A New Large Hyainailourine from the Bartonian of Europe and Its Bearings on the Evolution and Ecology of Massive Hyaenodonts (Mammalia)

We describe a new large-sized species of hypercarnivorous hyainailourine–Kerberos langebadreae gen. & sp. nov.–from the Bartonian (MP16) locality of Montespieu (Tarn, France). These specimens consist of a skull, two hemimandibles and several hind limb elements (fibula, astragalus, calcaneum, metatarsals, and phalanges). Size estimates suggest K. langebadreae may have weighed up to 140 kg, revealing this species as the largest carnivorous mammal in Europe at that time. Besides its very large size, K. langebadreae possesses an interesting combination of primitive and derived features. The distinctive skull morphology of K. langebadreae reflects a powerful bite force. The postcranial elements, which are rarely associated with hyainailourine specimens, indicate an animal capable of a plantigrade stance and adapted for terrestrial locomotion. We performed the first phylogenetic analysis of hyainailourines to determine the systematic position of K. langebadreae and to understand the evolution of the group that includes other massive carnivores. The analysis demonstrates that Hemipsalodon, a North American taxon, is a hyainailourine and is closely related to European Paroxyaena. Based on this analysis we hypothesize the biogeographic history of the Hyainailourinae. The group appeared in Africa with a first migration to Europe during the Bartonian that likely included the ancestors of Kerberos, Paroxyaena and Hemipsalodon, which further dispersed into North America at this time. We propose that the hyainailourines dispersed into Europe also during the Priabonian. These migrants have no ecological equivalent in Europe during these intervals and likely did not conflict with the endemic hyaenodont proviverrines. The discovery of K. langebadreae shows that large body size appears early in the evolution of hyainailourines. Surprisingly, the late Miocene Hyainailouros shares a more recent common ancestor with small-bodied hyainailourines (below 15 kg). Finally, our study supports a close relationship between the Hyainailourinae and Apterodontinae and we propose the new clade: Hyainailouridae.

Introduction all the taxa and characters, while Leakitherium is excluded in the second and third analyses because it is very poorly known.
The genus Leakitherium is only known from two maxillary fragments that display P 4 , M 1 and M 2 [3]. Morales et al. [41] referred a fragment of an M 2 to the genus. The taxonomic status of Leakitherium among hyainailourines is unresolved [41]. They hypothesized that Leakitherium is a synonym for Isohyaenodon andrewsi. In this analysis Leakitherium is closely related to Miocene hyainailourines such as Megistotherium and Isohyaenodon, but the hypothesis of Morales et al. [41] is not supported here.
Finally, the first character, which concerns body mass (S2 Text), is excluded in the third analysis because it could correspond to ecologic convergence rather than reflecting a phylogenetic signal.
After each analysis, each node of the strict consensus tree was assigned a Bremer support calculated with TNT for ten supplementary steps.
Notes. As in other specialized carnivorous mammals, hyaenodonts possess secant teeth. These meat-slicing teeth, also called carnassial teeth, are located in M 1-2 and M 2-3 position in Hyaenodonta [45]. Hyaenodonta was traditionally placed in the extinct order Creodonta together with the Oxyaenidae. Some workers have suggested a diphyletic origin for Creodonta [1,38,[49][50][51]. While the separate origins of Oxyaenidae and Hyaenodontidae have not been established in a cladistic framework, Solé [51] proposed to raise these two families to the ordinal level-Oxyaenodonta and Hyaenodonta, and not Hyaenodontida as mistakenly used by several authors [6,[51][52][53]-to highlight the possible separate evolutionary origins of these two extinct carnivorous mammalian lineages. It is worth mentioning that Spaulding et al. [54] recovered a monophyletic Creodonta, but their phylogenetic analyses, which addressed the relationships of Cetacea among mammals, only included four creodonts (three hyaenodonts and only one oxyaenid).
The systematic position of Koholiinae among Hyaenodonta is presently uncertain because cranial material is unknown (see below).
Discussion of the relationships within Hyaenodonta. Polly [38] established that hypercarnivory (characterized by reduction of the metaconids, simplification of the talonid, extension of the metastyle, reduction of the protocone, among other features) arose at least twice within hyaenodonts: among Hyainailourinae (Pterodontinae sensu Polly) and Hyaenodontinae. The distinction between the two subfamilies includes several cranial features. Polly noted that hyainailourines are characterized by a nuchal crest that does not extend laterally toward the mastoid processes and a circular subarcuate fossa on the petrosal. Hyaenodontines are characterized by a nuchal crest that extends toward the mastoid process, a robust bridge over the foramen stylomastoid primitivum, the absence of a bony ridge dividing the posterior petrosal sinus from the foramen stylomastoid primitivum, the presence of an inflated posterior petrosal sinus, and the presence of a horse-shoe shaped subarcuate fossa.
The cranial morphology of the hyainailourines is similar to that of the European and African hyaenodont subfamily Apterodontinae, notably in the short lateral extension of the nuchal crest (Fig 3, Feature 6; Table 1). Moreover, based on the present study, we can add the following features that seem to unite Hyainailouridae (Hyainailourinae+Apterodontinae): presence of a preglenoid crest (Figs 4 and 5, Feature 7; Table 1), early obliteration of the suture between the frontal and parietal bones (Figs 3 and 5, Feature 2; Table 1), anteroposteriorly extended suture between the jugal and squamosal (Fig 5, Feature 11; Table 1), a lateral expansion of the squamosal posterior to the zygomatic arch (Fig 3, Feature 4; Table 1), and large occipital condyles (Figs 4 and 5, Feature 9; Table 1). We can also note that apterodontines and hyainailourines often display more than two mental foramina (Fig 5, Feature 13; Table 1), while hyaenodontids very rarely display more than two, and have a weak ventral concavity anterior to the angular process ( Fig 5, Feature 14; Table 1). The younger representatives of the clade (e.g., Hyainailouros) are characterized by a transverse elongation of the mastoid process (Fig 4, Feature 8; Table 1), rather than an anteroposterior elongation (Figs 3 and 4, Feature 5; Table 1) as observed in the earliest species in the order Hyaenodonta. Finally, the narrowest part of the braincase is immediately behind the postorbital processes in Hyaenodontidae, while this constriction is more posterior in Hyainailouridae (at least at the middle of the parietals) (Fig 3,  Feature 3; Table 1).
Some representatives of hyainailourines have lost the P 1 [9]. This feature is one of the main characteristics of Koholiinae [1]. Koholiines and hyainailourines share the latter feature, as well as the presence of pronounced secant teeth and extensive radiations in Africa. However, the koholiines differ from hyainailourines in featuring a robust angular process that is preceded anteriorly by a concavity (Fig 5, Feature 14; Table 1). It is thus presently difficult to refer the koholiines to either Hyainailouridae or Hyaenodontidae.
The cranial morphology of Hyaenodontidae is on the whole similar to Hyaenodontinae in Polly [38]. The clade is characterized by the morphology of the nuchal crest, which extends toward the mastoid process (Figs 3 and 5, Feature 6; Table 1). We can add several features not described previously to distinguish Hyaenodontidae from Hyainailouridae. Hyaenodontids generally possess: a well-developed postorbital process (Figs 3 and 5, Feature 1; Table 1), a short extension of the jugal/squamosal suture anteroposteriorly (Fig 5, Feature 11; Table 1),   Table 1), a process at the junction between the maxillary and the jugal (Fig 5, Feature 10; Table 1), and a small occipital condyle (Figs 4 and 5, Feature 9; Table 1). Unlike Hyainailouridae, Hyaenodontidae exhibit a clear suture between the parietal and frontal (Figs 3 and 5, Feature 2; Table 1) and the suture is located close to the anterior constriction of the braincase (Fig 3, Feature 3; Table 1). Hyaenodontids all possess a single-or double-rooted p1 and very rarely display more than two mental foramina ( Fig 5, Feature 13; Table 1). Hyaenodontid pterygoids are generally less developed ventrally and posteriorly than hyainailourid pterygoids (Fig 5, Feature 12; Table 1). In almost all hyaenodontids the mandible possesses a concavity on the ventral margin anterior to the angular process and the latter is generally robust and projects more ventrally in Hyaenodontidae (Fig 5, Feature 14; Table 1). We tentatively refer the Teratodontinae as recently defined by Solé et al. [6] to Hyaenodontidae based on the skull of Dissopsalis described by Colbert [59] and mandibles available for the teratodontine genera. The fragmentary skull of Dissopsalis seems to display developed postorbital process (Figs 3 and 5, Feature 1; Table 1), constriction before the braincase (Fig 3, Feature  3; Table 1), and a distally elongated mastoido-paroccipital apophysis (Figs 3 and 4, Feature 5; Table 1). Colbert [59] noted that Dissopsalis possesses a developed preglenoid crest (Figs 4 and 5, Feature 7; Table 1), a feature of Hyainailouridae; however, based on his drawings, we assess that the morphology of the mandibular fossa is similar to that of Hyaenodontidae rather than to that of Hyainailouridae. Finally, teratodontines display a concavity anterior to a robust angular process (Fig 5, Feature 14; Table 1) as in other Hyaenodontidae.
The Indohyaenodontidae seem to be endemic in South (India, Pakistan) and Southeast (Myanmar) of Asia [6]. Unfortunately, only few cranial remains are known for this subfamily [60][61]. Cranial and dental specimens show typical features of Hyaenodontidae such as the presence of a concavity anterior to the angular process ( Fig 5, Feature 14 Table 1). We therefore include this subfamily in Hyaenodontidae.
Machaeroidinae has been often considered to be closely related to Limnocyoninae [62]. Machaeroidinae and Limnocyoninae notably exhibit an extension of the nuchal crest towards the mastoid, as seen in other species in Hyaenodontidae. Species in both subfamilies have moreover lost M 3 and M 3 (except in their oldest representative Prolimnocyon), supporting a close relationship between these families and the inclusion of both Machaeroidinae and Limnocyoninae in Hyaenodonta. Several authors (e.g. Muizon & Lange-Badré [63]) however referred the Machaeroidinae to Oxyaenodonta; Zack [64] recently proposed arguments based on postcranial elements to support this hypothesis.
The two cranial morphotypes described and discussed here may be useful for clarifying the relationships within Hyaenodonta, though these features await rigorous evaluation in a cladistic analysis to establish plesiomorphic and apomorphic states of characters. Because the two morphotypes imply distinct recruitment of the cranial muscles and both morphotypes seem to be a combination of primitive and derived features, we hypothesize that these two cranial morphotypes appeared early in the radiation of Hyaenodonta. Based on the presence of hyaenodontids in the Paleocene of Asia, we place the radiation of Hyaenodonta and divergence of Hyaenodontidae and Hyainailouridae during the Selandian, if not earlier [2].
Concerning the dispersals of these two families, it is worth mentioning that apterodontines dispersed into Europe and hyainailourines dispersed into North America, Europe and Asia [9,[65][66]. Among Hyaenodontidae, only Teratodontinae and Indohyaenodontinae are known exclusively from Gondwanan landmasses (Africa and South/South-East of Asia). If Hyaenodontidae appeared in Laurasia, these two subfamilies may have dispersed into Africa during the earliest Eocene.
Hyaenodontidae Leidy, 1869 emend. [46] Diagnosis. The members of this family share a cranial pattern characterized by the presence of a visible suture between the parietal and frontal and a well-developed postorbital process, a constriction anterior to the braincase, pterygoid generally shorter than in Hyainailouridae, a short extension of the jugal/squamosal suture anteroposteriorly, the absence of preglenoid crest, a distally elongated mastoido-paroccipital apophysis, a nuchal crest extending to the mastoid process, small occipital condyles, the presence of only two mental foramina, and a concavity anterior to the angular process (less pronounced in hyaenodontines). The P 1 (single-or double-rooted) is usually present.
Note. The systematic position of the saber-toothed Machaeroidinae among Hyaenodonta or Oxyaenodonta is presently uncertain (see Gunnell [45]). Hyainailouridae Pilgrim, 1832 emend. [47] Diagnosis. The members of this family share a cranial pattern characterized by a massive skull, the absence of visible suture between the parietal and frontal, a weak postorbital process, a pterygoid generally extended ventrally and distally, an anteroposteriorly extended suture of the jugal/squamosal, the presence of a preglenoid crest, a lateral expansion of the squamosal posterior to the zygomatic arch, a transversally expanded mastoid process (in later forms), a nuchal crest that does not extend laterally to mastoid process, large occipital condyles, and the presence two or more mental foramina. The P 1 , if present, is small (single-rooted).
Relationships between the Hyainailourinae and Apterodontinae. Hyainailourinae and Apterodontinae are diverse during the Paleogene in Africa [3,52,[69][70]. The two groups are also recorded in Europe: the Hyainailourinae are recorded in the Late Eocene and Early to Middle Miocene, while the Apterodotinae are known in the Early Oligocene [9,33,66,71]. Hyainailourinae is also known in Early to Middle Miocene of Asia [47].
Recent discoveries in North Africa place the youngest possible origin of Hyainailourinae in the late Early or early Middle Eocene [6] and the youngest possible origin of Apterodontinae in the late Middle Eocene [52], implying an Afro-Arabian origin for these subfamilies. The presence of Hyainailourinae and Apterodontinae in Eurasia is likely a consequence of Northern immigrations (see below).
Apterodontinae is a peculiar group among Hyaenodonta [37,38,52,57]. Apterodontines are characterized by the loss of the metaconid, the reduction of the paraconid, and retention of a prominent talonid on the lower molars where most hyaenodonts with reduced metaconids have also reduced the talonid and increased the relative size of the paraconid. Based on postcranial material, Grohé et al. [52] proposed semi-aquatic habits for Apterodon. They suggested that the unusual dentition of Apterodon may have been related to a semi-aquatic lifestyle with simplified and tall teeth, employed for catching aquatic prey such as 'fish' and/or 'invertebrates' [52].
In contrast to apterodontines, large hyainailourines were likely hunters and scavengers and were probably ecologically analogous to the extant hyenas and the extinct borophagines; the latter however differ from large hyainailourines in exhibiting cursorial adaptations (see [72] and [73]).
The possibility of a close relationship between apterodontines and hyainailourines based on dental and cranial characters has been suggested before [6,38,52,61]. Here we explicitly identify numerous cranial features (e.g., absence of visible suture between the parietal and frontal, weak postorbital process, presence of a preglenoid crest, large occipital condyles) that group Apterodontinae and Hyainailourinae within the family Hyainailouridae (see above). The strongest similarities between apterodontines and hyainailourines lie in the posterior region of the skull related to the characteristic rearrangement of cervical musculature first described by Polly [38] in Hyainailourinae (Pterodontinae in Polly [38]).
The skull of Apterodon is anteroposteriorly elongate. The oldest hyainailourines such as Pterodon and Kerberos also exhibit anteroposteriorly elongate skulls, while others, such as the later hyainailourines Paroxyaena, Hemipsalodon, and Hyainailouros, differ from Apterodon and early hyainailoruines in featuring anteroposteriorly short rostral and basicranial regions. This suggests that Kerberos and Apterodon possess a primitive overall cranial morphology for Hyainailouridae.
Apterodon differs from hyainailourines in several features, which include a more robust zygomatic arch. This feature provides greater surface area for the origin of larger masseter muscles in apterodontines. On the mandible, the masseteric fossa is shallow but anteroposteriorly elongate, providing a long insertion for the masseter relative to the masseteric fossa in Kerberos and Pterodon, which further supports the important role of the masseter muscles in apterodontines.
Apterodon also differs from hyainailourines in the arrangement of the upper tooth rows that are close to parallel along the palate while nearly all other hyaenodonts have tooth rows that are distinctively diverging posteriorly. This parallel alignment may be related to semi-aquatic feeding as well. The facial morphology of apterodontines also differs from hyainailourines that have a strong narrowing of the facial morphology posterior to the canines [8,9]. Besides the differences in the masseteric fossa morphology, the mandibles of apterodontines and hyainailourines are generally similar. It is interesting to note that Apterodon macrognathus and Kerberos share the presence of a dorsally inflected, thin and pointed angular process, and an extended coronoid crest.
Concerning the postcranium, the only elements that can be compared between Apterontinae and Hyainailourinae are the distal part of the fibula, the astragalus, and the calcaneum. The distal part of the fibula is very similar in these two groups. However, there are differences in the tarsal bones. The tuber calcaneus of Kerberos and Hyainailouros is shorter and wider than that of Apterodon and the ectal facet of the hyainailourines, though poorly preserved on the specimen referred to Kerberos, seems to be shorter than the ectal facet of Apterodon. The neck of the astragalus of Kerberos and Hyainailouros is also shorter and narrower than that of Apterodon, though this difference in morphology may be related to the larger body size of the hyainailourine species. Apterodoninae and Hyainailourinae share the shallow groove of the tibial facet. This feature suggests that both groups maintained a plantigrade foot posture, though this morphology seems to be primitive in Hyaenodonta. The differences suggest a more weight-bearing mode of life in Kerberos, while the adaptations of Apterodon may suggest adaptations towards semi-aquatic mode of life.
In conclusion, apterodontines and primitive hyainailourines share a characteristic cranial construction. Broad attachments for nuchal musculature could have been adaptive for scavenging/hunting in terrestrial and aquatic environments, but this hypothesis needs to be tested in the future. However, numerous differences are present in the dentition and postcranium, and we advocate retaining separate subfamily designations for these lineages.
Hyainailourinae Pilgrim, 1932 [47] Emended diagnosis (emended after Holroyd [69]). Hyainailouridae with high and secant paraconid, connate metacone and paracone on M 1 -M 2 , weak to absent P 3 lingual cingulum, P 4 lacking continuous lingual cingulum, relatively large anterior keels on lower molars, M 3 talonid reduced relative to that of M 1 -M 2 , lower molar protoconids and paraconids subequal in length, facial region constricted lateromedially at P 2 and abruptly expanded caudal to P 4 , infraorbital process poorly developed, massive zygomatic arch, anterior margin of choanae located at the level of M 3 , and circular subarcuate fossa present on petrosal.
However, agreement on which species should be assigned to different hyainailourine genera is not firmly established. We follow Morlo et al. [78] [3]) and Isohyaenodon pilgrimi Savage, 1965 [3]; however Leakitherium is not synonymized with Isohyaenodon andrewsi herein. Moreover, our analysis shows that 'Pterodon' might not be a natural group but represents a polyphyletic genus.
Distribution. Africa and Europe; Bartonian (Eocene) to Serravallian (Miocene). Diagnosis. This large-sized hyainailourine displays a particular combination of primitive features, namely an anteriorly narrow mandible and a small diastema between P 1 and P 2 , and derived features, namely a reduced parastylar area on upper molars, a large protocone on P 3 , and premaxilla narrow throughout its whole length. It differs from Paroxyaena by having a more closely appressed paracone and metacone, and more reduced protocone on upper molars, and a shorter and narrower talonid on the lower molars. Kerberos differs from Pterodon dasyuroides by possessing larger P 1 and P 1 , and from Parapterodon by retaining the P 1 , exhibiting a more reduced parastylar area, and a more reduced protocone on upper molars.
Etymology. Dedicated to Dr. Lange-Badré, who greatly improved our knowledge of Eocene carnivorous mammals.
Type horizon and age. Measurements. Tables 2 and 3.
Description of the skull, mandible and dentition. The skull of the holotype is distorted transversally but is almost complete (Fig 7). It is also riddled with postmortem cracks and breaks, which partly obliterate the sutures (see Fig 13 for drawings of the skull with indications of the sutures). The auditory region, which is poorly preserved, will not be described thoroughly. The taxa chosen for comparison assessed from original specimens housed in the MNHN and literature represent the different subfamilies of Hyaenodonta (i.e., the Proviverrinae, Hyainailourinae, Hyaenodontinae, and Limnocyoninae), which are known from the Bartonian, Priabonian and Oligocene.
The most striking feature of the skull is its tremendous size compared to contemporaneous European proviverrines. The condylobasal length (~35 cm) approaches that of a female Ursus arctos [82]. The rostrum is very short, even shorter than that of the brevirostrate Hyaenodon brachyrhynchus. The stoutness of the rostrum contrasts with the relatively long ethmoid region. Even though the basicranium is crushed transversally, it seems to have been mediolaterally extensive as in Pterodon dasyuroides.
Premaxilla (Fig 13, Premax.). In lateral view (Fig 7), the opening of the nasal cavity forms an angle of approximately 45°with the anteroposterior lengthening of the skull, which approaches the condition of H. brachyrhynchus. The nasal aperture forms a more opened angle in the proviverrine Cynohyaenodon cayluxi and a slightly sharper angle in P. dasyuroides. The premaxilla differs from that of other hyaenodonts as the ascending ramus is narrow along its entire length. The posterior portion of the premaxilla is not extensive, reaching just posterior to the dorsal edge of the opening of the nasal cavity. The general hyaenodont condition is an extended premaxilla that reaches more posterior to the nasal aperture. The incisive foramina (Fig 13, Inc. F.) are anteroposteriorly shorter than those of P. dasyuroides.
Maxilla (Fig 13, Max.). Dorsal to the canine, the maxilla shows a weak anteroposterior convexity surrounding the long root of the tooth. At the level of P 2 , the maxillae are conspicuously constricted transversally. The constriction extends to the palate ventrally. Posteriorly the lateral borders of the maxillae flare, reaching more than twice the width of that of the anterior constriction. This morphology is similar to the condition observed in P. dasyuroides. The anterior opening of the infraorbital canal (Fig 13, Infr. F.) is at the level of the mesial root of P 4 , as in P. dasyuroides, but unlike the more anterior condition at the level of the interradicular space of P 3 seen in Hyaenodon [9]. Anterior to the infraorbital canal there is a shallow and round concavity, as described in C. cayluxi [9]. Superior to M 1 and M 2 is a well-marked fossa for the origin of nasolabialis. The inferior border of the orbit is horizontal, as in C. cayluxi. The posterior margin of the orbit does not slope dorsally as in P. dasyuroides. The orbit is relatively open on its dorsal and ventral margins, as in C. cayluxi, and differing from that of P. dasyuroides and H. brachyrhynchus.
Nasal (Fig 13, Na.). The nasal is a narrow and slightly triangular bone in dorsal view. Its posterior process terminates posterior to the anterior border of the orbit. Similarly long nasals are found in P. dasyuroides, contrasting with the shorter nasals of the proviverrines Quercytherium and C. cayluxi, the hyaenodontine H. brachyrhynchus, and the limnocyonine Thereutherium. There is no lateral expansion of the nasals directed toward the lacrimal as can be found in P. dasyuroides or H. brachyrhynchus. The nasal is mediolaterally convex at its anterior end, and flattens posteriorly.
Lacrimal (Fig 13, La.). The lacrimal is a large bone with a prominent anterior excursion onto the face beyond the medial margin of the orbit. The dorsal suture is on the dorsal border of the orbit, while the ventral suture extends onto the anterior margin of the zygomatic arch at the ventral border of the orbit. Unfortunately, the posterior border of the lacrimal is obscured by the fractures. Jugal (Fig 13, J.). The fossa for nasolabialis described on the maxilla extends on the anterior portion of the jugal. It is weakly concave ventrally and dorsally. The jugal is almost straight with little lateral flare. Its lateroventral border is a rugose surface. This rugosity indicates a robust origin for the superficial layer of the masseter muscle, and suggests that the taxon was capable of powerful adduction of the mandible [9]. The suture with the squamosal gently slopes posteroventrally and the jugal almost reaches the preglenoid process (  otic region is severely crushed and hence hardly describable. The anteroposteriorly short promontorium is preserved and is isolated from the basioccipital by a large carotid foramen. A similar condition is found in P. dasyuroides. Palatine (Fig 13, Pal.). The anterior border of the choana (posteriormost ventral contact between the palatines) almost reaches the level of the posterior end of the teeth row. Among Hyaenodonta, this feature is shared with the proviverrine Allopterodon and the limnocyonine Thinocyon. The position of the choana contrasts with the fused palatines in P. dasyuroides and Apterodon macrognathus, which finally open to the choana at the mid-point of the orbit. As in P. dasyuroides, there is a well-developed torus palatinus (Fig 13, To. Pal.) at the level of M 3 . Anterior to M 3 , the palatine is deeply excavated, for the occlusion with M 3 . The palatines are as mesiodistally extended as they are in P. dasyuroides.
Alisphenoid (Fig 13, Al.). The morphology of the alisphenoid is difficult to define because of the numerous fractures in the specimen. However, the suture between the alisphenoid and palatine is clearly preserved at the posterior edge of the nasopalatine structure. A single foramen is preserved in the orbital region and we tentatively identify this as the optic foramen ( Fig  13, Op. F.). Basisphenoid (Fig 13, Bas.). The contacts between the alisphenoid, basisphenoid and squamosal are situated close to the foramen oval (Fig 13, F. Ov.). The basisphenoid is almost rectangular in ventral view ( Fig 13B). The contact with the pterygoid bones is obscured by matrix.
Mastoid process (Fig 13, Mas. Pr.). The mastoid process is well preserved on the skull of Kerberos. However, the structures that contribute to its composition are difficult to distinguish. We tentatively delimitate its three components: the retrotympanic process of the squamosal, the mastoid apophysis (Mas. Ap.), and the paroccipital apophysis (Par. Ap.). The paroccipital apophysis extends distally.
Occipital region (Fig 13, Oc.). The supraoccipital, exoccipital, and basioccipital bones are all thoroughly fused to each other. The lambdoidal crests are incomplete and do not extend to the mastoid process, a feature characteristic of Hyainailouridae (see above). Instead, the lambdoidal crests converge ventrally, forming a concave, oval posterior surface above the foramen magnum (Fig 8). They are directed ventrally toward the foramen magnum (Fig 13, F. M.), forming a clover-leaf-like shape (Fig 8). This morphology, notably found in P. dasyuroides, is used by Polly [38] to diagnose the Hyainailourinae ("Pterodontinae" sensu Polly [38] = Hyainailouridae). The anterior condyloid foramina are clearly visible and distal to the foramina, the occipital condyles (Fig 13, Oc. C.) are large and positioned around the foramen magnum.
Upper dentition. The skull (MNHN.F.EBA 517) displays three upper incisors. The three incisors are compressed transversally. The I 1 is the smallest. I 2 and I 3 are simple and conical. I 3 is distinctly larger than the two other incisors and is semicaniniform.
Unfortunately, the crowns of the canines are not preserved. The P 1 , which is double-rooted, is formed by a single pointed cusp (= paracone) (Fig 9). No parastyle is present. The metastyle is very poorly developed. The crown of the P 2 is unknown but the alveoli indicate that it is double-rooted like the P 1 . The P 3 is triple-rooted with a heavily worn parastyle. The metastyle is short and low. A small protocone is present; it is lingually located relative to the paracone. The P 4 is the largest premolar. The parastylar region is worn out as on the previous premolars, but it was certainly well developed, which is typical of the hyainailourine P 4 . The protocone is much more developed than on the P 3 . The protocone does not bear accessory cusps.
The three upper molars are known. The M 1 is characterized by a short parastyle, a small, mesially located protocone relative to the paracone and metacone, and a long, mesiodistally aligned metastyle. The paracone and metacone are fused into an amphicone [9]. No groove delimitates the paracone from the metacone though the apex of each cusp would have been distinct. The M 1 has no cingulum, which is also the case on the M 2 and M 3 . The M 2 has a generally similar morphology to the M 1 . However, the parastylar region and the metastyle are slightly more developed with heavy distal abrasion where the M 3 would have shorn against the metastyle. The M 3 is transversally elongate. The preparacrista is long. The metacone is absent. The protocone is short and narrow. The P 4 , M 1 and M 2 are implanted obliquely. Rather than  being positioned parallel to the buccal margin, the cusps of these teeth angle slightly posteriorly and laterally. Such a condition is found in the hyainailourine P. dasuryoides.
Dentary. One mental foramen is present below the P 1 and another is present below the P 4 (Fig 10). The symphysis extends posteriorly up to the distal root of the P 3 . The ventral margin of the mandible is slightly convex, especially posterior to the tooth row. The mandibular condyle is slightly higher than the tooth row. It is cylindrical and mediolaterally elongate. The coronoid crest arises at a 45°angle relative to the horizontal ramus. A deep fossa for insertion of the temporal muscle is present along the anterior margin of the crest. This fossa is particularly prominent at the base and extends below the distal root of the M 3 . The angular process is curved sharply dorsally and is relatively thin. The masseteric fossa is deep and wide.
Lower dentition. The holotype does not preserve the lower incisors. The crown of P 1 is not preserved, but its presence is confirmed by the presence of its single alveolus. The P 1 is separated from the P 2 by a small diastema. The P 2 is double-rooted and is only slightly longer than the P 1 ( Table 2). The double-rooted P 3 is known by one worn tooth. The presence of a slight precingulid indicates the probable presence of a small paraconid. The talonid is large and is constituted by the hypoconid. The P 4 is larger than the previous premolars. The relatively small paraconid and the large, hypoconid-bearing talonid recall the morphology of P 3 .  The M 1 is unknown but was clearly smaller than the P 4 and the distal molars. The metaconid is absent on the M 2 and M 3 . The talonid on M 2 is simple; only the hypoconulid is clearly visible at the distal part of the talonid. A small anterior keel is visible on the buccal aspect of the paraconid. The M 3 differs from the M 2 by its larger size and shorter talonid. The talonid of M 3 is as simple as the talonid of M 2 .
Description of the postcranial elements. Fibula. The entire fibula is preserved, including both proximal and distal epiphyses. The fibula is partially reconstructed, but this reconstruction does not have any impact on the features described herein. There is no indication that the fibula was fused in any way to the tibia (Fig 11). A tendinal sulcus can be traced on the shaft. The proximal head of the fibula exhibits a semilunar facet for articulation with the tibia. The distal head is broad and possesses two facets that articulate with the calcaneus and the astragalus. The former calcaneo-fibular joint is characteristic of "Creodonta" (e.g., in Hyaenodon and Oxyaena).
Calcaneus. The calcaneus is not well preserved (Fig 11). The distal portion is particularly crushed and difficult to interpret accurately. The tuber calcaneum is long and thickened. The facet for the articulation with the cuboid seems to have been laterally sloping (as in Hyaenodon), because its distal portion is diagonal in proximal view. The sustentacular facet is situated plantar to the astragalar facet. The shape of the sustentacular facet cannot be described, because the bone is damaged in this region. The astragalar facet is elongate and ovoid with a slight proximal inflection that extends onto the tuber calcanei. The astragalar facet is not kidneyshaped as in Hyaenodon.
Astragalus. The astragalus is crushed, especially on its plantar aspect (Fig 11). The trochleaas well as the head of the astragalus-are asymmetrical. The trochlea is grooved, but rather shallow compared to that of Hyaenodon. The head of the astragalus is widened and transversally flat. In proportion to the trochlea, the head appears unusually large even when compared with the astragalus of the large bodied hyainailourine Hyainailouros. Compared to Hyaenodon the caput astragali is not rounded, but transversally flattened, a condition also found in Hyainailouros. The neck is robust and short. The astragalar neck is relatively shorter and the trochlea shallower than in Hyaenodon. The plantar articular facets for the calcaneus are too poorly preserved to allow any comment.
Metatarsal I. The robust first metatarsal is reduced in length relative to the other metatarsals (Fig 12).
Metatarsal II. The second metatarsal, which is known for both the right and left feet, is subequal in length to the third metatarsal and is short and robust. The proximal articulation facet is proximo-distally elongated. It is asymmetrical with a medial indentation of its shaft. The trochlea is slightly curved. This bone is shorter and wider than the second metatarsal of Hyaenodon. The trochlea is not globular as in Hyaenodon and all recent carnivorans.
Metatarsal III. The third metatarsal is broadly comparable to the second metatarsal except for the morphology of the proximal articulation. The plantar portion of the proximal articulation facet is laterally compressed while the dorsal portion expands dorsally with lateral and medial facets that would have overlapped the metatarsals lateral and medial to the fourth metatarsal.
In general, the columnar shafts of the metatarsals and globular proximal articulations are similar to the metatarsals of Hyainailouros [4] and are completely unlike the more gracile metatarsals present in the digitigrade foot of Hyaenodon [83]. When articulated, the three metatarsals are slightly divergent distally.
The dental formula of the new taxon is reminiscent of the earliest proviverrines, sinopines, arfiines, and indohyaenodontines-as well as the earliest placentals-in the presence of three upper incisors, four premolars and three molars. However, the reduced molar talonids of the new taxon clearly distinguishes it from the contemporaneous proviverrines, which retain three-cusped, basined talonids. The reduction of the talonid is shared with the hyaenodontines. The specimen differs from the hyaenodontines in the morphology of its nuchal crest, which does not extend to the mastoid process, and differs from Hyaenodon in the subequal length of the paraconid and protoconid. These features are characteristic of the Hyainailouridae [38]. The new taxon also displays several dental features of the Hyainailourinae such as are more reduced talonid on M 3 relative to that of M 2 , the absence of metaconids on the molars, and the connate paracone and metacone [38,70].
Kerberos notably differs from Metapterodon by the presence of a P 1 , a short and low metastyle The Eocene-Oligocene hyainailourines are represented in Africa by four distinct genera: Furodon and Parvavorodon from late Early or early Middle Eocene of Algeria, and Akhnatenavus and Pterodon from the Fayum Depression of Egypt [6,37,69]. Three hyainailourine genera are known for the same period in Europe: Paroxyaena, Parapterodon, and Pterodon. Only one hyainailourine, Hemipsalodon, is known in the Eocene of North America (see below for discussion referring it to Hyainailourinae).
Kerberos shares with Furodon (late Early or early Middle Eocene; Algeria), the oldest hyainailourine, the presence of a single-rooted P 1 ; this tooth is larger than P 1 in the younger Pterodon and Akhnatenavus. However, Kerberos differs from Parvavorodon and Furodon by the loss of the metaconid and a more reduced talonid on its lower molars and more connate paracone and metacone on its upper molars. These features are shared with the Late Eocene and Oligocene hyainailourines (e.g., Akhnatenavus Paroxyaena, Parapterodon, and Pterodon). Kerberos and Furodon share a similar mandibular morphology with the mandibular condyle located superior to the tooth row, a relatively superior placement of the angular process, and a steeply inclined ascending ramus and deepened fossa for the insertion of the temporal muscle on the anterior portion of coronoid crest (Fig 14). The African genus Akhnatenavus is presently represented by only one species: the Oligocene Akhnatenavus leptognathus [69]. Kerberos differs from Akhnatenavus by exhibiting less pronounced diastemata between the premolars than Akhnatenavus, and more simplified molars (e.g., reduced talonid). All these features are derived in Hyainailourinae. Based on dental measurements, Kerberos is 40% larger than Akhnatenavus. Kerberos thus appears more derived than Akhnatenavus. However, Kerberos also differs from Akhnatenavus in the retention of an important primitive feature: a more developed P 1 .
Kerberos and Pterodon dasyuroides are the only Eocene hyainailourines currently represented by skulls. Both have anteroposteriorly elongate skulls with relatively long rostra and neurocrania. Pterodon includes four European and African species: Pterodon africanus Andrews, 1903 [84] (Oligocene), Pterodon phiomensis Osborn, 1909 [56] (Oligocene), Pterodon syrtos Holroyd, 1999 [69] (Oligocene), and the European Pterodon dasyuroides Blainville, 1839 [77] (Late Eocene), which is the type species. In the analysis presented here, the genus is paraphyletic (see below). The teeth of Kerberos are generally similar to those of "Pterodon", but Kerberos differs in its larger P 1 , P 1 , and M 3 . The P 1 is often absent in P. dasyuroides [9] and small in "P." africanus and "P." phiomensis. The talonids on the molars of Kerberos are relatively longer, and the diastema between P 1 and P 2 is longer than in "Pterodon" species. The mandible of Kerberos is more primitive than that of "Pterodon" species in having a more dorsal mandibular condyle, a more dorsal angular process, a more steeply inclined and taller coronoid crest, and deeper insertions of the temporal muscle on the anterior margin of the coronoid process (Fig 14). The mandibular features are reminiscent of those of Furodon, which is-with Parvavorodon-the oldest hyainailourine presently known (Fig 14). Finally, the metacone and paracone on the molars are more fused, a more derived feature than the less complete fusion present in P. dasyuroides.
Kerberos further differs from the European Pterodon dasyuroides in the presence of large P 1 and P 1 , and an unreduced number of upper incisors (only two upper incisors are present in P. dasyuroides); these features are primitive among hyainailourines. Kerberos also differs from P. dasyuroides in its larger protocone on P 3 , and reduced parastylar area; these are derived features in hypercarnivores. Finally, based on dental measurements and skull measurements, Kerberos differs from P. dasyuroides by its larger size (10-14%).
In her study of the European hyaenodontids from the Late Eocene to Late Oligocene, Lange-Badré [9] described three genera that she considered to be close to Pterodon: Parapterodon, Paroxyaena, and Schizophagus. Since this study, Schizophagus has been synonymized with Paroxyaena, which is considered to be a peculiar hyainailourine [8,85]. Paroxyaena is known in Robiac's fauna (reference locality of MP16 reference-level) [27]. Consequently, Kerberos and Paroxyaena are nearly coeval. Sudre [28] noted that the "Castrais" fauna could be slightly older than the Robiac fauna, contra Astruc et al. [29].
Kerberos shares with Parapterodon and P. dasyuroides but not with Paroxyaena the presence of a high, sharp P 4 , and a more fused paracone and metacone on the upper molars. Kerberos differs from Parapterodon in the primitive presence of a large P 1 -this tooth is even vestigial in Parapterodon-and a primitively large M 3 . Kerberos further differs from Parapterodon by having a more reduced parastylar region, and a more reduced protocone on upper molars, generally derived features among lineages that acquire hypercarnivorous dentition [63,86]. Kerberos shares with Paroxyaena the presence of three upper incisors, but Kerberos differs from Paroxyaena by the presence of P 1 , a less reduced P 1 , a less developed protocone on P 3 , larger M 3 (all considered primitive features), more reduced talonids on lower molars, fusion of the paracone and metacone, and reduced protocones on the upper molars (generally derived features).
Kerberos differs from the North American Hemipsalodon in its smaller size (the skull of Hemipsalodon is 45 cm long), its mesiodistally elongate skull (notably in the rostrum and the neurocranium), the distal elongation of the mastoid process (primitive features), and a more closely appressed paracone and metacone on the molars (derived feature).
In summary, the taxon described here departs from other members of the hyainailourine subfamily in featuring a distinctive mixture of primitive and derived features. Thus, we refer this taxon to the new genus, Kerberos. Its discovery increases the taxonomic and morphological diversity of the hyainailourines, especially in Europe.

Ecology of Kerberos
The body size of an animal is one of the most important factors in determining its role in its ecosystem. Body size influences dietary preference, predatory behavior, and niche partitioning [87][88]. It is directly related to an organism's biomechanical and physiological demands. Several methodologies have been proposed for determining the body mass in fossil mammals. Several of them have been applied to hyaenodonts. Depending on the methodology, they correspond to regressions based on dental, cranial or postcranial dimensions [21,[89][90][91][92]. Most of these studies use extant taxa to correlate known body size to osteological measurements that can be collected from related extinct taxa. The reconstruction of body sizes for Hyaenodonta has been problematic because the entire group is extinct, making direct regressions difficult to apply to the group. Moreover, as mentioned by Van Valkenburgh [89], dentally or cranially derived body size estimates based on extant carnivorans produce unreasonably large values in hyaenodonts because they have relatively large crania compared to carnivorans.
With these caveats in mind, we estimate the body mass of Kerberos langebadreae using three methodologies. (1) The methodology of Morlo [21], which is based on dental dimensions (length of the molars), provides an estimation of 85-90 kg (Table 4). (2) The methodology of Van Valkenburgh [89], based on the skull length, estimates the body mass of K. langebadreae to be about 277 kg. (3) The methodology of Tsubamoto [92], which is based on the dimensions of the astragalus results in ranges from 60 kg to 269 kg (length Li1) and from 49 to 199 kg (length Ar1). The median body mass of these estimates is 140 kg.
We present a comparison of the estimated body mass of Kerberos with those of proviverrines from MP7 to MP19 and contemporaneous hyainailourines and hyaenodontines (Fig 15;  Table 4). We only use the methodology of Morlo [21] for this comparison because it is based on dental measurements, and cranial and astragalar dimensions are not available for the entire sample. During their evolution, proviverrines never have exceeded 20 kg (see Solé et al. [17]). The two oldest hyainailourines recorded in Europe, Kerberos and Paroxyaena, are clearly distinguished from endemic proviverrines by their much larger size. Thus, neither Kerberos nor Paroxyaena was likely in direct competition with proviverrines. Conceivably, the smaller proviverrines may have even been their prey.
It is noteworthy that the body mass range of contemporaneous European hyaenodontines (4-39 kg) partially overlaps with that of proviverrines (0.03-20 kg) (Fig 15). Morlo [21] reconstructed the large proviverrines (e.g., Oxyaenoides) as cursorial hypercarnivorous predators; because this corresponds to the locomotion of Hyaenodon, they thus may have belonged to the same ecological niche. Additionally, one can note that the appearance of Hyaenodontinae in Europe is roughly coeval with the disappearance of the large and hypercarnivorous proviverrines. The body mass range of the hyainailourines (46-98 kg), as noted above, sets them in a very different size class than either hyaenodontines and proviverrines (Fig 15). Consequently, hyaenodontines may have had a greater impact on the evolution of proviverrines than hyainailourines, which did not move into a previously occupied niche space (based on body mass).
The association of the mandible and skull allows detailed discussion of the masticatory musculature in Kerberos and a reconstruction of its feeding habits (see [94][95][96][97][98] for a description of the role and position of each masticatory muscle in carnivorous mammals). It should be noted that the temporal group musculature is more developed than the masseter group as usually observed in carnivorous mammals [9,94]. The anterior base of the coronoid process has a remarkably deep fossa with a significant anterior extension, producing a large insertion for the anterior fibers of the temporal muscle (pars orbitalis). Because this fossa is deeper than in "Pterodon", it would have allowed Kerberos a more powerful clenching action. As in "Pterodon", the squamosal and temporal areas, in which the deep masseter, zygomaticomandibularis and temporal muscle (pars temporalis) originate, are important. These three muscles are extensively developed; this observation correlates with the large masseteric fossa observed on the two dentaries. As indicated above, the anterior border of the choana is more anterior than the choana in Pterodon dasyuroides, providing a reduced surface area for the origin of the medial pterygoid muscle. The delicate angular process (relative to other hyainailourines) further supports the feeble contribution of the medial pterygoid to mandibular adduction. The enlargement of the pterygoid muscle in "Pterodon" was noted by Lange-Badré [9]. The important mastoid process, which is larger than in "Pterodon", suggests that the digastric muscle, an abductor and protruder of the mandible, was also powerful. However, the mastoid is also the site of origin for much of the ventral cervical musculature and the separate sites of origin for  Apterodontinae Apterodon langebadreae and Apterodon indet. Middle Eocene Semi-aquatic Plantigrade [52] *Lesmesodon is only represented by juveniles [24].
doi:10.1371/journal.pone.0135698.t005 these muscles are difficult to interpret (see [99][100] for a description of the muscle attachment sites in the mastoid region in mammals).
Although the cranial morphology of Kerberos is similar to that of "Pterodon", it has several more primitive features. The younger hyainailourine species expanded the role of the pterygoid muscles and of the masseter muscles, a shift evinced in the more developed angular process and reduced insertion for the temporal in "Pterodon" and younger species. Finally, the emphasis on the temporal musculature and digastric muscles in Kerberos indicates that this taxon would have been capable of powerful slicing and crushing near the molars, but the dentary was only capable of minimal lateral movements. Savage [80] described a complete skull of Megistotherium osteothlastes (= Hyainailouros bugtiensis) and noticed that the skull is "moulded for maximum efficiency in feeding" (Savage,[80]: p.503). He particularly emphasized the importance of the temporal musculature that originated from the long, tall sagittal crest, a feature shared with Kerberos. Hyainailouros bugtiensis differs from Kerberos in having a shorter rostrum. Savage [80] interpreted the shortened face as important for stabilizing the roots of the canines during food capture. With a more extensive rostrum than M. ostethlastes, the canines may not have been used as actively in Kerberos.
The cervical vertebrae are unknown for Kerberos. However, several of these elements are known for Hyainailourinae and because the basicranial portion of the skull is similar for all known hyainailourines, the cervical vertebrae were likely similar across the subfamily. Savage [80] described the atlas of Megistotherium (= Hyainailouros bugtiensis) and Ginsburg [4] described two cervical vertebrae of Hyainailouros sulzeri. Savage [80] noted that the large size and broad transverse processes of the atlas imply significant extension of the head. The lateral cervical musculature associated with the large, transverse cervical processes indicates the animal was capable of strong lateral flexion of the head, which is particularly useful for restraining struggling prey. Concerning the cervical vertebrae of H. sulzeri, Ginsburg [4] noted the relatively short neural spine on the seventh cervical vertebra. This spine, which is taller and more robust in Hyaenodon, implies that Hyainailouros bore his head low, aligned with the thoracic vertebrae.
The secant molars M 2-3 and M 1-2 of Hyainailourinae exhibit little morphology that would aid in crushing (e.g., protocone and talonid on molars) and puncturing (e.g., metaconid), but the premolars are robust and display horizontal wear, which indicates they served a separate function from the slicing molars-the addition of a well-developed P 3 protocone in hyainailourines illustrates the crushing function hypothesized for the premolars. Based on this dental morphology Ginsburg [4,33] argued that "Pterodon" and Hyainailouros were bone crushers comparable to recent hyaenids, which also possess secant dentition. Stefen [101] noted that Pterodon dasyuroides and Hyainailouros possess only zigzag Hunter-Schreger-Bands (HSB), while limnocyonines Prolimnocyon (probably with the exception of P. elisabethae) and Thinocyon, and sinopines Prototomus and Sinopa display undulating HSB. A transitional state is found in Apterodon and some Hyaenodon and "Pterodon" species: they actually display a transition from undulating HSB at the enamel base through acute angled HSB to zigzag HSB at the tips of the teeth. She concluded that the development of zigzag HSB in the hyainailourines correlates with ossiphagy. The transition from undulating to zigzag HSB is notably found in most European Hyaenodon species-which also exhibits a pattern of wear-suggesting that bone crushing was also part of Hyaenodon´s diet [102].
Consequently, based on the taxon's robust premolars, the powerful cranial musculature and the phylogenetically bracketed presence of zigzag HSB, we hypothesize that Kerberos was capable of scavenging like its closest relatives. Extant mammal scavengers are also capable hunters, a mode of food acquisition likely partially utilized by Kerberos and hyainailourines more generally. Werdelin [103] distinguishes two morphotypes among ossiphageous carnivores: the bone-crackers (e.g., spotted hyaenas), which break bones with their premolars, and bone-crushers (e.g., wolves) that break bones with their post-carnassial molars. Because Kerberos has extensive premolar wear, it may be considered a bone-cracker with a diet analogous to that of the extant hyenas (except the aardwolf).
The postcranial bones allow a partial reconstruction of the locomotion of Kerberos. The fibula exhibits interesting features related to locomotion. The fibula was not fused to the tibia. Furthermore, the fibula has a large distal head and large facet for contact with the calcaneus and a large proximal head in contact with the tibia. These features indicate a fibula capable of significant rotation for progression over rocky or uneven surfaces [104]. Among extant carnivores, such a rotatory fibula is known in bears and cats for instance and, among Hyaenodontidae, a rotary fibula is known in Hyaenodon, although Hyaenodon shows clear adaptations to cursoriality as well [83,105]. Other postcranial features indicate a plantigrade posture for Kerberos including: a transversally flattened caput astragali, a short astragalar neck, short and robust metatarsals and a short and robust calcaneus [106][107]. The sustentaculum tali has a plantar position, which is an osteological correlate for terrestrial locomotion [104]. The grooving of the astragalar trochlea is a well-known indicator for locomotion with cursors tending to have more deeply grooved astragali, while ambulators have more shallowly grooved astragali [108][109]. The relative depth of the astragalar trochlea also contains a strong phylogenetic signal [89]. The astragalar trochlea of Kerberos is shallow, but grooved. Its development suggests terrestrial locomotion, but clearly not in the cursorial posture indicated by the astragalar trochlea of Hyaenodon [105]. Because O'Leary et al. [110] recently proposed that scansorial features are basal for placental mammals, one must keep in mind that some of the features of Kerberos are potentially primitive, as they are found in many early Tertiary mammals (for instance see [111]).
To conclude, its postcranial anatomy suggests that Kerberos was a plantigrade, terrestrial mammal; we use the term "terrestrial" in the sense of Egi [91], referring to the following particular locomotion: "climbs rarely/never, scarcely running". This feature is important because it ecologically distinguishes Kerberos from two carnivoran groups that occupied the ossiphageous niches, the Borophaginae and Hyaenidae: these carnivorans are reconstructed to be cursorial [72,73]. In this regard, Kerberos appears more similar to the oxyaenid Palaeonictis because the latter displays dental ossiphageous adaptations [62] and a terrestrial locomotion [23,112].
The locomotor styles of hyaenodonts are poorly known: the postcranial morphology of only a few taxa has been studied, and most of these taxa are from North America (Table 5). Interestingly, a wide range of locomotion has been hypothesized: the limnocyonine Thinocyon is a probable semi-fossorial form, Hyaenodon is cursorial, while Apterodon shows semi-aquatic adaptations [52,83,115]. This lack of knowledge and the apparent diversity of locomotor adaptations among hyaenodonts prevent a rigorous reconstruction of the ancestral locomotor state for this group. However, based on the postcranial elements found in North America we are able to posit a possible primitive locomotor pattern for hyaenodonts. The oldest North American hyaenodonts have adaptations that would have allowed them to move in both arboreal and terrestrial environments. Rose ([113]: p.166) even noted that the "Wasatchian hyaenodontid [an]s, probably [were] reasonably facile in the trees as well as on the ground." Among North American hyaenodonts, the Sinopinae and Arfiinae differ from the Limnocyoninae in being more terrestrial. For instance, Arfia displays a mosaic combination of cursorial forelimb adaptations and scansorial hindlimb adaptations [34]. The ability to exploit both the trees and the ground was probably the ancestral condition of the hyaenodonts. Finally, the presence of plantigrady in numerous, primitive hyaenodonts allows us to hypothesize a plantigrade foot at the origin of hyaenodonts.
The sole hyainailourine for which the locomotion has been studied is Hyainailouros sulzeri. Ginsburg [4,33] reconstructed Hyainailouros not as a fast runner, but as a capable jumper. He also provided arguments for a digitigrade stance. According to our analysis, Hyainailouros is more derived than Kerberos and the postcranial morphology of the new taxon is consistent with the retention of an archaic locomotion.
In conclusion, Kerberos was a heavily built hyaenodont, with a plantigrade posture that predominantly moved across terrestrial substrates. Because of its size and the absence of obvious adaptations to cursoriality, Kerberos was likely not a fast runner but rather complemented active predation with scavenging. Carbone et al. [88,116] demonstrated that extant carnivorous mammals that weigh above 21.5-25 kg generally hunt prey as large or larger than themselves. Because K. langebadreae exceeds this size bracket, it probably focused on large prey such as large artiodactyls (e.g., Choeropotamus) and perissodactyls (e.g., Lophiodon and Palaeotherium). Due to its large body mass, Kerberos langebadreae was probably an important, apex predator among the European faunas of the Bartonian. However, as noted by several authors [117][118], the reconstruction of the ecomorphology of extinct carnivorous mammals is difficult especially when a taxon like Kerberos displays unique character combination, which are not any analogous to those found in extant carnivorans: Kerberos with features similar both to Palaeonictis and several living hyenas, clearly illustrates this issue.

Phylogeny of the Hyainailourinae
As indicated above, Hyainailourinae is characterized by a secant dentition as Hyaenodontinae. Polly [38] demonstrated that the two groups are not closely related and convergently acquired similar shearing dentitions. He provided cranial and dental characters for differentiating the two hypercarnivorous groups. He included Pterodon, Hyainailouros, Hemipsalodon, Sivapterodon, and Megistotherium among hyainailourines. Holroyd [37] added the genus Akhnatenavus and Metapterodon. Peigné et al. [119] removed Dissopsalis and Anasinopa from the "Afro-Asian proviverrines" of Egi et al. [60], and referred them to Hyainailourinae as well. Dissopsalis and Anasinopa differ from Pterodon and its closely affiliated taxa by the presence of a wide talonid and the retention of the metaconid. Lewis & Morlo [70] used this work as a basis for the establishment of a list of African hyainailourines. Hyainailourinae in Morlo et al. [53] further added African "Sinopa" and Metasinopa to the group.
In an effort to better understand the relationships within Hyainailourinae, we performed phylogenetic analyses of the group including closely related taxa. A study of hyainailourines was partially undertaken by Holroyd in her PhD thesis [37]. However, she included several taxa that have been excluded from Hyainailourinae in subsequent work (e.g., Matthodon, Paenoxyaenoides, and Oxyaenoides, which are now referred to Proviverrinae, and Metapterodon, which is referred to Koholiinae), and she restricted her study to the Eocene and Early Oligocene hyainailourines.
Our study includes almost all hyainailourines that are known by a complete lower or upper dentition. We did not include either Parvavorodon or Pterodon sp. [69], because the former is only known by two isolated teeth, and because the maxillary fragment of the latter is much abraded. We also included the apterodontines Apterodon langebadreae and A. macrognathus, as well as the koholiines Lahimia and Boualitomus, because the two subfamilies share many characters with Hyainailourinae, which can potentially help polarizing character transformations within the in-group. The purpose of this study is not to test all hyaenodont relationships. Several taxa, notably species of Hyaenodontidae, were not included and will be considered in future evaluations of the phylogenetic relationships within Hyaenodonta.
Three distinct analyses have been performed (see Material and methods section). The first analysis (all the taxa and characters included) yielded 201 equally parsimonious trees, with a tree length of 117 steps, consistency index (CI) of 0.60 and retention index (RI) of 0.77. The strict consensus tree is 135 steps long (CI = 0.52 and RI = 0.68) (Fig 16A). The majority rule consensus is 120 steps long (CI = 0.59 and = RI 0.75; Fig 16B). The Hyainailourinae includes the genera Paroxyaena and Hemipsalodon. In this analysis the genus Pterodon is not monophyletic and will be closely examined in a future work. The second analysis (Leakitherium deleted) yielded 7 equally parsimonious trees, with a tree length of 116 steps, consistency index (CI) of 0.61 and retention index (RI) of 0.77. The strict consensus tree is 120 steps long (CI = 0.59 and RI = 0.75) (Fig 17A). The majority rule consensus is 117 steps long (CI = 0.60 and RI = 0.76; Fig 17B). The third analysis (Leakitherium deleted and character related to body mass excluded) resulted in 8 equally parsimonious trees (L = 106; CI = 0.63; RI = 0.78; the strict consensus has a length of 111 steps for a CI of 0.60 and a RI of 0.76) (Fig 18). Fig 19 depicts the consensus tree of the third analysis with stratigraphical and geographical information.
The proposed topology does not show temporal incoherence. However, the topology implies a significant ghost lineage for Pterodon dasyuroides. As demonstrated by Holroyd [69], the hyainailourines experienced a significant diversification during the Late Eocene. This diversification should be interpreted cautiously as there are significant gaps in the fossil record of hyainailourines, particularly in the early Paleogene record of Africa. Interestingly, the two Neogene groups-the small hyainailourines represented by Isohyaenodon and the very large Hyainailouros and Sivapterodon-are rooted in Late Eocene taxa.
The consensus of the third analysis differs from the consensus trees of the second analysis by the absence of close relationship between the Apterodontinae and Hyainailourinae although some relationships remain unresolved. Moreover, the Bremer support values are slightly higher. In the second analysis (Fig 17), the clade formed by the Apterodontinae and the Hyainailourinae is supported by the presence of large premolar diastemata [35 (2) . The close relationships between these two subfamilies was hypothesized by Grohé et al. [52] and Rana et al. [61]-together with African Teratodontinae in the latter study-but not by Solé et al. [6]. Our expanded taxonomic sampling allows more complete examination of this relationship. Moreover, as discussed above, the two subfamilies share numerous postcranial and cranial features that would likely lend support to the Apterodontinae+Hyainailourinae clade defined as Hyainailouridae. We describe in the following paragraphs the consensus tree of the third analysis because the analysis was not disturbed by Leakitherium and not influenced by the body masses (Fig 18). The node A is supported by a weak and short paraconule crista on the molar [16 (1)], the reduction of the P 2 and P 3 relative to P 4 [29(1)] and the reduction of the metaconid [37 (1)]. The Koholiinae (node B) are supported by the loss of the P 1 [28 (2)] and a molar series longer than the premolar series [36(1)]. The Apterodontinae (node C) are grouped based on the following features: reduction of the metastyle on P 4 (1)]. Grohé et al. [52] hypothesized that Apterodon could be a semi-aquatic hyaenodont based on postcranial features. The peculiar morphology of apterodontine dentitions-homogenized premolars and simplified molars-could be an adaptation for a piscivorous diet. This homogenized, simplified dentition is reminiscent of pinniped dentition.
The Hyainailourinae (node D) are supported by a small, mesially located protocone on the molars [24(1), 26 (1) Parvavorodon has not been included in the phylogenetic analysis (see above) and has been placed at the node C (Fig 19) with an unresolved relationship with other hyainailourines. This position for Furodon contradicts the hypothesis of Solé et al. [6] that Furodon could be closely related to Akhnatenavus.
Numerous features support node E: the absence of the metaconule on the molars [17 (1)], the length of the molars equivalent to that of the premolars [36 (2)], the loss of the metaconid [37 (2) It is interesting to note that these two genera have a basal position among hyainailourines in our topology. This early origination agrees with the pleisiomorphic features retained in the group such as the separated paracone and metacone on molars. Surprisingly, Paroxyaena pavlovi is closer to Hemipsalodon than to Paroxyaena galliae (node G); however, it should be noted that Paroxyaena galliae is currently only represented by a juvenile specimen [8], and its permanent dentition is not completely known. The relationship between Paroxyaena and Hemipsalodon allows the origin of the North American genus to be reconstructed. Contrary to Mellett [65], Gustafson [120] considered a close relationship between Pterodon and Hemipsalodon possible because of similarities in the basicranial regions of both genera. We follow the opinion of Gustafson [120] in considering Hemipsalodon as a hyainailourine. As shown by the phylogenetic analysis, several dental features such as the separation of the paracone and metacone of equal height and the development of a large protocone on P 3 support a sister-relationship with Paroxyaena. Moreover, we can add the following features: the presence of three upper incisors, presence of a lingual cingulum on the premolars and molars, the non-reduction of the M 3 (compared to other hyainailourines, except Kerberos), the shortening of the rostral part of the skull, and the presence of a depression on the lateral part of the maxilla.
To emphasize the peculiar morphology of Paroxyaena among hyainailourines, Lavrov [8] recognized the tribe Paroxyaenini. The Paroxyaenini are characterized by the fusion of the orbital fissure and the foramen rotundum. Due to this fusion, the Paroxyaenini have only three foramina in the posterior part of the orbital region. Gustafson [120] noted the presence of three foramina instead of four in Hemipsalodon, but he misidentified them; thus, from anterior to posterior, Hemipsalodon has an ethmoid foramen, an optic foramen, and a fused fissura orbitalis and foramen rotundum. As no feature excludes Hemipsalodon from Paroxyaenini, we include this genus in the tribe. The presence of Hemipsalodon among Paroxyaenini has important biogeographic implications as this genus is the only hyainailourine recorded in North America (see below).  fragmentary fossils of hyainailourines have been reported from this period (see for instance Rasmussen & Gutierrez [121]). Consequently, this lack of fossils could explain the irresolution of the relationships within this clade. This node is comprised of two distinct size groups: node O corresponds to the small Neogene hyainailourines (Isohyaenodon), while node P includes the largest hyainailourines ever recorded (Sivapterodon and Hyainailouros). Consequently, the separation between these two ecological groups occurred during the Oligocene or earlier.
Akhnatenavus is distinguished by the absence of a distal cingulid on P 3 [31 (1)], reduction of the talonid on P 4 [34 (1)], presence of large diastemata between the premolars [35(0)], and a less reduced talonid on M 2 [45(1)]. Isohyaenodon is not monophyletic in the strict consensus tree of the third analysis, but it was monophyletic in the majority rule consensuses of the first and second analyses. However, the relationships within the Isohyaenodon genus do differ between the first and second analyses. Isohyaenodon is a good example of the trend among hyainailourines towards a simplification of the molars. Isohyaenodon andrewsi-the largest species of the genus-is characterized by the presence of a talonid on M 1 (1) FO]. The two species of Hyainailouros form a clade (node P), which supports the monophyly of the genus and the distinctive diagnosis of Sivapterodon (see Pilgrim [47] and Ginsburg [4]); this group is however only supported by a low and broad P 4

Evolution of size among hyainailourines
The large size of Kerberos was unexpected: it is larger than any other contemporaneous European hyaenodont (Fig 15). With the inclusion of North American Hemipsalodon in Hyainailourinae, there were at least two large-bodied Hyainailourines in Laurasia during the Bartonian. The large body size, exemplified by Bartonian Kerberos and Hemipsalodon, is a common evolutionary tendency in hyainailourines (Fig 19). Hyainailourinae includes one of the largest carnivorous mammals ever known: Megistotherium (= Hyainailourus bugtiensis). The skull of Megistotherium is about 66 cm long-nearly twice as long as that of Kerberosand the animal may have weighed as much as 800 kg [80].
In his description of Megistotherium, Savage [80] observed that the face of hyainailourines is relatively long when compared to that of carnivorans and Hyaenodontidae. He proposed two hypotheses for the conservation of a long mandible: (1) it is non-adaptive and corresponds to a conserved genetic trait, or (2) it is selective and related to the size of preferred prey (e.g., anthracotheres, pigs, rhinoceroses, mastodonts and deinotheres). The second hypothesis is supported by the abundance of proboscideans in Megistotherium's fauna. Savage [80] estimated that Megistotherium-due to the length of its jaw-may have had 30 cm of clearance between the tips of the canines, which would allow the animal encircle and bite a proboscidean limb. Rasmussen et al. [122] hypothesized that large hyainailourines, which mainly radiated in Africa, evolved originally as specialized predators or scavengers of the massive herbivores endemic to Africa, such as embrithopods and proboscideans. Ginsburg [4] supported this hypothesis and noted that Hyainailouros appeared in Europe together with the proboscideans (Proboscidean datum; see [123]). Large body size is an ancient feature among proboscideans with the oldest representatives of group already relatively large [124]. Numidotherium, which is known from the latest Early Eocene of Algeria, was one of the largest terrestrial mammals of its time, and initiated an impressive size increase among proboscideans [125]. Following Rasmussen et al. [122], we therefore hypothesize that the large size of hyainailourines may result from co-evolution with endemic African herbivores such as proboscideans, though this hypothesis must be more rigorously tested. However, as demonstrated at a broader mammalian scale, large mammals tend to have longer faces than closely related but smaller ones [126]. Therefore, another hypothesis would be that the long face of Megistotherium is related to the constraints of body size.
The large body size of the hyainailourines, exemplified by Kerberos and by Hemipsalodon, appears early in the history of the subfamily. The large size of the Miocene Hyainailouros and Sivapterodon could result from either conservation or convergence on this trait. Because Hyainailouros and Sivapterodon are close to the small Isohyaenodon in the phylogenetic analyses we performed, it is possible that gigantism evolved at least twice in hyainailourines.
In the same faunas occupied by the gigantic Hyainailouros species, there were also smaller hyainailourines. The smaller hyainailourines are known from the Late Eocene (Akhnatenavus) to Middle Miocene (Isohyaenodon) (Fig 19). The largest one (I. andrewsi) only weighed around 14 kg. Holroyd [69] hypothesized that these differences in sized among hyainailourines are suggestive of the exploitation of different ecological niches. Based on Carbone et al. [88,116], these smaller taxa are below the threshold for hunting prey equal to or larger than their body size. Conversely, they likely exploited prey smaller than them. Unfortunately, the skull and postcranium are unknown for these small hyainailourines, which prevents a clear evaluation of their ecology (i.e. locomotion, choice of prey). The presence of these small-bodied hyainailourines suggests that the hyainailourine bauplan was not specifically adapted for large-bodied predation.
The hypothesis of several reductions of the body mass among hyainailourines (and apterodontines) is poorly supported by the analyses of body size evolution among carnivorous mammals. Van Valkenburgh et al. [127] tested the Cope's rule in extinct canids (borophagines and hesperocyonines) and noticed a clear evolutionary trend toward larger body size in these groups. They argued that energetic constraints and pervasive selection for larger size in carnivorous mammals (e.g., killing and feeding on large preys) also lead to dietary specialization (hypercarnivory). These factors increase vulnerability to extinction, explaining why large hypercarnivorous mammals appear to have been limited in their temporal appearance relative to smaller and more conservative species: a greater vulnerability to extinction ultimately results from evolution towards a larger body size. Based on this remarkable study, the possibility that large sizes evolved at least twice in hyainailourines is more reasonable than several cases of convergent reductions of the body mass. Finally, it is worth reminding that the oldest and most primitive hyainailourine and apterodontines are only of small sizes [6,52].

Paleobiogeographic implications
Based on the discovery of Parvavorodon and Furodon, Solé et al. [6] hypothesized that hyainailourines originated in Africa during the Early Eocene (contra Egi et al. [128]). The presence of a new hyainailourine in the Bartonian of Europe has implications for the terrestrial connection between Europe and Africa during the Eocene. The earliest appearance of the Hyainailourinae in Europe was Paroxyaena galliae, which is known from the MP16, MP17a and MP17b reference-levels [9,27,85] (Fig 1). The presence of Kerberos is consistent with the appearance of Hyainailouridae in Europe at MP 16 (Fig 1). Pterodon dasyuroides is recorded in the MP18 reference-level but is unknown after MP20 reference-level. The stratigraphic position of Parapterodon is also presently uncertain but is probably contemporaneous with Pterodon [9] (Fig 1). As noted by Crochet et al. [85], Paroxyaena and Pterodon were not present in Europe at the same time. The description Paroxyaena pavlovi by Lavrov [8] could contradict this statement, but the stratigraphic position of this species is still uncertain.
These stratigraphic partitions raise the question of the dispersal of Hyainailourinae into Europe and the subfamily's diversification on the European continent. Did the Hyainailourinae disperse once or several times from Africa to Europe? Did the Hyainailourinae diversify in Europe?
The primitive features of Kerberos (e.g., presence of three upper incisors, large P 1 and P 1 ) conforms with an older age for Kerberos compared with the more derived features of "Pterodon" and Parapterodon (Fig 1). Paroxyaena, which is also recorded in the Bartonian of Europe, shares with Kerberos an unreduced number of upper incisors. However, the two genera are clearly different taxa (notably in the reduction of the P 1 and P 1 in Paroxyaena and the separation of the paracone and metacone). Paroxyaena clearly differs from Kerberos, "Pterodon", and Parapterodon and represents a distinct lineage (see the phylogenetic analysis). This supports a diversification of the Hyainailourinae during the Bartonian or earlier. However, this diversification is poorly understood due to the lack of fossils, and may have occurred either in Africa or in Europe.
As for the relationships among Kerberos, Parapterodon and "Pterodon", the new genus is distinguished by its larger size, more derived molars (e.g., parastylar reduced) and a more developed protocone on P 3 . Because of these differences, Kerberos does not seem to be ancestral to Pterodon dasyuroides or Parapterodon. Consequently, it appears that the Hyainailourini did not diversify and radiate in Europe, but rather dispersed at least twice into Europe. The Paroxyaenini seem to have diversified in Laurasia, but this needs further scrutiny.
The hyainailourines are unknown in Asia until the Miocene. The species Orienspterodon dahkoensis described by Egi et al. [128] from late Middle Eocene of Asia is now considered part of Hyaenodontidae and Indohyaenodontinae rather than to Hyainailouridae [70]. This absence implies that the Bartonian hyainailourines may have dispersed directly from Africa to Europe via the Iberian or Apulian route.
The history of trans-Tethyan Paleogene mammal dispersals is characterized at least by four dispersal phases: during the Early Thanetian, and by the Thanetian/Ypresian, Bartonian/Priabonian, and Priabonian/Rupelian transitions. Two other dispersals between Africa and Laurasia (by the Ypresian/Lutetian and by the Lutetian/Bartonian transitions) have less support [129].
Here we refer to the dispersals of hyainailourines between Africa and Europe as the "Out of Africa event" (= OoA).
Gheerbrant & Rage [129] considered the Lutetian/Bartonian Dispersal Phase as doubtful and minor, but it could explain the dispersal of amphipithecids from Africa to Laurasia. The presence of Kerberos and Paroxyaena in the earliest part of Bartonian (Fig 19) further substantiates this dispersal phase hypothesis. The Lutetian/Bartonian Dispersal Phase corresponds to the first migration of hyainailourines to Laurasia (OoA-1 on Figs 19 and 20). The appearance of the hyainailourines in Europe is related to a modification of the European mammal fauna: the second intra-Eocene mammal turnover of Franzen [130]. This turnover seems to be related to dispersals between Central European Island and Iberian Peninsula [130][131]. Based on this exchange, we propose that the Bartonian hyainailourines may have entered into Europe through the Iberian Peninsula.
The presence of Hemipsalodon in the Duchesnean NALMA is very interesting. It appeared together with other new migrants. Emry [132] noted that the artiodactyls Brachyhyops and Simimeryx, as well as the hyaenodontine Hyaenodon, were probably migrants from Asia during the Duchesnean. Because Hemipsalodon is not present in all Duchesnean localities, it has been used as an index taxon for separating early and late Duchesnean with Hemipsalodon not present in the early Duchesnean. Because Hemipsalodon and Hyaenodon do not appear in North American localities at the same time, and because there is a close relationship between Paroxyaena and Hemipsalodon (see the phylogenetic analysis), the Paroxyaenini may have dispersed from Europe to North America during the late Duchesnean (Fig 20).
Concerning this issue, it should be emphasized that Orienspterodon displays several features that are reminiscent of Paroxyaena such as the presence of cingulids on premolars. However, the Asian species differ from Paroxyaena in double-rooted P 3 , better fusion of paracone and metacone on M 1 , small metaconid on M 3 and vestigial metaconid on M 1-2 [128]. A close relationship between Orienspterodon and Paroxyaenini would support a dispersal of these hyainailourines to Asia. Such a dispersal represents a competing hypothesis to the dispersal of Paroxyaenini in North America (i.e., a dispersal from Asia to North America as reconstructed for the hyaenodontines). Only the discovery of new material for Orienspterodon will allow its systematic position to be precisely tested and further discussion of these paleobiogeographic scenarios. The presence of the hyainailourine taxa "Pterodon"and Parapterodon in the Late Eocene of Europe has previously been considered support for a Bartonian/Priabonian Dispersal Phase [6] (OoA-2 on Figs 19 and 20). As noted by Gheerbrant & Rage [129], the Bartonian/Priabonian Dispersal Phase is characterized by the immigration of the "baluchimyine" rodent Protophiomys into Africa. Consequently, the hyainailourines could represent the sole case of a northward dispersal during this phase.
The hypothesized existence of two possible dispersal events results from the lack of support for endemic diversification of the hyainailourines in Europe during the Bartonian and Priabonian. It also implies a turnover, which affected the Europe hyainailourines between these stages. This could result from the inability of hyainailourines to cope with some aspects of European environments. Indeed, "Pterodon" also rapidly disappeared from Europe and is only recorded from MP18-MP20. This is especially conspicuous when compared to the extensive diversification and long stratigraphic record of European Hyaenodon and development of a separate lineage within Europe [133].
Unlike Apterodontinae [66], the hyainailourines did not disperse to Laurasia during the Oligocene, but did so during the Miocene. Two distinct phases can be defined. The OoA-3 corresponds to a migration from Africa to Asia. It resulted in the presence of Hyainailouros and Sivapterodon in Asia (Figs 19 and 20). This dispersal occurred at 19.6 Ma (Mammal Neogene level, MN3). The second dispersal of Hyainailouros occurred only slightly later, when it dispersed to Europe at MN4 (Figs 19 and 20).
By the Early-Middle Miocene, the widely dispersed Hyainailouros and African Isohyaenodon were the last representatives of hyainailourines. Their size differences show that this group remained ecologically and morphologically diverse. The only other hyaenodonts known at that time were the teratodontine Dissopsalis, which was present in Africa and Asia [134] and the koholiine Metapterodon, which was possibly present in Asia [135]. These four taxa are the swan song of Hyaenodonta and they witnessed the ultimate disappearance of their once diverse and successful order.

Conclusion
In this contribution we describe a new hyainailourid hyainailourine, Kerberos langebadreae gen. & sp. nov., based on a skull, mandible, and a few elements of the hind limb. Associated dental and cranial or postcranial elements of hyaenodonts is rare, especially among the hyainailourines, making this description of one of the oldest known hyainailourines particularly important for understanding the evolution of Hyaenodonta.
The skull of Kerberos already displays features typical of the hyainailourines-and by extension of the Hyainailouridae. The hind limb of Kerberos exhibits features consisted with a preference for terrestrial substrates and a plantigrade posture. Because of its size and the absence of adaptations to cursoriality, the new taxon was probably not only an active predator, but also a successful opportunistic scavenger.
The powerful masticatory musculature supports the scavenger behavioral hypothesis usually evoked for Miocene hyainailourines. This contrasts with the contemporaneous Hyaenodon. These hyaenodontines, which probably originated in Asia and were represented by numerous species in Europe [9,136], exhibit terrestrial (semi-plantigrady/digitigrady) and cursorial postcranial features [33,74,137]. Kerberos was ecologically more similar to the striped and spotted hyenas than to contemporaneous hyaenodontines except that Kerberos lacked the cursorial adaptations of these extant hyaenids.
The respective endemic evolution of hyaenodontines (in Asia) and hyainailourines (in Africa) explains both similarities and differences between these two clades, which became competitors in Europe and North America, but our knowledge of this history still needs to be improved.