Reassessment of the enigmatic ruminant Miocene genus Amphimoschus Bourgeois, 1873 (Mammalia, Artiodactyla, Pecora)

Amphimoschus is an extinct Eurasian ruminant genus, mostly recorded in Europe, without a close living relative and, hence, an unknown systematic position. This genus is known from around 50 localities from the late early to the middle Miocene. Two species were described during 180 years, but since their first description during the late 19th century and early 20th century, hardly any detailed taxonomic work has been done on the genus. Over the years, extensive collecting and excavating activities have enriched collections with more and more complete material of this still rare and enigmatic animal. Most interestingly, a number of skull remains have been unearthed and are promising in terms of providing phylogenetic information. In the present paper, we describe cranial material, the bony labyrinth, the dentition through 780 teeth and five skulls from different ontogenetic stages. We cannot find a clear morphometric distinction between the supposedly smaller and older species Amphimoschus artenensis and the supposedly younger and larger species A. ponteleviensis. Accordingly, we have no reason to retain the two species and propose, following the principle of priority (ICZN chapter 6 article 23), that only A. ponteleviensis Bourgeois, 1873 is valid. Our studies on the ontogenetic variation of Amphimoschus does reveal that the sagittal crest may increase in size and a supraorbital ridge may appear with age. Despite the abundant material, the family affiliation is still uncertain.


Introduction
The late early Miocene was a time period of huge changes in the European mammalian communities. A deep faunal renewal associated with a peak of first appearance data at the genus and family levels is observed during the MN3 biozone (early Orleanian, from ca. 19.5 to 17.2 Mya; e.g., [1,2]). A tectonically driven land bridge emerged between Eurasia, Africa and Anatolia permitting large-scale dispersals to occur [1][2][3]. Western European ruminants were a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 (NMB), the Paläontologische Museum der Universität Zürich, Switzerland (PIMUZ), the Musée d'Histoire Naturelle Toulouse, France (MHNT), the Museum National d'Histoire Naturelle de Paris, France (MNHN), the Musée des Sciences Naturelles d'Orléans, France (MSNO), the Staatliche Naturwissenschaftliche Sammlungen Bayerns-Bayerische Staatssammlung für Paläontologie und Geologie, Germany (SNSB-BSPG), the Staatliches Museum für Naturkunde Stuttgart, Germany (SMNS), and from the private collection of Dr. D. Kälin. "No permits were required for the described study, which complied with all relevant regulations." All measurements were taken with a precision of 0.1 mm. 3D models of the skull MNHN.F.Ar3466 (based on a scanning of the surface) and the ear regions of the skulls MNHN.F.Ar3466 and SMNS 40693 (based on a CT scanner data) are available online [37].

Synonymy list
Conventional indications used before the year in the synonymy list follow Matthews [40]: � , the work validates the species; v, the authors have seen the original material of the reference; non, the reference actually does not belong to the species under discussion; p.p., in part; no indications, the authors were unable to check the validity of the reference.

Statistics
We performed an analysis of variance (ANOVA) to test if the size of the specimens from the various biozones were significantly different, as suggested in previous articles. Then we separated specimens from MN3 and MN4 (representing the supposed smaller species Amphimoschus artenensis), and those from MN5 and MN6 (representing the supposed larger species Amphimoschus ponteleviensis). The specimens from the Falun Formation were not included in this analysis because they may come from different biozones due to reworking [4,41,42]. The analysis was run on the lower third molar since it is the most diagnostic tooth of the genus. Phylogenetic analysis. Phylogenetic analyses were based on a data matrix encompassing 65 characters (dentition, cranium, and postcrania), and 19 taxa. This matrix is based on the work of Aiglstorfer et al. [43], Sánchez et al. [44], Heckberg [45], and new characters were added (see S1 Data). Several characters were constrained as ordered irreversible (0!1): 12, 29-35, 44-45, and 64. The remaining characters were treated as unordered. Maximum parsimony analyses of this updated character-taxon matrix were performed with PAUP version 4.0a [46] using a heuristic search with 10 000 replicates and 100 trees saved by replication, and ACCTRAN character state optimization. Recent molecular-based phylogenetics suggest that Antilocapridae and Giraffidae are the most basal families of living Pecora, while the Bovidae and Moschidae form a clade which is a sister group to the Cervidae (e.g. [47][48][49]). However, no morphologically-based analysis has so far produced similar results mostly because of strong homoplasy in all these lineages (for alternative topologies see [44,50,51]). In order to take into account the molecular phylogenies, we applied a topological constraint to our phylogenetic analysis with a molecular backbone of the living species included in this study (Antilocapra americana, Giraffa camelopardis, Muntiacus muntiac, Cervus elaphus, Capreolus capreolus, Boselaphus tragocamelus, Eudorcas rufifrons, Moschus moschiferus) based on the phylogeny provided by Hassanin et al. [47]. The tragulids Hyemoschus aquaticus and the Miocene Dorcatherium crassum were used as non-pecoran outgroups. The position of the early Miocene species Procervulus dichotomus has been constrained as a stem Cervidae as it possesses antlers, the main synapomorphy of cervids [16,32,52], and it has been widely considered as the oldest known Cervidae [5,10,25,32,43,45,[52][53][54][55][56]. Setting this taxon at the base of cervids may help orientate the polarity of the characters in early Miocene pecorans. The first ascertained bovids (Eotragus and Namacerus), the moschid Micromeryx, as well as the extinct antilocaprid Stockoceros, are also included in the analysis as well as the early Miocene ruminants Amphitragulus and Dremotherium as members of the early Miocene radiation of the Pecora without a welldefined family attribution [2], and Ampelomeryx as an early member of the giraffomorph Palaeomerycidae [2].

New diagnosis
Pecoran ruminant possessing enlarged upper canines; a high and strong, obliquely arranged ectostylid on lower molars without external postprotocristid; and the third lower molar possessing a very well developed entoconulid forming a double crescent or lobe with the hypoconulid. The mesostyles of the upper molars, in labial view, and the metastylids of the lower molars, in lingual view, are inclined anteriorly. The tooth enamel is wrinkled. The tooth crowns are high in comparison to all known contemporaneous European ruminants. The diastema between the upper/lower canine and the upper/lower cheek teeth is almost equal to the antero-distal length of P/p2-4. The maxillary interalveolar crest is concave in ventral view. The cranium and the mandible are massive. The maxillary possesses a strong facial crest at the level of the orbits. The auditory bullae are elongate with an elongate neck, a posterior styloid process, and an anteriorly placed lamina vaginalis. Amphimoschus has two lacrimal orifices, and no external lacrimal fossa. The bony labyrinth shows an advanced state of the cochlea, elongated in comparison with contemporaneous ruminants (2.75 turns), detached from the vestibule, and a relatively similar thickness of the different turns. In adults, the petrosal bone has a large epitympanic wing that encircles the promontorium. The apex is blunt. The fossa for the tensor tympani muscle is large and round. The basicapsular groove is present and dorsal.

Etymology
Genus name means close to the genus Moschus in Greek.

Nomenclatural acts
In 1838, von Meyer [19] created the species Cervus lunatus from Käpfnach (MN5). However, he did not designate a type specimen, neither did he publish an illustration, nor make a description. He only stated that it cannot be distinguished from the present day deer Cervus. Actually, some of the specimens designated as Cervus lunatus indeed belong to Cervidae (PIMUZ A/ V0334, PIMUZ A/V0349, and PIMUZ A/V0364). Following this, Stehlin [18] synonymized this species with Amphimoschus pontileviensis Bourgeois (with the typo). In its original publication, Bourgeois [4] did not designate a type specimen. Four dental remains and fragments of both a humerus and metapodials found in Thenay were illustrated. Stehlin [18] considered that only the three illustrated mandibles should be types without any specific designation. Three of these illustrated specimens were recovered in the collection of the  [4]) and Mayet (1908, Fig 92 [9]) since this portion of mandible presents diagnostic characters for the genus "Amphimoschus":the elongated diastema between c and p2, the p4 without a developed mesolingual conid, and a molar with a prominent ectostylid, without an external postprotocristid, a closed posterior basin, and a prehypocristid that does not fuse with any other cristid. Consequently, MNHN.FP.1070 (Bourgeois 1878, Pl. X, Fig 1 [ [9]) are considered as paralectotypes, thus completing the diagnosis of the genus, including the shape of the third lobe of the m3. It is interesting to note that in the original article of Bourgeois [4], the specimens are mentioned as being found in "Pont-Levoy" (= Pontlevoy), while the precise locality is further described later in the text as "Thenay", a village close to Pontlevoy. Given this, Thenay should be considered as the type locality. Moreover, in this original description of A. ponteleviensis [4], Bourgeois specifies that the illustrated fossils have been found in the fluvial sands belonging to the Sables de l'Orléanais Formation (MN4) and some reworked specimens can also be discovered in the abovelying layers of the marine deposits from the Falun Formation (MN5b). In Mayet [9], the specimens originally reported by Bourgeois were considered to be found in the Falun-type beds, which served to justify the occurrence of a smaller species, A. artenensis, in the older locality (MN4) of Artenay. However, the fluviatile beds exposed in Thenay-Pontlevoy are not the Sables de l'Orléanais Formation but the Sables et Marnes du Blésois that are MN5a [8].

Etymology
The species name refers to the Pontlevoy (Loir-et-Cher, France), the closest city to the Thenay locality where the species was first discovered.

Diagnosis
As for the genus Amphimoschus.

Skull SMNS 40693 (Figs 4 and 5)
The skull SMNS 40693 is a cranium of a young individual with left and right deciduous tooth rows, M2 has totally emerged and M3 is erupting. The skull suffers strong dorsoventral flattening with a small amount of distortion. This skull is almost complete. Only the premaxillary bones are missing. The upper part of the skull is broken into small pieces. It is very hard to identify the bone sutures. The right basilar area is the best-preserved part of the hornless skull.
Facial region. The maxillary bones are preserved from the anterior part of the left diastema and the right D2. The left canine is partially erupted. Only its elongated root is partially preserved on the skull. Its apex is separated from the skull and not worn at all. The maxillary bulge above the left canine root makes a groove on its upper part separating the maxilla from the nasal bone. The nasal bones are broken along their anteroposterior length. The lacrimal foramina and the infraorbital foramina are not visible due to the poor preservation of the specimen in that area. There is no lacrimal fossa. The maxillary interalveolar crest is preserved on the left side of the skull. It is highly concave and slightly deformed by distortion. It joins the anterior part of D4. The left greater palatal foramen is located at the level of M1. The medial palatal suture is quite flat, with a small crest in the middle.
Orbital region. Both orbits are complete, but are flattened and disarticulated. The orbits are quadrangular with a flat upper and lower surface. Both zygomatic arches are very strong and bear a huge facial crest. Both zygomatic processes of the frontal bones are preserved, but dislocated. The frontal processes of the zygomatic bones are tilted slightly backwardly. The dorsal part of the frontal bones is fragmented. Both supraorbital foramina are preserved with a large supraorbital groove. Behind the orbits, the frontal bones possess an oblique temporal line, which is linked to the sagittal crest. The lacrimal bones cannot be distinguished. This fossil differs from PIMUZ A/V4656 (Fig 7A) in lacking the facial crest above the orbits.
Dorsocranial-temporal region. The left parietal bone is fragmentary. The right temporal bone is the best preserved. The retroarticular process is well developed. It lines the anterior part of the auditory bullae, just at the level of the external auditory meatus. The alisphenoid wing, which is just in the axis of the retroarticular process, forms another crest. The posterior

PLOS ONE
part of the temporal bone is poorly preserved. Only the left mastoid process is preserved. Its crest joins the well-developed nuchal crest. The mandibular fossa is shallow. The zygomatic process of the temporal bone exhibits a small tuberosity on its ventral part.
Nuchal region. The occipital bone is well preserved. The nuchal crest is well developed. The protuberance formed by this crest is prominent. The nuchal tuberosity is cracked on its upper part. The foramen magnum is circular. The occipital condyles are triangular in caudal and ventral views. They extend by a rather large tuberosity. On the basilar part of the occipital bone, there is a spine protruding from the axis of the bone. The hypoglossal nerve canals are both preserved. They are large and oval, positioned at the level of the basilar bend. They cross the occipital condyles and can be observed on the proximal part of the ventral condylar fossa. Another foramen, probably the vascular foramen, is situated at the end of the vascular fossa.
Basicranial region. The basioccipital bone is rectangular in shape. The left paroccipital process is well preserved. It is triangular in shape in posterior view, flat ventrally, rounded dorsally. Close to the suture with the basisphenoid bone, there are the muscular tuberosities of the basilar process. The basisphenoid bone is short. The pterygoid process borders it. The left pterygoid hamulus is preserved. On the alisphenoid bone, there is a small oval foramen just between the styloid process and the basisphenoid bone. The palatine bones are preserved but cannot be distinguished from maxillary bones. No foramen can be observed. The choanae are preserved in three dimensions and rectangular in shape. The start of the vomer is located at the middle part of the choanae.
Auditory region. The left auditory bulla is almost in anatomical position. The right bulla is broken and the petrosal bone has been moved to the top of the skull roof, through the frontal bones. The left external auditory meatus has a partially preserved vaginal process and is broken. The right external auditory meatus is well preserved. It is circular, with a notch on its ventral part. The left part of the skull exhibits the mastoid process, the external auditory meatus, the retroarticular process and the mandibular fossa. The two petrosal bones can be seen on the dorsal part of the skull. They are triangular in section due to their insertion with the occipital bone just below the ventral condylar process. The petrosal crest can be observed on the dorsal part of the skull. The petrosal bones are described later in detail. The caudal border of the left bulla is at the level of the rostral border of the basioccipital bone. On its posterior part, the bulla touches the paroccipital process, forming a narrow and triangular depression between the bulla and the basioccipital. The external auditory meatus is elongate, straight, and ended by the bulla, which is oriented at an acute angle. The bullae are bean-shaped (larger dorsoventral dimension than rostrocaudal) and have a smooth surface. The right bulla is slightly compressed anteriorly. The right vaginal lamina extends to the middle of the bulla. Due to poor preservation, the styloid process cannot be seen, but it is located on the posterior part of the bulla. The caudal ends of both bullae are tapered and the muscular processes are lacking. The left external auditory meatus is circular in section with two notches. The vaginal process begins on the median part of the external auditory meatus orifice and runs rostrally, enclosing a groove between the auditory bulla and itself. This groove cannot be seen on the left side, because the bulla is flattened towards the vaginal process.
Skull SMNS 45592 (Fig 6) The specimen SMNS 45592 belongs to a young individual. The first molars are erupting. The mandibles are well preserved with erupting D3-4. The left one also possesses an I3 and D2 in place. The right premaxilla is associated with the right mandible. It is toothless, high and triangular in lateral view. In inner view, the palatal fissure is elongated, narrow like the palatal process, and rounded at its extremity. The premaxillary bones are elongated and slightly enlarged on their rostral part. The nasal cavity is wide and separated in two parts by a fragmented ethmoidal bone. The left maxillary bone has D3-4 associated with the zygomatic bone, which has a facial crest. The right maxillary bone has a D4 and an erupting M1. Other skull bones are associated to this specimen: fragments of the frontal bone, the zygomatic bone with the frontal process, the temporal process, the orbital face and a very strong zygomatic crest. (Fig 7A) PIMUZ A/V4656 is a cranium of an aged individual with full postcanine dentition except left and right P4, which are missing (pathological condition). The crowns of the teeth that are present are highly worn. The caudal part of the skull is missing, i.e. broken transversally, from rostral of the left orbit and at the level of the missing P4 to the middle part of the right orbit and caudal of the right M3. This fossil is preserved in 3D, with a small deformation of the right maxillary bone. This skull is robust and globular.

Skull PIMUZ A/V4656
Facial region. The premaxillary bones are missing. The maxillary bones are preserved from the caudal edge of the canine alveolus opening. The alveolus of the enlarged upper canine runs in an arch along the dorsal edge of the rostral maxilla. The nasal bones are missing their rostral elongation. In rostral view, they are long, regularly laterally curved, relatively narrow, and straight. Their contact with frontal bones is straight. The infraorbital foramens are large and oval. There is no lacrimal fossa. The maxillar interalveolar crest on the diastema between canine and P2 is preserved on both sides of the skull. It is strongly arched towards medial, such as in Bos taurus Linnaeus, 1758, and the extinct Bedenomeryx Jehenne, 1988 [79], which differs from the extinct Micromeryx Lartet, 1851 [80] (e.g. [43]) and the living Moschus Linnaeus, 1758 [59] where the crest is straight and parallel to the sagittal plane. The greater palatal foramina are small, located at the level of the posterior part of the M1, with a clear groove just rostral of it running up to the level of P3. In ventral view, the palate is transversally concavely arched, with a weak medial palate suture crest formed by the suture of these bones. Orbital region. Only the right orbit is partially preserved. The ventral and dorsal portions of the orbit are flat. The frontal bones are well preserved in the area of the articulation with nasal and lacrimal bones. In lateral view, their dorsal profile is flat as is the case with the nasal bones. There is a prominent supraorbital ridge from the nasal bone forming the dorsal edge of the orbit. At the rostral end of this crest, a sigmoid supraorbital groove joins a large supraorbital foramen. There is the typical double lacrimal orifice of cervoids, in dorsoventral arrangement. The lacrimal bones are flat and arranged in a sub-sagittal plane. The lacrimal bones do not to reach the nasal bones due to the presence of a small antorbital vacuity and the lack of a caudal lacrimal process. The facial crest is well marked below the orbit. (Fig 7B) PIMUZ A/V1783 consists of two fragments of skull belonging to a young adult. All the teeth are little worn. Only the left and right maxillary and zygomatic bones are partially preserved. All permanent teeth are erupted; the right P2 is missing.

Skull PIMUZ A/V1783
Orbital region. The facial crest begins at the level of the orbits and continues on the huge and sigmoidal zygomatic arch. The oval infraorbital foramen is large and dorsoventrally elongated. Its external opening is located directly rostral, just in front of P2 and internally it opens between P2 and P3. The bulge running along the rostro-dorsal edge of the maxillary bones houses the canine root and its alveolus. The interalveolar crest between C and P2 is partially preserved and arched towards medial in palatal view. The maxillary bone does not extend caudally beyond M3. Skull MNHN.F.Ar3266 (Fig 8) The cranium MNHN.F.Ar3266 is an almost complete skull. Similarly to PIMUZ A/V1783, all of the teeth are little worn. Only its rostral part and auditory region are damaged. The bone sutures are fully fused and cannot be traced because of both the age of the individual and the preservation.

PLOS ONE
Orbital region. Both orbits are quadrangular with flat dorsal, ventral, and caudal edges. Both zygomatic arches are very strong and possess a huge facial crest. The frontal processes of the zygomatic bones are perpendicularly arranged. The left supraorbital foramen is preserved with a small and weak supraorbital gulley. Caudal to the orbits, the frontal bones show a temporal line, which joins the sagittal crest. This fossil differs from PIMUZ A/V4656 (Fig 7A) in lacking the supraorbital ridge. There is no lacrimal fossa. Only one lacrimal foramen is apparent, but this portion of the skull is poorly preserved, and two foramina may have been present. The frontal bones are flat in shape.
Nuchal region. The supraoccipital bone is high and narrow. A very strong sagittal crest joins the preeminent nuchal crests to form a knot. The nuchal crests are preeminent and backwardly projected, like in the early bovid Eotragus Pilgrim, 1939 [81], but are less marked than in the early crown Pecora as in Dremotherium Geoffroy Saint-Hilaire, 1833 [82] and Bedenomeryx. The paroccipital process is prominent, extending ventrally.
Basicranial region. The basioccipital bone is of triangular outline in palatine view, elongate, and exhibits two weakly distinguished muscular tuberosities. Like in Eotragus, the basioccipital bones and the basisphenoid are slightly tilted ventrally compared to the rest of the skull axis. In comparison, the stem cervid Procervulus Gaudry, 1878 [83], for example, has more horizontally arranged basioccipital and basisphenoid bones.
Petrosal bones (Fig 9) Ventrolateral surface. The promontorium shows two different portions: the caudal one with an unusually bulged drop-like shape and strong medio-lateral elongation, and the much flatter rostral part separated by a deep valley-like structure. The epitympanic wing is very broad and has a rather blunt apex. The caudomedial flange of the promontorium is widely expanded and continuous with the epitympanic wing. A faint but wide and curved transpromontorial sulcus is visible on the flat part of the promontorium. The fenestra cochleae (i.e., round window) is much larger, at least twice as large as the oval-shaped fenestra vestibuli (i.e., oval window). The former is slightly more caudally positioned than the fenestra vestibuli and faces laterally. A broad crista interfenestralis separates both windows. The fossa for the tensor tympani muscle is very large, taking up about half the size of the promontorium. It is round to bean-shaped and protrudes into the tegmen tympani. A large secondary facial foramen is visible next to the fenestra vestibuli and posterior to the fossa for the tensor tympani. It opens on the wide facial sulcus separated from the round window by a large, high, and thick wall of the promontorium. A rather elongated stapedial muscle fossa is also clearly marked. The ventrolateral tuberosity is large.
Dorsolateral surface. The tegmen tympani is inflated and has an elongated triangular shape. Its surface is rather smooth. The rostral process of the tegmen tympani is short and blunt. The hiatus Fallopii is a very distinct hole at the base of a long and wide groove emanating from the promontorial side of the petrosal bone.
Dorsomedial surface. No prefacial commissure fossa (sensu O'Leary [84]) is visible. The internal acoustic meatus sits in the middle of the surface and shows both foramina acusticus (superius and inferius) separated by a thin crista transversa. A basicapsular groove appears to run on the lateral side of the petrosal (or "occipital surface" sensu Mallet and Guadelli [85]) down to more or less the level of the internal acoustic meatus. The subarcuate fossa is shallow, as in all pecoran ruminants, and wide. The opening for the vestibular aqueduct lies lateral to the subarcuate fossa. It is a large slit situated posteriorly to the cochlear aqueduct lying on the ventromedial surface of the petrosal bone. The mastoid region is very large and wedge-shaped.

PLOS ONE
Ventromedial surface. The basicapsular groove is visible and it is more dorsal than ventral. It runs down to the cochlear aqueduct. The latter is a large and distinct hole linking four grooves: the basicapsular groove coming from the apex, two grooves coming from the ventrolateral surface (mastoid fossa) and from the dorsomedial surface, and one from the mastoid area.
Bony labyrinth (Fig 9) The cochlea makes 2.75 turns. The aspect ratio (total height divided by total width) is 0.55 classifying the cochlea as a rather high one (sensu Gray [86]). The cochlear spiral is tight; the turns are relatively thin and asymmetric in section. The second turn is slightly keeled. The apical turn is rather flat. The secondary bony lamina runs on the dorsal surface of the basal turn over more than one turn. The basal turn begins caudal of the fenestra cochleae and shows a slight lateral expansion on the scala tympani. The basal turn is thus slightly inflated. The cochlear aqueduct is long and thin with a circular cross section. It ends in a thick triangular pouch-like structure. Both fenestrae are well visible. The relatively small fenestra cochleae is slightly more posteriorly positioned than the fenestra vestibuli. The fenestra vestibuli is not very ellipsoid in shape, with a stapedial ratio of 1.42 (long axis divided by short axis). On the vestibule, both the utriculus and sacculus are well defined, the sacculus being round in shape while the utriculus is elongated and ellipsoid in shape. The anterior semicircular canal is the most dorsally expanded canal; it expands above the level of the common crus while the posterior semicircular canal extends only slightly above the common crus. The common crus is straight and elongated. The lateral semicircular canal is straight over its entire course even close to the posterior ampulla. It branches on the posterior semicircular canal creating a short secondary common crus, a characteristic seen in basal artiodactyls [87], but also in basal pecoran ruminants. The ampullae at the base of the semi-circular canals are well defined and globular. The vestibular aqueduct is a rather thick canal detached from the common crus. It originates at the base of the common crus in a thick pear-like insertion, and very much anterolaterally to it. It runs up the dorsal most extent of the common crus. The endolymphatic sac at the end of the vestibular aqueduct is a small thin and flat triangular structure ending at about the dorsal most extent of the anterior semicircular canal.

Mandible (Figs 10 and 11)
The corpus is generally the best-preserved part of the mandibular remains (Figs 10 and 11). This portion is deep and massive. It regularly enlarges from p2 to m3. The masseteric fossa is shallow and flat, marked by a barely distinguishable rostral ridge (Fig 10A). The mandibular foramen is large and deep. It is located on the ramus in line with the molar row. A curved mylohyoid groove crosses a deep and well-marked pterygoid fossa. On the medial part of the mandible, between the corpus mandibularis and the mandibular angle, there is a fairly prominent bony crest for the insertion of the masseteric muscle. The space between the m3 and the ramus mandibulae space may be short. The incisura vasorum is marked and rounded. The corpus is comparatively large and massive, with a slightly concave margo ventralis. In occlusal view, the mandible narrows at the level of the diastema, just rostral of p2. The large and oval mental foramen is located at the level of the mandibular symphysis. The long diastema separating the canine and p2 forms a sharp crest dorsally (Figs 10 and 11). The lingual groove covers the caudal half of the mandible from m3. The ramus is large and oriented dorsal to subdorsal. The coronoid process is quite large and straight. The condylar process is in the axis of the ramus in the ontogenetically youngest and slightly off-centred in individuals of older individual age. There is a clear incisura mandibulae between the condylar process and the beginning of the coronoid process. The coronoid process is well rounded.
Lower dentition (Figs 10 and 11) Deciduous lower dentition. di1 is slightly rounded in its apical portion and is the smallest incisor. di2 is larger, slightly rounded in its posterior part, and straight in its anterior part. di3, the largest incisor, is procumbent, trapezoid, and strongly asymmetric with the posterior lateral section being the longest. The d2 and d3 resemble their successors in the permanent dentition (p2 and p3), d3 being longer than d2. The mesolabial conid is the highest cusp and centrally situated; and this mesolabial conid is more labial on d3. The anterior conid forms the anterior edge of the tooth. The transverse cristid is lingual, short, and posteriorly oriented on d2. It connects the mesolabial conid to a small oblique mesolingual conid on d3. The

PLOS ONE
oriented. The latter does not join the premetacristid, but it can be fused with the posterior cristid of the anterolabial conid. The labial cuspids are relatively aligned together. The metaconid is smaller than the entoconid, but it possesses the same orientation. The metaconid rib is well formed, and the metacristids are straight. The premetacristid ends within the anterior fossa. Between the protoconid and the anterolabial conid, the anterior ectostylid is large. The anterior cristid of the anterolabial conid joins the anterior cristid and the anterolingual conid, forming an angle on the anterior part of the tooth. The anterolingual conid is flat and anteroposteriorly oriented. There is no cingulid.
Permanent lower dentition. The enamel of the tooth crown is wrinkled. p1 is absent. p2-4 have nearly the same occlusal pattern, the premolars being more molarized and more elongate from p2 to p4. They are triangular in labial view due to a prominent posterolabial conid, well distinguished, with a groove on its anterolabial part. p2 lacks the anterior conid and the mesolingual conid. On p2-3, the posterior stylid is absent. The posterolabial conid becomes smaller from p2 to p4, whereas the mesolabial conid becomes larger. The mesolabial conid is the highest cusp of the premolar, central, but generally poorly distinguished. The anterior stylid and the anterior conid are well developed on p3-4. On p4, the mesolingual conid is barely distinguishable, with a well-developed backwardly-oriented transverse cristid. The molars are selenodont with relatively high cuspids on unworn teeth compared to other contemporaneous pecorans. The enamel is wrinkled. All the lingual cuspids are aligned forming a relatively flat lingual wall. There is a very weak metastylid. The protoconid, the hypoconid and the hypoconulid are large. The anterior and posterior fossae can be lingually closed by the preprotocristid joining the premetacristid and the posthypocristid joining the postentocristid respectively. No external postprotocristid can be observed. The prehypocristid is relatively transverse, and touches neither the entoconid nor the internal postprotocristid. The posthypocristid is relatively labio-lingually oriented. The ectostylid is well individualized, forming a small column between the protoconid and the hypoconid. Its base can be circular, ovoid, or elongated. The entoconulid is a well-developed and individualised cusp. It is as large as the metaconid and the entoconid, forming a double crescent third lobe of m3. The back fossa of m3 is bounded by a high posthypoconulidcristid delimitating a large basin. The molar crown has a well-marked anterior cingulid.

Upper dentition (Figs 4-8)
Deciduous upper dentition. The upper deciduous premolars are relatively compact and molarized in missing an anterior cone associated to an anterolingual crista. The size of the protocone and the metaconule widely differs, D2 having both cusps reduced with a larger metaconule than the protocone, D3 having only the metaconule, and D4 having both cusps well developed giving the tooth a molar form. Regarding the cristae, there is a clear complexification of the crest pattern from D2 to D4. The preprotocrista extends labially and tends to reach the parastyle. On D4, the postprotocrista can be bifurcated. Between the protocone and the metaconule, the entostyle is relatively high. The premetaconulecrista is always bifurcated on D4 and sometimes on D3 as well. These crests remain, however, isolated without any fusion with other crests. A metaconule fold, antero-posteriorly oriented, crosses the posterior fossa on D3 and D4 from the posterolabial part of the metacone to the center of the postmetaconulecrista. The labial wall is relatively flat. The paracone is well developed and possesses a labial rib with an anterior groove. The metacone becomes larger on D4 and its labial side displays a weak rib. The parastyle and the metastyle are relatively globular and salient. However, only the D4 possesses a mesostyle, which is extremely reduced.
Permanent upper dentition. The upper canine is long and regularly curved. Its anteroposterior section is triangular at its apex. The upper premolars share a similar basic morphology, but are increasingly molarised from P2 to P4. In this context, they increase in length and width from P2 to P4, and change from an ovoid shape in P2, a half-moon shape in P3, to a more triangular shape in P4. The anterolabial cone is the highest cone, becoming more central from P2 to P4, while the lingual cone increases in size from P2 to P4, and becoming more crescentic on P4. The anterolabial crista becomes more anteriorly elongated associated to a less prominent anterolabial cone ridge. The anterior style is salient, but weak. A posterior style is absent. The posterolabial cone is well formed on P3, as large as the anterolabial cone, and less visible on the other premolars. The posterolabial crista and the posterolingual crista do not fuse, leaving a gap on the posterior part of the teeth. The central fold joins the middle part of the posterolingual crista to the center of the lingual cone framing a large basin separated from the fossa by this fold, the anterior basin being larger.
The upper molars increase in size, but decrease in complexity from M1 to M3. The pre-and postprotocrista are labially oriented. On M2-3, the preprotocrista reaches the parastyle. The premetaconulecrista never fuses with the postprotocrista. The latter is generally labially oriented with a distal end anteriorly oriented section. The premetaconulecrista is also oblique and ends between the metacone and the paracone by means of a bifurcation, which remains unconnected to any adjacent cusp and crest. The bifurcation becomes less prominent on M2 and M3. The postmetaconulecrista is labially oriented and it reaches the base of the metastyle. The metaconule fold becomes less marked from M1 to M3. The metaconule fold can be isolated, connected, or bifurcated. On the anterior part of the metaconule, the entostyle forms a rounded small column, which may be elongated. The metacone and paracone are well developed, relatively flattened labially, without prominent ribs. The labial side of the metaconecomplex becomes more concave from M1 to M3. The metastyle does not form a small column. The mesostyle and the parastyle are small and sharp. The pre-and postmetacrista are slightly curved and join the mesostyle and the metastyle respectively. The post-and preparacrista are straight and aligned.

Phylogenetical results
PAUP recovered 39 equally parsimonious trees (MPT) of 230 steps, a consistency index (CI) of 0.417 and a retention index (RI) of 0.575. The distribution of character states for the internal nodes and the list of autapomorphies are presented in the S1 Data.
The strict consensus is poorly resolved with a basal polytomy from which only the clade bovids+moschids is preserved. Thus, we here discuss the majority rule consensus (50%) (Fig  12): the values on the branches corresponding to the frequencies for all observed bipartitions. In the majority consensus, Amphitragulus is basally placed as a sister taxon of the rest of the in-group as in Sánchez et al. [44] and Mennecart et al. [51]. Then, we observe a general polytomy separating a clade formed by Dremotherium + Ampelomeryx, Antilocapridae (extant and fossil) + Amphimoschus, Giraffa, Cervidae (extant and fossil), and Bovidae (extant and fossil) + Moschidae (extant and fossil). Here, Ampelomeryx is not grouped with Giraffa within the Giraffomorpha, as opposed to Sánchez et al. [44] and Mennecart et al. [51], and Dremotherium is not closely related to the "Cervidae" [44] nor to the Bovidae + Moschidae [51]. The clades (Moschus + Micromeryx) and (bovids + moschids) are well supported. Amphimoschus appears as a sister group of the well supported clade encompassing Antilocapra and Stockoceros. The position of Amphimoschus close to antilocaprids is supported by 12 synapomorphies (S1 Data), even though four of them are not coded for Amphimoschus (characters 19, 53, 54, and 56). Amphimoschus possesses three autapomorphies: the anterior position of the maxilla-palatine-suture line (21), the position of the foramen ovale with respect to the infratemporal fossa (39), and the bicuspidate third lobe of the m3 (44, constrained as irreversible in the analysis). The clade Antilocapridae is also supported by 12 synapomorphies (characters 17,18,22,23,25,28,34,40,46,47,50, and 51; see S1 Data).

How many species of Amphimoschus?
Two species of Amphimoschus are frequently cited in the European Orleanian and earliest Astaracian deposits: Amphimoschus ponteleviensis Bourgeois, 1873 [4] from Thenay Sables et Marnes du Blésois (MN5a) (misattributed to Thenay Sables de l'Orléanais Formation MN4 [4] and Pontlevoy Falun Formation MN5b [9]) and Amphimoschus artenensis Mayet, 1908 [9] from Artenay (MN4a). The stratigraphic range of Amphimoschus artenensis covers the early Orleanian whereas that of A. ponteleviensis includes the late Orleanian to earliest Astaracian [16,32,35]. The distinction between these two species is based on the height of the crowns, the shape of the ectostylid base, the more bulged metaconid and the overall tooth-size [9,16,64]. It is interesting to note that Mayet [9] published significantly larger measurements of the type specimens from Thenay than what we observed, with a discrepancy in size of 1 to 3 mm (for example MNHN.FP.876 m1: 12.2 vs 15; MNHN.FP.1070 p3: 10.1 vs 13, the first measurement being our data and the second from Mayet [9] respectively).
Based on wear features of the tooth crowns, the large species, A. ponteleviensis, was classified as a browser, while the smaller (A. artenensis) was considered to be a mixed feeder [35]. Stehlin [18] had previously explained, that the size difference between the two species are not

PLOS ONE
that clear, as there are huge size overlaps when considering larger samples of the two "mutations", as he puts it. However, in absence of further material, no conclusion was drawn.
In this study, the teeth of 250 specimens have been measured (S2 Data) and 780 teeth have been observed at the locality level (S3 Data). Tooth morphology varies little within an assemblage (i.e. a horizon of a fossil locality). For example, the morphology of the ectostylid can be highly variable (column with a circular, triangular, or even extended shape) in a same locality. The postentocristid and the posthypocristid may or may not be fused depending on the age of the individual and the wear state of the teeth. Likewise, the fusion of the premetacristid and preprotocristid may vary in an assemblage. The cranial characters remain difficult to interpret due to different ontogenetic stages and no clear sexual distinction.
The lower m3 is the easiest recognisable tooth (because of the uncommon feature of a double crescent in the third lobe). This is the most frequently identified tooth, with 87 specimens observed by the late Léonard Ginsburg when he started his unfinished study of Amphimoschus. Regarding size variation, important overlaps exist between fossil materials from different biochronological zones from MN3 to MN5. In this study (S2 Data), we compared 65 third molars from different localities and ages. The results of the ANOVA show that the Amphimoschus chronological assemblages cannot be distinguished based on the length of the m3 from the three biozones: MN3, MN4, and MN5 (p = 0.5586; main variance differences being observed comparing the data from MN5 to the other ages). Tooth crown height and metaconid shape are also similar across the whole dataset, we could not observe the differences that Mayet [9] reported. Accordingly, size differences and tooth morphology cannot provide an empirical basis to justify the validity of both species. Hence, we conclude that there was only one species of Amphimoschus, which roamed Europe during the late early to the middle Miocene. The nomenclatural priority goes to Amphimoschus ponteleviensis Bourgeois, 1873 [4].
Looking at the size and shape of the teeth, we agree with Stehlin [18] in attributing the specimens ascribed to Cervus lunatus by von Meyer [19] from Käpfnach (MN5) to A. ponteleviensis. Mayet [9] mentioned the presence of an Amphimoschus most likely similar to the smaller form from Artenay in Neuville aux Bois (MN3). Unfortunately, the specimen (MSNO 224) was lost during the early 20 th century [9]. He also mentioned a specimen "b" not included in our analysis since the published sizes are much larger than observed in Amphimoschus ponteleviensis. Raia et al. [88] referred to an "Amphimoschus elegans" based on their geometric morphometrics analysis. This undescribed species might actually be the result of confusion between the genera Amphimoschus and Amphitragulus. The mandibles of these genera are different in shape [6]. The mandible described by Raia et al. [88] looks very similar to those of Amphitragulus elegans of Mennecart et al. [6].
Based on this information, only the species Amphimoschus ponteleviensis Bourgeois, 1873 [4] should be considered as valid in the European fossil record.

Ontogeny of Amphimoschus
The studied crania belong to different ontogenetic stages. The youngest one (SMNS 45592, Fig  6) only possesses a d4, a D4 and an erupting M1. No wear can be observed on the occlusal surface of teeth. By comparing to age classes from living ruminants, the specimen would not be older than three months [89]. In another specimen (SMNS 40693, Fig 4), all deciduous premolars and molars are preserved and the m3 is erupting, suggesting they are from an animal in its second year. The sagittal crest of the latter is not as developed as in specimens belonging to older individuals (e.g., MNHN.F.Ar3266, Fig 8). In specimen PIMUZ A/V4656 (Fig 7A), all the molars are heavily worn. In addition, this cranium possesses some unique features. Both P4s are missing, which could be due to a palaeopathology. The teeth may have been retained in a bud stage, or they were already lost due to the advanced age of the individual. However, the bilateral absence of these two teeth and the reduction of the gap between P3 and M1 may indicate the original absence of these structures. Moreover, this specimen has very strong supraorbital ridges, which are absent in the other specimens. In extant ruminants, such a feature is only observed in Muntiacus Rafinesque, 1815 [90] and may be associated with cranial appendages. However, so far no cranial appendages have been found associated with Amphimoschus.

Phylogenetic position of Amphimoschus
The absence of cranial appendages in the described Amphimoschus crania makes its phylogenetic affinities uncertain. All authors are in agreement about pecoran affinities based on the fully fused metapodial bones (observed in Bourgeois 1873 [4]), massive and complex p4, relatively high crowned molars in comparison to the contemporaneous European ruminants, fully selenodont teeth, and the shallow subarcuate fossa on the petrosal. However, its phylogenetic position within pecoran ruminants remains highly debated. Bourgeois (1873) [4] described Amphimoschus as closely related to the hornless Moschus (Moschidae). However, except for Hydropotes inermis (the only living ruminant where the absence of appendages is known as secondarily lost), the absence of cranial appendages is a plesiomorphic character.
Amphimoschus is recovered in our phylogenetic analysis as the sister taxon of the Antilocapridae (the pronghorns). This family is known by a relatively rich Neogene fossil record in North America where the family is known today by only a single extant representative (Antilocapra americana) [91]. Paracosoryx and Merycodus, the earliest known antilocaprids, are dated to ca. 18.8 My [91] which is about the age of Amphimoschus occurrences in Europe (see Table 1). Since no older Pecora have been found in Northern America, it is admitted that Antilocapridae most likely originated in the Old World and subsequently migrated to North America [91]. Similarly to extant and fossil Antilocapridae, Amphimoschus possesses a short vestibular aqueduct (15), a lateral semicircular canal partly fused with the posterior ampulla (16), absence of contact between the postglenoid process and the external acoustic meatus (27), a small postglenoid foramen (28), the parieto-squamosal suture located in the middle of the upper and lower borders of the temporal fossa (36), a highly developed postentocristid that fuses with the posthypocristid, distally closing the lower molars (43), and a bicuspidate third lobe of the m3 (45). However, we refrain in our Systematic Paleontology section from classifying Amphimoschus with Antilocapridae. A short vestibular aqueduct may be convergent in different lineages [55], while the relatively low position of the lateral semicircular canal is found in Pecora other than crown Bovidae and crown Moschidae [43,56,92]. It has recently been shown that the morphology of the bony labyrinth bears a strong phylogenetical signal in artiodactyls [55,[93][94][95]. This bony structure is species-specific, and varies little [96,97]. Very little is currently known on the bony labyrinth morphology in Miocene ruminants [42,43,55,56,92]. Amphimoschus shows a mosaic of apomorphic and plesiomorphic traits of its bony labyrinth with the plesiomorphic condition of a decentred vestibular aqueduct, but the derived features of the cochlea showing a high number of turns (0.75 turns more than stem early Miocene cervids, see [55,56]). The cochlea, which is detached from the vestibule (not seen in stem cervids, see [55,56]), has turns of roughly the same thickness (again see [55,56] for a contrasting condition in stem cervid). Moreover, all the cranial characters uniting Amphimoschus with the Antilocapridae can be observed in other lineages (27/1 in Giraffoidea and Bovidae, 28/0 Giraffoidea and Procervulus, 36/0 in Cervidae and Bovidae). Leinders [20], Janis and Scott [5], McKenna and Bell [22], and Mazza and Rustioni [21] saw a close relationship of this genus with the enigmatic Hoplitomeryx Leinders, 1983 [20] (late Miocene-Pliocene, Gargano, Italy), based on the shape of the tympanic bullae, the bicuspidate third lobe of m3, and the morphology of the p4. Van de Geer et al. (page 78 [98]) state, there is not "[. . .] any evidence to consider the hornless extinct moschids Micromeryx and Amphimoschus as sister-taxa of Hoplitomeryx". They argue that "Hoplitomeryx differs from this genus [Amphimoschus] owing to the presence of cranial appendages, loss of lower p2, a non-bifurcated protocone, weakly developed entostyle and ectostylid". Likewise, the morphology of the third lobe of the m3, (bicuspidate and with well-developed post-entoconulidcristid) has appeared convergently in Amphimoschus, Hoplitomeryx spp., Lagomeryx pumilio, and moschids. Having more hypsodont teeth to cope with more abrasive food items may induce a reduction or loss of the external postprotocristid (Palaeomeryx fold) [5], the absence of a p1, and a robust corpus mandibularis [6,99] which are all also observed convergently in Amphimoschus, Antilocapridae, and Bovidae. Consequently, the position of Amphimoschus suggested by our phylogenetic analysis should be taken with caution. Moreover, despite the backbone phylogenetic tool applied on extant species and the Miocene Cervidae Procervulus, the relative systematic position of the different Crown pecoran families appears poorly resolved. If the molecular phylogenies of pecoran ruminants are congruent (e.g. [47][48][49]), those based on the morphoanatomical characters remain conflicting with the former. This is accentuated when fossils are included in the phylogenetic analysis, suggesting that convergent evolution is important among pecorans: Moschidae as a sister taxa of Cervidae in a supertree [50]; Antilocapridae closely related to Moschidae and Bovidae in a Bayesian analysis [44]; Cervidae as the basalmost pecoran in a heuristic analysis [51]. Hence, the phylogenetic position of the early Miocene and hornless Pecora remains problematic. Our phylogenetic hypothesis will have to be tested with a larger dataset, including additional taxa and new morphological features, especially for numerous extinct taxa that remain poorly documented. Augsburg, Germany (Dr. Rummel), the Musée d'Histoire Naturelle Toulouse, France (Dr. Duranthon), Dr. Kälin for giving access to his private collection, and all the people who helped us in this study. BM and LC are grateful to George Schulz and Bert Müller for providing access to the CT scanner facility. We are grateful to the reviewers Christine M Janis and Nikos Solounias and to the editors Matthew C. Mihlbachler and Thierry Smith who helped us to improve our manuscript. Their detailed and precise questions and corrections were extremely useful.