A new family of diprotodontian marsupials from the latest Oligocene of Australia and the evolution of wombats, koalas, and their relatives (Vombatiformes)

We describe the partial cranium and skeleton of a new diprotodontian marsupial from the late Oligocene (~26–25 Ma) Namba Formation of South Australia. This is one of the oldest Australian marsupial fossils known from an associated skeleton and it reveals previously unsuspected morphological diversity within Vombatiformes, the clade that includes wombats (Vombatidae), koalas (Phascolarctidae) and several extinct families. Several aspects of the skull and teeth of the new taxon, which we refer to a new family, are intermediate between members of the fossil family Wynyardiidae and wombats. Its postcranial skeleton exhibits features associated with scratch-digging, but it is unlikely to have been a true burrower. Body mass estimates based on postcranial dimensions range between 143 and 171 kg, suggesting that it was ~5 times larger than living wombats. Phylogenetic analysis based on 79 craniodental and 20 postcranial characters places the new taxon as sister to vombatids, with which it forms the superfamily Vombatoidea as defined here. It suggests that the highly derived vombatids evolved from wynyardiid-like ancestors, and that scratch-digging adaptations evolved in vombatoids prior to the appearance of the ever-growing (hypselodont) molars that are a characteristic feature of all post-Miocene vombatids. Ancestral state reconstructions on our preferred phylogeny suggest that bunolophodont molars are plesiomorphic for vombatiforms, with full lophodonty (characteristic of diprotodontoids) evolving from a selenodont morphology that was retained by phascolarctids and ilariids, and wynyardiids and vombatoids retaining an intermediate selenolophodont condition. There appear to have been at least six independent acquisitions of very large (>100 kg) body size within Vombatiformes, several having already occurred by the late Oligocene.

Systematic palaeontology. Order Diprotodontia Owen, 1866 New Definition (see Table 1) Suborder Vombatiformes Woodburne, 1984 New Definition (see Table 1) Infraorder Vombatomorphia Aplin and Archer, 1987 New Definition (see Table 1) Superfamily Vombatoidea Kirsch, 1968 New Definition (see Table 1) Family Mukupirnidae Type genus: Mukupirna Included taxa: Mukupirna nambensis new species Generic diagnosis: As for the only known species type species. Mukupirna nambensis gen. et. sp. nov. Differential diagnosis: differs from known members of Wynyardiidae in possessing a P3 that lacks a posterolingual cusp (="hypocone"), less well-developed selenodonty, a less well-developed masseteric process, palatal vacuities entirely enclosed by the palatines, a proportionately longer deltopectoral crest and broader distal end of the humerus (Epicondylar index = 0.44 22 ), a proportionately longer olecranon of the ulna (Index of Fossorial Ability = 0.42 23 ), and a much larger body size (estimated body mass based on postcranial measurements = 143-171 kg); differs from vombatids in lacking bilobate molars (molars are only slightly bilobate in Nimbavombatus, Rhizophascolonus and Warendja, but strongly bilobate in other vombatids); differs from all vombatids except Nimbavombatus in retaining three upper incisors and the upper canine; differs from Nimbavombatus in larger size, more bicuspid P3, and palatal vacuities entirely enclosed by the palatines; differs from vombatids known from postcranial remains in lacking a laterally extensive deltopectoral crest of the humerus; differs from hypselodont vombatids in having closed premolar and molar roots; differs from known members of Thylacoleonidae in retaining only a single upper premolar (P3), with this tooth not as elongate or bladelike, lacking a marked reduction in molar size posteriorly, having a proportionately longer deltopectoral crest and broader distal end of the humerus, and having a proportionately longer olecranon of the ulna; differs from known members of Phascolarctidae in lacking strongly selenodont molars, having a less well-developed masseteric process, a proportionately longer deltopectoral crest and broader distal end of the humerus, and a proportionately longer olecranon of the ulna; differs from known members of Ilariidae in lacking posterobuccal and lingual cusps on P3, in lacking strongly selenodont molars, and in lacking a well-developed masseteric process; differs from known members of Diprotodontidae and Palorchestidae in lacking a molariform P3, molars not strongly bilophodont, in lacking a well-developed masseteric process, and in retaining palatal vacuities. Mukupirna nambensis cannot be compared directly with Marada arcanum (the only known representative of the vombatiform family Maradidae), because Mu. nambensis is only known from the cranium and upper dentition whereas Ma. arcanum is known only from the lower dentition, and it is possible that they represent the same taxon or are closely related (see the supplementary information). Based on the shapes of the preserved alveoli (Fig. 1b), the upper first incisor of Mukupirna was proportionately larger than those of Namilamadeta and Muramura, but the second and third incisors were still present. In vombatids, I1 is very large and (with the probable exception of Nimbavombatus 20 ) is the only upper incisor present 17 . Also based on alveolar evidence (Fig. 1b), Mukupirna retained a large, single-rooted upper canine, which is a plesiomorphic feature seen in wynyardiids and several other vombatiforms, including phascolarctids, ilariids, thylacoleonids and some diprotodontoids [27][28][29][30] . However, an upper canine is absent in all vombatids described to date except Nimbavombatus 20 .
Interestingly, the unworn P3 of the extant common wombat Vombatus ursinus (assuming that this is indeed P3 and not a retained dP3) is also bicuspid and without evidence of lingual cusps (Fig. 2c), although it differs in being proportionately much smaller and more distinctly bicuspid, and in lacking vertical ridging. However, the P3 (again, if it is not dP3) of Lasiorhinus latifrons and of fossil vombatids is sub-triangular and not bicuspid, and a small posterolingual cusp is typically present 20,21,33,34 . The P3 of vombatids is also usually smaller than the anterior upper molars, in contrast to Mukupirna (Fig. 2b,c) 20,21,33,34 . Thus, the bicuspid P3 of Mukupirna and Vombatus may be the result of homoplasy; overall, the P3 of Mukupirna most closely resembles that of the wynyardiid Namilamadeta ( Fig. 2b) 28,30 . However, the enamel on the labial surface of P3 of Mukupirna extends onto the root of the tooth, which is a distinctive feature of most vombatids but apparently absent in other vombatiforms, including wynyardiids.
The molar morphology of Mukupirna (Fig. 2b) is very similar to that of the wynyardiids Namilamadeta and Muramura (Fig. 2a) in being selenolophodont, i.e., exhibiting both lophodont and selenodont features [27][28][29][30] . Prominent stylar cusps are present along the labial margin of the molars, with the paracone and metacone located further lingually (Fig. 2b), but they are not as centrally placed on the tooth crown as they are in wynyardiids (Fig. 2a). A weak selenodont pattern is apparent in Mukupirna, at least on M1-2, with identifiable pre-and postparacristae and pre-and postmetacristae extending from the paracone and metacone respectively (Fig. 2b); these crests are also present, but much better developed, in wynyardiids (Fig. 2a). Weak lophs connect the paracone to the protocone and the metacone to the metaconular hypocone in Mukupirna (Fig. 2b), again as in wynyardiids (Fig. 2a). The upper molars of most other vombatiforms are either fully lophodont (diprotodontids, palorchestids), fully selenodont (ilariids, phascolarctids), or bunodont-bunolophodont (thylacoleonids) 3,5,9 . Intriguingly, however, the same basic occlusal morphology seen in Mukupirna occurs in unworn upper molars of the plesiomorphic fossil vombatids Nimbavombatus and Rhizophascolonus and juveniles of the living wombats Vombatus and Lasiorhinus 20,21 , in which homologues of the paracone and metacone retain traces of a selenodont pattern, but are also connected by weak lophs to the protocone and metaconular hypocone respectively (Fig. 2c). In contrast to wynyardiids and most other vombatiforms (in which a distinct cervix usually separates the root and crown), there is no clear distinction between the root and crown of the upper molars of Mukupirna (Fig. 1a). This is another feature also seen in vombatids, although Mukupirna differs from vombatids in that the enamel of its molars does not extend down the lingual surface of the roots. The molar roots of Mukupirna are long, and the lingual roots extend ventrally far beyond the molar alveoli, whereas the labial roots are not visible, again as in vombatids. However, Mukupirna has closed molar roots, and so in this respect is unlike all known vombatids except the early Miocene Nimbavombatus and Rhizophascolonus 17, 21,35,36 . The occlusal surface of early Miocene vombatids was subject to moderate to severe amounts of wear (particularly towards the anterior end of the molar row). The most extreme wear is seen in R. crowcrofti where the occlusal morphology appears to have been obliterated relatively early in the animal's life, leaving an enamel perimeter surrounding a dentine wear surface 35,36 , as is the case in hypselodont vombatids. By contrast, the molars of Mukupirna evidently retained their original occlusal morphology into at least early adulthood, with no evidence of accelerated wear in the M1 position.
Although the anterior part of the zygomatic arch is poorly preserved in AMNH FM 102646, the masseteric process appears to be weakly developed or absent (Fig. 1a), as it is in vombatids 17 ; by contrast, this process is distinct in most other vombatiforms, including wynyardiids 3,5,27-30 , although it is also absent in most thylacoleonids 37,38 . In most vombatids (with the notable exceptions of Warendja and Nimbavombatus 17,34 ), a very large fossa extends across the lateral surface of the maxilla and jugal, at the anterior end of the zygomatic arch; in extant wombats, this fossa has been shown to house a greatly enlarged superficial masseter 17,39 , which generates high occlusal forces during the medially-directed power stroke of the lower jaw 17,39,40 . The wynyardiids Muramura and Namilamadeta have a similarly-positioned but much smaller fossa 27,28] , and the preserved part of the jugal of Mukupirna also preserves a shallow but distinct fossa on its lateral surface. The presence of this fossa in wynyardiids and Mukupirna might represent a precursor of the much larger fossa seen in most vombatids, in which case, its absence in Warendja and Nimbavombatus is presumably secondary; alternatively, it may instead reflect the presence of enlarged snout musculature in wynyardiids and Mukupirna 1 . Palatal vacuities are present in Mukupirna and appear to be entirely enclosed by the palatine bones, as in most vombatids and also the modern koala Phascolarctos cinereus. In wynyardiids, thylacoleonids, fossil phascolarctids and the vombatid Nimbavombatus, these vacuities are between the maxillae and palatines, which is probably the plesiomorphic condition within Marsupialia 41 . Diprotodontids and palorchestids lack palatal vacuities 3,5 .
The glenoid fossa of Mukupirna is planar, and the postglenoid process also appears to have been either very weakly developed or entirely absent, although the region may be damaged in AMNH FM 102646 (Fig. 1c). By contrast, in most vombatiforms (including wynyardiids), the glenoid fossa has a raised articular eminence anteriorly and distinct mandibular fossa posteriorly, and the postglenoid process is well-developed 3,5,16 . The overall morphology of the glenoid region of Mukupirna somewhat resembles that of the plesiomorphic vombatid Warendja wakefieldi, which is also planar with a very weakly developed postglenoid process 17, 34 . The glenoid region of other vombatids known from cranial remains is highly specialised, with a mediolaterally broad and convex glenoid fossa that lacks any trace of a postglenoid process 16 . The auditory region of AMNH FM 102646 is also damaged, but it appears that the zygomatic epitympanic sinus of the squamosal was either absent or very shallow (Fig. 1c); vombatids are also characterised by a shallow and laterally open zygomatic epitympanic sinus, in contrast to most other diprotodontians in which this sinus is largely enclosed and invades deep into the zygomatic process of the squamosal 16 .
The postcranium of Mukupirna (Figs. 3, 4) exhibits a number of features that are indicative of digging behaviour 42,43 . The humerus is broad distally, giving an Epicondylar Index (= humeral epicondylar width/humeral length) 23 of 0.44, similar to that of modern wombat species Lasiorhinus latifrons and Vombatus ursinus ( Fig. 4a and Table 1). The olecranon process of the ulna is elongate in Mukupirna, providing increased mechanical advantage to the triceps brachii when extending the forearm, which again is common in digging mammals. The Index of Fossorial Ability (=olecranon length/[total ulnar length-olecranon length]) of Mukupirna s 0.42, which is similar to Vombatus ursinus but somewhat less than in Lasiorhinus latifrons ( Fig. 4b and Table 2). A distinct third trochanter is present on the femur, which is unusual among marsupials, and which indicates the presence of well-developed gluteal musculature that may be connected with digging behaviour 44,45 . Most of these features are absent or much less well-developed in other vombatiforms (including wynyardiids), but they are present in vombatids 17,46,47 . The wide distal end of the humerus of Mukupirna and vombatids is largely due to a prominent medial epicondyle, which reflects the presence of powerful extensors and pronators of the forearm 17,42,43 . The manual and pedal phalanges of Mukupirna are strikingly similar to those of vombatids and also the ilariid Ilaria 46 , and are strongly indicative of digging behaviour: they are dorsoventrally flattened (especially the unguals), and their distal ends taper dorsoventrally. In addition, manual phalanx V of Mukupirna exhibits a medial twist at its distal end, and lateral buttressing of its proximal end, which is a distinctive feature also seen in vombatids; this morphology serves to position digit V close to digits I-IV, and in living wombats allows the manus to be used as a shovel during digging.
However, the humerus of Mukupirna (Fig. 4a) differs from those of vombatids in lacking a hypertrophied deltopectoral crest that extends laterally beyond the edge of the humeral shaft and forms a tunnel-like fossa for the origin of the brachialis muscle 17,46 .
phylogenetic relationships of Mukupirna nambensis and other vombatiforms and morphological character evolution within Vombatiformes. Phylogenetic analysis of 79 craniodental and 20 postcranial characters using undated Bayesian inference (using the Mkv model and an eight category lognormal distribution to accommodate rate heterogeneity between characters) places Mukupirna as sister to Vombatidae, with high support (Bayesian posterior probability = 0.98; Fig. 5), and hence a member of Vombatoidea as defined here (Table 1). Maximum parsimony analysis of the same matrix also recovers a Mukupirna + Vombatidae clade, although with relatively low support (bootstrap = 45%; see supplementary information). The Bayesian (2020) 10:9741 | https://doi.org/10.1038/s41598-020-66425-8 www.nature.com/scientificreports www.nature.com/scientificreports/ analysis and the maximum parsimony analysis identify the same four features as unambiguous synapomorphies of Mukupirna + Vombatidae: prominent lingual cusp on P3 absent (character 10); enamel extending down the buccal surface of P3 and onto the root present (character 11); articular eminence of glenoid fossa planar or concave, and mandibular fossa absent or indistinct (character 63); postglenoid process absent or weakly developed (character 64).
Monophyly of all currently recognised vombatiform families represented by two or more terminals is supported by BPP > 0.5, except Wynardiidae, with BPP = 0.36. Thylacoleonidae is sister to the remaining vombatiforms, rather than (as in most previous studies) a member of Vombatomorphia; however, this relationship was also found by Gillespie et al. 48    ASRs for presence or absence of P1 and P2 on the same topology and branch lengths using a single rate Mk model in Mesquite provide strong support for the hypothesis that loss of P1 and P2 occurred within Vombatiformes after the divergence of thylacoleonids (with loss of P2 occurring independently within Thylacoleonidae), but before the split between Vombatomorphia and Phascolarctidae ( Table 3). ASR of molar type using the same Mk model indicates that bunolophodonty is ancestral for Vombatiformes, with subsequent acquisition of selenodonty by the Vombatomorphia+Phascolarctidae lineage, which was retained by phascolarctids and ilariids. After the divergence of ilariids, the remaining vombatomorphians evolved selenolophodonty, which was retained by wynyardiids and vombatoids (including Mukupirna), with diprotodontoids subsequently evolving fully lophodont molars (Table 3). This scenario for the evolution of different molar types within Vombatiformes is attractive in its simplicity, but we acknowledge that it is partially dependent on some very weakly supported relationships, specifically the position of Wynyardiidae and Ilariidae as successive sister taxa to Diprotodontoidea + Vombatoidea.

Estimated size of Mukupirna nambensis and body mass evolution within Vombatiformes.
Estimated body mass of Mukupirna using the "total skull length" regression equation of Myers 49 is 46 kg, which seems implausibly low given the size of the postcranium. We note that the largest specimen used by Myers 49 in calculating his regression equations was a 70 kg Macropus individual (species unspecified), and so use of these equations to estimate body masses of likely larger extinct taxa involves extrapolation beyond the data used to calculate them. The overall proportions of Mukupirna and several other extinct vombatiforms (e.g. diprotodontoids, thylacoleonids) also appear very different from the extant marsupial species used by Myers 49 to produce the regression equations. We have therefore used the postcranial regression equations presented by Richards et al. 13 , which were produced using a dataset of mammalian and non-mammalian taxa with body masses that collectively span from 51 g to 6.4 tonnes. These give body mass estimates for Mukupirna of 143 kg based on femoral circumference only, 160 kg based on combined humeral and femoral circumference, and 171 kg based on humeral circumference only. The humerus of Mukupirna is very robust, and so the estimates that incorporate humeral circumference might be inflated, as Richards et al. 13 also suggested for the vombatiform Palorchestes; nevertheless, it seems likely that Mukupirna exceeded 100 kg. This is compared to an average weight of 32 kg for the largest living vombatiform, the northern hairy-nosed wombat (Lasiorhinus kreffttii). www.nature.com/scientificreports www.nature.com/scientificreports/ Compared to estimated body masses for selected other late Oligocene vombatimorphians (see supplementary material), Mukupirna is much larger than the wynyardiid Muramura williamsi (16-20 kg), slightly larger than the diprototodontid Ngapakaldia bonythoni (~119 kg), but somewhat smaller than the ilariid Ilaria illumidens (~215 kg). Ancestral state reconstruction of body mass on the Bayesian majority rule consensus retaining compatible partitions with BPP < 0.5 (in which branch lengths are proportional to the estimated amount of change in the morphological characters used to infer the phylogeny) using StableTraits suggest a median body mass of 5.5 kg for the last common ancestor of vombatiforms, which is only slightly smaller than the modern koala Phascolarctos cinereus ( Table 2). The thylacoleonid Microleo attenboroughi from the early Miocene of Riversleigh World Heritage Area has not been included here due to its comparative incompleteness; however, it has an estimated body mass of 590 g 48 , and, if it is sister to all other known thylacoleonids (as found by Gillespie et al. 48 ), then it implies an even smaller ancestral body mass for Vombatiformes, possibly <1 kg. Our StableTraits analysis of body mass (Fig. 6) indicates independent evolution of very large (>100 kg) size at least six times within Vombatiformes. For the thylacoleonid Thylacoleo carnifex, we have used the mean of estimates based on postcranial measurements (= 57.2 kg 13 ); however, we note other studies have found that body mass of some T. carnifex individuals may have exceeded 100 kg 50,51 , in which case it would represent a seventh independent evolution of >100 kg body mass within Vombatiformes.

Discussion
Our phylogenetic results imply that postcranial digging adaptations, such as large EI and IFA, evolved in Vombatoidea in an ancestor that still retained a somewhat wynyardiid-like, selenolophodont molar dentition (as seen in Mukupirna and also the plesiomorphic vombatids Nimbavombatus and Rhizophascolonus), rather than the specialised, hypselodont molars characteristic of later vombatids. However, Mukupirna lacks a laterally extensive, flange-like deltopectoral crest, which is present in all vombatids known from associated postcranial material 17 . This modified deltopectoral crest creates an enclosed, tunnel-like fossa for the origin of the brachialis muscle, and is likely a specialised fossorial adaptation 17 . Absence of this feature in Mukupirna, together with its  Table 2. Epicondylar Index (=humeral epicondylar width/humeral length) and Index of Fossorial Ability (=olecranon length/(total ulnar length-olecranon length) for Mukupirna nambensis and selected other vombatiforms and other marsupials, plus ancestral state reconstructions for Vombatiformes and selected vombatiform subclades (see Table 1 and Figs. 5, 6). Ancestral state reconstructions (indicated by ASR) were inferred using StableTraits on the majority rule consensus with all compatible partitions of the post-burnin trees from a Bayesian analysis of our 99 character morphological matrix using the Mkv model; the median estimates and 95% confidence intervals (shown in brackets) from the StableTraits analyses are presented here. a Estimated (see Munson 46  www.nature.com/scientificreports www.nature.com/scientificreports/ large size (>100 kg) means that Mukupirna may not have been capable of the true burrowing behaviour of modern wombats 17,46,52 . Instead, it may have used scratch-digging to access subterranean food items, such as roots and tubers, as has also been proposed for Ilaria 46 and Rhizophascolonus 36 .
It has been suggested, based on the ecology and relatively close relationship of modern wombats and koalas, that vombatiforms (and also other diprotodontians) have been characterised by "long-term maintenance of ecological niche differentiation" 53 . However, the evidence from the fossil record of Vombatiformes clearly demonstrates that this is an artefact of the relictual nature of the three extant representatives: known fossil vombatiforms range in size from small (<5 kg) Oligo-Miocene phascolarctids 32 and thylacoleonids 48,54 , to the rhino-sized (> 2 tonne) Diprotodon from the Pleistocene, and collectively span a diverse range of morphologies and ecologies, including several that lack obvious modern analogues (at least in the Australian mammal fauna). The >100 kg Mukupirna, together with similarly sized contemporaneous taxa such as the ilariid Ilaria and the diprotodontid Ngapakaldia, demonstrates that multiple vombatiform lineages had already evolved very large (>100 kg) body size by the late Oligocene, and possibly considerably earlier. In total, our ancestral state reconstructions indicate at least six independent origins of body masses>100 kg within Vombatiformes, from an ancestor that is estimated to have been between 1 and 5.5 kg, the upper bound being slightly smaller than the mean mass of the living koala, Phascolarctos cinereus. In this respect, vombatiforms resemble another mammalian clade with relictual modern diversity but for which far more diverse fossil representatives are known, namely sloths 55 .  20 . All multistate characters representing putative morphoclines were specified as ordered.

Material and
Undated Bayesian analysis of the morphological dataset was carried out using MrBayes 3.2.7 56 , using the Lewis 57 Mk model with the assumption that only variable characters were scored (i.e. the Mkv model), and with rate heterogeneity between characters modelled using an eight category lognormal distribution 58 . The Bayesian www.nature.com/scientificreports www.nature.com/scientificreports/ analysis were run for 10 × 10 6 generations, using four independent runs of four chains (one cold and three heated chains, with the temperature of the heated chains set to the default value of 0.1), and sampling trees every 5000 generations. Tracer 1.7 59 was then used to identify an appropriate burn-in period for the trees saved for each run. The post-burn-in trees were then summarised using MrBayes in the form of a majority rule consensus that retains compatible partitions with BPP < 0.5, using the "contype = allcompat" comand. The matrix was analysed using maximum parsimony, in TNT version 1.5, with the tree search comprising an initial "new technology" Figure 6. StableTraits ancestral state reconstruction (ASR) of body mass estimates for Vombatiformes. The tree used for ASR is the majority rule consensus that retains compatible partitions with BPP < 0.5 from our undated Bayesian analysis (branching topology same as in Fig. 5), with branch lengths proportional to the total estimated amount of change in our morphological characters.  www.nature.com/scientificreports www.nature.com/scientificreports/ search with sectorial search, ratchet, drift and tree fusing that was run until the same minimum tree length was found 1000 times, followed by a "traditional" search within the trees save from the first stage, using the tree bisection resection (TBR) swapping algorithm. Multiple most parsimonious trees were combined using strict consensus. Support values were calculated using 2000 standard bootstrap replicates, implemented using a "traditional" search, with results output as absolute frequencies.
estimation of body mass. To estimate body mass of Mukupirna nambensis and for other fossil vombatiforms that lack published estimates but for which appropriate postcranial material was available, we used the regression equations presented by Richards et al. 13 , which are based on humeral circumference, femoral circumference, and combined humeral and femoral circumference. Where possible, we have preferred estimates based on these measurements rather than those based on craniodental measurements (e.g., using the regression equations of Myers 49 ), as it seems more likely that circumference of the limb bones will more accurately reflect body mass than will dimensions of the skull and teeth 60 . Where at least one humerus and one femur was available for a particular taxon, we used all three of the Richards et al. 13 equations, and calculated a mean value.
For taxa for which postcranial material was not available, we used the craniodental regressions of Myers 49 , using his "diprotodontians" dataset. For each taxon, we used up to four of the highest ranking (as measured by total rank) equations that could be calculated based on available material, incorporating the relevant smearing estimates, and then used these estimates to calculate a mean value.
Ancestral state reconstructions of continuous traits. For ancestral state reconstruction of continuous variables -namely Epicondylar Index (EI), Index of Fossorial Ability (IFA) and body mass -, we used the "stable" model implemented StableTraits 1.5 61 , which allows occasional "jumps" in trait values, and which is likely to outperform a standard Brownian motion model 61,62 . StableTraits requires a fully resolved phylogeny with branch lengths, and so we used a majority rule consensus that retains compatible partitions with BPP < 0.5 that resulted from our undated Bayesian analysis, in which branch length is proportional to the total estimated amount of change in our morphological characters. StableTraits cannot be used with zero-length branches, and so we increased the length of all such branches by a trivial amount (10 −6 ). For each StableTraits analysis, taxa that lacked data were deleted. Values for EI and IFA (which are ratios) were used "as is", but body mass was log10-transformed prior to analysis. Each StableTraits analysis was run using default settings: two independent runs of 1 × 10 6 generations, in 40 stages of 2.5 × 10 4 generations. The first 5 × 10 5 generations (i.e. 50%) were discarded as burn-in; the proportional scale reduction factor PRSF, 63 was <1.1 (indicating convergence) before this point.
Ancestral state reconstructions of discrete dental traits. We also used the majority rule consensus that retains compatible partitions with BPP < 0.5 from our undated Bayesian analysis, with branch lengths proportional to the total estimated amount of change in our morphological characters, to implement ancestral state reconstruction of the following discrete dental traits: presence or absence of P1, presence or absence of P2, and molar type. The first two traits were taken directly from the morphological dataset we used in our phylogenetic analysis (Characters 4-5), whilst the third reflects overall molar morphology, with the following states: tribosphenic, bunodont (i.e., distinct crests absent), bunolophodont (i.e., weakly developed lophs present), fully lophodont (i.e., tall, well-developed lophs present), selenodont, and selenolophodont (i.e., both selenodont and lophodont features present). Ancestral state reconstruction was done in Mesquite version 3.61 64 , using the Mk1 (symmetrical) model. nomenclatural acts. This published work and the nomenclatural acts it contains have been registered in ZooBank, the proposed online registration system for the International Code of Zoological Nomenclature (ICZN). The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix "http://zoobank.org/". The LSIDs for this publication are: urn:lsid:zoobank.org:act:7303DC9B-AD03-4BB1-985D-35F2BBE52EBE; urn:lsid:zoobank.org:act:6281C654-D711-459D-93E8-955449B645FD; urn:lsid:zoobank. org:act:CC66EA70-9914-4826-A491-09DBFAF4C00A.