First nimravid skull from Asia

Maofelis cantonensis gen. and sp. nov. is described based on a complete cranium from the middle-upper Eocene Youganwo Formation of Maoming Basin, Guangdong Province, China. The new taxon has characters diagnostic for Nimravidae such as a short cat-like skull, short palate, ventral surface of petrosal dorsal to that of basioccipital, serrations on the distal carina of canine, reduced anterior premolars, and absence of posterior molars (M2-3). It is plesiomorphic nimravid taxon similar to Nimravidae indet. from Quercy (France) in having the glenoid pedicle and mastoid process without ventral projections, a planar basicranium in which the lateral rim is not ventrally buttressed, and P1 present. The upper canine is less flattened than in other Nimravidae. Maofelis cantonensis gen. and sp. nov. exemplifies the earliest stage of development of sabertooth specialization characteristic of Nimravidae. This taxon, together with other middle-late Eocene nimravid records in South Asia, suggests origin and initial diversification of Nimravidae in Asia. We propose that this group dispersed to North America in the late Eocene and to Europe in the early Oligocene. The subsequent Oligocene diversification of Nimravidae took place in North America and Europe, while in Asia this group declined in the Oligocene, likely because of the earlier development of open habitats on that continent.

Holotype: Collection from the Maoming Basin in the School of Life Sciences, Sun Yat-sen University, Guangzhou, China (SYSU-M) 2, almost complete cranium (skull without mandible), with most of the dentition preserved.
Diagnosis: Maofelis is referred to Nimravidae based on the following combination of characters that is diagnostic for this group 1,2 : cat-like skull with shortened rostrum and mesocranium; palate short, does not extend posterior to the toothrow; walls of the basipharyngeal canal converged posteriorly; ventral surface of petrosal significantly dorsal to that of basioccipital; hypoglossal foramen separated from the posterior lacerate foramen; paraoccipital process moderately large, posteriorly projecting; incisors with reduced lingual cingula; serrations on the distal carina of canine; anterior premolars reduced; P4 without a parastyle; posterior molars (M2-3) absent.
Differential Diagnosis: Maofelis is similar to Nimravidae indet. from Quercy (Muséum d'Histoire naturelle de Montauban, France (MA-PHQ) 348) and differs from the remaining nimravids in having no ventral projection of the glenoid pedicle or mastoid process, a planar basicranium with the lateral rim not or weakly buttressed ventrally, and P1 present (P1 is variably present in Nimravus and Eofelis). Maofelis differs from all known nimravids except Nimravidae indet. from Quercy (MA-PHQ 348) and Dinaelurus in having upper canines that are only slightly flattened (length/width ratio ~1.3); from Nanosmilus, Hoplophoneus, and Eusmilus by having sutural contact between the jugal and the lacrimal; from Nimravus, Dinictis, Hoplophoneus, and Eusmilus by the presence of a strongly developed sagittal crest; from Hoplophoneus and Eusmilus by the presence of a sagittal crest that is rectilinear (not concave) in lateral view; from Nimravus and Dinaelurus by the lack of a fossa on the ventromedial face of the zygomatic arch, below the postorbital process; from Dinaelurus, Nanosmilus, Hoplophoneus, and Eusmilus by the presence of a double-rooted P2; from Dinaelurus, Nimravus, Pogonodon, Dinictis, Nanosmilus, Hoplophoneus, and Eusmilus by the presence of a three-rooted P3; from Hoplophoneus and Eusmilus by the P4 protocone projecting strongly mesiolingually, with its mesial end at or anterior to the mesial end of the paracone; from Dinaelurus, Nimravus, Nanosmilus, Hoplophoneus, and Eusmilus by the presence of a transversely elongated M1, in which the protocone is prominent and widely separated from the paracone; from Dinaelurus, Nimravus, and Eusmilus by the M1 crown not being vertically orientated.
Comments: Maofelis is most similar to the primitive nimravid represented by a skull from Quercy (MA-PHQ 348) which has been identified originally as cf. Eofelis sp. 28 . This identification is not supported by our phylogenetic analysis, which did not place MA-PHQ 348 with Eofelis (see section on the phylogenetic analysis). MA-PHQ  348 is identified here as Nimravidae indet. Its similarity with Maofelis is based on plesiomorphic characters and does not suggest a close relationships between these taxa.
Description. The rostral part of the skull is about 65% of the condylobasal skull length ( Fig. 2; Supplementary   Fig. 1). The braincase is about 43% of the condylobasal skull length. The rest of the skull is occupied by the orbitotemporal region. In lateral view, the dorsal margin of the skull is convex, with its highest point at the postorbital constriction. The greatest width across the postcanine teeth is 1.25 times greater than the greatest width across the braincase (posterior to the external acoustic meatus).
The external nasal aperture is 1.4 times taller than wide. In dorsal view, the lateral sides of the rostrum are subparallel, convex at the canine alveolus and concave between the canine alveolus and the anterior end of the zygomatic arch. The premaxillae project anteriorly beyond the anterior end of the nasals. The anterior end of the nasals is situated at the level of the middle of canine alveolus. The nasals are almost horizontal in lateral view and convex. The infraorbital foramen is large and placed at the root of the zygomatic process of the maxilla above the mesial root of P3.
The bony surface of the palate is poorly preserved. The maxillary-premaxillary suture is only partially preserved. The incisive foramina are relatively short and round, confined to the level of the canine distal half. Anteriorly, there are distinct parallel grooves leading to these foramina. The palatine extends anteriorly to the level of P3 protoconal swelling.
The zygomatic arch is gently curving, tall and narrow. In lateral view the zygomatic arch is slightly arching dorsally and placed along the ventral margin of the skull.
The orbital fossa is about half the length of, and lower than, the temporal fossa. The orbit is mostly open posteriorly, delimited by short postorbital processes of the frontal and zygomatic arch. The jugal contacts the lacrimal. The lacrimal foramen is posteriorly facing and not visible in lateral view.
The postorbital constriction is located at the middle of the temporal fossa, well posterior to the frontal postorbital process. The width at the postorbital constriction is about 50% of the width at the frontal postorbital processes (reconstructed).
There are prominent sagittal and nuchal crests on the braincase. The sagittal crest is very high, with a slightly convex dorsal margin. There are weak temporal ridges between the postorbital processes and the sagittal crest. The temporal ridges converge to form the sagittal crest posterior to the postorbital constriction. The basicranium region is poorly preserved (see Supplementary Fig. 2). There are no bullar elements attached to the skull. The petrosal is placed in a deep depression and its ventral surface is distinctly dorsal to the ventral surface of the adjacent basisphenoid and basioccipital. There is no ventral process of the promontorium. The paroccipital process is short and placed just posterior to the mastoid region. The posterior lacerate foramen is anteromedial to the paroccipital process and confluent with the petrobasilar canal. A somewhat smaller hypoglossal foramen is posteromedial to the latter foramen. The posterior opening of the alisphenoid canal and the foramen ovale open in a common groove bordered by a ventromedial rim, just medial to the glenoid fossa. There is no postglenoid pedicle of the squamosal. The postglenoid foramen cannot be identified. There is no median ridge on the ventral side of the basioccipital. The basisphenoid and basioccipital have no lateral buttressing, suggesting that the bullar elements were not strongly attached to the basicranium, if present.
The choanae are similar in width to the foramen magnum. The pterygoid processes are converge slightly posteriorly.
The nuchal crest overhangs the occiput and extends posteriorly beyond the occipital condyles (Fig. 3). The occiput faces posteroventrally. The foramen magnum is wider than high and extends dorsally slightly beyond the dorsal margin of the occipital condyles.
The incisors are in gently curving arch and closely appressed (Fig. 4). There is a rather complete left I3 and badly damaged left I2 and right I2-3. The I1, judging from its alveolus, was somewhat smaller than I2. The I3 is about twice as large as I2. The crown of I3 is conical. There is a sizable diastema between I3 and C1.  The canine (length 16.6 mm, width 12.7 mm, measured at the enamel-dentine junction) is dagger-like, ventrally projecting, and moderately constricted laterally (length/width ratio is 1.31). The canine cross-section is asymmetrical, with the mesial carina placed at the mesiolingual corner, an almost flat lingual margin between the mesial and distal carinae, and a strongly convex lateral margin. Although the canine surface is poorly preserved, the small serrations along the distal carina can be recognized (supplementary Fig. 3). The right canine alveolus length is 18.3 mm, width is 14.3 mm. The postcanine teeth diverge posteriorly.
The P1 was small and one-rooted (Fig. 4). Its alveolus length is 3.4 mm (left) and 4.6 mm (right), and its alveolus width is 4.2 mm (left) and 3.5 mm (right). The P2 was two-rooted, with alveolus length 9.3 mm (left) and 9.8 mm (right) and alveolus width 5.8 mm (left) and 5.3 mm (right). The P3 is three-rooted (length 16.2 mm, 16.5 mm; width 11.0 mm, 10.8 mm for the left and right tooth respectively). The crown of P3 is submolariform, with a large paracone occupying half of the labial side and a blade-like metacone. The paracone and metacone are separated by a shallow notch. There is a large protoconal swelling with a small protocone. There is a distinct mesial cingulum but no parastyle. The P4 is a typical carnassial tooth with a large paracone, blade-like metacone, narrow talon, and small protocone placed mesiolingual to the paracone (length 19. There is a single molar, M1, preserved on the right side (length 6.4 mm; width 15.9 mm; Fig. 4). It has a typical tooth shape of the ultimate eutherian molar [43][44][45] , with a reduced metacone and metastylar blade and labially directed parastylar lobe. The paracone is relatively large. The talon has a distinct basin and a small protocone. The lingual margin of the M1 talon is placed more labial compared with the P4 talon. The M1 is two-rooted, with the lingual root larger.
For the skull measurements see Supplementary Fig. 4

Discussion
Saber-tooth specialization of Maofelis. The saber-tooth specialization of Maofelis is less pronounced compared with other known nimravid taxa. In Maofelis this specialization is expressed by a large and caniniform I3 and a relatively large upper canine which is only weakly laterally compressed (length/width ratio is 1.3) and has serrated distal carina. Most nimravid taxa have more compressed upper canines, with length/width ratio above 1.5. In most specialized taxa the upper canines are very flattened, with length/width ratio 2.7-3.2 2 . The canine cross-section is similar to that of Pseudaelurus and intermediate between those of non-sabertooth and sabertooth felids 15 : Figs 1-3. The glenoid, mastoid, and paroccipital processes are all weakly developed in Miaofelis. The strong development of these processes is correlated with the ventral displacement of the mandibular articulation, which allows a wider gape needed for long upper canines, and development of powerful neck musculature 2,46 . The zygomatic arch of Maofelis is nearly horizontal in lateral view, while in derived nimravids it is strongly arched dorsally.
Phylogenetic position of Maofelis. Our phylogeny of Nimravidae ( Fig. 5 and Supplementary Fig. 5) is generally similar to that presented by Peigné 2 , differing in having greater resolution within the Hoplophoneus-Eusmilus clade, a more basal position for Dinailurictis, and absence of a sister group relationship between Dinictis and Pogonodon. Hoplophoneus is paraphyletic with regard to Eusmilus. Maofelis and Nimravidae indet. from Quercy (MA-PHQ 348) are the most basal nimravids, forming a polytomy with a clade containing all other nimravid taxa.
Evolutionary history of Nimravidae. Nimravids represent the first radiation of cat-like hypercarnivores.
They inhabited closed forest habitats and their decline in the late Oligocene in North America has been associated with the spread of grassland ecosystems 25 51 . All known vertebrate remains from the Maoming Basin were found in dark brown oil shales in the upper part of the Youganwo Formation, which were deposited in lacustrine conditions. The Youganwo Formation has been dated based on magnetostratigraphy and palynomorph assemblages as upper Eocene 52,53 , or middle-upper Eocene 54 . The vertebrate assemblage of the Youganwo Formation includes cyprinid fishes, carettochelyid, adocid, trionychid, and geoemydid turtles, tomistomine and alligatorid crocodiles, and chalicotherioid and amynodontid perissodactyls 55-65 . Phylogenetic analysis. For the phylogenetic analysis we used the data matrix presented by Peigné 2 that included all nimravid genera. We changed the coding of character 24 (proportion and size of the protocone on M1) for MA-PHQ 348, a skull of primitive nimravid from Quercy, from "1" to "?" because M1 is absent on this specimen. Maofelis can be coded for 16 of the 33 characters (48.5%). The coding for Maofelis is the following: 0??0?? ?001000???000101010????????. Ten thousands repetitions of the parsimony ratchet (island hopper) algorithm of NONA version 2.0 66 run with Winclada version 1.00.08 interface 67 produced six most parsimonious trees with a length of 87 steps, a consistency index of 0.72, and a retention index of 0.89. The strict consensus of these six most parsimonious trees is present in Fig. 5 and supplementary Fig. 5.