3) and with a prominent labial buttress of alveolar bone (fig. 21A)—in all examined metatherians with unambiguously distinguishable antemolar tooth homologies with the exception of Myrmecobius, Dromiciops (fig. 21B), and †Microbiotherium (Hershkovitz, 1982, 1995, 1999).17 A single specimen of †Pucadelphys was originally reported to have an unstaggered i2 (Marshall and Muizon, 1995: 68), but subsequently studied material clearly shows that i2 was staggered in this taxon (Muizon, 1998: 81; R.M.D.B., personal obs.).

Conflicting opinions have been expressed regarding the presence or absence of a staggered i2 in paucituberculatans (Hershkovitz, 1982, 1995; Abello, 2007, 2013; Voss and Jansa, 2009). A crucial preliminary issue is whether i2 can be unambiguously identified in these taxa. All of the caenolestids included in our analysis (namely Caenolestes, Rhyncholestes, Lestoros, and †Stilotherium; table 2) are represented by at least one specimen preserving an unambiguously identifiable double-rooted p2 and an intact anterior part of the dentary; in these, four unicuspids are typically present between p2 and i1, of which at least two must be incisors. Assuming that incisors are lost from the end of the series (char. 103), then the unicuspid immediately posterior to i1 can be identified as i2 (see also Voss and Jansa, 2009: 55). Among our other pauctuberculatan terminals, it is unclear whether p2 is present or not in †Acdestis and †Palaeothentes (char. 154), although p3 can be unambiguously identified in both taxa. However, four unicuspids are present between i1 and p3 in both †Acdestis (Rae et al., 1996: fig. 9; Engelman et al., 2016) and †Palaeothentes (e.g., MACN A 5671), of which one must be an incisor for the same reasons outlined above (at most, the remaining three could potentially represent c1, p1, and p2); again assuming that incisors are lost from the end of the series, the unicuspid immediately following i1 in †Palaeothentes can also be identified as i2. Thus, Caenolestes, Rhyncholestes, Lestoros, †Stilotherium, †Acdestis, and †Palaeothentes can all be scored for this character.

In no specimen of Caenolestes, Rhyncholestes, or Lestoros that we examined is the alveolus of putative i2 staggered (contra Hershkovitz, 1995): there is no evidence of a buttress of alveolar bone displacing the alveolus of i2 dorsally relative to those of the more posterior unicuspids, nor does i2 appear to be lingually displaced (Voss and Jansa, 2009: 55; Abello, 2013). Therefore, we scored these taxa as state 1. Abello (2007: 249–250; 2013) considered i2 (her “i3”) to be staggered in †Palaeothentes and †Stilotherium. We agree that i2 is indeed staggered (state 0) in †Stilotherium dissimile, based on specimens MACN A 8466 (see Abello, 2013: fig. 5A) and 8467. However, whereas two specimens of †Palaeothentes minutus (MACN A 8347-8354b and 8331) also appear to preserve a staggered i2, two others that we examined (MACN A 5671 and 8376) do not show clear evidence of staggering at this locus: the labial margin of the i2 alveolus is not dorsally raised, and this alveolus appears to be approximately in line (rather than lingually displaced) with those of the following three uncuispids when viewed dorsally. Thus, we scored †Palaeothentes as polymorphic (“0+1”). Engelman et al. (2016: 6, 8) reported that i2 (their “u1”) is “slightly staggered” in †Acdestis maddeni, but in UF 99687 this tooth appears to be approximately in line with the succeeding unicuspids in dorsal view (based on Morpho-bank media number M406548), so we scored †Acdestis as state 1. Staggering of i2 has been reported in several other fossil paucituberculatans not included in the current analysis (Sinclair, 1906: 420, fig. 7; Abello, 2007; 2013).

Some diprotodontians (phascolarctids, wynyardiids, vombatids, diprotodontids, and most macropodiforms) lack any teeth between i1 and p2 (in the case of macropodiforms) or p3 (the others); these taxa were therefore scored as inapplicable (“-”) for this character. In Tarsipes too, there are no teeth between i1 and the undifferentiated postcanine dentition, so we also scored this taxon as inapplicable. In other diprotodontians, however, between one and three unicuspids are typically present between i1 and p3; although one of these unicuspids might be i2, there is no way to be sure, so we scored all such taxa as unknown (“?”). Similarly, in known mandibular specimens of †Yalkaparidon there are alveoli for between one and four unicuspids between i1 and a tooth that we identify as m1 (“p3” of Archer et al., 1988; see Beck et al., 2014); because it is unclear whether i2 is present, we also scored †Yalkaparidon as unknown.

In Notoryctes, there are a maximum of five lower teeth anterior to the large premolariform tooth that we identify as p3. Of these, the fifth (immediately anterior to p3) is a tiny spicule when it is present at all (e.g., on the right side of AMNH 198651). Again, on the assumption that there can be at most seven teeth anterior to p3 (i1–4, c1, p1–2), at least two incisors must be present in Notoryctes specimens with five. We therefore tentatively identify the two anteriormost lower teeth as i1 and i2. Because the alveolus of the putative i2 does not appear to be staggered in this taxon, we scored it as state 1.

Character 150. Third lower incisor (i3) not bilobed (0); or bilobed (1). The third lower incisor of most metatherians that possess this tooth has a simple, undivided cutting edge (state 0; e.g., Dromiciops [fig. 21B]), but the cutting edge of i3 is conspicuously bilobed in peramelemorphians (state 1; e.g., Echymipera [fig. 21A]), a potential synapomorphy that has been remarked by several authors (e.g., Muirhead and Filan, 1995). However, this trait also occurs in dasyurids, among which it is most commonly observed on unworn teeth (Bensley, 1903; Archer, 1976c; Voss and Jansa, 2009). We therefore scored i3 as bilobed if this morphology was distinct in examined juvenile or subadult specimens. Besides peramelemorphians (in which this trait is consistently expressed), we observed a bilobed i3 in most dasyurids (albeit with the posterior lobe sometimes very weakly developed, as in Planigale), with the exception of Dasyuroides, Dasyurus, Myoictis, Parantechinus, Phascogale, and Sarcophilus; a bilobed i3 is variably present in Dasycercus, which we scored as polymorphic (“0+1”). A distinctly bilobed i3 is also present in juvenile specimens of Thylacinus (e.g., UMZC A6.7/10, and SAM M1956 and M1957), but it is absent in all examined specimens of Myrmecobius, including juveniles. Our scoring of this feature in dasyuromorphians differs from that of Archer (1976c: S39)—who stated that only Phascolosorex, Neophascogale, and Sminthopsis granulipes have a bilobed i3—and that of Wroe et al. (2000: char. 27) and Wroe and Musser (2001: char. 27)—who stated that Murexia, Antechinus, and Thylacinus lack a bilobed i3—perhaps because we were able to examine a greater range of juvenile and subadult material.

Taxa in which i3 is absent (those scored as state 2 for char. 148) were scored as inapplicable (“-”) for this character, whereas those in which the presence or absence of i3 cannot be unequivocally determined were scored as unknown (“?”).

Character 151. Tall lingual cusp of lower incisors (subequal in height to labial cusp) absent (0); or present (1). Voss and Jansa (2003: char. 62, fig. 14; 2009) identified the presence or absence of a small but distinct lingual cusp or heel on unworn lower incisors as distinguishing members of the tribe Didelphini (Chironectes, Didelphis, Lutreolina, Philander)—in which this structure is absent—from other didelphids, in which it is present. However, we were unable to consistently score this character in other metatherians, for three reasons. First, we observed a continuous range of intermediate morphologies, sometimes including variation along the toothrow of a single individual (for example, a distinct lingual cusp is absent on i1 and i2 but present on i3 in AMNH 109536, a juvenile specimen of Neophascogale lorentzii). Second, this cusp is often lost with minimal wear (Voss and Jansa [2009: 55] reported a lingual cusp or heel to be absent in Dasyurus, but it is present in a number of juvenile and subadult specimens that we examined [e.g., AM M35911]). Third, we encountered various scoring ambiguities among taxa with gliriform teeth (e.g., the distinctively “bulbous” [sensu Roberts, 2008: 377] morphology of the first lower incisor in Pseudochirops archeri [see Archer, 1984c: fig. 186c] and P. cupreus [fig. 51]).

A qualitatively distinct morphology that we scored for this analysis, however, is seen in Thylacinus, in which a lingual cusp that is as tall as the major, labial cusp (from which it is separated by a distinct fissure) is present on all three lower incisors. State 1 is therefore an autapomorphy of Thylacinus in the current analysis.

Lower Canine

The lower canine (c1) is a nonproblematic tooth in most polyprotodont taxa, easily recognized by virtue of its position between the incisor and premolar series and by its occlusion with a similarly shaped tooth that usually occupies the premaxillary-maxillary suture in the upper dentition. Unlike C1 (see char. 109), c1 is always single-rooted in metatherians. Although c1 exhibits noteworthy taxonomic variation in crown morphology in some clades (e.g., didelphids; Voss and Jansa, 2009: fig. 23), such variation is correlated with similar modifications of C1 such that occlusal transformations of these teeth seem to be nonindependent. By contrast, c1 is absent in some marsupial taxa that consistently retain C1, clear evidence of independent dental expression at the upper and lower canine loci.

Character 152. Lower canine present (0); or absent (1). The lower canine is present (state 0) in most metatherians, but it is unambiguously absent (state 1) in Phascolarctos (fig. 44), †Nimiokoala, vombatids (including †Warendja; Hope and Wilkinson, 1982), diprotodontids, wynyardiids, and most macropodiforms (e.g., Lagostrophus [fig. 53]), all of which lack teeth between the gliriform lower incisor and large multirooted teeth at the second or third premolar locus (see below). Tarsipes (fig. 49) also lacks any teeth between its gliriform i1 and its vestigial postcanine dentition. We scored other diprotodontians (i.e., thylacoleonids, burramyids, phalangerids, pseudocheirids, petaurids, acrobatids, and macropodiforms that retain a unicuspid posterior to i1), paucituberculatans, and †Yalkaparidon as unknown (“?”) because the homologies of single-rooted unicuspids that occur behind the gliriform lower incisor are uncertain in these taxa (see char. 148). Based on examined specimens it is unclear whether Notoryctes retains a lower canine or not (Gadow, 1892, reported that presence of the lower canine is variable in marsupial moles), so we also scored this taxon as unknown.

Lower Premolars

The number of lower premolars in metatherian taxa examined for this study ranges from one to three. Complete complements are present in the nonmarsupial metatherians included here, didelphids, microbiotheriids, peramelemorphians, and most dasyuromorphians, but one or more lower premolars are clearly missing or are assumed to be missing in other marsupial taxa. Because assumptions made in previous studies about which lower premolar loci are present and which are absent in different taxa are often not clearly justified, we discuss the basis for our scoring decisions in substantive detail. The two or three lower postcanine teeth present in Tarsipes are vestigial and of uncertain homology (see char. 113; Parker, 1890; Archer, 1984c: 690, fig. 133; Russell and Renfree, 1989; Rosenberg and Richardson, 1995), so we scored characters relating to the lower premolars as unknown (“?”) or inapplicable (“-“) in this taxon.

Character 153. First lower premolar present (0); or absent (1). We scored the first lower premolar (p1) as present (state 0) in all metatherians with three unambiguously recognizable premolars; these taxa include †Mayulestes, †Allqokirus, †Pucadelphys, †Mimoperadectes, †Herpetotherium, didelphids, microbiotheriids, peramelemorphians, thylacinids, Myrmecobius, and most dasyurids. In the didelphid Caluromysiops, P1 is only variably present (see char. 114), whereas p1 (although small; Voss and Jansa, 2009: 57) is consistently present in the specimens we examined, indicating that presence or absence of of one of these teeth is at least partially independent of the other.

We coded p1 as present in dasyurids that typically retain only two lower premolars (Dasycercus, Dasykaluta, Dasyuroides, Dasyurus, Pseudantechinus, and Sarcophilus), because none of these exhibit any evidence of dental replacement at the more posterior locus (suggesting that p3 has been lost). In effect, the same dental locus appears to have been suppressed in both the upper and lower dentitions of these taxa (see accounts for chars. 114, 122, above). Myoictis is polymorphic for presence of p3 (char. 155; as noted by Tate, 1947; Woolley, 2005a), and this tooth is also occasionally present as a single-rooted vestigial element in Pseudantechinus (e.g., AMNH 196786; see also Spencer, 1896: 28; Ride, 1964: 62; Woolley, 2011: 58, appendix 3). By contrast, p1 and p2 are well-developed, double-rooted premolariform teeth in all examined dasyurids. Although not conclusive, we consider this additional supportive evidence that it is p3 that has been lost in two-premolared dasyurids.

In the fossil phalangerid †Onirocuscus, there are alveoli for two double-rooted teeth (inferred from two narrowly separated pairs of alveoli) anterior to p3 (Flannery and Archer, 1987a; Crosby, 2007); to our knowledge, lower incisors and lower canines are consistently single-rooted in metatherians, so we interpret these alveoli in †Onirocuscus as housing p1 and p2 and score this taxon as state 0 here. We coded p1 as absent (state 1) in macropodiforms that lack teeth between the gliriform i1 and the tooth that we interpret as p2 (see chars. 148, 154). We also coded p1 as absent in phascolarctids, diprotodontids, wynyardiids, and vombatids, all of which lack teeth between the gliriform lower incisor and a large tooth appressed to the molar series that we interpret as p3 (see char. 155).

We coded this character as unknown (“?”) in other diprotodontians and in all paucituberculatans, because the homologies of the single-rooted, unicuspid teeth that occur between the gliriform lower incisor and p2 and/or p3 in these taxa are unclear (char. 148). By contrast, Martin (2007: 400, fig. 4) tentatively identified the presence of p1 (which he referred to as “dp1,” following Luckett [1993b] and Luckett and Hong [2000]) in Rhyncholestes, based on the presence of a “slightly developed talonid” on a tooth immediately anterior to p2 in young specimens with unworn dentitions. We also observed a talonidlike structure on the tooth immediately anterior to p2 in a specimen of Rhyncholestes (BMNH 1975.1723), but the tooth anterior to that has an almost identical crown morphology (likewise with a “talonid”); therefore, this structure does not seem sufficient to identify p1 in Rhyncholestes, which we scored as unknown (“?”). We also scored †Yalkaparidon and Notoryctes as unknown due to the uncertain homologies of the antemolar dentition in both taxa (see chars. 148, 152).

Character 154. Second lower premolar present (0); or absent (1). We coded the second lower premolar (p2) as present in all examined taxa with three lower premolars and in dasyurids with two lower premolars for the reasons previously discussed (see char. 153). We also identify as p2 the double-rooted tooth immediately anterior to p3 in most paucituberculatans, including Caenolestes, Rhyncholestes, and Lestoros; however, we scored †Acdestis and †Palaeothentes as unknown because the tooth immediately anterior to p3 in these taxa is single-rooted and hence of uncertain homology (see chars. 148, 149, and 152).

The acrobatids Acrobates and Distoechurus (fig. 48) exhibit a conventionally premolariform p2, the tallest tooth in the lower jaws of these taxa (Archer, 1984c: fig. 143). We also identify as p2 the double-rooted tooth (inferred from a narrowly separated pair of alveoli) immediately anterior to p3 in †Onirocuscus (see char. 153 and Flannery and Archer, 1987a). Available juvenile specimens of macropodiforms have a large two-rooted sectorial tooth appressed to a deciduous tooth that is clearly dp3; we interpret this sectorial tooth as p2, in agreement with previous authors (e.g., Thomas, 1888; Tate, 1948; Ride, 1961; Bartholomai, 1971b; Kirkpatrick, 1978; Archer, 1984c; Flannery, 1987; Kear and Cooke, 2001; but note that some of these authors followed Thomas's (1888) hypothesis of dental homologies and referred to this tooth as “p3”; see table 12). As described above (char. 117), both the macropodiform p2 and its occluding partner in the maxillary dentition (P2) are shed when the permanent third premolars erupt (or shortly thereafter), except in †Ekaltadeta (see Wroe and Archer, 1995).

To avoid a priori assumptions of macropodiform monophyly, we coded all macropodiform and nonmacropodiform taxa in which adult specimens lack p2 but for which juvenile specimens (which would reveal whether p2 is present earlier in ontogeny) are unavailable as unknown (“?”); such taxa include †Rhizosthenurus, †Balbaroo, †Ganawamaya, †Neohelos, †Silvabestius, †Ilaria, †Muramura, †Warendja, and †Nimiokoala. By contrast, p2 is unequivocally absent in adult and juvenile specimens of †Ngapakaldia and †Nimbadon (Black et al., 2010), †Namilamadeta (Pledge, 2005), Phascolarctos, Vombatus, and Lasiorhinus. We scored all remaining diprotodontians (thylacoleonids, pseudocheirids, burramyids, petaurids, Tarsipes, and all phalangerids except †Onirocuscus) as well as Notoryctes and †Yalkaparidon as unknown due to uncertain dental homologies.

Character 155. Third lower premolar present (0); or absent (1). Although p3 is present (state 0) in all metatherians that unambiguously exhibit three lower premolars (including examined nonmarsupial metatherians, didelphids, peramelemorphians, and most dasyuromorphians), we scored p3 as absent (state 1) in two-premolared dasyurids (Dasycercus, Dasykaluta, Dasyuroides, Dasyurus, Pseudantechinus, and Sarcophilus; see discussion under char. 153). As noted earlier, Myoictis is polymorphic (“0+1”) for the presence of p3 (Tate, 1947; Woolley, 2005a). All other metatherians in which premolar homologies are uncontroversial consistently retain p3, with the exception of Distoechurus, in which this tooth is only variably present (Archer, 1984b: 695), and which we also scored as polymorphic (the specimen illustrated in fig. 48 lacks p3).

Obviously, presence or absence of p3 could be expected to show some degree of covariation with presence or absence of P3 (char. 122). However, both Distoechurus and Myoictis consistently retain P3 but sometimes lack p3, whereas Dasyuroides retains P3 and lacks p3. Some specimens of Pseudantechinus (e.g., AMNH 196786; see also Spencer, 1896: 28; Ride, 1964b: 62; Woolley, 2011: 58, appendix 3) and Dasycercus (Archer, 1976c: table 1; 1982b: table 2; Woolley, 2005b; Woolley et al., 2013) also retain P3 and lack p3. Collectively, this implies some degree of character independence. Thus, we scored the presence or absence of these loci as separate characters.

Character 156. Second lower premolar (p2) distinctly taller than p3 (0); or p2 and p3 subequal in height (1); or p3 distinctly taller than p2 (2). The relative heights of the second and third premolars (as indicated by the tips of fully erupted but unworn or lightly worn teeth when the alveolar plane of the mandible is oriented horizontally) exhibit consistent taxonomic variation in Marsupialia. Descriptions of character states and an account of their distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 63). Among other metatherians, p2 is taller than p3 (state 0) in †Thylophorops, many dasyurids (including †Barinya; Wroe, 1999: text-fig. 3), †Badjcinus (Muirhead and Wroe, 1998: fig. 1D), some peramelemorphians (e.g., Chaeropus [fig. 37]), and Acrobates. Most specimens of Pseudantechinus and Distoechurus lack p3, and this tooth is only variably present in Myoictis (char. 155); where present, however, p3 is smaller than p2 in these three taxa, so we scored them all as state 0. By contrast, these teeth are subequal in height (state 1) in †Sparassocynus, several peramelemorphians (e.g., Macrotis [fig. 38]), some dasyurids (e.g., Antechinomys, Murexia), and Myrmecobius (fig. 41). Lastly, p3 is taller than p2 (state 2) in †Pucadelphys (Marshall and Muizon, 1995: fig. 6D), †Herpetotherium, all paucituberculatans in which p2 can be unambiguously identified (e.g., Caenolestes [fig. 35]), microbiotheriids (e.g., Dromiciops [fig. 43]), Thylacinus (fig. 40), †Nimbacinus, †Ekaltadeta (Wroe and Archer, 1995), and Hypsiprymnodon moschatus (Ride, 1961: plate I, fig. 3).

We scored this character as inapplicable (“-”) in taxa that entirely lack p2 and/or p3, and in those where p2 and p3 do not cooccur (i.e., those macropodiforms in which p3 simultaneously evicts p2 and dp3). Taxa in which the presence or absence of p2 cannot be determined with certainty (char. 154) were scored as unknown (“?”).

FIG. 22.

Occlusal views of right deciduous third lower premolars (dp3) of Sminthopsis crassicaudata (A, AMNH 196686) and Isoodon macrourus (B, AMNH 160085). Alternative states of character 157 (see main text for description of this character and character states) are illustrated as follows: Sminthopsis 157(0); Isoodon 157(1). Teeth are not shown to the same scale.

img-z154-1_01.jpg

Character 157. Deciduous lower third premolar (dp3) with distinct trigonid and talonid (0); or without distinct trigonid and talonid (1). The deciduous third lower premolar is a large and more or less molariform tooth (state 0) in most metatherians, but it has been reduced to a functionless vestige (state 1) in several groups. This character is modified from Voss and Jansa (2003: char. 64), because we were unable to consistently distinguish between a complete (tricuspid) and incomplete (bicuspid or unicuspid) trigonid of dp3 among the taxa in this study. Instead, we simply distinguish between taxa in which dp3 comprises an obviously distinct trigonid (regardless of the number of cusps) and talonid, and those in which a distinct trigonid and talonid cannot be identified.

Juvenile specimens retaining dp3 were not available for all the terminals included in our analysis. However, dp3 is large and exhibits distinct trigonid and talonid moieties in most Recent didelphids (except Hyladelphys; Voss et al., 2001), †Sparassocynus (Beck and Taglioretti, 2020), Dromiciops (Hershkovitz [1999: fig. 30B, 30C], but note that dp3 is incorrectly labelled as p3 in fig. 30C), the dasyurids Sminthopsis (fig. 22A) and Antechinomys, †Namilamadeta, †Nimbadon (Black and Hand, 2010), †Ngapakaldia, and all macropodiforms and phalangerids for which specimens preserving dp3 were available for examination.

As noted above (char. 121), Myrmecobius appears to retain dp3 into adulthood. The morphology and size of this tooth varies considerably among examined specimens, but it is typically tricuspid, with the cusps arranged anteroposteriorly (as in, e.g., BMNH 1844.1.22.21 and 1844.9.2.18); we tentatively identify these cusps as representing (from anterior to posterior) the paraconid, either the metaconid or the fused metaconid + protoconid, and the entoconid. We consider this morphology to correspond most closely to state 0 and scored Myrmecobius accordingly.

By contrast, dp3 is vestigial and occlusally featureless in Caenolestes (Luckett and Hong, 2000), the only paucituberculatan for which we were able to examine juvenile material. This tooth is also small and lacks a distinct trigonid and talonid in peramelemorphians (e.g., Isoodon; fig. 22B), most three-premolared dasyurids (with the notable exceptions of Sminthopsis and Antechinomys; see above), Thylacinus (Flower, 1867; Luckett et al., 2019), and several diprotodontians (including Phascolarctos [Thomas, 1887a], Distoechurus, Petaurus, Dactylopsila, Dactylonax, Pseudochirops cupreus, and Petropseudes).

As previously discussed (see chars. 120, 155), we assume that the tooth at the third premolar locus in adult vombatids is a successor tooth (P3 in the upper dentition, p3 in the lower dentition). Examination of young juvenile specimens of Vombatus and Lasiorhinus did not reveal the presence of an obvious dp3, and indicated that p3 erupts very early in ontogeny (Thomas, 1887b). As for the upper dentition (see char. 120), we consider this evidence that dp3 was either absent or vestigial and lost very early in development, and score both Vombatus and Lasiorhinus as state 1. However, in the absence of juvenile material, we score †Warendja as unknown (“?”). The deciduous dentition of thylacoleonids has never been identified, and the permanent third premolar erupts at a very early stage (before m1) in †Thylacoleo carnifex (see Finch, 1971; Pledge, 1975). These facts have been interpreted as evidence that this taxon may have lacked deciduous premolars (Finch, 1971; Pledge, 1975). We therefore scored †Thylacoleo as state 1. Juvenile specimens are unavailable for †Lekanoleo and †Wakaleo, however, which have therefore been scored as unknown (“?”).

Lower Molars

Tribosphenic lower molars include three anterior cuspids (paraconid, protoconid, and metaconid) that form an elevated wedge-shaped array (the trigonid) with the protoconid at its labial apex, and three posterior cuspids (hypoconid, hypoconulid, and entoconid) that surround the talonid basin (figs. 14, 23A–C; see also Bown and Kraus, 1979: fig. 9-1; Reig et al., 1987: fig. 1; Wroe, 1999: text-fig. 1A; Goin and Candela, 2004: fig. 2A; Kielan-Jaworowska et al., 2004: fig. 11.1; Williamson et al., 2014: fig. 3). Typically (but not invariably), these cusps are linked by enamelled crests, notably the paracristid (connecting the paraconid and protoconid), the metacristid (connecting the protoconid and metaconid),18 and the cristid obliqua (connecting the hypoconid with the metacristid). In many (but not all) tribosphenic taxa, the paracristid and the metacristid are prominently notched. Other lower-molar structures common to many tribosphenic taxa include shelflike anterior and posterior cingulids.

In general, lower-molar cuspid homologies among nontribosphenic metatherians are less controversial than the upper-molar cusp homologies discussed in the preceding accounts. The anterior lower molars (m1–3) of caenolestids (fig. 23B) and other paucituberculatans (Abello, 2013: figs. 24), for example, include all the usual six tribosphenic cuspids, although the paraconid is not always clearly identifiable when the paracristid is unnotched. The lower molars of peramelemorphians (fig. 23C) preserve all of the usual tribosphenic cuspids, differing in only minor details from the plesiomorphic condition (e.g., by the absence of a hypoconulid notch in the anterior cingulid and by the presence in some forms of a neomorphic labial cuspid in the hypoflexid region).

By contrast, zalambdodont taxa have radically modified lower molars. Correlated with reduction or loss of the occluding protocone on the upper molars, the talonid is greatly reduced in both Notoryctes (fig. 23D; Asher and Sánchez-Villagra, 2005: fig. 6A; Archer et al., 2011: fig. 1) and †Yalkaparidon (Archer et al., 1988; Beck, 2009; Beck et al., 2014). As a consequence, only the trigonid moiety—complete with paraconid, protoconid, metaconid, and the usual connecting cristids—remains as a fully functional component of zalambdodont metatherian lower molars.19

Selenodont lower molars appear strikingly different from the tribosphenic condition because the cuspids and cristids are modified as crescentic blades (“selenes”), but four of the six main cuspids of the tribosphenic bauplan remain, at least on the posterior molars (m2–4), in both phascolarctids (fig. 24A) and pseudocheirids (fig. 24B). Only the paraconid is consistently missing in both selenodont groups, although the hypoconulid is also indistinct or absent in phascolarctids (compare Archer, 1984b: figs. 62, 178D-I). In both groups, however, the m1 trigonid exhibits additional nontribosphenic features (see below).

FIG. 23.

Occlusal views of right lower molars of Metachirus nudicaudatus (A, AMNH 266453 [m2]), Caenolestes fuliginosus (B, UMMZ 155575 [m3]), Echimipera kalubu (C, AMNH 221654 [m2]), and Notoryctes typhlops (D, AMNH 198651 [m2]). General features of lower molar crown morphology discussed in the text are illustrated, as are characters 158, 161, 162, 163, 166, 167, 168, 170, 171, 172, 174, 175, 176, 177, 178, 179, and 180 (see main text for descriptions of these characters and character states), as follows: Metachirus 158(0), 161(0), 162(1), 163(0), 166(0), 167(0), 168(0), 170(0), 171(0), 172(0), 174(0), 175(0), 176(0), 177(0), 178(0), 179(0), 180(0); Caenolestes 158(0), 161(0), 162(0), 163(0), 166(0), 167(0), 168(0), 170(0), 171(0), 172(0), 174(0), 175(0), 176(0), 177(1), 178(0), 179(0), 180(0); Echimipera 158(1), 161(0), 162(1), 163(0), 166(0), 167(0), 168(0), 170(0), 171(1), 172(0), 174(0), 175(0), 176(0), 177(0), 178(0), 179(0), 180(0); Notoryctes 158(-), 161(0), 162(-), 163(0), 166(1), 167(-), 168(-), 170(-), 171(-), 172(-), 174(-), 175(-), 176(-), 177(-), 178(-), 179(-), 180(-). Numerical labels refer to character states defined in the text. Teeth are not drawn to the same scale.

img-z156-1_01.jpg

Most nonselenodont diprotodontians consistently have only four well-developed lower-molar cusps, the paraconid and the hypoconulid having atrophied or disappeared completely. Although the lower molars of bunodont, bunolophodont, and fully lophodont taxa differ in many other occlusal details, the principal contrast among these dentitions concerns the development of transverse lophids, of which the anteriormost is called the protolophid and the posteriormost the hypolophid (fig. 25A–D). However, because these lophids occlude with the transverse lophs of upper molars, we did not score incipient or fully developed lophodonty in the lower dentition as a separate character. Similar to characters relating to the lower premolars (see above), we scored lower molar characters as or unknown (“?”) or inapplicable (“-”) in Tarsipes, due to the uncertain homology of its two or three vestigial lower postcanine teeth (char. 113; Parker, 1890; Archer, 1984c: 690, fig. 133; Russell and Renfree, 1989; Rosenberg and Richardson, 1995).

FIG. 24.

Occlusal views of right second lower molars (m2) of Phascolarctos cinereus (A, USNM 534818) and Pseudochirops cupreus (B, AMNH 109633). General features of lower molar crown morphology discussed in the text are illustrated, as are characters 158, 161, 162, 166, 167, 168, 170, 171, 172, 174, 175, 176, 177, 178, 179, and 180 (see main text for descriptions of these characters and character states), as follows: Phascolarctos 158(-), 161(0), 162(-), 166(0), 167(0), 168(0), 170(0), 171(1), 172(0), 174(1), 175(0), 176(0), 177(1), 178(1), 179(-), 180(0); Pseudochirops cupreus 158(-), 161(0), 162(2), 166(0), 167(0), 168(0), 170(0), 171(1), 172(0), 174(1), 175(0), 176(0), 177(1), 178(0), 179(0), 180(0). Abbreviations: encd, entoconid; hycd, hypoconid; mecd, metaconid; prcd, protoconid. Numerical labels refer to character states defined in the text. Teeth are not drawn to the same scale.

img-z157-1_01.jpg

Character 158. Hypoconulid notch in anterior cingulid of m2–4 present (0); or absent (1). The anterior cingulid of m2–4 in most tribosphenic metatherians has a distinct notch that receives the hypoconulid of the preceding molar (state 0; figs. 14, 23A, B). However, the hypoconulid notch is absent (state 1) in peramelemorphians, in which the hypoconulid is directly posterior to the entoconid, and the anterior cingulid of the following tooth extends without interruption to the lingual margin of the tooth (fig. 23C). Absence of a hypoconulid notch has been interpreted as a peramelemorphian synapomorphy (Wroe, 1997b; Muirhead, 2000), but this indentation is also absent in Sarcophilus (see Wroe, 1997b; 1997a; 1999; Wroe et al., 2000: char. 30; Wroe and Musser, 2001: char. 30).

We note some discrepancies between our own observations and those previously reported in the literature about this character. Wroe (1997b, 1997a, 1999) and Muirhead and Wroe (1998) stated that a hypoconulid notch is absent in Thylacinus, and Wroe (1997b, 1999) also claimed that it is absent in microbiotheriids; nevertheless, Wroe et al. (2000: char. 30) and Wroe and Musser (2001: char. 30) scored the hypoconulid notch as present in both Thylacinus and Dromiciops. We observed a distinct hypoconulid notch on m2–4 of all examined specimens of Thylacinus cynocephalus (see also Warburton et al., 2019: figs. 1113), Dromiciops gliroides, and †Microbiotherium tehuelchum, so we scored these three taxa as state 0. Wroe (1997b) also stated that the hypoconulid notch is absent in Myrmecobius, but because we could not identify an anterior cingulid in this taxon, we scored it as inapplicable (“-”). Other taxa that we scored as inapplicable—because they lack an anterior cingulid, a hypoconulid (see char. 178), or both—include most examined diprotodontians, †Acdestis, Notoryctes, and †Yalkaparidon.

FIG. 25.

Occlusal views of right lower molars of Distoechurus pennatus (A, AMNH 104058 [m2]), Hypsiprymnodon moschatus (B, AMNH 160120 [m2]), Trichosurus vulpecula (C, AMNH 65547 [m1]), and Lagostrophus fasciatus (D, AMNH 155105 [m1]). General features of lower molar crown morphology discussed in the text are illustrated, as are characters 158, 161, 162, 163, 166, 167, 168, 170, 171, 172, 174, 175, 176, 177, 178, 179, and 180 (see main text for descriptions of these characters and character states), as follows: Distoechurus 158(-), 161(0), 162(-), 163(0), 166(0), 167(0), 168(0), 170(0), 171(0), 172(0), 174(0), 175(1), 176(0), 177(0), 178(1), 179(-), 180(0); Hypsiprymnodon moschatus 158(-), 161(0), 162(-), 163(0), 166(0), 167(0), 168(1), 170(0), 171(0), 172(0), 174(0), 175(0), 176(0), 177(0), 178(1), 179(-), 180(0); Trichosurus vulpecula 158(-), 161(0), 162(-), 163(0), 166(0), 167(0), 168(1), 170(0), 171(0), 172(0), 174(-), 175(0), 176(0), 177(2), 178(1), 179(-), 180(0); Lagostrophus 158(-), 161(0), 162(-), 163(0), 166(0), 167(0), 168(0), 170(0), 171(0), 172(0), 174(-), 175(0), 176(0), 177(1), 178(1), 179(-), 180(0). Abbreviations: encd, entoconid; hycd, hypoconid; mecd, metaconid; prcd, protoconid. Numerical labels refer to character states defined in the text. Teeth are not drawn to the same scale.

img-z158-1_01.jpg

Character 159. Lower first molar (m1) paracristid present and distinct (0); or both paracristid and paraconid indistinct or absent (1); or paraconid well developed but paracristid indistinct or absent (2). Both the paracristid and paraconid are distinct occlusal features (state 0) in most tribosphenic metatherians, including †Pucadelphys, †Allqokirus, †Mayulestes, †Herpetotherium, †Mimoperadectes, didelphids (figs. 14, 23A), microbiotheriids, peramelemorphians (fig. 23C), and most dasyuromorphians (Wroe, 1996a). They are also distinct in some nontribosphenic taxa, including Notoryctes (fig. 23D), †Yalkaparidon, phascolarctids (fig. 24A), wynyardiids, and most pseudocheirids. However, in many diprotodontians (e.g., petaurids, diprotodontids, thylacoleonids, some macropodiforms) the paracristid is clearly identifiable as a distinct crest extending anterolingually or anteriorly from the protoconid (the major trigonid cusp), but it does not always terminate in a distinct, cusplike paraconid. Examined specimens of Vombatus were too worn to score for this character, but Brewer et al. (2018: fig. 13.2) illustrated a juvenile specimen that suggests that a short, but distinct paracristid is present in this taxon, which we therefore scored as state 0. The paracristid is also clearly present on m1 in paucituberculatans, although this crest does not always end in a distinct, cusplike paraconid (Abello, 2013: char. 15). Similarly, in several dasyurids (Dasycercus, Dasyuroides, Dasyurus, Myoictis, Sarcophilus, and some specimens of Murexechinus and Pseudantechinus) the paracristid is clearly present, but the paraconid is either vestigial or absent; in at least some of these taxa (e.g., Pseudantechinus, Murexechinus), the paraconid appears to have fused with the anterior cingulid. Because the degree of development of the paraconid appears to vary continuously among these taxa and is noticeably affected by even slight wear, we scored all examined metatherians with a distinct paracristid as state 0 whether or not this crest terminates in a distinct paraconid. Taxa without a distinct paraconid typically lack a paracristid notch, a trait that we score separately (see char. 160).

By contrast, both the paraconid and paracristid appear to be entirely absent from m1 (state 1) in Burramys (Archer, 1984b: fig. 147E), Acrobates (Archer, 1984b: fig. 143), and in all specimens of Macrotis that we examined (but see Bensley, 1903: 115; Ride, 1964b: 112, fig. 9; and Archer and Kirsch, 1977: 22 for conflicting reports regarding the distribution of this character in Macrotis). In Macrotis at least, absence of a distinct paraconid and paracristid appears to be the result of fusion of the paraconid with the anterior flank of the metaconid (e.g., on WAM M632). Distoechurus is polymorphic for this character, with the m1 paracristid present in some specimens (e.g., AMNH 191045) but absent in others (e.g., AMNH 104058), so we scored this taxon as polymorphic (“0+1”). We observed a third distinct morphology (state 2) in Myrmecobius, which has a well-developed paraconid in the usual tribosphenic position (anterolingual to the protoconid and anterior to the metaconid), but which lacks a crest linking this cuspid to the protoconid. Although all specimens of Lasiorhinus we examined were too worn to score this character, a juvenile specimen of L. latifrons illustrated by Brewer et al. (2018: fig. 13) demonstrates that both the paraconid and paracristid are also absent in this taxon, which we scored as state 1.

We scored this character as inapplicable (“-”) in Tarsipes (due to its vestigial postcanine dentition) and as unknown (“?”) in the fossil vombatid †Warendja, because all available specimens of this taxon were too worn to allow confident scoring. As defined, the states of this character do not represent an obvious morphocline, so we did not order them in any of our analyses.

Character 160. Lower first molar (m1) paracristid notched (0); or m1 paracristid unnotched (1). In tribosphenic metatherians that retain a distinct paracristid terminating in a well-developed paraconid on m1, this crest is usually notched (state 0). This morphology characterizes †Herpetotherium, †Pucadelphys, †Allqokirus, †Mayulestes, fossil and Recent didelphids (Voss and Jansa, 2009: figs. 24, 25), microbiotheriids, peramelemorphians, and most dasyuromorphians. The m1 paracristid is also notched in zalambdodont marsupials (Notoryctes and †Yalkaparidon). By contrast, in all examined paucituberculatan lower dentitions (i.e., of Recent caenolestids, †Acdestis, †Palaeothentes, †Pichipilus, and †Stilotherium) the m1 paracristid is unnotched, sloping gently without interruption from the protoconid to the anterolingual corner of the tooth (state 1; see Marshall, 1980: fig. 6; Rae et al., 1996: fig. 9; Goin et al., 2007a: char. 11; 2009a; char. 11; Abello, 2007: char. 15; 2013: char. 14, fig. 3K–L). The paracristid of m1 is similarly unnotched in dasyurids that lack a distinct paraconid on that tooth (namely, Dasyuroides, Dasyurus, Myoictis, Sarcophilus, Murexechinus, and Pseudantechinus), whereas more posterior teeth have notched paracristids; we also scored these taxa as state 1. In two other dasyurids (Dasycercus and Phascolosorex) the m1 paracristid is unnotched in some specimens but not others, so we scored them as polymorphic (“0+1”).

The m1 paracristid is consistently notched in some pseudocheirids (Pseudochirops archeri, Pseudochirulus, and Petropseudes), but this notch is only variably present in other pseudocheirids (Hemibelideus, Petauroides, and Pseudochirops cupreus), Phalanger, Ailurops, and Trichosurus, which we scored as polymorphic (“0+1”). In all other diprotodontians in which the m1 paracristid is identifiable, this crest is unnotched. We scored this character as inapplicable (“-”) in Burramys, Acrobates, Myrmecobius, Lasiorhinus and Tarsipes, none of which exhibit a distinct paracristid (char. 159), and we scored it as unknown (“?”) in †Warendja because the first lower molars of available specimens were too worn to score confidently.

Character 161. Second and third lower molars (m2, m3) with distinct paracristids (0); or both paraconids and paracristids indistinct or absent from m2 and m3 (1); or paraconid well developed but paracristid indistinct or absent on m2 and m3 (2). As on m1 (see char. 159), the presence of distinct paracristids and paraconids on m2 and m3 appears to be plesiomorphic for metatherians, as seen (for example) in †Herpetotherium, †Pucadelphys, †Allqokirus, †Mayulestes, †Mimoperadectes, didelphids, microbiotheriids, dasyuromorphians, Notoryctes, †Yalkaparidon, and most peramelemorphians and paucituberculatans. In Macrotis, the pauctituberculatans †Acdestis and †Palaeothentes (see Marshall, 1980: 124; Marshall et al., 1990: fig. 2, node 33; Abello, 2007: char. 24, 2013: char. 23), and many diprotodontians (e.g., petaurids, pseudocheirids, burramyids, most macropodiforms), the paracristid is clearly identifiable on m2 and m3, but it does not terminate in a distinct, cusplike paraconid. However, because the degree of development of the paraconid on these teeth appears to vary continuously among examined taxa, we scored all of these as state 0 based on the presence or absence of the paracristid only, regardless of whether a distinct paraconid is present.

By contrast, both the paraconid and paracristid are absent from m2 and m3 in Acrobates and diprotodontids (state 1). In Myrmecobius, m2 and m3 resemble m1 (see char. 159) in possessing a well-developed paraconid anterolingual to the protoconid and anterior to the metaconid, but these teeth lack a paracristid linking the paraconid to the protoconid (state 2). As defined, the states of this character do not represent an obvious morphocline, so we did not order them in any of our analyses.

Character 162. Paracristids of m2 and m3 deflected posterolingually by hypoconulid of preceding tooth (0); or paracristids of m2 and m3 undeflected (1); or paracristids of m2 and m3 terminate labial to the hypoconulid (2). In most tribosphenic metatherians, the paracristid extends anterolingually in a more or less straight line from the protoconid to the paraconid (state 1), as in †Herpetotherium, †Pucadelphys, †Allqokirus, †Mayulestes, didelphids (fig. 23A), microbiotheriids, peramelemorphians (fig. 23C), and most dasyuromorphians. However, in all examined paucituberculatan lower dentitions, the paracristids of m2 and m3 are abruptly deflected posterolingually by the hypoconulids of m1 and m2, respectively (fig. 23B; state 0).20 Lower dental material referable to †Evolestes hadrommatos has not been described, but Goin et al. (2010: 75) reported that in †Evolestes sp. “the paracristid is not straight: there is a ∼90° angle at the meeting of the postparacristid [their term for the lingual segment of the paracristid] and the preprotocristid [their term for the labial segment of the paracristid],” indicating that the paracristid is deflected as we observed in other paucituberculatans, so we also scored †Evolestes as state 0.

In pseudocheirids the paracristid extends anterolingually from the protoconid to terminate near the midpoint of the anterior margin of m2–4, well labial to the hypoconulids of the preceding teeth (Archer, 1984a [fig. 178D-H]; Roberts et al., 2008 [fig. 2B], 2009 [text-fig. 1B]), an apparently distinctive morphology (state 2).

We scored this character as inapplicable in taxa that lack an identifiable paracristid on m2 and m3 (Myrmecobius; see char. 161), or that lack an identifiable hypoconulid (†Yalkaparidon, Notoryctes (fig. 23D), most diprotodontians; see char. 178), or that lack both structures (diprotodontids, Acrobates).

Character 163. Metaconid present on all lower molars (0); or absent on m1 only (1); or absent on all lower molars (2). The metaconid is consistently present on m1–4 of most tribosphenic metatherians, including †Pucadelphys, †Allqokirus, †Mayulestes, †Herpetotherium, †Mimoperadectes, didelphids (figs. 14, 23A), paucituberculatans (fig. 23B), microbiotheriids, peramelemorphians (fig. 23C), Notoryctes (fig. 23D), and most dasyuromorphians (state 0). However, the metaconid is absent on all four lower molars of Thylacinus (state 2; see, e.g., Ride, 1964a: fig. 7; Archer, 1984c: fig. 29; Muizon and Lange-Badré, 1997: fig. 4C, D; Archer et al., 2016b: fig. 9F; Solé and Ladevèze, 2017: fig. 4). An apparently intermediate morphology is seen in †Badjcinus (Muirhead and Wroe, 1998) and Sarcophilus (Ride, 1964a: fig. 7; Archer, 1984c: fig. 15B; Archer et al., 2016b: fig. 9E), in which the metaconid is absent on m1 but present on m2–4 (state 1). The metaconid is absent from m1 in some specimens of Pseudantechinus but present in others, whereas it is consistently present on m2–4; therefore, we scored this taxon as polymorphic (0+1). Although m2–4 of †Mutpuracinus clearly possess metaconids (Murray and Megirian, 2006a), m1 is unknown for this taxon, which we scored using ambiguity coding (as 0/1).

A distinct metaconid is present on all four lower molars of Gymnobelideus, most pseudocheirids (with the exception of Hemibelideus and some specimens of Pseudochirops cupreus), burramyids, some phalangerids (e.g., †Onirocuscus, Phalanger, Spilocuscus, Trichosurus, “Trichosurus” †dicksoni, and Wyulda), all modern and fossil macropodiforms, wynyardiids, diprotodontids, and phascolarctids. Among lophodont and bunolophodont diprotodontians, the metaconid forms the lingual end of the protolophid. Acrobatids have only three upper and lower molars (see char. 130), but a distinct metaconid is present on all three lower molars, so these terminals were scored as 0. Examined specimens of the vombatids Vombatus and Lasiorhinus were too worn to confidently score this character, but juvenile specimens illustrated by Brewer et al. (2018: fig. 13) suggest that the metaconid is present on m1–4 of both taxa, which have been scored as state 0.

The m1 trigonid of thylacoleonids comprises a single cuspid, which we consider to be homologous with the protoconid following Gillespie (2007), but contra Woods (1956), Clemens and Plane (1974), and Megirian (1986), all of whom identified it as the metaconid. In †Lekanoleo (which has four lower molars; Gillespie, 2007: fig. 8.8) and †Wakaleo (which has only three lower molars; Gillespie et al., 2014: fig. 5C), a distinct metaconid is present on m2. The morphology of the more posterior lower molars (m3–4) is currently unknown in †Lekanoleo (Gillespie, 2007; Gillespie et al., 2020), and the m3 of †Wakaleo is occlusally simplified (Gillespie et al., 2014); nevertheless, we scored both terminals as state 1 based on the condition in m1–2. The m2 of †Thylacoleo is very small and occlusally simple, so it is unclear whether a metaconid is present on this tooth; given that the metaconid is unequivocally absent from m1 in this taxon (see above), we scored †Thylacoleo as ambiguous (1/2) for this character. A distinct metaconid is also absent from m1 in Dactylopsila (Archer, 1984c: fig. 208), Dactylonax, Hemibelideus, and Strigocuscus, but the metaconid is clearly identifiable on at least one more posterior molar in all these taxa, which we therefore scored as state 1. A metaconid is variably present on m1 in Petaurus, Pseudochirops cupreus, and Ailurops, all of which we scored as polymorphic (“0+1”). The homology of the posterolingual cusp on the lower molars of †Yalkaparidon is unclear: it might either be the metaconid (as in Notoryctes) or a remnant of the talonid (Beck et al., 2014: 155). We therefore scored †Yalkaparidon as unknown (“?”).

The states of this character represent a plausible morphocline and have been so ordered in all of our phylogenetic analyses.

Character 164. Trigonid of m4 resembles that of m3 (0); or m4 trigonid structurally simplified, only the metaconid distinct (1); or m4 trigonid structurally simplified, only the protoconid distinct (2). In most metatherians the trigonid of m4 resembles that of m3 (state 0). However, in living caenolestids and †Acdestis only a single cuspid is distinct on the m4 trigonid; based on its position and comparisons with m1–3, this cuspid appears to be the metaconid (state 1; Voss and Jansa, 2009: 59–60). In Burramys, the trigonid of m4 is also reduced to a single distinct cuspid, but comparison with more anterior molars suggest that in this case it is the protoconid (state 2). We scored this character as inapplicable (“-”) in taxa lacking m4 (acrobatids, †Wakaleo, and †Thylacoleo). As defined, these states do not form a plausible morphocline, so we left this character as unordered in all our phylogenetic analyses.

Character 165. Additional cuspid labial to m1 protoconid absent (0); or present, forming a cusplike protostylid (1); or present, forming a vertically directed crest (2). In most metatherians, the protoconid is the labialmost trigonid cuspid (state 0). However, in a number of diprotodontians the m1 trigonid is strongly compressed labiolingually and a neomorphic cuspid—the protostylid sensu Flannery and Rich (1986)—is present labial to the protoconid (state 1). This obviously derived morphology is seen in pseudocheirids (in which the protostylid is particularly large; Archer, 1984c: figs. 178E–H, 180B, F), phascolarctids (Black et al., 2012a), Phalanger (Flannery and Rich, 1986: fig. 13. 1), Hypisprymnodon moschatus (Cooke, 1997a: fig. 6F), †Ganawamaya (Kear et al., 2007; Butler et al., 2018), and †Ekaltadeta (Archer and Flannery, 1985). Some specimens of Spilocuscus exhibit a distinct cusplike protostylid on m1, but in others this cusp is absent (Cooke, 1997b: 278), and it is only variably present in †Ilaria spp. (present in †I. illumidens, absent in †I. lawsoni; Tedford and Woodburne, 1987: 410), so we scored both of these taxa as polymorphic (“0+1”). Archer (1984c: fig. 143) and Aplin and Archer (1987: lviii) suggested that the major labial cuspid of the trigonid in the acrobatids Acrobates and Distoechurus might represent a protostylid, but unpublished fossil acrobatid material from late Oligocene to middle Miocene sites at Riversleigh World Heritage Area convincingly demonstrate that it is the protoconid (Fabian, 2012).

In the thylacoleonids †Lekanoleo (Gillespie, 2007: fig. 8.8), †Wakaleo (Gillespie et al., 2014: fig. 5A, C), and †Thylacoleo there is a well-developed vertical crest (rather than a cuspid) buttressing the labial face of m1 (state 2). Because these states do not form a plausible transformation series, we did not order this character in any of our phylogenetic analyses.

Character 166. Talonid present (0); or greatly reduced or absent (1). A distinct talonid is present posterior to the trigonid (state 0) in most metatherians, but in Notoryctes the talonid is absent or reduced to a tiny, cingulidlike remnant posterior to the metaconid (state 1; Archer, 1984c: fig. 57F, G; Asher and Sánchez-Villagra, 2005: fig. 6A; Archer et al., 2011: fig. 1). As discussed earlier (see char. 163), the homology of the part of the lower molars posterior to the protoconid in †Yalkaparidon is unclear, representing either the metaconid or a remnant of the talonid (Beck et al., 2014: 155); whichever it is, the talonid is clearly reduced to a vestige in this taxon (Archer et al., 1988; Beck, 2009; Beck et al., 2014), which we also scored as state 1. Due to the absence of a distinct talonid in both Notoryctes and †Yalkaparidon, we scored all subsequent characters relating to talonid structures (i.e., chars. 167–180) as inapplicable (“-”) for these two taxa. We also scored this character as inapplicable in Tarsipes due to the vestigial nature of its postcanine dentition (see char. 113).

Character 167. Cristid obliqua present (0); or absent or indistinct (1). The cristid obliqua is present (state 0) in most tribosphenic metatherians as a prominent crest that extends anterolingually from the hypoconid (figs. 14, 23A–C). The cristid obliqua is also present in most fully and incipiently lophodont diprotodontians (fig. 25); in many of these forms, the cristid obliqua connects the protolophid and hypolophid, and hence is often referred to as the “midlink.” In phalangerids, a crest that is probably homologous with the cristid obliqua is strongly kinked (see char. 168). However, the cristid obliqua is entirely absent from the lower molars of †Neohelos (Murray et al., 2000a; Black et al., 2013: fig. 6C), †Silvabestius (Black and Archer, 1997a: fig. 3), and †Nimbadon (Hand et al., 1993: fig. 3; Black et al., 2010: figs. 7, 10M, N; Black and Hand, 2010: fig. 8); in these taxa the protolophid (homologous with the metacristid of tribosphenic molars) and hypolophid are separated by a crestless valley. In †Ngapakaldia, however, the cristid obliqua is variably present (Black, 2010, referred to this crest as the “midlink”), so we scored this taxon as polymorphic (“0+1”). The cristid obliqua is consistently absent in Myrmecobius, which we also scored as state 1.

We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166). We also scored it as inapplicable in †Thylacoleo, in which the occlusal structures of the talonid are extremely simplified.

Character 168. Cristid obliqua not strongly kinked (0); or with strongly developed buccal kink (1). Flannery et al. (1987: 482–483) identified the presence of a buccal kink in the cristid obliqua—a morphology referred to as an “angulate cristid obliqua” by Woodburne (1984b: 65) and several other authors (e.g., Woodburne and Clemens, 1986a, 1986b; Black and Archer, 1997b; Crosby and Archer, 2000)—as a distinctive feature of Recent phalangerids, some fossil phalangeroids, and the fossil macropodoid Hypsiprymnodonbartholomaii (see also Flannery and Archer, 1987c). Despite its potential phylogenetic significance, most authors have not provided a clear description of this trait, although it was well illustrated by Flannery and Rich (1986: fig. 2-2). Here we employ a modified version of the definition provided by Crosby (2002a: 337), namely, that this crest is kinked when it extends for a short distance anteriorly from the hypoconid before changing direction sharply lingually, resulting in a distinctly angled deflection in occlusal view (fig. 25B-C).

This kinked crest has been universally identified as the cristid obliqua by previous authors (e.g., Woodburne, 1984b; Woodburne and Clemens, 1986a, 1986b; Archer et al., 1987; Flannery and Archer, 1987a, 1987c; Flannery et al., 1987; Woodburne et al., 1987a; Marshall et al., 1990; Black and Archer, 1997b; Brammall, 1998; Crosby and Archer, 2000; Crosby, 2002a, 2002b; Helgen and Flannery, 2004; Schwartz, 2006a). However, we observed two separate, well-developed crests extending anteriorly from the hypoconid on the unworn m2 and m3 of young specimens of Burramys (e.g., NMV C26919). The more labial of these two crests is slightly kinked, somewhat resembling the morphology seen in phalangerids, whereas the more lingual crest resembles the uninflected cristid obliqua of, for example, Cercartetus (Archer, 1984c: figs. 154–156), petaurids, and acrobatids. Two crests, resembling those we observed in Burramys, can also be seen on an unworn m2 of “Trichosurus” †dicksoni (QM F13061), but on this tooth the more lingual, unkinked crest (referred to as an “accessory ridge” by Flannery and Archer, 1987b: 534) is vestigial and extends only a short distance from the hypoconid, whereas the kinked crest is well developed (see Flannery and Archer, 1987b: fig. 3F). The presence of two crests extending from the hypoconid in Burramys and “T.” †dicksoni raises the possibility that the kinked crest is not homologous with the cristid obliqua, but instead represents a neomorph. Alternatively, it could be that the unkinked crest seen in these two taxa is neomorphic. However, in the absence of more compelling data, we follow traditional usage here and refer to the kinked crest as the cristid obliqua.

The cristid obliqua is strongly kinked (state 1) in phalangerids, Dactylopsila (e.g., AMNH 157140), Hypsiprymnodonbartholomaii (based on QM F13053, an unworn left m3; Flannery and Archer, 1987c: fig. 4B), and H. moschatus (e.g., AMNH 160120). Because we are uncertain which of the two crests extending from the hypoconid in Burramys is the homolog of the cristid obliqua (see above), we scored this taxon as unknown (“?”). Some pseudocheirids (e.g., Pseudochirulus) exhibit a distinctly kinked crest extending from the hypoconid on m4, but not on more anterior molars, so we scored these taxa as state 0. This character is inapplicable for Myrmecobius (in which the cristid obliqua is absent; char. 167) and †Thylacoleo (in which the occlusal structures of the talonid are greatly reduced; char. 167). We scored all other examined taxa with nonvestigial talonids as state 0. (Note, however, that a kinked cristid obliqua has also been reported in a number of fossil diprotodontians not included here; Crosby and Archer [2000].)

Character 169. Cristid obliqua of m1 contacts metaconid (0); or contacts metacristid labial to metaconid (1); or terminates labial to protoconid (at protostylid) (2). In most metatherians in which the cristid obliqua is present (see char. 167), this crest extends anterolingually from the hypoconid to contact the metacristid (or protolophid), at a point labial to the metaconid (state 1). Minor positional differences in the point of contact between the cristid obliqua and metacristid (specifically, whether it is labial or lingual to the “carnassial notch” of the metacristid) have been coded in several previous analyses of metatherian phylogeny (e.g., Springer et al., 1997 [char. 23]; Horovitz and Sánchez-Villagra, 2003 [char. 160]), but we observed continuous variation in this trait among the taxa examined for the present study. However, a qualitatively distinct morphology (state 0) is seen in most pseudocheirids, where the m1 cristid obliqua extend lingually past the metacristid to connect directly with the metaconid (Archer, 1984c: fig. 178E–H; Roberts et al., 2008: fig. 2A). An m1 cristid obliqua that contacts the metaconid has been identified as a potential synapomorphy of Pseudocheiridae (e.g., Archer, 1984c; Archer et al., 1987; Woodburne et al., 1987b), but the cristid obliqua consistently terminates at the metacristid (anterolabial to the metaconid) in Pseudochirulus; Pseudochirops archeri is polymorphic (“0+1”) for these traits; and Hemibelideus cannot be meaningfully scored for this character because it lacks a distinct m1 metaconid (see char. 163). Additionally, we observed state 0 in a number of other metatherians, namely Burramys (Archer, 1984c: fig. 147E), acrobatids (Archer, 1984c: fig. 143), †Acdestis (Bown and Fleagle, 1993: figs. 14, 15.5; Abello, 2007: lámina 11; Abello, 2013: fig. 3I), †Palaeothentes (Bown and Fleagle, 1993: fig. 11; Abello, 2007: lámina 9F–L; Abello, 2013: fig. 3G–H), and Chaeropus (Travouillon, 2016: fig. 2i, j).

In Hypsiprymnodon moschatus (see Cooke, 1997a: fig. 6F) and †Ekaltadeta (Archer and Flannery, 1985: fig. 7.4), the m1 cristid obliqua terminates at the protostylid, labial to the protoconid, a distinctive morphology that we scored as an additional state (state 2; see char. 165).

We scored this character as inapplicable (“-”) in taxa that lack an identifiable m1 metaconid (e.g., Hemibelideus, thylacoleonids, some dasyuromorphians; char. 163) or that lack an m1 cristid obliqua (i.e., Myrmecobius, †Neohelos, †Nimbadon, †Silvabestius; char. 167). We also scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166). As arranged here, the states comprising this character comprise a plausible morphocline, reflecting a progressive labial shift in the point of termination of the m1 cristid obliqua. Thus, we ordered this character in all our phylogenetic analyses.

Character 170. Anterior terminus of cristid obliqua of m2–4 contacts protoconid or metacristid (0); or bypasses metacristid to terminate at entoconid or entocristid (1); or terminates at distinct fissure (2). In the vast majority of metatherians in which the cristid obliqua of m2–4 is present, this crest terminates at the protoconid, or slightly lingual to this, at the metacristid or protolophid (figs. 23A–C, 24; state 0). However, the cristid obliqua of m2–4 bypasses the metacristid to terminate at the entoconid or entocristid (state 1) in Chaeropus (Travouillon, 2016: fig. 2i, j)21 and †Nimiokoala. The bandicoot Isoodon is polymorphic for this character (“0+1”). A qualitatively distinct morphology is observed in some phalangerids (Ailurops, Phalanger, Spilocuscus, Strigocuscus [Flannery and Rich, 1986: fig. 2.2], and †Onirocuscus), in which a distinct fissure separates the anterior terminus of the cristid obliqua from other structures on the occlusal surface of m2–4 (state 2). The morphology of the cristid obliqua can be obscured by wear, so we only scored it from specimens with unworn or lightly worn teeth. We scored diprotodontids (with the exception of †Ngapakaladia) and Myrmecobius as inapplicable (“-”) due to the consistent absence of a cristid obliqua on m2–4 in these taxa (see char. 167). We also scored †Thylacoleo as inapplicable (due to its highly reduced m2), and we scored Notoryctes, †Yalkaparidon, and Tarsipes as inapplicable for reasons previously discussed (see char. 166). As defined, these states do not represent an obvious morphocline, so we left this character as unordered in all our phylogenetic analyses.

Character 171. Neomorphic cuspid(s) in hypoflexid region absent (0); or present (1). Where present, the hypoflexid of metatherians is a shallow labial indentation of the tooth crown adjacent to the cristid obliqua, between the protoconid and hypoconid (fig. 14). In most metatherians in which it is identifiable, this region is essentially featureless (state 0), although a small cingulid is sometimes present (fig. 23A-B). However, one or more distinct, neomorphic cuspids (ectostylids sensu Van Valen, 1966: fig. 1B) are present within the hypoflexid region (state 1) of the peramelemorphians Echymipera, Isoodon, and Macrotis (see Travouillon et al., 2010 [char. 23] and subsequent papers that used versions of their matrix for the distribution of this character in peramelemorphians), as well as the diprotodontians Phascolarctos, Hemibelideus, Pseudochirops cupreus, and Petropseudes. Hypoflexid cuspids are also variably present in Ailurops, Lestoros, and Lestodelphys (e.g., MMNH 17171), all of which we scored as polymorphic (“0+1”). We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

Character 172. Hypoconid large and distinct (0); or small and indistinct (1). The hypoconid is the prominent posterolabial cuspid of the lower molars in most metatherians that retain a talonid (figs. 14, 23A–C, 24; state 0). However, as described by Bensley (1903: 103; see also Waterhouse, 1841: 150; Tate, 1951: 4; Archer and Kirsch, 1977: 20), both of the labial cusps (protoconid and hypoconid) are reduced relative to the lingual cusps (paraconid, metaconid, and entoconid) on the lower molars of Myrmecobius, with the reduction of the hypoconid particularly striking (state 1). The hypoconid is also small and indistinct on m1 and m2 of †Wakaleo (Gillespie et al., 2014: fig. 5) and †Thylacoleo, and on all lower molars of †Sparassocynus (Beck and Taglioretti, 2020); we therefore also scored these three terminals as state 1. We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

Character 173. Lower third molar hypoconid labially salient, projecting beyond the protoconid or level with it (0); or m3 hypoconid lingual to salient protoconid (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 65). This character is equivalent to characters in other studies that considered the relative widths of the trigonid and talonid (e.g., Springer et al., 1997: char. 13; Rougier et al., 1998: char. 50; Horovitz and Sánchez-Villagra, 2003: char. 158). We corroborated the scoring of Voss and Jansa (2003, 2009), except that we observed the m3 hypoconid to be approximately level with the protoconid in Lutreolina (e.g., on AMNH 133255, 210421; see also Martinez-Lanfranco et al., 2014: fig. 7D), which we therefore scored as state 0. In effect, the only Recent didelphids in which the m3 hypoconid is lingual to the protoconid are Chacodelphys, Lestodelphys, and Monodelphis.

Among other metatherians, the m3 hypoconid is labially salient in †Herpetotherium (Korth, 2018: fig. 6b), paucituberculatans, microbiotheriids, some peramelemorphians (e.g., Echymipera, Perameles, Peroryctes, Rhynchomeles, and †Galadi [Travouillon et al., 2010: fig. 11]), †Litokoala (Black and Archer, 1997b: fig. 6A, C), Phascolarctos (Black et al., 2014a: figs. 2C, D, 11), †Namilamadeta (Pledge, 2005), Recent vombatids, several macropodiforms (including most macropodids), most phalangerids (with the exception of Wyulda and “Trichosurus” †dicksoni), pseudocheirids, petaurids, and Burramys (Archer, 1984c: fig. 147E).

By contrast, the m3 hypoconid is consistently lingual to the protoconid in †Pucadelphys (Marshall and Muizon, 1995: fig. 7E; Ladevèze et al., 2011: fig. 4), †Mayulestes (Muizon, 1998: figs. 3B, E, 4C), †Allqokirus (Muizon et al., 2018: figs. 1113), †Mimoperadectes (Bown and Rose, 1979: text-fig. 1A), †Sparassocynus (Abello et al., 2015: fig. 3A; Beck and Taglioretti, 2020: fig. 3b), †Hesperocynus (Forasiepi et al., 2009: fig. 7C), †Thylophorops (Goin et al., 2009b: fig. 4A), †Thylatheridium (Reig, 1952: fig. 11B), Macrotis, †Yarala, most examined dasyuromorphians, †Nimiokoala (Black and Archer, 1997b: figs. 2C, 3B), †Muramura (Pledge, 1987c: fig. 4D), diprotodontids, several macropodiforms (including Aepyprymnus, Caloprymnus, Potorous, Bettongia, and ?Bettongiamoyesi [Flannery and Archer, 1987a: figs. 1B, 3), Wyulda, “Trichosurus” †dicksoni (Flannery and Archer, 1987a: fig. 3F, K), acrobatids (Archer, 1984c: fig. 143), and Cercartetus (Archer, 1984c: figs. 154–156). We observed polymorphisms in several taxa—including Chaeropus, Isoodon, Microperoryctes, Micromurexia, Myrmecobius, Osphranter, and †Warendja—all of which we scored as “0+1.” We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166). The m3 of †Wakaleo is so reduced as to preclude meaningful scoring (Gillespie et al., 2014: figs. 5, 6A), so this taxon has also been scored as inapplicable (“-”).

Character 174. Entostylid labial to the entoconid absent (0); or present (1). In most nonlophodont metatherians that possess a distinct entoconid there is no structure immediately labial to this cuspid (state 0). However, a distinct cusplike structure that we refer to as an entostylid (after Archer, 1984c: figs. 178E–H, 185, 186D, 187; Roberts et al., 2008: fig. 2A) is present labial and slightly posterior to the entoconid (state 1) on the lower molars of Pseudochirops cupreus and P. archeri (see Archer, 1984c: fig. 178D–G, 186D, 187). A cusplike entostylid is also present in †Litokoala (Springer, 1987; Black and Archer, 1997b) and †Nimiokoala (Black and Archer, 1997b). An entostylid is clearly identifiable on the unworn m1 posterolabial to the entoconid in Phascolarctos (see Black et al. [2014a:], who referred to it as an “entostylid ridge”), but it is less obvious on more posterior molars; however, we also scored Phascolarctos as state 1. Following Tedford and Woodburne's (1987) homologies for the lower molar cuspids of †Ilaria, a neomorphic “central” cuspid is present between the entoconid and hypoconid; we tentatively identify this as an entostylid, and score †Ilaria as state 1.

We scored most incipiently or fully lophodont diprotodontians (i.e., those scored as state 1 or state 2 for char. 144) as inapplicable (“-”) for this character, because it is unclear whether an entostylid is genuinely absent or has been incorporated into the hypolophid of such taxa. However, this character can be meaningfully scored in a few incipiently lophodont diprotodontians—including Petaurus, Gymnobelideus, Distoechurus, Cercartetus, and Hypispymnodon moschatus—all of which have been scored as state 0. By contrast, a cusplike entostylid is clearly identifiable on the unworn lower molars of †Namilamadeta (e.g., QM F51342; Pledge, 2005: fig. 7B), although this structure has been partially incorporated into the hypolophid.

This character was scored as inapplicable in †Thylacoleo (due to the extremely simplified morphology of its m1, the only lower molar in this taxon) and in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166). As defined, these states do not comprise a plausible morphocline, and hence this character has been left unordered in all analyses presented here.

Character 175. Entoconid large and well developed on m1–3 (0); or very small or indistinct (1). Among metatherians that retain a fully developed talonid, the entoconid is usually a large lingual cuspid (state 0). However, this cusp is very small or indistinct (state 1) in a number of metatherians. Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 66, fig. 16; 2009), who observed that the entoconid is very small (subequal to or smaller than the hypoconulid) in Chacodelphys and Monodelphis.

Among our other terminals, we observed an entoconid that is subequal to or smaller than the hypoconulid in †Mayulestes (Muizon, 1998: figs. 3C, D, 4B), †Allqokirus (Muizon et al., 2018: figs. 1113), †Mimoperadectes (Bown and Rose, 1979: text-fig. 1A), †Sparassocynus (Beck and Taglioretti, 2020), †Hesperocynus (Forasiepi et al., 2009: fig. 7D), †Thylatheridium (Reig, 1952: fig. 10B), Thylacinus (Muizon and Lange-Badré, 1997: fig. 4C; Archer et al., 2016b: fig. 9F; Warburton et al., 2019: figs. 1013), and several dasyurids (Antechinomys, Dasycercus, Dasykaluta, Dasyuroides, Ningaui, Parantechinus, Planigale, and Pseudantechinus). In Sarcophilus, the entoconid appears to have been lost entirely (Archer et al., 2016b), as also seems to be the case in some sparassodonts (Forasiepi et al., 2014a). Archer (1981) observed that the degree of development of the entoconid in dasyurids appears to be correlated with habitat, with arid-adapted taxa typically exhibiting smaller entoconids than those living in areas of higher rainfall. Although Distoechurus, †Wakaleo (Gillespie et al., 2014: fig. 5), and †Thylacoleo lack identifiable hypoconulids for relative size comparisons, the entoconid is clearly very small in these taxa, which we also scored as state 1. We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

Character 176. Entocristid present (0); or indistinct or absent (1). In most metatherians, the entocristid is a well-developed crest that extends anteriorly or anterolabially (see char. 177) from the entoconid (state 0). However, the entocristid is indistinct or absent (state 1) in Myrmecobius, some peramelemorphians (Chaeropus, Macrotis, Perameles [fig. 26A], †Yarala [Muirhead and Filan, 1995: fig. 1.3], some specimens of Isoodon), Sminthopsis, and some diprotodontians (e.g., †diprotodontids, most Recent macropodids). In other diprotodontians—for example, Dactylopsila (in which an additional, presumably neomorphic cuspid [= entoconulid sensu Van Valen, 1966: fig. 1B] is present between the metaconid and entoconid on m2–4), †Namilamadeta (Pledge, 2005: fig. 1), †Nimbadon Hand, et al., 1993: 201), †Bulungamaya (Travouillon et al., 2014b: fig. 4.7), †Ganguroo (Travouillon et al., 2014b: figs. 2.5, 2.7, 4.8), and Dorcopsulus—a distinct entocristid is present on m1 but absent from m2–4; we scored these taxa as state 1 based on the condition observed for m2–4. Because the entocristid extends from the entoconid, we scored this character as inapplicable (“-”) for taxa in which the entoconid is unidentifiable, absent, or so indistinct as to preclude meaningful scoring (i.e., Tarsipes, †Thylacoleo, Thylacinus, Dasykaluta, Ningaui, Planigale, Sarcophilus, Notoryctes, and †Yalkaparidon; see char. 175).

FIG. 26.

Occlusal views of lower molars of Perameles gunnii (A, MVZ 127060[left m2]) and Isoodon macrourus (B, AMNH 160085[right m3, not fully erupted]). General features of lower molar crown morphology discussed in the text are illustrated, as are characters 176 and 177 (see main text for descriptions of these characters and character states), as follows: Perameles 176(1), 177(-); Isoodon 176(0), 177(1). Abbreviations: encd, entoconid; encrd, entocristid. Teeth are not shown to the same scale.

img-z168-1_01.jpg

Character 177. Entocristid extends along the lingual margin of the tooth (0); or entocristid well labial of the lingual margin (1); or entocristid with distinct anterolabial kink (2). When present, the entocristid is usually a short, straight crest that extends anteriorly from the entoconid along the lingual edge of the talonid (figs. 14, 23A, C). However, as discussed by Voss and Jansa (2009: 60; see also Abello, 2007: char. 38; Goin et al., 2007b: char. 19; Goin et al., 2009a: char. 19; Abello, 2013: char. 37), the entocristid of Caenolestes and Rhyncholestes is long and deflected labially, terminating within the talonid basin (state 0; fig. 23B). Among other paucituberculatans, the same (labially deflected) entocristid morphology occurs in Lestoros and †Pichipilus, whereas †Acdestis, †Palaeothentes, and †Stilotherium are polymorphic (“0+1”) for this character. The entocristid also extends somewhat labially, away from the lingual margin of the tooth, on unworn molars of Isoodon (e.g., AMNH 160085; fig. 26B) and in several diprotodontians—including phascolarctids (e.g., Phascolarctos; fig. 24A), pseudocheirids (fig. 24B; Archer, 1984c: figs. 178E–H, 185, 186D; Roberts et al., 2008: fig. 2, 2009: text-fig. 1A, B), Cercartetus, Dorcopsis (e.g., AMNH 109396), and Lagostrophus (e.g., WAM M2432)—and the same trait occurs on m1 (only) of Dorcopsulus (e.g., AMNH 158998), and †Nimbadon (e.g., QM F53648b, F30784; see also Hand et al., 1993: 201). Although the entocristid in these taxa is usually not as long nor as strongly deflected as it is in paucituberculatrans, we also scored them as state 1.

On unworn molars of all examined Recent and fossil phalangerids the entocristid is short and distinctly kinked, somewhat resembling the kinked cristid obliqua of these taxa (see char. 168): the entocristid extends anteriorly and slightly lingually from the apex of the entoconid before changing direction sharply and extending anterolabially for a short distance toward the center of the tooth (state 2). We scored this character as inapplicable (“-”) in all taxa entirely lacking an entocristid (see char. 176), and we scored it as inapplicable in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166). As defined, these states do not represent an obvious morphocline, so we left this character as unordered in all our phylogenetic analyses.

Character 178. Hypoconulid present (0); or absent or indistinct (1). A distinct hypoconulid is present (state 0) in most metatherians with an identifiable talonid, including †Pucadelphys, †Mayulestes, †Allqokirus, †Mimoperadectes, †Herpetotherium, didelphids (fig. 14, 23A), paucituberculatans (fig. 23B), microbiotheriids, dasyuromorphians, and peramelemorphians (fig. 23C). By contrast, the hypoconulid is absent (state 1) from most examined diprotodontians that could be scored for this character. However, we observed a distinct hypoconulid in pseudocheirids, in which this cusp is clearly identifiable immediately posterior to the entoconid on m1 and m2 (fig. 24B). Springer (1993: fig. 1) referred to this cusp in pseudocheirids as the “entostylid,” but we follow Archer (1984b [fig. 178F–I]), Bassarova et al. (2001 [fig. 1b]), and Roberts et al. (2007 [fig. 1], 2008 [fig. 2]), who identified it as the hypoconulid and referred to the cusp labial to the entoconid as the entostylid (see char. 174).

In juvenile specimens of Vombatus (e.g., AMNH 65623; Brewer et al., 2018: fig. 13.2) two cuspids visible on the unworn m3 might represent the hypoconulid. The first of these is immediately posterior to and basally confluent with the entoconid, whereas the second is smaller and located in the middle of a low cristid that connects the hypoconid with the entoconid and defines the posterior margin of the talonid. Although it is unclear which of these two cuspids is homologous with the hypoconulid, it seems probable that one of them is, so we scored Vombatus as state 0. By contrast, there is no trace of a hypoconulid on any of the unworn lower molars of Lasiorhinus (Brewer et al., 2018: fig. 13.2).

We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

Character 179. Hypoconulid at or near posterolingual margin of talonid, twinned with entoconid on m1–3 (0); or at midline of tooth, approximately equidistant to hypoconid and entoconid, not “twinned” with the latter cusp (1). Descriptions of character states and an account of their taxonomic distribution among Recent didelphids were provided by Voss and Jansa (2003: char. 67; 2009). In most didelphids (including all the fossil didelphids included here, namely, †Hesperocynus, †Sparassocynus, †Thylatheridium, and †Thylophorops), the hypoconulid is positioned near the posterolingual margin of the talonid and is said to be twinned with the entoconid (state 0), but in Caluromysiops the hypoconulid is about equidistant to the entoconid and hypoconid, near the midline of the tooth (state 1). When present, the hypoconulid is also twinned with the entoconid in most examined nondidelphid metatherians, with the exceptions of Thylacinus and †Acdestis. In Thylacinus, the hypoconulid is distinctly cusplike but approximately on the dental midline (Muizon and Lange-Badré, 1997: fig. 4D; Archer et al., 2016b: fig. 9F; Solé and Ladevèze, 2017: fig. 4; Warburton et al., 2019: figs. 1113). In †Acdestis, the hypoconulid forms a cingulidlike structure (Abello, 2007: char. 46, 2013: char. 45; Goin et al., 2007b: char. 16, 2009a: char. 16), which is approximately equidistant between the entoconid and hypoconid (Bown and Fleagle, 1993: fig. 13; Abello, 2007: lámina 11; 2013: fig. 3I),22 Both of these taxa have been scored as state 1 here.

As remarked by several authors (e.g., Muirhead, 1994, 2000; Muirhead and Filan, 1995; Woodburne and Case, 1996; Wroe, 1996a, 1997b, 2003; Beck et al., 2008b), the hypoconulid is in a very lingual position in all Recent and fossil peramelemorphians, immediately posterior (rather than posterolabial) to the entoconid. We did not score this morphology as a distinct state (contra, e.g., Wroe et al., 2000: char. 28; Wroe and Musser, 2001: char. 28; Travouillon et al., 2010: char. 20) because we observed a continuous range of variation in the position of the hypoconulid (from posterolingual to lingual) among the taxa included in this study, and because the position of the hypoconulid in peramelemorphians appears to be at least partially connected with the absence of a hypoconulid notch, which we have already coded as a separate character (see char. 158).

We scored this character as inapplicable (“-”) in all taxa lacking an unambiguously identifiable hypoconulid (see char. 178). Although the hypoconulid is clearly present in Sarcophilus, the absence of an unambiguously identifiable entoconid (see Archer et al., 2016b) likewise precludes meaningful scoring of this character for this taxon. We also scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

Character 180. Lower molars without a posterior cingulid (0); or distinct posterior cingulid present (1). Many tribosphenic metatherians have a posterior cingulid (or postcingulid), a narrow but distinct basal shelf of enamel that extends along the posterolabial margin of each lower molar from the hypoconid to the hypoconulid (state 1). This is the morphology seen in †Allqokirus, †Mayulestes, †Pucadelphys (Ladevèze et al., 2011: fig. 4a, b), †Herpetotherium (Fox, 1983: figs. 1b, 2b; Korth, 2018), †Mimoperadectes (at least on m2–3; see Bown and Rose, 1979: plate 1 fig. 2), thylacinids, and most dasyurids. By contrast, posterior cingulids are absent (state 0) in all the Recent and fossil didelphids included in this study23 as well as in paucituberculatans, peramelemorphians (Muirhead, 2000: 513), and microbiotheriids (state 0). Sarcophilus is polymorphic (“0+1”) for this character (Archer, 1976c). We observed some discrepancies between our scoring and that of Wroe et al. (2000: char. 36) and Wroe and Musser (2001: char. 36), both of whom scored the posterior cingulid as present but reduced in Metachirus and absent in Dasyurus and Sarcophilus. In fact, none of the Metachirus specimens we examined exhibits any trace of a posterior cingulid and, although the posterior cingulid is absent in some (but not all) Sarcophilus specimens we examined (e.g., AMNH 65674), it is consistently present in Dasyurus (although vestigial on m1).

A posterior cingulid is unequivocally absent in selenodont diprotodontians (pseudocheirids and phascolarctids; fig. 24), in which there is no structure posterior to the elongate, labially directed posthypocristid (or postcristid) that forms the posterior margin of the tooth. It is also clearly absent in some lophodont diprotodontians (such as Recent macropodids), because there is no structure posterior to the hypolophid (which forms the posterior margin of the tooth). In other lophodont and bunolophodont forms, a cingulidlike structure is present posterior to the hypolophid, but in most cases it is clear that this structure corresponds to the posthypocristid (or posthypocristid + postentocristid)—as in Burramys, †Ekaltadeta, †Namilamadeta (Pledge, 2005: fig. 1), and phalangerids. The sole exception among the taxa we examined is †Balbaroo, in which a distinct posterior cingulid is identifiable posterior to the posthypocristid (Cooke, 1997a; Black et al., 2014c). In diprotodontids, a cingulidlike structure is present posterior to the hypolophid (e.g., in Nimbadon [Hand, et al., 1993: fig. 3; Black and Hand, 2010: fig. 10M, N], Ngapakaldia [Black, 2010: fig. 3A, D], and Neohelos [Black, et al., 2014: fig. 6C1]), but it is unclear whether this is a true posterior cingulid or whether it is formed by the posthypocristid and postentocristid; we therefore scored diprotodontids as unknown (“?”). The lower molars of Myrmecobius are too derived to score this character confidently, so we also scored it as unknown. We scored this character as inapplicable (“-”) in Notoryctes, †Yalkaparidon, and Tarsipes for reasons previously discussed (see char. 166).

RESULTS

Molecular Analyses

Analyses of sequence data from the combined nuclear loci (fig. 27) and from whole mitochondrial genomes (fig. 28) result in broadly similar topologies, with only a few noteworthy examples of phylogenetic conflict. Both analyses place Paucituberculata as sister to the remaining marsupial orders, with strong support for the sister-group relationship between Didelphimorphia and Australidelphia. Within Australidelphia, both analyses also recover a sister relationship between Diprotodontia and Microbiotheria with moderate to strong support.

At the interordinal level, the greatest difference between our analyses of the nuclear and mitochondrial datasets concerns the position of Peramelemorphia: in the nuclear analysis (fig. 27) it is strongly supported as sister to Dasyuromorphia within the clade Agreodontia (which also includes Notoryctemorphia), whereas in the mitochondrial analysis (fig. 28) it is sister to a strongly supported clade comprising the other australidelphian orders (Notoryctemorphia, Dasyuromorphia, Microbiotheria, and Diprotodontia).

Monophyly of the only extant paucitubeculatan family, Caenolestidae, is strongly supported by analyses of both datasets. Monophyly of the didelphimorphian crown clade (Didelphidae) is also strongly supported by both the nuclear and mitochondrial analyses, and both analyses likewise strongly support Caluromyinae as sister to the remaining didelphids. However, the nuclear analysis places Hyladelphinae (= Hyladelphys) and Glironiinae (= Glironia) as successive sister taxa to Didelphinae with strong support, whereas the mitochondrial analysis weakly supports Hyladelphinae and Glironiinae as sister to Metachirini (= Metachirus) + Didelphini. Within Didelphinae, both analyses recover monophyly of the tribes Didelphini and Thylamyini, but support values for relationships within Didelphinae are much lower in the mitochondrial analysis than in the nuclear analysis.

Monophyly of Peramelemorphia is strongly supported by both the nuclear and the mitochondrial analyses. Within Peramelemorphia, the mitochondrial analysis places Chaeropodidae (Chaeropus) as the sister to the remaining terminals, but nuclear sequences are unavailable for Chaeropus.24 Both analyses agree in placing Thylacomyidae (Macrotis) as the next taxon to diverge. The nuclear analysis supports the monophyly of Peramelinae (Isoodon and Perameles), Echymiperinae (Echymipera and Microperoryctes), and Echymiperinae + Peroryctinae (the latter represented by the sole modern genus Peroryctes), but the mitochondrial analysis does not.

Monophyly of Dasyuromorphia is strongly supported by both the nuclear and mitochondrial analyses. Within Dasyuromorphia, only mitochondrial sequences are available for Thylacinus, and the mitochondrial analysis weakly supports Myrmecobiidae + Thylacinidae to the exclusion of Dasyuridae. Within Dasyuridae, nuclear and mitochondrial analyses both strongly support the reciprocal monophyly of Sminthopsinae (Antechinomys, Ningaui, Planigale, and Sminthopsis) and Dasyurinae (containing the remaining modern genera). Analyses of both datasets also strongly support monophyly of Sminthopsini (Antechinomys, Ningaui, and Sminthopsis), Phascogalini (Antechinus, Micromurexia, Murexechinus, Murexia, Paramurexia, Phascogale, and Phascomurexia), and Dasyurini (comprising the remaining dasyurine genera).

FIG. 27.

Fifty-percent majority rule consensus of post-burn-in trees that results from Bayesian analysis of our nuclear sequence-only dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”).

img-z172-1_01.jpg

Monophyly of Diprotodontia is strongly supported by analyses of the nuclear and mitochondrial datasets, both of which also support a basal divergence between Vombatiformes (Lasiorhinus, Phascolarctos, and Vombatus) and Phalangerida (the remaining diprotodontian genera). The nuclear and mitochondrial datasets both strongly support monophyly of Vombatiformes and Vombatidae, but monophyly of Phalangerida receives strong support only from the nuclear dataset. Analyses of both datasets strongly support the existence of three major clades within Phalangerida: Phalangeridae + Burramyidae, Macropodoidea (Hypsiprymnodon, Potoroidae, and Macropodidae), and Petauroidea (Acrobatidae, Tarsipes, Petauridae, and Pseudocheiridae). Nuclear and mitochondrial analyses also place Macropodoidea and Petauroidea as sister taxa, to the exclusion of Phalangeridae + Burramyidae, but support for this arrangement is weak.

FIG. 28.

Fifty-percent majority rule consensus of post-burn-in trees that results from undated Bayesian analysis of our mitochondrial sequence-only dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”).

img-z173-1_01.jpg

Within Phalangeridae, analyses of both datasets strongly support a basal divergence between Trichosurus +Wyulda (= Trichosurinae) and the remaining genera, with strong support for monophyly of Trichosurinae, Phalanger + Spilocuscus (= Phalangerinae) and Ailurops + Strigocuscus (= Ailuropinae). Within Petauroidea, both analyses strongly support Acrobatidae as sister to the remaining families, as well as a Petauridae + Pseudocheiridae clade. Within Petauridae, the nuclear analysis strongly supports Gymnobelideus as sister to Dactylopsilinae (Dactylopsila and Dactylonax), whereas the mitochondrial analysis strongly supports a Gymnobelideus + Dactylopsila clade to the exclusion of Dactylonax. Within Pseudocheiridae, the nuclear and mitochondrial analyses strongly support the same overall topology, with three subclades—Pseudocheirinae (Pseudocheirus + Pseudochirulus), Hemibelidinae (Hemibelideus + Petauroides; represented by Petauroides only in the mitochondrial analysis) and Pseudochiropsinae (Pseudochirops + Petropseudes)—of which Pseudocheirinae and Hemibelidinae are recovered as sister taxa.

Monophyly of Macropodiformes is strongly supported by both the nuclear and the mitochondrial analyses, which also agree that Hypsiprymnodontidae (Hypsiprymnodon) is sister to Macropodidae + Potoroidae. Reciprocal monophyly of Macropodidae and Potoroidae is strongly supported by both datasets, which likewise agree that Lagostrophus (the only extant lagostrophine) is sister to the remaining modern macropodid genera (= Macropodinae).

Unsurprisingly, analyses of the combined-molecular (nuclear + mitochondrial) dataset recover nearly all of those relationships common to the separate analyses described above (fig. 29). For example, at the interordinal level, Paucituberculata is again placed as sister to the other orders, and there is again strong support for Didelphimorphia + Australidelphia, and for Microbiotheria + Diprotodontia. Where our nuclear and mitochondrial results differ, the combined-molecular topology is usually more similar to the nuclear results; for example, in providing strong support for monophyly of Agreodontia (with Notoryctemorphia sister to Dasyuromorphia + Peramelemorphia) and in recovering similar relationships within Didelphidae (with Caluromyinae sister to Glironiinae + Didelphinae).

Morphological Analyses

Maximum parsimony analysis (fig. 30) and the undated Bayesian analysis (fig. 31) of our craniodental dataset both place †Mimoperadectes outside Marsupialia. However, both analyses recover a polytomy at the base of Marsupialia that includes †Herpetotherium, and in which Didelphimorphia is not recovered as monophyletic. Dasyuromorphia, Diprotodontia, Microbiotheria, and Peramelemorphia are monophyletic in both analyses, but support for these clades is generally higher in the Bayesian analysis. In both analyses, the caenolestids Caenolestes, Rhyncholestes, and Lestoros form a clade with the fossil paucituberculatans †Acdestis, †Palaeothentes, †Pichipilus, and †Stilotherium with weak to moderate support, but †Evolestes is not recovered as a member of this clade.

Few relationships above the ordinal level are recovered in either the maximum parsimony or Bayesian analyses of these phenotypic data. However, both analyses weakly support the notoryctemorphian Notoryctes as sister to Microbiotheria. The Bayesian analysis weakly supports a clade that includes all our terminals known to have gliriform lower incisors (diprotodontians, paucituberculatans, †Yalkaparidon) plus †Evolestes (for which lower incisor morphology is currently unknown), and there is similarly weak support for a sister-taxon relationship between Paucituberculata (excluding †Evolestes) and Diprotodontia. The maximum parsimony analysis recovers a weakly supported superordinal clade that comprises †Yalkaparidon, †Evolestes, Dasyuromorphia, Peramelemorphia, Paucituberculata (excluding †Evolestes), and Diprotodontia, within which Paucituberculata (excluding †Evolestes) and Diprotodontia are sister taxa, but other interordinal relationships are unresolved.

Neither morphology-only analysis recovers monophyly of Didelphimorphia, but among the didelphimorphian clades found in both analyses are Caluromyinae, Thylamyini, and Didelphini (including †Thylophorops), although these clades are only weakly supported in the maximum parsimony analysis. The Bayesian analysis also places †Hesperocynus, †Sparassocynus, and †Thylatheridium in a clade with Monodelphis with moderate support, but this clade is not found in the maximum parsimony analysis.

Within Peramelemorphia, †Galadi, †Bulungu, and †Yarala are recovered outside the crown clade, but monophyly of the clade excluding these fossil taxa receives strong support only in the Bayesian analysis. Both analyses recover a clade comprising Chaeropus, Perameles, Isoodon, and Macrotis, with the last two taxa also forming a clade.

FIG. 29.

Fifty-percent majority rule consensus of post-burn-in trees that results from undated Bayesian analysis of our combined nuclear and mitochondrial sequence dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”).

img-z175-1_01.jpg

Within Dasyuromorphia, the Bayesian analysis strongly supports a monophyletic Dasyuridae, sister to a weakly supported clade comprising Myrmecobius, Thylacinidae, and the fossil dasyuromorphians †Barinya and †Mupuracinus. Thylacinidae (here comprising Thylacinus, †Badjcinus, and †Nimbacinus) is also strongly supported as monophyletic. The maximum parsimony analysis recovers a broadly similar topology (albeit with weak support), except that Dasyuridae is not monophyletic.

Within Paucituberculata (excluding †Evolestes), both analyses support a clade comprising †Pichipilus, Lestoros, Caenolestes, and Rhyncholestes.

FIG. 30.

Strict consensus of 99,999 most parsimonious trees (each of 949 steps, consistency index =0.269, retention index = 0.815) that result from maximum parsimony analysis of our morphological dataset. Black dots at nodes indicate ≥70% bootstrap support (“strong support”); dark gray dots indicate 50%–69% bootstrap support (“moderate support”); light gray dots indicate <50 bootstrap support (“weak support”).

img-z176-1_01.jpg

Within Diprotodontia, the maximum parsimony analysis recovers a basal split between Vombatiformes (†Thylacoleonidae, Phascolarctidae, Vombatidae, †Diprotodontidae, and †Wynyardiidae) and Phalangerida, although both of these clades receive weak support. The Bayesian analysis, meanwhile, places †Thylacoleonidae as sister to the remaining diprotodontians and fails to recover a monophyletic Phalangerida. Both analyses place Phascolarctidae as sister to a clade (= Vombatomorphia sensu Beck et al., 2020: table 1) that includes vombatids, diprotodontids, wynyardiids, and the ilariid †Ilaria. Both analyses strongly support monophyly of Dactylopsilinae, Acrobatidae, Pseudocheiridae, and Phalangeridae, and they weakly support monophyly of Macropodiformes, but they do not recover a monophyletic Petauridae, Petauroidea, Burramyidae, or Burramyidae + Phalangeridae.

Undated Total-Evidence Analysis

Undated Bayesian analysis of our total-evidence dataset (fig. 32) strongly supports a clade comprising our definitive marsupial terminals plus †Herpetotherium, to the exclusion of †Mimoperadectes and our outgroup taxa (†Pucadelphys, †Mayulestes, and †Allqokirus). However, similar to our morphology-only analyses (figs. 30, 31), Didelphimorphia is not recovered as monophyletic. Our remaining marsupial terminals form a strongly supported clade, within which there is a weakly supported monophyletic grouping comprising a trichotomy of †Evolestes, the remaining paucituberculatans, and †Yalkaparidon. There is weak support for a clade comprising the living paucituberculatans (the caenolestids Caenolestes, Rhyncholestes, and Lestoros) and †Pichipilus.

Our remaining terminals comprise a moderately supported clade that corresponds to Australidelphia. Within Australidelphia, the first divergence is between a strongly supported Microbiotheria + Diprotodontia clade and a moderately supported clade (= Agreodontia) comprising the remaining three orders, namely Dasyuromorphia, Notoryctemorphia, and Peramelemorphia. Within Diprotodontia, there is moderate support for monophyly of Vombatiformes, with †Thylacoleonidae the first family to diverge, followed by Phascolarctidae, and then a weakly supported clade comprising †Diprotodontidae, †Ilariidae (†Ilaria), Vombatidae, and †Wynyardiidae; within the last, †Wynyardiidae and †Ilariidae are weakly supported as sister taxa, with †Diprotodontidae sister to this clade. Monophyly of Phalangerida is also strongly supported. Relationships among our modern phalangeridan terminals are essentially identical to those recovered by our combined nuclear and mitochondrial sequence analyses, but placement of our fossil terminals in Phalangerida merits comment: within Phalangeridae, “Trichosurus” †dicksoni and †Onirocuscus are sister to Wyulda and Strigocuscus, respectively; within Hypsiprymnodontidae, Hypisprymnodonbartholomaii is sister to H. moschatus, and †Ekaltadeta is sister to this clade; within Potoroidae, ?Bettongiamoyesi is sister to Bettongia + Aepyprymnus; within Macropodidae, †Balbaroo, †Ganawamaya, †Gangaroo, †Hadronomas, and †Rhizosthenurus form a clade with Dorcopsulus, and within that clade, †Balbaroo and †Ganawamaya are sister taxa (=†Balbaridae).

Within Agreodontia, Dasyuromorphia and Peramelemorphia are strongly supported as sister taxa to the exclusion of Notoryctemorphia. Similar to the situation within Phalangerida, the relationships of extant dasyuromorphians and peramelemorphians are essentially identical to the results of our combined nuclear and mitochondrial sequence analyses. Monophyly of Dasyuromorphia is strongly supported and all our fossil taxa are placed within the crown clade (= Dasyuroidea). Within Dasyuroidea, Myrmecobiidae (Myrmecobius) and Dasyuridae are strongly supported as sister taxa to the exclusion of Thylacinidae. †Barinya, †Badjcinus, †Mutpuracinus, and †Nimbacinus are all placed within Thylacinidae, with †Badjcinus sister to Thylacinus and †Nimbacinus sister to †Badjcinus + Thylacinus. Monophyly of Peramelemorphia is also strongly supported and there is weak support for a clade comprising all our Recent peramelemorphian terminals (= Perameloidea), to the exclusion of †Galadi (which is sister to Perameloidea), and †Yarala + †Bulungu, which form a clade that is sister to all other peramelemorphians. Within Perameloidea, Chaeropodidae (Chaeropus) is the first family to diverge, followed by Thylacomyidae (Macrotis); although Peramelinae and Echymiperinae + Peroryctinae are recovered as sister taxa, this clade is only moderately supported.

FIG. 31.

Fifty-percent majority rule consensus of post-burn-in trees that results from undated Bayesian analysis of our morphological dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”).

img-z178-1_01.jpg

Dated Total-Evidence Analysis

Incorporating temporal information in the form of calibrating the ages of our terminals and selected nodes and then carrying out a Bayesian total-evidence dating analysis (fig. 33) has little impact on the relationships of our extant terminals compared to the equivalent undated analysis (fig. 32). There is, however, a major difference in the position of the root of Marsupialia, but we emphasize that this is due largely to the topological and temporal constraints we enforced in our dated analysis. In the undated analysis (fig. 32), the root of Marsupialia was positioned on a polytomy comprising †Herpetotherium, a nonmonophyletic Didelphimorphia, and a clade comprising our remaining ingroup terminals. By contrast, in the dated analysis (fig. 33), we constrained †Herpetotherium to fall outside crown-clade Marsupialia, congruent with most recent published phylogenetic analyses (e.g., Sánchez-Villagra et al., 2007; Horovitz et al., 2008; Horovitz et al., 2009; Ladevèze et al., 2020). We also constrained Didelphimorphia to be the sister to the remaining marsupial orders, given the statistically significant support for this topology provided by retrotransposon insertions (Gallus et al., 2015a), whereas our molecular analyses consistently place the marsupial root between Paucituberculata and Didelphimorphia + Australidelphia (figs. 2729).

FIG. 32.

Fifty-percent majority rule consensus of post-burn-in trees that results from undated Bayesian analysis of our total evidence dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”).

img-z179-1_01.jpg

Many of our fossil taxa take up very different positions in the dated total-evidence analysis (fig. 33) compared to the equivalent undated analysis (fig. 32), in most cases occupying deeper positions in the tree. This might be expected given that they predate (often by tens of millions of years) our Recent terminals (see, e.g., Lee and Yates, 2018; Beck and Taglioretti, 2020; King and Beck, 2020). Within Didelphimorphia, †Thylophorops moves from forming a weakly supported clade with Didelphis and Philander to forming a trichotomy with Chironectes and a clade comprising Didelphis, Philander, and Lutreolina. Within Paucituberculata, †Pichipilus moves from a weakly supported position as sister to the extant Lestoros, to a position outside a strongly supported clade comprising all three extant caenolestid genera. Within Dasyuromorphia, †Barinya, †Badjcinus, and †Mutpuracinus are no longer placed within Thylacinidae. Instead, †Badjcinus is placed in a trichotomy at the base of Dasyuromorphia, together with a moderately supported clade comprising Thylacinus and †Nimbacinus (= Thylacinidae) and a moderately supported clade comprising remaining dasyuromorphian terminals, whereas †Barinya and †Mutpuracinus form a weakly supported clade with Myrmecobius, which is in turn sister to a strongly supported Dasyuridae.

FIG. 33.

(opposite page and at right) Fifty-percent majority rule consensus of post-burn-in trees that results from dated Bayesian analysis (using combined tip-and-node dating and separate Independent Gamma Rates [IGR] clock models for the molecular and morphological partitions) of our total evidence dataset. Black dots at nodes indicate ≥0.95 Bayesian posterior probability (“strong support”); dark gray dots indicate 0.75–0.94 Bayesian posterior probability (“moderate support”); light gray dots indicate 0.50–0.74 Bayesian posterior probability (“weak support”). Nodes without dots were constrained a priori so that their ages could be calibrated (see tables 6, 13). Orange bars represent 95% Highest Posterior Density (HPD) intervals on the ages of nodes. For clarity, 95% HPD intervals are not shown for the ages of fossil terminals.

img-z180-1_01.jpg

Continued

img-A5ma_01.jpg

The biggest topological changes are seen within Diprotodontia. Perhaps most notably, thylacoleonids are not unambiguously supported as members of Vombatiformes in our dated total-evidence analysis, Instead, the first divergence within Diprotodontia in this analysis is a trichotomy comprising Phalangerida, †Thylacoleonidae, and a clade comprising the remaining terminals currently classified as members of Vombatiformes (diprotodontids, the ilariid †Ilaria, phascolarctids, wynyardiids, and vombatids). Within this last clade, Phascolarctidae is the sister to the remaining terminals and there is strong support for monophyly of Diprotodontidae and Vombatidae. The wynyardiids †Namilamadeta and †Muramura no longer form a clade; instead †Namilamadeta is sister to Vombatidae, with †Muramura sister to this clade, although support values for these relationships are weak.

Within Phalangeridae, †Onirocuscus and “Trichosurus” †dicksoni are now placed outside the phalangerid crown clade, the monophyly of which receives strong support. At the base of Macropodiformes, there is now a polytomy comprising †Ekaltadeta (which no longer forms a clade with Hypsiprymnodon moschatus and H.bartholomaii), Hypsiprymnodontinae (= Hypsiprymnodon moschatus + H.bartholomaii), †Balbaridae (= †Balbaroo + †Ganawamaya), and Macropodidae + Potoroidae. Within Potoroidae, ?Bettongiamoyesi moves outside the crown-clade, monophyly of which is weakly supported. Within Macropodidae, †Ganguroo, †Hadronomas, and †Rhizosthenurus are likewise placed outside the crown clade, with †Hadronomas and †Rhizosthenurus forming a weakly supported clade (equivalent to the subfamily †Sthenurinae; Kear, 2002; Kirkham, 2004; Prideaux, 2004). However, it should be noted that some of these macropodiform relationships are the result of topological constraints that were required to calibrate selected nodes within Macropodoidea (see Methods and appendix 2).

Our dated total-evidence analysis places the first divergence within Marsupialia (between Didelphimorphia and the remaining marsupial taxa, a relationship that was constrained and calibrated a priori; see above, table 6, and appendix 2) in the late Paleocene or earliest Eocene (median estimate: 56.2 Mya; 95% HPD: 54.7–58.6 Mya). The divergence between Paucituberculata and Australidelphia (another constrained and calibrated node) is estimated as close to the Paleocene-Eocene boundary (median estimate: 55.1 Mya; 95% HPD: 54.6–56.6 Mya), whereas the first split within Australidelphia (between Agreodontia and Microbiotheria + Diprotodontia) is ∼7 Myr later (median estimate: 48.0 Mya; 95% HPD: 44.3–50.9 Mya). The estimated divergence times between Notoryctemorphia and Dasyuromorphia + Peramelemorphia (median estimate: 45.7 Mya; 95% HPD: 42.3–49.2 Mya) and between Dasyuromorphia and Peramelemorphia (median estimate: 43.7 Mya; 95% HPD: 39.6–47.4 Mya) fall entirely within the Eocene, as does the divergence between Microbiotheria and Diprotodontia (median estimate: 45.6 Mya; 95% HPD: 41.4–48.8 Mya). The divergence between †Yalkaparidon and Paucituberculata (with monophyly of the latter clade enforced a priori), meanwhile, is estimated as Eocene or earliest Oligocene (median estimate: 39.9 Mya; 95% HPD: 30.7–49.7 Mya). Estimated divergence times of selected intraordinal clades are summarized in table 6 and discussed in the individual taxonomic accounts below.

DISCUSSION

Molecular Analyses

Our results document the presence of strong phylogenetic signal in the molecular data analyzed here. Phylogenetic congruence between the results of analyzing the nuclear (fig. 27) and mitochondrial (fig. 28) datasets separately seems particularly impressive despite discrepancies at a few problematic nodes. The results of our separate (figs. 27, 28) and combined analyses (fig. 29) of these molecular data are also broadly consistent with those reported by other recent, large-scale sequence-based analyses of marsupial phylogeny (Meredith et al., 2009a, 2011; Mitchell et al., 2014; May-Collado et al., 2015; Duchêne et al., 2018; Álvarez-Carretero et al., 2021), but there remain several noteworthy areas of conflict and uncertainty together with some incongruence with the results of retrotransposon-based analyses (Nilsson et al., 2010; Gallus et al., 2015a).

One conspicuous point of uncertainty concerns the position of the marsupial root node. Whereas all our molecular analyses and several previous studies (e.g., Meredith et al., 2009a, 2011; Mitchell et al., 2014; May-Collado et al., 2015; Álvarez-Carretero et al., 2021) place the root between Paucituberculata and the remaining Recent marsupial orders, sequence-based analyses reported by other researchers (e.g., Nilsson et al., 2003, 2004; Beck, 2008a; Meredith et al., 2008b, 2009c; reviewed by Eldridge et al., 2019: 806) have placed the marsupial root node on the branch leading to Didelphimorphia. The latter position also receives statistically significant support from retrotransposon insertions (five uncontradicted insertions, p = 0.001; Gallus et al., 2015a), so this was the root position we enforced to calibrate this node in our dated total-evidence analysis despite the conflicting signal in our own results.

Within Didelphimorphia, our mitochondrial analysis shows some conflict with other recent molecular studies (Jansa and Voss, 2000; Voss and Jansa, 2003, 2009; Jansa and Voss, 2005; Jansa et al., 2006, 2014; Gruber et al., 2007; May-Collado et al., 2015; Vilela et al., 2015; Álvarez-Carretero et al., 2021) by, for example, placing Hyladelphinae (Hyladelphys) and Glironiinae (Glironia) in a trichotomy with Didelphini + Metachirini, albeit with relatively weak support. However, our nuclear and combined-molecular analyses are in strong agreement with the current consensus view of subfamilial and tribal relationships within Didelphidae (Voss and Jansa, 2009; Amador and Giannini, 2016; Eldridge et al., 2019), as we discuss in more detail in the annotated classification that follows.

A major difference between the results of our mitochondrial and nuclear analyses concerns the position of Peramelemorphia. Whereas our mitochondrial analysis strongly supports Peramelemorphia as sister to a clade comprising the remaining four australidelphian orders (Dasyuromorphia, Diprotodontia, Microbiotheria, and Notoryctemorphia), our nuclear and combined-molecular analyses strongly support Peramelemorphia as sister to Dasyuromorphia, with which (together with Notoryctemorphia) it forms the clade Agreodontia. Several early studies based on mitochondrial genes (e.g., Krajewski et al., 1997; Burk et al., 1999; Palma and Spotorno, 1999) as well as some based on combined nuclear and mitochondrial data (e.g., Springer et al., 1998; Asher et al., 2004) have found Peramelemorphia in a relatively basal position within Marsupialia, often quite distant from the other Australian orders (but see Nilsson et al., 2004).25 However, Phillips et al. (2006; see also Beck, 2008a) showed convincingly that this position for Peramelemorphia is due to base-compositional bias in mitochondrial protein-coding and ribosomal genes; when this bias is corrected (e.g., by RY-coding or deleting a particularly biased partition, such as the third codon positions of protein-coding genes), Peramelemorphia forms a clade (= Agreodontia) with Dasyuromorphia and (when included) Notoryctemorphia, in agreement with the results of nuclear-sequence data analyses (Phillips et al., 2006; Beck, 2008a). However, only one retrotransposon insertion has been found that supports monophyly of Agreodontia (Nilsson et al., 2010; Gallus et al., 2015a).

Within Agreodontia, our nuclear and combined-molecular analyses provide moderate to strong support for Dasyuromorphia + Peramelemorphia to the exclusion of Notoryctemorphia. By contrast, the phylogenomic study of Duchêne et al. (2018), which used 867 kb of DNA sequence data from 1550 nuclear loci, found strong support for Peramelemorphia + Notoryctemorphia. Nevertheless, the short branches separating these divergences and the lack of retrotransposon insertion data supporting any specific branching pattern between Dasyuromorphia, Notoryctemorphia, and Peramelemorphia (Nilsson et al., 2010; Gallus et al., 2015a) suggest a rapid radiation at the base of Agreodontia that may approximate a “hard” polytomy (see also Duchêne et al., 2018: 408).

Within Dasyuromorphia, Thylacinus (for which only mitochondrial sequence data were used) is sister to Myrmecobius in the mitochondrial analysis, but sister to Mymecobius + Dasyuridae in the combined-molecular analyses; the latter topology is congruent with other recent molecular and total-evidence analyses (Miller et al., 2009; May-Collado et al., 2015; Westerman et al., 2016; Kealy and Beck, 2017). Feigin et al. (2018) found 11 retrotransposon insertions supporting a sister relationship between Myrmecobius and Dasyuridae, to the exclusion of Thylacinus, which represents statistically significant support for this clade, but the presence of a few conflicting insertion patterns suggests incomplete lineage sorting at the base of Dasyuromorphia. The relationships we recover within Dasyuridae are largely congruent between our mitochondrial and nuclear analyses and in agreement with recent molecular and total-evidence analyses (Miller et al., 2009; May-Collado et al., 2015; Westerman et al., 2016; Kealy and Beck, 2017; García-Navas et al., 2020; Álvarez-Carretero et al., 2021), all of which strongly support monophyly of the subfamilies Dasyurinae and Sminthopsinae and of the tribes Dasyurini, Phascogalini, and Sminthopsini.

Our molecular results for relationships within Peramelemorphia are congruent with those reported from several recently published molecular analyses (Westerman et al., 2012; Kear et al., 2016; Travouillon and Phillips, 2018) in supporting Chaeropus (represented by mitochondrial sequence data only) as sister to other Recent taxa, and in supporting Macrotis as sister to Peramelidae. However, we note that other molecular and total-evidence studies have found alternative resolutions for the position of Chaeropus, including as sister to Macrotis (Travouillon and Phillips, 2018: figs. 1E, 2) and within a paraphyletic Peramelidae (May-Collado et al., 2015; Travouillon et al., 2019). The results of our molecular analyses agree with those of most recent molecular and total-evidence studies (Meredith et al., 2008a; Westerman et al., 2012; Mitchell et al., 2014; May-Collado et al., 2015; Kear et al., 2016; Travouillon and Phillips, 2018; Álvarez-Carretero et al., 2021) in supporting the monophyly of Peramelinae (Isoodon + Perameles) and Echymiperinae (Echymipera + Rhynchomeles + Microperoryctes). Most of these studies also support a Peroryctinae + Echymiperinae clade (Meredith et al., 2008a; Mitchell et al., 2014; May-Collado et al., 2015; Kear et al., 2016; Travouillon and Phillips, 2018; Álvarez-Carretero et al., 2021), and some authors have recognized this clade as the family Peroryctidae, restricting Peramelidae to Isoodon and Perameles only (e.g., Travouillon and Phillips, 2018: see also Groves and Flannery, 1990; Eldridge et al., 2019). However, only our nuclear and and combined-molecular analyses support Peroryctinae + Echymiperinae. Given the inconsistent results found here and in other published molecular and total-evidence analyses (see especially Travouillon et al., 2019), it is clear that further research with even larger sequence datasets will be required to robustly resolve relationships among extant peramelemorpians, with retrotransposon insertion data also likely to prove useful.

A conspicuous result of our nuclear and combined-molecular analyses is consistently strong support for a sister-group relationship between Microbiotheria and Diprotodontia. This biogeographically anomalous clade has also been found in DNA-DNA hybridization studies (Kirsch et al., 1991, 1997) and the combined nuclear and mitochondrial sequence analysis of May-Collado et al. (2015). However, most other phylogenetic analyses that have included nuclear sequence data (Phillips et al., 2006; Beck, 2008a; Meredith et al., 2008b, 2009a, 2009c, 2011; Mitchell et al., 2014; Duchêne et al., 2018; Álvarez-Carretero et al., 2021) have found Microbiotheria to fall outside a clade comprising the four Australian orders (= Eomarsupialia sensu Beck et al., 2014), and mitochondrial data appear to favor the latter topology when corrected for compositional heterogeneity (Phillips et al., 2006). Some of these studies found only weak support for monophyly of Eomarsupialia over alternative topologies, but the phylogenomic study of Duchêne et al. (2018) strongly supported this clade, and four uncontradicted retrotransposon insertions also provide statistically significant support (p = 0.0123) for Eomarsupialia (Nilsson et al., 2010; Gallus et al., 2015a). Although this collective evidence persuades us that Eomarsupialia is probably monophyletic, we did not enforce monophyly of this clade in our dated total-evidence analysis to avoid a priori constraints on taxa with uncertain relationships—in particular †Yalkaparidon, which fell within Eomarsupialia in the total-evidence analysis of Beck et al. (2016), but not in other published analyses (Beck et al., 2014; Abello and Candela, 2019; Zimicz and Goin, 2020).

Within Diprotodontia, our nuclear, mitochondrial, and combined-molecular analyses are congruent with recent molecular studies in supporting a basal split between Vombatiformes and Phalangerida, and in strongly supporting monophyly of Petauroidea, Macropodoidea, Phalangeridae + Burramyidae, and all the families, subfamilies, and tribes recognized in table 13 (Beck, 2008a; Meredith et al., 2008b, 2009a, 2009c; Phillips and Pratt, 2008; Mitchell et al., 2014; May-Collado et al., 2015; Duchêne et al., 2018; Álvarez-Carretero et al., 2021). Within Phalangerida, our molecular analyses consistently recover Macropodoidea + Petauroidea and Phalangeridae + Burramyidae as reciprocally monophyletic clades, each with strong support. This topology agrees with some other recent molecular analyses (e.g., Meredith et al., 2011; Mitchell et al., 2014; May-Collado et al., 2015; Álvarez-Carretero et al., 2021). However, others (e.g., Beck, 2008a; Meredith et al., 2008b, 2009a, 2009c; Phillips and Pratt, 2008) have instead supported a Macropodoidea + (Phalangeridae + Burramyidae) clade, with Meredith et al. (2009a) going so far as to formally recognizing this clade with the name Australoplagiaulacoidea. The phylogenomic study of Duchêne et al. (2018) found that some individual gene trees for their 1550 nuclear loci supported Macropodoidea + Petauroidea whereas others supported Macropodoidea + (Phalangeridae + Burramyidae), suggesting that divergences among these three clades occurred close together in time, or involved large population sizes, or both. However, Duchêne et al. (2018) found that the clear majority (approximately two-thirds) of the gene trees for their individual loci supported Macropodoidea + Petauroidea, which was the topology that was consistently found here.

Morphological Analyses

In contrast to the strong and generally consistent phylogenetic signal observed in the preceeding molecular analyses, the results of our morphological analyses are characterized by much less resolution and generally lower support values (figs. 30, 31). Of the 44 clades at the tribe level or above that receive strong support in the undated Bayesian analyses of our combined-molecular data (table 13), only 16 (36.4%) are recovered in the MP analysis of our craniodental data and, of those, only five (Caluromyinae, Phalangeridae, Pseudocheiridae, Dactylopsilinae, and Acrobatidae) are strongly supported; the remainder are either unresolved (14/44 = 31.8%) or actively contradicted (14/44 = 31.8%). Undated Bayesian analysis of our craniodental data performs only marginally better: of the same 44 clades strongly supported by our combined-molecular analyses, 16 (36.4%) are recovered (the same proportion as in the MP analysis), of which 10 are strongly supported, whereas 16 (36.4%) are unresolved, and 12 (27.3%) are actively contradicted.

TABLE 13

Comparisons of Clade Support Values among Different Analyses

Dataset abbreviations: CM, combined molecular; Mo, morphological; Mt, mitochondrial; Nu, nuclear; TE, total evidence. Analytic abbreviations: DB; dated Bayesian; MP, maximum parsimony; UB, undated Bayesian. Bootstrap support values are tabulated for the Mo/MP analysis and Bayesian posterior probabilities for all others; bootstrap values and posterior probabilities interpreted as strong support are boldfaced. Other table entries: n/a, not applicable; no, actively contradicted; calibrated, constrained to be monophyletic; unresolved, clade not recovered but not contradicted.

img-z186-9_01.gif

continued

img-z187-2_01.gif

Although the lack of strong morphological support for many clades that are strongly supported by our analyses of molecular data seem disappointing, this result is not unexpected for several reasons. First, our morphological analyses are based on many fewer characters but include more taxa, so even in the absence of character conflict there are fewer state changes to apportion among more nodes. Second, including extinct taxa that are sister to crown groups with strong molecular support has the result of breaking up the long branches along which crown-group synapomorphies accumulated (Horovitz, 1999; Wilkinson, 2003; Cobbett et al., 2007; Bapst, 2013). And third, fossil taxa are almost always incompletely preserved, introducing abundant missing data to the morphological matrix.

There is considerable debate in the literature as to whether Bayesian analyses of discrete morphological data using Lewis's (2001) Mk model and its variants are likely to result in more accurate phylogenies than are parsimony-based approaches (Wright and Hillis, 2014; O'Reilly et al., 2016, 2018; Brown et al., 2017; Goloboff et al., 2018a, 2018b, 2019; Puttick et al., 2017a, 2017b; Sansom et al., 2018; Schrago et al., 2018; Goloboff and Arias, 2019; Smith, 2019). Here, we found no difference in the number of clades strongly supported by molecular data that are recovered by the two methods used to analyze our morphological data (table 13). However, we note that some molecular clades are recovered by one morphological analysis only and not by the other; for example, our undated Bayesian analysis of morphology strongly supports monophyly of Dasyuridae, whereas maximum parsimony analysis of the same morphological data fails to resolve dasyurids as monophyletic. Interestingly, of the clades that are strongly supported by our molecular data and that are recovered in both of our morphological analyses, more receive strong support in the undated Bayesian analysis than in the maximum parsimony analysis (table 13). Nevertheless, we recognize that meaningful comparisons of Bayesian posterior probabilities with bootstrap frequencies are difficult (Suzuki et al., 2002; Wilcox, 2002; Erixon et al., 2003; García-Sandoval, 2014), so the significance of this observation is unclear.

Maximum parsimony and undated Bayesian analyses of our morphological data recover monophyly of most marsupial orders represented by more than one terminal, namely Dasyuromorphia, Diprotodontia, Peramelemorphia, and Microbiotheria. Both analyses of our morphological dataset also recover a clade comprising all our paucituberculatan terminals except †Evolestes. However, Didelphimorphia is not resolved as a clade in either morphological analysis due to a lack of resolution at the base of Marsupialia. This is not surprising given that didelphids are morphologically plesiomorphic by comparison with other marsupials (Wible, 1990, 2003; Voss and Jansa, 2009, 2021; Williamson et al., 2014). Although didelphids are characterized by apomorphic fusion between the interparietal and supraoccipital early in postnatal life (char. 31; Nesslinger, 1956; Clark and Smith, 1993; Voss and Jansa, 2009), this feature could not be assessed in any of our outgroup terminals (†Pucadelphys, †Andinodelphys, †Allqokirus, and †Mimoperadectes) or in paucituberculatans (see chars. 25, 30), and it is inapplicable in †Herpetotherium (due to apparent absence of the interparietal in this taxon; chars. 30, 31), so precocious interparietal-supraoccipital fusion does not optimize as a didelphimorphian synapomorphy in these analyses. The apomorphic absence of a posterior cingulid on the lower molars (char. 180) is a dental feature that distinguishes most didelphids26 from otherwise dentally similar outgroup terminals, but, based on its distribution within Marsupialia and nonmarsupial metatherians, it is likely a synapomorphy of Marsupialia as a whole rather than of Didelphimorphia (Voss and Jansa, 2009).

More problematically, our craniodental analyses recover some higher-level relationships that are incongruent with our current understanding of marsupial phylogeny (Eldridge et al., 2019). In particular, Australidelphia is not supported in our craniodental results, which instead group all of our ingroup terminals characterized by gliriform lower anterior incisors (diprotodontians, paucituberculatans, and †Yalkaparidon) into a single clade despite compelling molecular evidence that gliriform lower anterior incisors originated independently in diprotodontians and paucituberculatans (Weisbecker and Beck, 2015). Our failure to recover Australidelphia from analyses of craniodental characters is not surprising, however, because morphological support for this clade seems to consist only of the tarsal characters described by Szalay (1982a). In fact, all studies of marsupial phylogeny based on the cranium and dentition previous to Szalay (1982a) consistently failed to link microbiotherians with Australian marsupials (reviewed by Archer, 1984c: 760–785). Although some analyses of craniodental data that postdate Szalay (1982a) have recovered Australidelphia (e.g., Sánchez-Villagra, 2001; Ladevèze and Muizon, 2007; Carneiro and Oliveira, 2017b; Carneiro, 2018; Engelman et al., 2020; Muizon et al., 2018), only those by Carneiro and coworkers have included both paucituberculatan and diprotodontian terminals (Carneiro and Oliveira, 2017a; Carneiro et al., 2018; Carneiro, 2019; Rangel et al., 2019). Similar to our results, several other analyses that have focused on craniodental characters have likewise failed to support australidelphian monophyly (Springer et al., 1997; Ladevèze and Muizon, 2010; Wilson et al., 2016).

Although our craniodental analyses fail to recover monophyly of Didelphimorphia (see above), they do recover several didelphid subclades that are well corroborated by previous molecular, morphological, and total-evidence studies (e.g., Steiner et al., 2005; Flores, 2009; Voss and Jansa, 2009; Jansa et al., 2014; Mitchell et al., 2014; May-Collado et al., 2015; Vilela et al., 2015; Amador and Giannini, 2016; Beck and Taglioretti, 2020; Álvarez-Carretero et al., 2021), including Didelphini, Metachirus + Didelphini, and Caluromyinae. Interestingly, they also recover monophyly of Thylamyini sensu Voss and Jansa (2009), whereas Voss and Jansa's (2009) own analysis of 129 nonmolecular (external morphology, craniodental, and karyotype) characters did not. Our craniodental analyses agree with recent studies of fossil didelphimorphians in consistently placing †Thylophorops within Didelphini, and in supporting the monophyly of †Sparassocynini (Forasiepi et al., 2009; Goin et al., 2009b; Voss and Jansa, 2009; Suárez Gómez, 2019; Beck and Taglioretti, 2020). However, only our Bayesian analysis supports a close relationship between †Sparassocynini, Monodelphis, and †Thylatheridium (in agreement with Beck and Taglioretti, 2020), whereas the relationships among these taxa is unresolved in our maximum parsimony analysis.

Within Peramelemorphia, our craniodental grouping of Perameles and Isoodon with Macrotis and Chaeropus agrees with Groves and Flannery's (1990) classification as well as with recent craniodental analyses by Travouillon and coauthors (Travouillon et al., 2010, 2013a, 2013b, 2015b, 2017, 2019, 2021; Gurovich et al., 2014; Chamberlain et al., 2015; Travouillon and Phillips, 2018). However, this result is notably incongruent with the results of most molecular analyses, including our own, which consistently support peramelid monophyly. As noted by several authors (Travouillon et al., 2010; Chamberlain et al., 2015; Kear et al., 2016; Warburton and Travouillon, 2016; Travouillon and Phillips, 2018), morphological support for grouping Perameles, Isoodon, Macrotis, and Chaeropus may reflect convergent adaptations to open, xeric habitats. By contrast, our analyses agree with all published morphological and total-evidence analyses of Peramelemorphia (Travouillon et al., 2010, 2013a, 2013b; 2015b, 2017, 2019, 2021; Gurovich et al., 2014; Chamberlain et al., 2015; Kear et al., 2016; Travouillon and Phillips, 2018) in placing the fossil taxa †Bulungu, †Galadi, and †Yarala outside the crown clade.

Within Dasyuromorphia, our craniodental analyses place Myrmecobius closer to Thylacinus than to Dasyuridae, whereas most recent molecular and total-evidence analyses have instead placed Myrmecobius closer to Dasyuridae (Miller et al., 2009; Mitchell et al., 2014; May-Collado et al., 2015; Kealy and Beck, 2017; Feigin et al., 2018). However, an undated Bayesian analysis of craniodental and postcranial characters by Kealy and Beck (2017: fig. 1b) also recovered Myrmecobius + Thylacinidae, and a maximum parsimony analysis of the same dataset did not resolve this relationship (Kealy and Beck, 2017: fig. 1a). Our analyses also place †Badjcinus, †Mutpuracinus, †Barinya, and †Nimbacinus in a clade with Thylacinus, whereas the relationships of these fossil taxa have varied in other analyses. †Nimbacinus has been consistently recovered as a thylacinid in recently published phylogenetic analyses (Murray and Megirian, 2006a; Kealy and Beck, 2017; Rovinsky et al., 2019), and †Badjcinus was also recovered as a thylacinid by Rovinsky et al. (2019) and by the morphological and undated total-evidence analyses of Kealy and Beck (2017), but a dated total-evidence analysis by the latter authors placed †Badjcinus as a stem-dasyuromorphian (Kealy and Beck, 2017: fig. 3). The positions of †Mutpuracinus and †Barinya within Dasyuromorphia were unresolved in the morphological analyses of Kealy and Beck (2017), but their undated total-evidence analysis placed †Barinya as sister to Myrmecobius, and their dated total-evidence analysis placed both †Mutpuracinus and †Barinya in a clade with Myrmecobius. However, most of the morphological analyses of Rovinsky et al. (2019) instead placed †Mutpuracinus and †Barinya together with crown-clade dasyurids.

Wroe (1999) identified four cranial features as synapomorphies of crown-clade Dasyuridae: “presence of a fully enclosed stylomastoid foramen that includes a periotic component; contact between the pars petrosa and a paroccipital tympanic process27 that fully encloses the paroccipital hypotympanic sinus ventrally; extensive dorsal enclosure of the internal jugular canal; and contact between the mastoid tympanic process [= caudal tympanic process of the petrosal here] and the pars petrosa” (Wroe, 1999: 501). We incorporated versions of the first and last of these features into our craniodental character set by scoring presence/absence and composition of the stylomastoid foramen (char. 79) and contact between the caudal tympanic process and the pars cochlearis of the petrosal (char. 68). Despite this, our maximum parsimony analysis fails to resolve monophyly of crown-clade Dasyuridae. In contrast, our undated Bayesian analysis of our craniodental characters does strongly support monophyly of crown-clade Dasyuridae.

Both of our craniodental analyses recover relationships within Paucituberculata that conflict with recent studies of paucituberculatan phylogeny (Abello, 2007, 2013; Goin et al., 2007b, 2009a; Forasiepi et al., 2013; Rincón et al., 2015; Engelman et al., 2016; Abello et al., 2020, 2021). In particular, neither of our analyses groups †Evolestes with other paucituberculatans, whereas this important fossil has been consistently recovered as a plesiomorphic paucituberculatan in other published analyses (Goin et al., 2007b; Abello, 2013; Rincón et al., 2015; Engelman et al., 2016; Abello et al., 2020, 2021). Our analyses also conflict with those studies in failing to group the fossil caenolestid †Stilotherium with extant caenolestids, instead placing †Pichipilus (a pichipilid palaeothentoid) within crown-clade Caenolestidae, and in failing to recover monophyly of †Palaeothentoidea (= †Acdestis + †Palaeothentes + †Pichipilus) or †Palaeothentidae (= †Acdestis + †Palaeothentes). The grouping of †Pichipilus with extant caenolestids may be due to the shared presence of an anteorbital vacuity, which is absent in †Acdestis and †Palaeothentes (the condition in †Stilotherium is unknown; char. 4). In addition, we included only the best-preserved fossil paucituberculatan taxa as terminals here, whereas other studies (Abello, 2007, 2013; Goin et al., 2007b, 2009a; Forasiepi et al., 2013; Rincón et al., 2015; Engelman et al., 2016; Abello et al., 2020, 2021) included a much denser sampling of fossil paucituberculatans (including taxa only known from fragmentary dental remains), and they used a wider range of paucituberculatan-specific morphological characters that could not be included here due to the difficulty of scoring them consistently across our broader taxonomic sampling.

To our knowledge, ours is the first published study using discrete morphological characters to include a dense sampling of modern and fossil terminals across all major subclades within Diprotodontia. Other published studies have included only a few diprotodontian terminals within the context of broader marsupial relationships—e.g., Horovitz and Sánchez-Villagra (2003), Carneiro et al. (2018), Carneiro (2019), and Rangel et al. (2019)—or have focused on specific subclades within Diprotodontia, such as Macropodiformes (Kear et al., 2007; Kear and Pledge, 2008; Prideaux and Warburton, 2010; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018) and Vombatiformes (Munson, 1992; Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020). In particular, sampling of “possums” (non-macropodiform phalangeridans) has been limited in previous studies.

Maximum parsimony and undated Bayesian analyses of our craniodental characters both recover monophyly of Diprotodontia, albeit with only weak support. Our maximum parsimony analysis recovers monophyly of Vombatiformes, with †Thylacoleonidae as the sister to the remaining vombatiforms, in agreement with Gillespie et al. (2016) and Beck et al. (2020). By contrast, our undated Bayesian analysis places †Thylacoleonidae as sister to the rest of Diprotodontia, a novel result that has not (to our knowledge) been proposed previously (see the taxonomic account for †Thylacoleonidae, below). Phascolarctidae is sister to the remaining vombatiform taxa (= Vombatomorphia sensu Beck et al., 2020: table 1) in both of our morphological analyses, a result that is congruent with those of other recent morphological studies (Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020). Neither of our morphological analyses robustly resolves the branching pattern between the vombatomorphian families †Diprotodontidae, Vombatidae, †Ilariidae, and †Wynyardiidae, but we note that there is no current consensus regarding their interrelationships based on previous studies (Archer, 1984c; Woodburne, 1984b; Aplin and Archer, 1987; Marshall et al., 1990; Munson, 1992; Murray, 1998; Gillespie, 2007; Black, 2008; Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020).

Although monophyly of Phalangerida is strongly corroborated by molecular sequence data (e.g., Meredith et al., 2009a, 2011; Mitchell et al., 2014; May-Collado et al., 2015; Duchêne et al., 2018; Álvarez-Carretero et al., 2021), only our maximum parsimony analysis recovers this clade; our undated Bayesian analysis instead finds a major polytomy at the base of Diprotodontia. Both our craniodental analyses recover monophyly of some widely accepted clades such as Macropodiformes, Pseudocheiridae, Acrobatidae, Phalangeridae and Dactylopsilinae, but not of others such as Burramyidae, Potoroidae, or a Phalangeridae + Burramyidae clade. Interestingly, our undated Bayesian analysis recovers a clade that comprises pseudocheirids, acrobatids, and Petaurus, a grouping that somewhat resembles the well-corroborated clade Petauroidea (e.g., Aplin and Archer, 1987; Phillips and Pratt, 2008; Meredith et al., 2009a, 2011; Mitchell et al., 2014; May-Collado et al., 2015; Duchêne et al., 2018; Álvarez-Carretero et al., 2021) but that lacks Tarsipes, Gymnobelideus, and dactylopsilines.

Within Macropodiformes, maximum parsimony and undated Bayesian analyses of our craniodental characters both place †Balbaridae in a clade with macropodids and potoroids, whereas other published analyses have placed balbarids outside Macropodidae + Potoridae (Kear et al., 2007; Kear and Pledge, 2008; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018). Our two craniodental analyses differ in the position of †Ekaltadeta: whereas our maximum parsimony analysis places this taxon as sister to all other macropodiforms, our undated Bayesian analysis places it as sister to Hypsiprymnodon. †Ekaltadeta and the two other genera currently recognized within the subfamily †Propleopinae (†Propleopus and †Jackmahoneya, not included here) have been placed in Hypsiprymnodontidae by some authors (Ride, 1993; Ride et al., 1997; Kear and Cooke, 2001), but Wroe et al. (1998) proposed that propleopines and balbarids form a clade to the exclusion of Hypsiprymnodon. The position of †Ekaltadeta has also varied in other published analyses (Kear et al., 2007; Kear and Pledge, 2008; Bates et al., 2014; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016, 2022; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018).

Undated Total-Evidence Analysis

Our undated total-evidence analysis (fig. 32) fails to unequivocally place †Herpetotherium outside Marsupialia; instead, †Herpetotherium forms a polytomy with various didelphid terminals (which do not form a clade) and a clade comprising our remaining marsupial terminals. Based on previous studies (e.g., Sánchez-Villagra et al., 2007; Horovitz et al., 2008, 2009; Ladevèze et al., 2020), it seems probable that †Herpetotherium does indeed fall outside Marsupialia, and we enforced this topology in our dated total-evidence analysis to enable us to calibrate the age of Marsupialia; however, we note that a few analyses have placed †Herpetotherium within the crown clade (Wilson et al., 2016; Maga and Beck, 2017: fig. 39).

Our failure to recover a didelphid clade in the undated total-evidence analysis is probably related to the overall plesiomorphic morphology of opossums as discussed above. Although our undated total-evidence analysis fails to recover didelphid monophyly, a position for our didelphid terminals outside a Paucituberculata + Australidelphia clade is in agreement with retroposon evidence (Gallus et al., 2015a) and contrasts with our molecular results, wherein the marsupial root node is consistently placed between Paucituberculata and Didephimorphia + Australidelphia (see above).

With the exception of didelphid nonmonophyly and the shifted position of the marsupial root node, relationships among Recent terminals in our undated total-evidence analysis (fig. 32) closely resemble those recovered by our combined-molecular analysis (fig. 29), suggesting that the addition of craniodental characters had little impact on the resultant topology. This is perhaps unsurprising given the relative sizes of the molecular and craniodental partitions—24.5 kb of molecular sequence data (representing 16,116 variable sites, of which 13,794 are parsimony informative) versus 180 morphological characters (all of which are variable, and 163 are parsimony informative)—and (as discussed above) the evidently much stronger phylogenetic signal in the molecular data (Scarpetta, 2020; but see Neumann et al., 2021).

The positions of fossil terminals in this analysis are generally similar to those recovered in the craniodental analyses. For fossil didelphids, the positions of †Thylophorops within Didelphini and of †Thylatheridium and the sparassocynins †Hesperocynus and †Sparassocynus in a clade with Monodelphis agree with the results of previous studies (Reig, 1958b; Goin and Rey, 1997; Voss and Jansa, 2009; Beck and Taglioretti, 2020). Within Paucituberculata, †Evolestes groups with the other paucituberculatan terminals, †Acdestis and †Palaeothentes form a clade (†Palaeothentidae), and †Stilotherium is sister to crown caenolestids, all of which are in closer agreement with other published studies (Abello, 2007, 2013; Goin et al., 2007b, 2009a; Forasiepi et al., 2013; Rincón et al., 2015; Engelman et al., 2016; Abello et al., 2020, 2021) than are our craniodental results. However, as in our craniodental analyses (see above), the presence of †Pichipilus within the paucituberculatan crown clade conflicts with several published studies that place †Pichipilus closer to †Palaeothentidae (represented here by †Acdestis and †Palaeothentes) within the superfamily †Palaeothentoidea. As in our craniodental analyses, this result is probably at least partly due to the presence of an anteorbital vacuity in †Pichipilus, a derived feature shared with extant caenolestids (the condition in †Stilotherium is unknown) but absent in †Acdestis and †Palaeothentes (see char. 4).

The presence of †Yalkaparidon in a trichotomy with †Evolestes and our remaining paucituberculatan terminals conflicts with the phylogenetic analyses of Beck et al. (2014, 2016), which suggest that †Yalkaparidon is probably an australidelphian (isolated tarsals referred to †Yalkaparidon by Beck et al. [2014] exhibit apparently australidelphian apomorphies). However, subsequent analyses by Beck (2017) and Zimicz and Goin (2020), which also included data from the putative †Yalkaparidon tarsals described by Beck et al. (2014), placed Yalkaparidon in a clade with the caenolestids Caenolestes and †Palaeothentes (and with argyrolagid terminals not included in this study), a result similar to that obtained in this analysis.

As in our craniodental analyses, †Bulungu, †Galadi, and †Yarala all fall outside crown-clade Peramelemorphia, in agreement with many recent studies (Travouillon et al., 2010, 2013a, 2013b, 2014a, 2015b, 2017, 2019, 2021; Gurovich et al., 2014; Chamberlain et al., 2015; Kear et al., 2016; Travouillon and Phillips, 2018). Our undated total-evidence analysis also resembles our craniodental analyses in placing the fossil dasyuromorphians †Barinya, †Mutpuracinus, †Badjcinus, and †Nimbacinus in a clade with Thylacinus; as already discussed (see above), the positions of †Barinya, †Mutpuracinus, and †Badjcinus has varied considerably in published phylogenetic analyses, and there is no current consensus regarding their affinities (Wroe et al., 2000; Wroe and Musser, 2001; Murray and Megirian, 2006a; Kealy and Beck, 2017; Rovinsky et al., 2019).

Within Diprotodontia, relationships within Vombatiformes (here including †Thylacoleonidae) are better resolved than in our craniodental analyses. As already discussed, the position of †Thylacoleonidae as sister to the remaining vombatiforms is congruent with some recent studies (Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020), as is the position of Phascolarctidae as the next family to diverge (Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020). The relationships among vombatomorphians—a †Wynyardiidae + †Ilaria clade sister to †Diprotodontidae, with Vombatidae outside this—found in our undated total-evidence analysis have not been recovered in any other published analysis of vombatiform relationships, but other studies are characterized by conflicting topologies (Archer, 1984c; Woodburne, 1984b; Aplin and Archer, 1987; Marshall et al., 1990; Munson, 1992; Murray, 1998; Gillespie, 2007; Black, 2008; Black et al., 2012a; Gillespie et al., 2016; Beck et al., 2020).

Within Phalangeridae, the position of “Trichosurus” †dicksoni within Trichosurinae and †Onirocuscus within Ailuropinae is broadly congruent with the original description of these fossil taxa by Flannery and Archer (1987a), who referred “T.” †dicksoni to the modern genus and identified †O. reidi as a fossil member of the ailuropine genus Strigocuscus. However, in an unpublished Ph.D. dissertation, Crosby (2002b) concluded that “T.” †dicksoni should be assigned to a different genus, and she subequently (Crosby, 2007: fig. 7) presented an informal phylogeny in which “T.” †dicksoni and †Onirocuscus were placed in a clade separate from Recent phalangerids.

Within Macropodiformes, the position of †Ekaltadeta as sister to Hypsiprymnodontinae is (as already discussed above) congruent with some of the phylogenetic analyses of den Boer and Kear (2018), but most published analyses instead place †Ekaltadeta in a clade with balbarids (Kear et al., 2007; Kear and Pledge, 2008; Bates et al., 2014; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018; see also Wroe et al., 1998). ?Bettongiamoyesi does not form a clade with our extant Bettongia terminal, contra Flannery and Archer (1987b), but nevertheless is recovered as a bettongin; the position of ?Bettongiamoyesi in other published analyses has varied and does not always fall within Potoroidae (Kear et al., 2007; Kear and Pledge, 2008; Bates et al., 2014; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018).

The balbarids †Balbaroo and †Ganawamaya form a clade with three fossil macropodids (†Gangaroo, †Hadronomas, and Rhizosthenurus) that is sister to the living macropodid Dorcopsulus, with Dorcopsis outside this. This arrangement is strongly incongruent with currently accepted relationships, as †Balbaridae is consistently supported as falling outside Macropodidae + Potoroidae, whereas †Gangaroo, †Hadronomas, and Rhizosthenurus (the last two placed together in †Sthenurinae) are likely stem macropodids (Kear and Cooke, 2001; Cooke, 2006; Kear et al., 2007; Kear and Pledge, 2008; Bates et al., 2014; Black et al., 2014c; Travouillon et al., 2014b, 2015a, 2016; Cooke et al., 2015; Butler et al., 2016, 2018; den Boer and Kear, 2018). The anomalous topology found in our undated total-evidence analysis may be due to (1) the fact that Dorcopsis and Dorcopsis are generally accepted to be the most plesiomorphic living macropodids (Prideaux and Warburton, 2010), and (2) our omission of several fragmentary fossil macropodiforms that Cooke (1997a, 1997b, 1997c, 2006) interpreted as evidence that the fully lophodont dentitions of balbarids and macropodids evolved convergently.

Dated Total-Evidence Analysis

Comparing the results of our undated (fig. 32) and dated total-evidence (fig. 33) analyses, the only major topological change among our Recent terminals concerns the position of the root—the direct result of imposing topological constraints so that the ages of Marsupialia and Didelphimorphia could be calibrated (see appendix 3). Our decision to enforce †Herpetotherium as falling outside Marsupialia is congruent with most recent published phylogenies (e.g., Sánchez-Villagra et al., 2007; Horovitz et al., 2008, 2009; Ladevèze et al., 2020). Similarly, our decision to enforce Didelphimorphia as sister to the remaining marsupial orders is congruent with statistically significant retrotransposon insertion evidence for this rooting position (Gallus et al., 2015a).

However, the incorporation of temporal information had a major effect on the positions of many fossil terminals, which moved to nodes deeper within the tree (a phenomenon previously reported in other studies; see, e.g., Lee and Yates, 2018; Beck and Taglioretti, 2020; King and Beck, 2020). This is presumably due to two interrelated factors. All else being equal, dated analyses using a Fossilized Birth-Death model are expected to reconstruct fossil terminals as originating from deeper nodes than modern terminals, particularly when phylogenetic signal in the character data that might otherwise overwhelm the temporal signal is weak (Lee and Yates, 2018; King, 2021). In addition, most current implementations of Fossilized Birth-Death models (including our own) assume constant rates of speciation, extinction, and fossilization/sampling through time and between lineages. This assumption means that long, unsampled branches will be disfavored, such that fossil terminals are shifted to deeper positions if fossil sampling is not consistent throughout the tree (Turner et al., 2017; King, 2021). Fossil sampling is undoubtedly not consistent through time or between lineages in this study. Most obviously, we did not include any crown-clade (marsupial) terminals older than the late Oligocene, because older taxa tend to be highly incomplete, typically known only from fragmentary dental remains. Indeed, for some clades we did not include any fossil representatives, either because fossil members described to date are highly incomplete (e.g., Pseudocheiridae, Petauridae, Burramyidae), or because no fossil members are currently known (e.g., Tarsipedidae, Myrmecobiidae; Long et al., 2002; Archer and Hand, 2006; Eldridge et al., 2019). Thus, the phylogenetic affinities of taxa that shift position between our undated and dated total-evidence analyses (e.g., †Onirocuscus and “Trichosurus” †dicksoni, which are within crown-clade Phalangeridae in the undated analysis, but outside it in the dated analysis) require further study.

In general, the divergence dates found in our dated total-evidence analysis are younger than those reported by most other molecular-clock studies of pan-marsupial relationships (table 6). In particular, several such studies estimated the first divergence within Marsupialia to have occurred during the Cretaceous, whereas we date this divergence to the late Paleocene or earliest Eocene. However, a recent species-level phylogenomic analysis of mammalian divergence times by Álvarez-Carretero et al. (2021) also recovered a late Paleocene to early Eocene age (58.9–49.3 Ma) for the first divergence within Marsupialia. There are several plausible, nonexclusive explanations for such discrepancies.

Although we have not undertaken a detailed reassessment of the fossil calibrations used by previous molecular-clock studies, we note that some calibrations appear to be problematic in the light of current evidence. For example, Meredith et al. (2009a) assigned a mean prior distribution of 75 Mya on the age of Marsupialia, whereas Meredith et al. (2011: table S3) specified a minimum of 65.18 Mya for the split between Didelphimorphia and Australidelphia. Both dates were predicated on the assumption that peradectids are stem didelphimorphians as proposed by Horovitz et al. (2009); however, most subsequently published analyses suggest that peradectids fall outside Marsupialia, as we also found in our morphological and undated total-evidence analyses. As summarized by Eldridge et al. (2019), the oldest definitive marsupials are from the early Eocene Tingamarra Local Fauna, and so have a minimum age of only 54.55 million years old (Godthelp et al., 1992; Beck et al., 2008). Álvarez-Carretero et al. (2021: supplementary information) did use these Tingamarran fossils to provide a minimum bound on the first divergence within Marsupialia, but they specified a date of 48.07 Mya (the minimum age of the Ypresian; Speijer et al. 2020) following Benton et al. (2015: 60) rather than the actual radiometric date reported for the Tingamarra site by Godthelp et al. (1992). Kealy and Beck (2017: 16, 17) provided evidence that some of the calibrations used by Mitchell et al. (2014)—and subsequently by Duchêne et al. (2018)—are also inappropriate. Lastly, different dating programs, including MrBayes (used here) and MCMCTree (used by Meredith et al. 2011, Mitchell et al., 2014, Duchêne et al., 2018, and Álvarez-Carretero et al., 2021), are known to give different posterior estimates of divergence times even when provided with the same data and fossil calibrations (Barba-Montoya et al., 2017).

A relevant methodological difference between our study and previous molecular-dating studies based on Recent taxa (which necessarily used only node calibrations) is that we used a total-evidence dataset that included both extant and fossil terminals (allowing the use of tip calibrations), which we analyzed using the Fossilised Birth-Death model. Total-evidence dating arguably makes the best use of paleontological data for constructing evolutionary timescales (O'Reilly and Donoghue, 2016), and combining both tip and node calibrations (as done here) appears to resolve the issue of tip-dating-only analyses recovering unrealistically ancient dates (O'Reilly et al., 2015; O'Reilly and Donoghue, 2016; Ronquist et al., 2016). In particular, tip calibrations can interact with node calibrations to more objectively define maximum bounds on the ages of calibrated nodes (O'Reilly and Donoghue, 2016), whereas molecular-clock studies that use only node calibrations often specify highly conservative maximum bounds (or no maximum bound at all on some calibrated nodes), which tend to favor older divergence dates.

However, the use of morphological data (either alone or in the context of a total-evidence analysis) to directly estimate divergence times has been criticized (O'Reilly et al., 2015; Ronquist et al., 2016; Parins-Fukuchi and Brown, 2017). Much of this criticism surrounds the realism (or lack thereof) of the Lewis (2001) Mk model for discrete morphological character data (but see Klopfstein et al., 2019). Simulation studies also suggest that accurate estimates of divergence times using the Fossilized Birth-Death model require the inclusion of as much fossil occurrence data as possible, including highly incomplete taxa (Klopfstein et al., 2019; Luo et al., 2020; O'Reilly and Donoghue, 2020), whereas we included only well-preserved taxa among our terminals. Nevertheless, molecular-clock analyses are also sensitive to several assumptions (Warnock et al., 2012, 2015; Condamine et al., 2015; Bromham, 2019; Budd and Mann, 2020; Klopfstein, 2020) that have seldom been convincingly addressed in the marsupial literature.

In summary, although there is reason to believe that our date estimates are based on more appropriate calibration points and improved estimation methods than those employed in previous molecular-clock studies, all phylogenetic date estimates should be viewed with caution, and ours are no exception. In particular, we are concerned that our estimates for the the split between Didelphimorphia and our remaining terminals (median age: 56.2 Mya; 95% HPD: 54.7–58.6 Mya) and the split between Paucituberculata + Yalkaparidon and Australidelphia (median age: 55.1 Mya; 95% HPD: 54.6–56.6 Mya) are close to the minimum bound of the calibration applied to these nodes (54.55 Mya), suggesting that there may be some incongruence between this calibration and the overall clock model (which would presumably favor younger dates for these nodes if they were not calibrated). Additionally, this 54.55 Mya calibration is based on a K-Ar date for the Tingamarra Local Fauna that needs to be confirmed with improved geochronological methods (e.g., Ar-Ar dating). If the currently accepted age for Tingamarra is too old, then the crown age of Marsupialia may be even younger than estimated here. However, if the Tingamarran fossils are discounted, then the oldest definitive marsupials are probably australidelphian-type tarsals from the La Barda locality in Argentina, which have a minimum age of 43 Mya (Tejedor et al., 2009; Lorente et al., 2016).

Ultimately, however, the highly incomplete nature of the marsupial geological record—especially the very limited Australian record prior to the late Oligocene (Black et al., 2012b; Woodhead et al., 2014; Beck, 2017b; Bennett et al., 2018; Eldridge et al., 2019)—means that a wide range of potential divergence dates are congruent with the available fossil evidence. In the taxonomic accounts that follow, we consider the plausibility of our estimated divergence times given the known fossil record of the taxon in question and note any instances of obvious incongruence.

TAXONOMIC ACCOUNTS

The following classification of Marsupialia formalizes clade membership at the Linnaean rank of family and above and summarizes our conclusions regarding what appear to be well-corroborated groups. It is largely based on the results of our dated total-evidence analysis (fig. 33), because this shows the greatest congruence with other recent phylogenetic analyses of marsupials, and because it includes both fossil and extant terminals. Except where noted, we follow Voss and Jansa (2009) for names of American marsupial clades and Jackson and Groves (2015) for the names of Sahulian marsupials. For each taxon, we list its contents (restricted to taxa included in this study) and unambiguous craniodental synapomorphies (i.e., those identified under both Accelerated and Delayed Transformations; see Methods) as optimized by parsimony on our dated total-evidence topology (fig. 33).28 We also list estimated divergence times (median estimates and 95% highest posterior density, HPD) for their divergence from their sister taxa (stem age) and (if comprising more than one terminal) for their first diversification (crown age), also based on the results of our dated total-evidence analysis. Under Comments we compare these age estimates with the fossil record and (where relevant) other studies that have estimated divergence dates within Marsupialia using molecular or total-evidence clock analyses.

Marsupialia, Illiger 1811

  • Contents: Dasyuromorphia, Didelphimorphia, Diprotodontia, Microbiotheria, Notoryctemorphia, Paucituberculata, Peramelemorphia, and †Yalkaparidontia.

  • Stem Age: 59.7 Mya (95% HPD: 55.6–65.8 Mya).

  • Crown Age: 56.2 Mya (95% HPD: 54.7–58.6 Mya).

  • Unambiguous Craniodental Synapomorphies: Lower molars without a posterior cingulid (char. 180: 1→0; ci = 0.333).

  • Comments: As noted in our introduction, most recent authors restrict Marsupialia to the metatherian crown clade, and we follow the node-based definition of Marsupialia proposed by Beck et al. (2014: 131): the least inclusive clade containing Didelphis marsupialis, Caenolestes fuliginosus, and Phalanger orientalis. By contrast, †Pucadelphys, †Mayulestes, and †Allqokirus have consistently fallen outside Marsupialia in most published analyses (Rougier et al., 1998, 2004, 2015; Luo et al., 2003, 2011; Ladevèze and Muizon, 2007; Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Forasiepi, 2009; Ladevèze and Muizon, 2010; Beck, 2012, 2017a; Williamson et al., 2012, 2014; Engelman and Croft, 2014; Forasiepi et al., 2014a; Wilson et al., 2016; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Bi et al., 2018; Carneiro, 2018, 2019; Carneiro et al., 2018; Muizon et al., 2018; Rangel et al., 2019; Ladevèze et al., 2020; Muizon and Ladevèze, 2020. Notable exceptions to this widespread consensus (Goin et al., 2006; Velazco et al., 2022) are discussed in the account for †Pucadelphys in appendix 1).

  • Mimoperadectes also falls outside Marsupialia in all our craniodental (figs. 30, 31) and total-evidence (figs. 32, 33) analyses. This result conflicts with Horovitz et al. (2009), who recovered †Mimoperadectes as a stem didelphimorphian, but it agrees with most subsequent studies (Forasiepi, 2009; Beck, 2012; Beck et al., 2014; Engelman and Croft, 2014; Forasiepi et al., 2014a; Suarez et al., 2015; Carneiro, 2018, 2019; Carneiro et al., 2018; Muizon et al., 2018; Rangel et al., 2019; Engelman et al., 2020; Ladevèze et al., 2020; Muizon and Ladevèze, 2020), which have consistently found that †Mimoperadectes and other alleged peradectids are outside the metatherian crown clade.29 In particular, as we discussed above (see char. 55), †Mimoperadectes appears to lack any kind of tympanic process or bulla enclosing the hypotympanic sinus ventrally, as do our outgroup terminals †Pucadelphys, †Mayulestes, and †Allqokirus, plus several other nonmarsupial metatherians not included here (Muizon, 1999; Forasiepi, 2009; Bi et al., 2015; Beck, 2017a; Muizon et al., 2018; Muizon and Ladevèze, 2020). However, the polarity of this feature is debated, and there is clear evidence of homoplasy within Metatheria (Kielan-Jaworowska and Nessov, 1990: 7; Forasiepi, 2009: char. 50; Wilson et al., 2016: supplementary information; Beck, 2017a: 402; Muizon et al., 2018; Muizon and Ladevèze, 2020: 683–684).

  • Our craniodental and undated total-evidence analyses (figs. 3032) place †Herpetotherium in an unresolved polytomy with Recent marsupials. In contrast to our outgroup terminals and †Mimoperadectes, †Herpetotherium has an ossified hypotympanic sinus floor, which optimizes as an unambiguous synapomorphy of †Herpetotherium + Marsupialia in our results (see supplementary file S3 in the online supplement:  https://doi.org/10.5531/sd.sp.54). However, most recent published phylogenetic analyses place †Herpetotherium outside Marsupialia (Ladevèze and Muizon, 2007; Sánchez-Villagra et al., 2007; Beck et al., 2008a; Horovitz et al., 2008, 2009; Forasiepi, 2009; Ladevèze and Muizon, 2010; Beck, 2012; Williamson et al., 2012, 2014; Engelman and Croft, 2014; Forasiepi et al., 2014a; Beck, 2017a; Carneiro and Oliveira, 2017a, 2017b; Maga and Beck, 2017; Carneiro, 2018, 2019; Carneiro et al., 2018; Muizon et al., 2018; Rangel et al., 2019; Muizon and Ladevèze, 2020). Based on these studies, and to enable us to calibrate the age of Marsupialia, we constrained †Herpetotherium to fall outside the crown clade in our dated total-evidence analysis.

  • With †Herpetotherium constrained to fall outside Marsupialia, only loss of the posterior cingulid optimizes as an unambiguous synapomorphy for the crown clade. Although presence or absence of a posterior cingulid might be a useful criterion for distinguishing marsupials from nonmarsupial marsupialiforms (Voss and Jansa, 2009: 87), this structure was subsequently reacquired by dasyuromorphians (see below) and by the balbarid macropodiform †Balbaroo (and other balbarids not included here; see Cooke, 1997a: fig. 7, who referred to the posterior cingulid as the “hypocingulid”). Three fossil Australian taxa that have not been included here due to their incompleteness—†Djarthia murgonensis from the early Eocene Tingamarra fauna in northeastern Australia and its probable relatives †Ankotarinja tirarensis and †Keeuna woodburnei from the late Oligocene Etadunna Formation of central Australia—also have a posterior cingulid and appear to comprise a clade of stem australidelphians (Archer, 1976d; Godthelp et al., 1999; Beck et al., 2008a; Kealy and Beck, 2017), in which case they represent another lineage that independently reacquired this trait. As noted earlier, more or less distinct posterior cingulids have recently been reported in a few extant didelphids (