Ocepeia (Middle Paleocene of Morocco): The Oldest Skull of an Afrotherian Mammal

While key early(iest) fossils were recently discovered for several crown afrotherian mammal orders, basal afrotherians, e.g., early Cenozoic species that comprise sister taxa to Paenungulata, Afroinsectiphilia or Afrotheria, are nearly unknown, especially in Africa. Possible stem condylarth-like relatives of the Paenungulata (hyraxes, sea-cows, elephants) include only Abdounodus hamdii and Ocepeia daouiensis from the Selandian of Ouled Abdoun Basin, Morocco, both previously only documented by lower teeth. Here, we describe new fossils of Ocepeia, including O.grandis n. sp., and a sub-complete skull of O. daouiensis, the first known before the Eocene for African placentals. O.daouiensis skull displays a remarkable mosaic of autapomophic, ungulate-like and generalized eutherian-like characters. Autapomorphies include striking anthropoid-like characters of the rostrum and dentition. Besides having a basically eutherian-like skull construction, Ocepeia daouiensis is characterized by ungulate-like, and especially paenungulate-like characters of skull and dentition (e.g., selenodonty). However, some plesiomorphies such as absence of hypocone exclude Ocepeia from crown Paenungulata. Such a combination of plesiomorphic and derived characters best fits with a stem position of Ocepeia relative to Paenungulata. In our cladistic analyses Ocepeia is included in Afrotheria, but its shared derived characters with paenungulates are not optimized as exclusive synapomorphies. Rather, within Afrotheria Ocepeia is reconstructed as more closely related to insectivore-like afroinsectiphilians (i.e., aardvarks, sengis, tenrecs, and golden moles) than to paenungulates. This results from conflict with undetected convergences of Paenungulata and Perissodactyla in our cladistic analysis, such as the shared bilophodonty. The selenodont pattern best supports the stem paenungulate position of Ocepeia; that, however, needs further support. The remarkable character mosaic of Ocepeia makes it the first known “transitional fossil” between insectivore-like and ungulate-like afrotherians. In addition, the autapomorphic family Ocepeiidae supports the old – earliest Tertiary or Cretaceous – endemic evolution of placentals in Africa, in contrast to hypotheses rooting afrotherians in Paleogene Laurasian “condylarths”.


Introduction
The "condylarths" -or archaic ungulates -are the main and most spectacular radiation of the placental mammals at the beginning of the Tertiary, after the demise of the non-avian dinosaurs. This non-monophyletic group includes various ungulate-grade lineages, some of which became extinct more or less quickly by the Eocene, and some other of which succeeded and gave rise to the extant flourishing ungulate orders such as Perissodactyla and Artiodactyla.
The fossil record of the ''condylarths'' is heterogeneous, especially geographically. They are well known in the Early Tertiary of the Laurasian continents such as North America, Asia, and Europe, where they diversified in various lineages, including primitive and modern taxa. North American and European ''condylarths'' were known since the turn of the late 19th century, and have been the subject of several classical studies [1][2][3][4][5][6][7].
Together with India, the Arabo-African province is the most poorly known continental center of evolution for ungulate-grade and other placental mammals [25,26]. This is related to the general problem of the poor Cretaceous and early Paleogene fossil record of Africa. Consequently, our knowledge of the origin and early evolution of the African ungulate-like afrotherians, the Diagnosis (a: autapomorphies). Skull (unknown in O. grandis): Short and broad rostrum with very robust construction; short frontal with reduced orbitotemporal process; meso-and postero-cranial region elongated, especially the parietal; parietals with two oblique bony ridges diverging anteriorly from the mid part of the sagittal crest; extensive pneumatization of the skull bones, especially of the supraoccipital; middle ear and inner ear (pars cochlearis and pars canalicularis) remarkably small; tegmen tympani large and inflated.
Dentary: Symphysis short and partially fused (a); condyle significantly higher than the tooth row; corpus high and transversely inflated.
Molar pattern bunoselenodont and brachydont with well developed labial shearing crests (associated with semi-lunar shearing wear facets). Molars size increasing slightly from M1 to M3.
Crown of lower premolars and molars inflated labially (a). P 3-4 simple and trenchant. Lower molars: paraconid bulbous and lingual in M 1-2 , but median transversely in M 3 ; paraconid with small mesial crest; protoconid low, close in height to metaconid and paraconid; postmetacristid and metastylid present; mesoconid large; small entoconulid; hypoconid very large, low and broad (a); entoconid with a short but functional entolophid linking the lingual flank of the hypoconid; hypoconulid reduced and lingual in M 1-2 and postcristid very long, in relation to the selenodont pattern; short premetacristid; postcristid bearing several cuspules; no labial cingula. M 2 wider than M 1 and M 3 ; M 3 at least as long as M 2 , with expanded hypoconulid lobe, and with oblique distal root.
Upper cheek teeth more or less homodont (a): upper premolars extended transversely with well-developed protocone.
Holotype. CPSGM-MA1, fragment of right dentary with P 4 , M 1 . New referred material. Most specimens from lower dentition of O. daouiensis were reported and described by Gheerbrant et al. [26,39]. In this paper we refer and describe new specimens representing mostly the skull and its upper dentition that were previously unknown.
MNHN.F PM45, a partial skull broken on the left side (left maxillary and zygomatic arch absent)and bearing the right P 3-4 , M 1-2 ); a small individual, possibly a female; Ouled Abdoun basin, quarries of Sidi Chennane, Selandian.
MNHN.F PM54, skull rostrum preserving right and left maxillary with C 1 , P 3-4 , M 1-3 , the frontal and anteriormost part of the parietal, a fragment of left premaxillary (with a large incisor); probably a large male individual; Ouled Abdoun basin, unknown exact locality, Selandian.
Diagnosis. See Gheerbrant [26] and generic diagnosis; Ocepeia daouiensis differs mainly from O. grandis n. sp., the only other known species of the genus (see below), by its smaller size.

Description: Cranium
Material. The new material reported here allows the first description of the skull and upper dental morphology of Ocepeia daouiensis. Two partial skulls of Ocepeia daouiensis, MNHN.F PM45 and MNHN.F PM54, are known. They show important differences in size and general robustness related to intraspecific individual variations.
MNHN.F PM45, a relatively complete cranium, is the best preserved specimen. This is the only and best known mammal skull from the Paleocene of Arabo-Africa. It preserves most part of the cranium, except for a small part of left side (left maxillary, palatine, jugal), the premaxillae, the distal part of the right zygomatic arch (jugal) and the right maxillary tuberosity bearing M 3 . MNHN.F PM45 belongs to a small individual with low sagittal crest. The preserved teeth (right P 3-4 , M 1-2 ) are unworn and it is possible that M 3 was still in crypt or just erupting in this specimen, indicating a young individual. However, the sutures are strongly fused and in many cases hardly distinct; together with the low sagittal crest, it might suggest a young adult female.
MNHN.F PM 54 corresponds to most of the rostrum of a cranium preserving the permanent cheek teeth (left and right C 1 , P 2-4 , M 1-3 ), the anteriormost part of the parietal, the nasal, the frontal, a fragment of left premaxillary (with a large incisor), most of the right maxillary and the palatine, and part of the left maxillary. It belongs to a larger individual than MNHN.F PM45. It is also characterized by a more robust general construction with much stronger sagittal and temporal crests, and a large canine, all features probably related to sexual dimorphism. These characters suggest that MNHN.F PM54 belongs to a large male adult, in contrast to the individual represented by MNHN.F PM45. The lower jaw PM41 fits better in occlusion with MNHN.F PM 54 than with MNHN.F PM45, which might indicate that it also belongs to a male (young individual).
General morphology. Our reconstruction of the cranium of Ocepeia daouiensis is based on the specimen MNHN.F PM45 (Figs. 1-2, Fig. S1), with help of 3D digital modeling of tomographic image data. It should be noted that this specimen is most likely female individual with a more gracile morphology than MNHN.F PM54 (Fig. 3). The size of the skull is close to that of Meniscotherium chamense (estimated length 90 mm; width = 65 mm).
The overall cranium shape is noticeably robust, with a broad outline in dorsal view (W/L = 70%). The most remarkable feature is that the rostrum is short and wide, whereas the cranial region (especially mesocranial region) is elongated. This is illustrated by a noticeably high ratio (82%) of the width of the snout at the canine level versus the preorbital length. This primate-like robust short and broad snout is autapomorphic in Ocepeia. The short rostrum might be related to the occurrence of enlarged upper incisors, as illustrated by specimen MNHN.F PM 54 for I 2 . The face (rostrum) is about 1/3 the total length, and the frontal and nasal are short. A short frontal is known is some Paleocene taxa such as Meniscotherium (e.g., Gazin [44], pl. 2) and Pleuraspidotherium, but in Ocepeia the orbital process of the frontal is unusually reduced. For instance, the orbitosphenoid has an original long suture with the elongated parietal and only a small antero-dorsal suture with the frontal. The elongation of the braincase and the mesocranial region is illustrated by the long basisphenoids and pterygoids, by the choanae located very anterior relative to the skull length, by the very long parietal and sagittal crest, and by the anterior position of the pars cochlearis of the petrosal. The postorbital constriction of the skull is strong as illustrated especially in PM 54; the skull and braincase are narrowest near the mid skull length.
Another remarkable feature of the cranium of Ocepeia daouiensis, that is evidenced by the CT scan sections, is the extensive pneumatization of the bones such as the supraoccipital, the parietal, the basisphenoid, the basioccipital and the periotic. This is most striking for the supraoccipital which is considerably thickened and expanded below the parietal, and which shows extensive diploë (Fig. 4). The pars mastoidea of the petrosal, the basisphenoid and the basioccipital are also noticeably pneumatized. This is a probable autapomorphy of Ocepeia.
After correction of the post-mortem distortion (based on 3D digital model and drawings; see 14), the zygomatic arches appear not strong flared laterally (see reconstruction and Fig. S2), in contrast to tethytheres, Phosphatherium, and also Namatherium. There is no postorbital process, so that the orbit is poorly individualized and widely confluent with the temporal fossa. The tooth row does not extend behind the skull mid-length (it extends about 45% of skull length). The skull is low and the dorsal profile is long and straight, as in pantolestids and in contrast to the generalized eutherian construction. The sagittal crest is moderate (MNHN.F PM45) to strong (MNHN.F PM 54), but it does not form a high blade-like structure; the nuchal (lambdoid) crests are strong and salient dorsally and laterally.
Dorsal view. The most striking feature is the short rostrum relative to the long cranial region. This is illustrated in dorsal view (Fig. 1) by a very short frontal and a contrastingly very long parietal that extends close to the orbit level and that bears a sagittal crest nearly as long as half the skull length. The anterior extension of the parietal in the skull roof resembles that of Pleuraspidotherium [7] and the primate Adapis; Pleuraspidotherium differs however by having much more developed orbito-temporal process of the frontal and by having the orbits that are more anterior with respect to the parietal; Adapis has also more anterior orbits, but it is closer to Ocepeia in the somewhat reduced orbitotemporal process of the frontal. The CT scan sections show that in fact the parietal extends anteriorly considerably above the frontal, on an estimated length of 10 mm.
The squamosal (pars squamosa) is also elongated. The premaxillary was probably short (see reconstruction). The nasal cavity seen in MNHN.F PM54 is high and wide, as in primitive proboscideans; this is probably a paenungulate trait. The nasals are broadened posteriorly. They do not penetrate deep in the frontals. The naso-frontal suture is transverse; it is located at the level between M 1 and M 2 , as in the generalized eutherian  condition. The maxillae are well-developed, with a high processus frontalis as in primitive proboscideans; the maxillo-nasal suture is long. The maxillary tuberosity was small and located close to the anterior rim of the orbit. The frontal is flat; it extends anteriorly from the temporal line, up to about 1 cm in front the maxillary tuberosity (M 3 location).
The zygomatic process of the maxilla is well-developed laterally and it extends posteriorly on the medial part of the jugal up to the level of the beginning of the sagittal crest. The temporal fossa and zygomatic arches are longer than wide. The lacrimal is broken but its development can be inferred from surrounding preserved bones in MNHN.F PM45. The lacrimal forms the upper rim of the orbit with a large facial process (pars facialis), in the generalized eutherian condition. A slight bone inflation might indicate the presence of a lacrimal tubercle close to the jugal. The lacrimal has sutures with frontal and nasal dorsally. It is not clear if the lacrimal has a suture with the jugal or with the maxillary along the orbit rim. However, the jugal is narrowed in its anterior extremity (MNHN.F PM 45) as a short and flat bony blade above the zygomatic process of the maxillary. This suggests that the contact between the lacrimal and jugal was either slight or interrupted by a short intermediate extension of the maxillary onto the orbital rim; in other words, the jugal was slightly withdrawn posteriorly. The condition of Ocepeia is not far from that of basal paenungulates such as hyracoids where the lacrimal it is separated from the jugal by a narrow orbital process of the maxillary. The lacrimal foramen is present (MNHN.F PM 54) and located in the orbit above the maxillary foramen of the infraorbital canal. The infraorbital canal extends from the P 3 to M 2 levels, above the tooth row.
The postorbital constriction is strong and maximal at the transverse level where the temporal crests meet and unite with the sagittal crest. The parietals are remarkably long, even longer than in Pleuraspidotherium and Hyopsodus. The fronto-parietal suture follows the temporal crest. The postorbital process of the frontal is strikingly absent, as in Hyopsodus and primitive eutherians. The temporal fossa is very large. The zygomatic arches are moderately expanded laterally. There are two discrete but distinct ridges on the posterior part of the parietals that diverges anteriorly on the skull roof from the sagittal crest near the junction with the nuchal crest; this original feature might be autapomorphic. The braincase is well-developed (e.g., wide transversely).
Lateral view. The skull roof profile is very straight. The nasal and the muzzle are high. The maxilla extends high above the tooth row. The infraorbital foramen is much smaller than in Phosphatherium and in Eritherium. It is located not very high above P 3 , in the primitive eutherian condition. The orbit is noticeably posterior, with its anterior rim located at the level of M 2-3 , as in Hyopsodus and in contrast to primitive proboscideans (Eritherium: P 4 ) and even hyracoids (Dimaitherium: P 4 ; Saghatherium: distal part of M 2 ). The jugal extends below the orbit but does not contact the lacrimal, because of the presence of a short intermediate process of the maxillary. The maxillary participates in the zygomatic apophysis, as a short medial process extending 1,5 cm posteriorly. There is no submaxillary fossa. The zygomatic apophysis diverges low with respect to the dental row. There is a robust (wide) ventral postorbital process of the zygomatic arch, that is made mostly by the jugal; this is reminiscent of primitive proboscideans such as Numidotherium. The jugal extends anteriorly up to the level of M 2 parastyle. It forms a high and narrow vertical bony blade. The distal part of the zygomatic arch, broken in MNHN.F PM54, remains unknown in Ocepeia.
The parietal extends very anteriorly at the expense of the frontal. The area of the orbitotemporal fossa, medial to the zygomatic arch (sphenoid region), is not completely preserved, making difficult the identification of bones and their topographic relations. This is especially true of the orbital process of the frontal, the orbital process of the palatine, the orbitosphenoid, and the alisphenoid that is nearly entirely broken. The orbitosphenoid is small and located anteriorly, as in some primitive eutherians such as Uchkudukodon nessovi [45]; it extends dorsally above the postpalatine torus and the choana. The orbital process of the frontal is reduced in relation to the anterior development of the parietal; this is illustrated by a long parietal-orbitosphenoid suture and an anteriorly restricted orbitosphenoid-frontal suture. Among extant afrotherians, the condition of a parietal-orbitosphenoid suture associated with a reduced orbitosphenoid-frontal suture occurs in macroscelidids, and variably in Orycteropus [46]. Although broken, the alisphenoid was probably elongated, sharing a long suture with the parietal, in contrast for instance to Phosphatherium which has a long suture between the frontal and alisphenoid. Such extended relation of the parietal with at least the alisphenoid and possibly also with the orbitosphenoid is an original construction among the eutherians and the placentals (e.g., [46]). At the base of the orbitosphenoid, five millimeters behind the choana and above it, there is a large semi-lunar and anteriorly oriented opening which might be either the optic foramen or the sphenorbital fissure (f. lacerus anterius), or both. Its large size would agree with a common opening for both the sphenorbital fissure and the optic foramen (and possibly also for the foramen rotundum) as in Phosphatherium; it resembles in shape that of Phosphatherium, but it is located much more anteriorly. According to Cox [46], the optic foramen is located anteriorly on the orbitosphenoid in most extant afrotherians, except in tenrecids. Among extant afrotherians, the fused sphenorbital fissure and optic foramen is a derived condition known in chrysochlorids [46]. A small subcircular foramen is located above the latter foramen, probably the ethmoidal foramen; it is located below an oblique bony crest corresponding to the posterior extension of the crista orbitotemporalis, as in Phosphatherium.
The distal part of basicranium is not elevated with respect to the anterior part of the skull (mesocranium), in contrast to Phosphatherium and other proboscideans. The squamosal dermal part is elongated, and shares long subhorizontal suture with the parietal. There is no contact between the squamosal and the frontal, as in primitive eutherians, and in contrast to Phosphatherium. The alisphenoid canal is probably absent in contrast to Phosphatherium; this is the primitive eutherian state. The postglenoid process is well-developed and higher than the posttympanic process of the squamosal which is closely appressed to the mastoid process. The external auditory meatus is widely opened ventrally (not compressed). The nuchal crests are well-developed and salient posteriorly behind the occiput. However, they are not so expanded posteriorly as in many primitive eutherians such as arctocyonids.
The nuchal crests are formed by the supraoccipital which extends anteriorly for a short distance (3-5 mm).
Ventral view. The palate is wide. The occlusal surface of the cheek tooth series appears relatively large (e.g. breadth extension) with respect to the palatal area, which emphasizes the prominence of the dentition in Ocepeia daouiensis and indicates this is a megadont mammal. This is related to the herbivorous diet and this is another primate (anthropoid) convergence. Comparison of predictive equations of body mass based on various dental measurement (length, width, area) shows that the teeth are especially enlarged transversely. This derived morphology is well distinctive from many artocyonids such as Tricentes and Arctocyon; it is possibly autapomorphic.
The tooth row remains restricted mesially on the skull; it does not extend behind the orbit and the middle skull length. The cheek tooth row is aligned closely parallel to the longitudinal axis. Although broken apart in the known material, the premaxillaries are reconstructed as short and wide, by reference to the occlusion of the lower (anterior) dentition. No palatine fenestra are present in the maxillary. The posterior palatine foramen (f. palatinus majus) is located in front of M 2 . From this foramen, a well-developed palatine groove extends anteriorly up to the P 3 level, similarly to primitive hyraxes; it housed the greater palatine nerve (V2) and blood vessels. The palatines are very short (comparable to the premaxillae) and posterior, in the primitive eutherian condition. They do not extend more anteriorly than M 2 , in contrast to Eritherium (M 1 , anterior part), and to Pleuraspidotherium (P 3 ). They are even slightly more posterior than in Hyopsodus. The choana opens at mid-length level of M 3 . Its mesial rim is inflated as a liplike structure indicating the presence of a postpalatine torus (attachment for tensor palate). The zygomatic process of the maxilla is well-developed laterally but short. It diverges between midlength of M 2 and M 3 , and it extends as a thin a bony blade medial to the anterior part of the jugal, and well behind M 3 . The temporal fossa and zygomatic arches are wide laterally and also long. The zygomatic process of the squamosal does not diverge very posteriorly, but at the level of the basisphenoid-basioccipital suture (i.e., 27% of skull length).
The pterygoid processes are well-developed, very long (e.g., versus Pleuraspidotherium) and widely separated. Consequently, the pterygoid fossa is wide and large, similarly to Meniscotherium. The large development of this fossa, which accommodates the medial pterygoid muscle, is in agreement with the enlarged angle of the dentary in the species. Just behind the pterygoid processes, opens the small foramen ovale within a longitudinal groove at the basis of the pterygoid crest. It is located anterior to the basisphenoidbasioccipital suture and to the postglenoid process, in contrast to Pleuraspidotherium. The pterygoid, basisphenoid and basioccipital are elongated, resulting in the overall long basicranial region of Ocepeia. The basiphenoid is especially long (e.g., versus Hyopsodus); it extends posteriorly to the pterygoid process. The basioccipital area is large and long, although wider than long. The presphenoid and vomer are poorly distinct. However, the CT scans suggest that the presphenoid was short, high and pneumatized, and that the vomer is crossed by a large canal.
The glenoid surface of the squamosal is poorly excavated, more extended transversely than longitudinally and opened anteriorly which makes possible wide antero-posterior movements of the dentary in Ocepeia. This morphology, known in Paleogene hyracoids and proboscideans, is paenungulate-like. The glenoid fossa is located partly on the braincase, in contrast for instance to Arctocyon. The postglenoid process is tall, higher than in Phosphatherium. It is mainly posterior to the glenoid fossa. The postglenoid foramen opens medially to the high postglenoid process; it is more medial than in Phosphatherium, in the primitive eutherian condition seen in Daulestes, Maelestes, Zalambdalestes [47]; contra [37]). The external auditory meatus is large, with a broadly concave suprameatal surface of the squamosal for the auditory tube. The posttympanic process of the squamosal is present but small. It is much smaller and lower than the postglenoid process, in contrast to Phosphatherium. It is widely separated from the paroccipital apophysis by the elongated mastoid region that is exposed ventrally (proboscidean feature).
The basioccipital is convex with a ventrally salient median keel which is also seen in several other condylarth-like taxa such as Meniscotherium and Hyoposodus. The median crest vanishes before joining the occipital condyles, but this is probably related to postmortem damage. The hypoglossal foramen houses two small openings; it is smaller than the jugular foramen, and it is located more posteriorly from the jugular foramen than in Phosphatherium; it is 2.5 mm anterior to the occipital condyle. This eutherian-like construction differs from advanced proboscideans in which the two foramens tend to fuse. The hypoglossal foramen has an oval shape with a long axis parallel to the petrosal-basioccipital suture. The paroccipital process is low and located very posterior (behind the hypoglossal foramen and foramen magnum), as in Phosphatherium.
The braincase is wide, especially caudally (up to the alisphenoid-squamosal suture level), in contrast for instance to the arctocyonids and phenacodontids.
Periotic . Both the right and left petrosals are present in MNHN.F PM45, although the structural details in ventral view are poorly preserved. The petrosal of Ocepeia is large and especially long: 1) the pars cochlearis is small and very anterior with a strong rostral tympanic process extending to the level of the anterior margin of the postglenoid process; 2) the pars mastoidea is very long and widely exposed ventrally between posttympanic (squamosal) and paroccipital processes (both separated from about 10 mm). The enlarged pars mastoidea is a remarkable feature of Ocepeia that is shared with proboscideans. The pars cochlearis and promontorium are ovoid, long, and slightly oblique with respect to the longitudinal axis. The promontorium extends posteriorly to the postglenoid process level, and it is more ventral than the basioccipital. It is inflated and located anteriorly. Its almond shape resembles that of Hyopsodus (p). The promontorium has a smooth surface and it bears a stout ridge that is oblique postero-laterally. This ridge is bounded antero-laterally by a sulcus probably for the internal carotid artery (d). The promontorium sulcus for stapedial artery is absent. The rostral tympanic process of the petrosal is salient anteriorly to the promontorium. The epitympanic wing of the periotic medial to the promontorium is absent (p). The fenestra vestibuli (oval window) is very small (area = 0.312 mm 2 ), and much smaller than the fenestra cochleae. The size of the fenestra vestibuli of Ocepeia daouiensis, and correlatively of its stapes, is well below the regression line for fenestra vestibuli area versus body mass in modern mammals [48]. In fact, the small size of the fenestra vestibuli of Ocepeia daouiensis is in close proportion to Numidotherium koholense [49]. This is a possible derived feature shared with paenungulates, although sirenians show an early divergent trend to hypertrophy of the ear ossicles. It should be noted that the small size of the f. vestibuli correlates with that of the promontorium and inner ear (see below). The fenestra vestibuli is elliptical with a high stapedial ratio (2.05) which is generalized in placentals [42,50].
The fenestra cochleae (round window) is located postero-medial to the fenestra vestibuli, in the eutherian state. It is larger than the fenestra vestibuli, but smaller than the jugular foramen. It is vertical and elongated transversely (p), and it faces posteriorly as in Phosphatherium and the primitive eutherian or placental condition. The presence of the cochlear canaliculus (aquaeductus cochleae) is evidenced by the CT scan observations. It opens, in the primitive condition, ventro-medially in the jugular fossa, below and more medially than the cochlear fenestra; this differs from Phosphatherium where it is more ventral and anterior. This canal is unusually large in section and its opening is dilated transversely. The stylomastoid foramen for n. faciale (VII) is distinct postero-laterally to the fenestra cochleae. The jugular foramen (f. lacerus posterius for glossopharyngeal nerve, IX) is the largest foramen of the basicranium; it is located just behind the promontorium. Its shape is oval with an oblique postero-lateral to antero-medial long axis. The caudal tympanic process of the petrosal is well-developed, which is generalized in eutherians according to McPhee [51]. However, a large and inflated caudal tympanic process is a trait of the Paenungulata, known for instance in Phosphatherium and Numidotherium.
The tympanohyal is large. The tegmen tympani is remarkably inflated and pneumatized; it forms a large and robust barrel-like bony structure located antero-lateral to the promontorium, and it partially covers the epitympanic recess. An inflated tegmen tympani is a derived feature among Eutheria [51,52,53]. It was considered as an ungulate [53] and a tethytherian feature [54,55]. The barrel-vault like shape of the tegmen tympani is known in tethytheres. The hyperinflated tegmen tympani is known in Meniscotherium, Mesonyx, anthracotheriids, hippopotamids and cetaceans [56,57]. The tympanohyal is large. No trace of an ossified tympanic bulla is seen in MNHN.F PM45.
In dorsal view (CT scan 3D model: Fig. 6C, Video S2), the internal meatus and the subarcuate fossa are well distinct and comparable in size. The subarcuate fossa is moderately large and deep. The internal meatus shows two distinct openings separated by a short crista transversa: the f. acusticum superius for the facial nerve (VII), and the f. acusticum inferius for the vestibulo-cochlear nerve (VIII). The f. acusticum inferius is posterior and circular, and it shows distinct link with the cochlea (cochlear canal) close to f. cochleae. The f. acusticum superius is anterior and elongated (beanlike outline). In dorsal to dorso-lateral view, there is a large opening that extends medially as a canal in the inflated tegmen tympani and that might correspond to the foramen for the ramus superior of the stapedial artery. Its opening is as large as that of the internal acoustic meatus.
Inner ear (Figs. 6B, 7). The endocast of the bony labyrinth of Ocepeia daouiensis was successfully reconstructed with the help of the computed X-ray micro-tomography analysis of the skull MNHN.F PM45. We reconstructed the whole inner ear endocast of the left petrosal ( Fig. 6B), as well as also the semicircular canals of the right petrosal. The perfect alignment of the semicircular canals of the right and left petrosal in lateral view indicates that the bony labyrinth is well preserved and not distorted in the skull MNHN.F PM45. The detailed description of the bony labyrinth system of Ocepeia daouiensis will be presented separately with a functional analysis (ms in prep); here we report a comparative description of the main morphological traits of the inner ear that are of systematic and phylogenetic values. The number of coils of the cochlea measured following method of West [43] is 2.12 (765u). Most placentals have at least 2 turns, with some exceptions such as erinaceids and some sirenians [58]; Cretaceous eutherians are plesiomorphic with fewer turns (1 to 1.5 turns) than crown placentals [58,59]. Numidotherium has a smaller number of turns than Ocepeia according to Court [60] (1.5) and Benoit et al. [61] (1.62). The spiral plane of the cochlea is much more oblique antero-dorsally with respect to the horizontal plane (50-60u) than in Mesozoic eutherian mammals. The coiling is subplanar as in the primitive eutherian condition seen for instance in the zhelestids [59]. In the coiling plane, the spiral turns are well separated from each other as in primitive Late Cretaceous eutherians such as zhelestids [59]. The ventral coil has a diameter comparable to the dorsal coil, and the apex of the cochlear canal (helecotrema) is large. This is distinct from Notostylops, but of unknown polarity among eutherians. The secondary bony spiral lamina is present on the basal coil (340 to 360u from f. vestibuli). In several late Cretaceous eutherians [58] and in Notostylops [50], it is primitively restricted to half of the basal coil. However, the secondary bony spiral lamina extends to the whole basal coil in some other primitive Mesozoic eutherians such as zhelestids, as in Ocepeia. The anterior semicircular canal is the larger semicircular canal as in most eutherians. The posterior segment of the lateral semi-circular canal is coalescent with the posterior ampula of the posterior semicircular canal on a short distance from the vestibule. However, the lateral semi-circular canal separates from the posterior ampula before its ends, and the lumen of both lateral semicircular canal and posterior semicircular canal PSC remains distinct, which indicates that there is no true secondary crus commune. The absence of a secondary crus commune is a generalized condition shared by crown placentals in contrast to stem eutherians that have a secondary crus commune [59]. However, it is noticeable that several placentals retain a secondary crus commune, such as the aardvark, some carnivorans, the primitive macroscelidean Chambius [61], and the primitive proboscidean Numidotherium [62]. The crus commune of the anterior semicircular canal and posterior semicircular canal is less long than in generalized eutherians such as zhelestids and Zalambdalestes [59].
Occipital view. The supraoccipital is well-developed. Transverse and longitudinal CT sections show that it is very thickened and pneumatized (Fig. 4). It extends anteriorly below the parietal, not far from the level of the basisphenoid-basioccipital suture. The foramen magnum is large, comparable in size to ''condylarths'' such as Hyopsodus; it is wider than high. The occipital condyles are salient posteriorly, and they join ventrally as in Meniscotherium (and Zalambdalestes). The nuchal crest is salient posteriorly all around the occiput; it does not form the peculiar posterior upper expansion seen in many eutherians and primitive placentals such as arctocyonids, pleuraspidotheriids, pantodonts, pantolestids and others. Below the nuchal crest, the occiput appears slightly inclined anteriorly (eutherian trait).
The occipital surface is depressed between the nuchal crest and the condyles. There is a small condylar foramen. The paroccipital process is small. There is no mastoid exposure in the occipital side (amastoid structure).
Incisors (Fig. 8). A broken piece of left premaxillary of the specimen MNHN.F PM54 bears an enlarged root belonging to a broken incisor that lacks the crown (Fig. 8). The root is hypertrophied and compressed laterally; it is larger than the canine root (L = 7 mm). It indicates an enlarged incisor that is identified as a left I 2 (e.g., no trace symphysis in associated preserved bone). The crown was probably procumbent. It is inferred that this is the larger upper incisor as in early proboscideans such as Numidotherium and Moeritherium.
As in the lower dentition known in O. daouiensis [26], the third upper incisor was probably very small and vestigial.
Canine (Figs. 3, 10). The canine preserved in MNHN.F PM54 is separated from the I 2 by a small diastema. The canine is stout, low, but sharp. The root was inflated and larger than the crown. The crown is convex laterally and concave lingually. It bears a small anterior crest and a longer (higher) distal crest. A faint trace of lingual cingulum is present, but it is much smaller than in the lower canine which is more anthropoid-like in this respect.
Premolars (Figs. 9-10). The two first premolars are absent, as in the lower dentition. They are not replaced by a significant diastema in the tooth row: P 3 is very close to C 1 . P 3 and P 4 are sub-molarized with a developed protocone, three roots, and a more or less transverse occlusal outline. As a consequence, the whole cheek dentition is rather homodont.
The labial cingulum is inflated, especially at anterior and distal ends. Traces of stylar cusps are present, most distinctly as a metastyle, and more variably as a D cusp. The paracone is the main labial cusp: it is large, low and nearly as wide as long. It bears long V-shaped shearing anterior and posterior crests that are underlined by a distinct wear facet. Shearing function with trigonid of occlusal relative teeth P [3][4]

was enhanced in Ocepeia.
There is no trace of metacone. The protocone is much smaller than the paracone. It bears symmetric and well-developed preand post-protocristae that enclose a well delimited protofossa. The preprotocrista joins a very small parastyle; the postprotocrista, well distinct in both P 3 and P 4 , joins the postparacrista. One or more variable cuspules are inflated on the postprotocrista behind the protocone. Their position is reminiscent of the molar metaconule. A small but distinct transverse crest crosses the protofossa in its anterior part and links the internal flank of paracone and protocone. The lingual flank of the protocone is convex. A very slight enamel bulging of the crown below and behind the protocone of P 4 suggests trace of postcingulum. P 3 and P 4 are very similar in morphology, but the former is slightly smaller, less transverse and has a smaller protocone.
The styles are remarkably well-developed and inflated, and located labial to paracone and metacone. Parastyle and mesostyle are the larger styles, they form inflated cusps on the labial side of the crown. Additional small stylar crests and cuspules are variably present in front of the paracone and metacone (e.g., D cusp). The stylar shelf is present but not so wide. The paracone and metacone are low and linked by a dilambdodont ectoloph that is typically joined to the large mesostyle. The two arms of the centrocrista are united labially just before joining the mesostyle.
The preparacrista joins the parastyle; the postmetacrista, welldeveloped, joins the metastyle. The labial flank of the paracone and metacone is convex. The paracone is larger than the metacone, and more so from M 1 to M 3 . The protocone is remarkably low and expanded mesio-distally; its crests are widely divergent. Its labial flank is convex. The protofossa is vast, and especially long, and shallow. A very short transverse crest is present at the base of the lingual flank of the paracone. The preprotocrista joins the parastyle below the preparacrista. The postmetacrista joins the base of the postmetacrista below the metacone. The paracingulum is much more developed than the metacingulum. A metaconule is more or less inflated on the postprotocrista where it is salient disto-lingually; it is more developed on M 1 than on M 2 . The metaconule is doubled lingually by a small additional conular cuspule. There is also a small paraconule on the preprotocrista; it is located more labially than the metaconule. The lingual flank of the protocone is strongly canted labially. A more or less distinct trace of continuous lingual cingulum is present on M 1-3 , around the protocone, but it lacks a hypocone. Three roots are present, including in M 3 .
The selenodont pattern of the ectoloph is functionally underlined by well-developed related semi-lunar shearing facets such as wear facets 3 and 4, 1a and 1b (nomenclature of Crompton [63]).
The occlusal outline is more or less square in M 1 , and wider transversely in M 2 and M 3 . The molar size increases slightly from M 1 to M 3 .

Body mass estimation (Table 4)
Our estimations of the body mass of Ocepeia daouiensis (Table 4) vary noticeably with the skeletal measurements: 1) Body mass estimates are higher based on anterior molars (M3 provides the smallest estimates); 2) Body mass estimates based on tooth area are higher than those based on tooth length.
For ungulates, it is agreed that tooth lengths are better predictors than tooth widths or areas that vary more with dietary specialisations [64]. Moreover, body mass estimates based of skull measurements are considered more reliable for primates [65].
Body mass estimates of Ocepeia daouiensis based on M1 and M2 area are in the higher range (predictive equation for all euungulates (Artiodactyla, Perissodactyla), mean M1 and M2 = 8.5 kg) and are obviously overestimated. This is related to the fact that M1 and, less markedly, M2 are larger in Ocepeia daouiensis relative to modern euungulates on which are based the predictive equations [64]; modern euungulates have a greater size ratio from M1 to M3 than in Ocepeia. The overestimation of the body mass based on anterior molars seems to confirm that Ocepeia is a megadont form with large anterior molars (see description).   Body mass estimate based on the skull length of Ocepeia daouiensis is in the lower range (2.9 kg) and is close to the estimates based on the length of M1-3 and M3. The relative size of M3 (larger relative to M2 and M1) in Ocepeia daouiensis is closer to that of modern mammals; this might explain why M3 provides more reliable body mass estimates that are close to those based on the length of M1-3 and of the skull. The mean body mass estimate of Ocepeia daouiensis based on the length of M1-3, M3 and the skull is 3.5 kg (predictive equations for all ungulates). It is comparable to extant hyraxes and to Meniscotherium chamense (3.4 kg [67]) which are close in skull size, and in dental morphology and specialisations.
Kondrashov & Lucas [66] have recently shown than dental measurements of early condylarths such as the phenacodont Tetraclaenodon puercensis ''pliciferus'' (Torrejonian), of skull size (110 mm) close to that of Ocepeia, give much larger body mass estimation (14 kg) than postcranial measurements (3.2 kg), similarly indicating they are megadont primitive ungulates. The cheek dentition of Ocepeia is actually proportionally much larger with respect to the palate and skull size than in Tetraclaenodon puercensis.
Reconstruction of the skull and the head of O. daouiensis:

Figs. 11-13
The reconstruction of the skull O. daouiensis, made by 3 of us (EG, FG, CL), is based primarily on the most complete specimen, MNHN.F PM45. However, because this specimen is incomplete and slightly crushed dorso-ventrally, some details of the reconstruction (e.g., dentition, I 2 C 1 , M 3 ) were completed based on comparison with MNHN.F PM54. However, the reconstruction of the whole skull is not a chimera of the two specimens. It is indeed important to note that specimen MNHN.F PM45 is more distorted than other specimens, and it is a probable female with much more gracile overall morphology and smaller size than MNHN.F PM54. Hence, because of the sexual dimorphism of O. daouiensis, our reconstruction is only partially representative of the species, i.e. it is not representative of male individuals which would appear significantly more robust (with stronger sagittal and temporal crests), as exemplified by MNHN.F PM54.
The lower jaw reconstructed in Figure 12 is based primarily on specimen PM41, and other specimens for the anterior dentition (see [26]).
The reconstruction of the skull was made using the 3D models computed from tomographic data and processing (by FG) with  help of programs Mimics Innovation Suite (Materialize, release 16), VG studio Max (Volume Graphics, release 2.2) and Cinema 4D (Maxon, release 13). The 3D model of the specimen MNHN.F PM45 was segmented following bone sutures, lines of distortion and breaks and best preserved part; segments were re-assembled after correction of distortions in the three main anatomical orientations (Fig. S2), by comparison with available material and using natural symmetry in the arrangement of skull bones.
The main difficulty in our reconstruction was the plastic distortion of individual bone that could not be retrodeformed; this especially true for the dorso-ventral compression of MNHN.F PM 45 that could not be evaluated precisely, and for which we propose an estimated correction in the Figures 11-13 based on the comparison with MNHN.F PM 54.
The final drawing of the reconstruction was based on the reconstructed 3D digital model (Fig. S2), with help of the original specimens and their casts, of superposed photographic views and camera lucida drawings (e.g., occlusal view of dentition, ventral view of petrosal), and of osteological measurements. Drawings (by CL) were made by hand and on computer with help of specialized computed programs (Adobe Creative Suite 6).

Comments
Many features of Ocepeia daouiensis are generalized eutherians traits (Table 5), which indicates it is very primitive.
Several symplesiomorphic resemblances are noted with ''condylarths'' such as arctocyonids (e.g., orbit location and construction) and hyopsodontids. The arctocyonids such as Arctocyon also share some remarkable derived features such as the shortened frontal, although the structure is more derived in Ocepeia, with for instance a strongly reduced orbito-temporal process of the frontal. However, the skull of Ocepeia daouiensis has closest resemblances with pleuraspidotheriids, including in derived features such as the elongated parietal and basicranium, the large pterygoid fossa, and the basioccipital shape. The lower jaw of Ocepeia also resembles pleuraspidotheriids in the wide angle. However many other differences, including divergent specialized traits, show that this is result of superficial resemblances related to convergences and symplesiomorphies.
Interestingly, a dental resemblance of Ocepeia is noted with dilambdodont pantodonts such as Haplolambda and Leptolambda. It includes the selenodont molars, but also the absence of hypocone, presence of a continuous lingual cingulum on upper molar, presence of a paraconid on lower molars, overall shape of the dentary with enlarged angle and deep and robust corpus, and short anterior dentition and symphysis. Similar skull features are also noted such as the flat and long skull roof, widely fused orbitotemporal fossa, and the long parietal and sagittal crest. However, Ocepeia and pantodonts differ by both derived and primitive features (orbit position, dental formula, zygomatic arch composition, etc…), indicating well distinct lineages that converged for some features.
Some skull features of Ocepeia daouiensis are known in semiaquatic mammals such as pantolestids: the short muzzle, the absence of postorbital process and related widely confluent orbitotemporal fossa, the long and flat skull roof, the skull shape broader than high, the strong and posteriorly salient lambdoid crests. However, many other features of Ocepeia are well distinct from pantolestids, for instance the small infraorbital foramen, the short frontal, the lower jaw morphology and the distinctive dentition.
The skull and dentition of Ocepeia daouiensis are characterized by a remarkable combination of primitive and derived features. The most remarkable derived features are summarized in Tables 5-7.
These derived features are of different taxonomic rank value. Few but significant ones are placental traits. They concern mostly the inner ear structure, such as the cochlea with at least two turns. Other derived traits are mostly ungulate-like (i.e., ungulate-grade like), or paenungulate-like, or autapomorphic features.
Paenungulate traits are much more significant than the phenacodont-like features previously discussed for Ocepeia [39]. They include skull and dental features (Tables 6, 7) that are discussed in the phylogenetic analysis.
Etymology. Grandis (Latin), in reference to the large size of the species with respect to the type-species O. daouiensis.
Age and occurrence. This Paleocene species comes from a phosphate level located above that of O. daouiensis: Thanetian upper Bone Bed (also known as the coprolites Thanetian Bone Bed) of Bed IIa of the local phosphate series from the Ouled Abdoun Basin (unknown quarries), Morocco [40].
Diagnosis. Species most closely related to O. daouiensis. It differs primarily from O. daouiensis in being 150% larger. Other few and discrete morphological features of Ocepeia grandis n. sp. that differs from O. daouiensis are the conules more developed, the hypoconid and the metacone slightly larger.

Description and comparisons
This species is known only by part of the cheek dentition and of the lower jaw. The upper dentition is poorly known, only by broken isolated teeth that perfectly occlude with associated lower teeth.
The dental morphology is very similar to O. daouiensis. Beside the much larger size, the few and most significant distinct morphological features are mentioned in the diagnosis. The mandibular symphysis was at least partially fused as in O. daouiensis (contra [26]). The mandibular corpus seems deeper in O. grandis than in O. daouiensis. The mental foramina show some variability in their position. A large mental foramina was present between P3 and C1, and another one at least between M1 and P4. MHNT PAL 2006.0.9 shows a small diastema between C1 and P3, but this is also a probable variable feature as in O. daouiensis.
By comparison to the entoconid, the molar hypoconid appears proportionally larger in O. grandis than in O. daouiensis.    (Fig. 20), and in the more marked enamel ornamentation with stronger enamel wrinkles. We interpret these features, especially conular development, as a more grinding-rasping functional specialization for harder food than in O.daouiensis. The labial fragment of upper premolar of specimen MNHN.F PM39, presents a distal crest and cusp more inflated behind the main labial cusp (paracone) suggesting a more distinct metacone or metastyle than in O. daouiensis. It also differs from O. daouiensis by the presence of a vertical groove instead of a crest on the anterior flank of the paracone.
Both the evolutionary stage (including in relative size) and the relative age O. daouiensis (Selandian) and O. grandis (Thanetian) agree with the hypothesis that they belong to a single lineage.

Dimensions (mm)
Teeth and lower jaw: see Tables 8-10 Body mass estimates (Table 11). Estimates based on the length of M1-3 cannot be calculated because none of the known specimens of O. grandis preserves well the molar series. The best body mass estimate of Ocepeia grandis based on available material (lower dentition) is slightly more than ten kilograms (Table 11). O. grandis was indeed close in size to Phosphatherium escuilliei, and much larger than O. daouiensis.

Phylogenetic Analysis of OCEPEIA
The new material of Ocepeia daouiensis discovered in the Paleocene of the Ouled Abdoun quarries provides the first data on the morphology of its skull and upper dentition (see reconstruction . O. daouiensis is now the best known mammal from the Paleocene of Africa. Its relationships were poorly known and are here investigated in a new cladistic analysis developed below (see also Text S1).

Phylogenetic matrix (Text S1, part II)
We investigated the ordinal and supra-ordinal cladistic relationships of Ocepeia with the help of the program TNT [68]. In this aim, we used a revised matrix of Gheerbrant et al. [37] and Gheerbrant [69] that was developed for the study of the earliest proboscideans Phosphatherium and Eritherium, both also known from the Ouled Abdoun phosphate basin. The matrix is extended to include comparisons with basal afrotherians and with other Laurasian ''condylarths''.
The new matrix includes 175 characters and 25 taxa that are detailed in Text S1 (Part I). The CT scan study of the skull of Ocepeia allowed especially the inclusion of several new characters of the petrosal (middle and inner ear).
For the most inclusive taxa such as Eutheria, Perissodactyla, Hyracoidea, Sirenia, Embrithopoda, Zhelestidae, our comparisons are based not on one particular genus or species, but on the most primitive known taxa (see details in Text S1, p.22) that are the best representatives of the hypothetical ancestral morphotype of the Table 5. Ocepeia daouiensis, main primitive (eutherian) features of the skull, dentition and lower jaw. clade/group, as presently documented. This is especially important in such autapomorphic taxa as the embrithopods, the tubulidentates, or even the modern hyracoids that have lost many of their supraordinal morphological synapomorphies. Some of these taxa such as the tubulidentates have a so poor fossil record (e.g., no Paleogene fossils) that many of their features cannot be used for comparisons (i.e., autapomorphies and ''inapplicable'' characters that cannot be homologized, see Table S1). We included the primitive embrithopod Namatherium [70] in our comparisons.
The whole matrix includes 3305 states of which about 30% are unknown or ''inapplicable'' (Table S1). 18 features are uninformative (Table S2) and where excluded from the analysis; they correspond mostly to generalized, and for some, poorly documented traits.

Cladistic analysis
We developed several analyses, all with the ''traditional search'' command of TNT, and with both the unweighted and standard implied weighting options. The matrix was analyzed under the following successive conditions: -All features (175) unordered; -152 features ordered; -152 features ordered and step matrix for 19 unordered features and 12 other unordered characters coded not reversible; -Partitioned matrix, separating dental and skull features; -Constrained analysis to test hypothesis of paenungulate position of Ocepeia: matrix with key features weighted (e.g., selenodonty).

Results and topologies
One remarkable point is that nearly all our analyses of the whole matrix result in the recovery of few most parsimonious trees (MPTs), i.e. a single tree or a well resolved consensus tree. In other words, we obtained few MPTs and all with similar topology. This means that there is some coherent signal among the various characters analyzed, i.e., they do not conflict significantly. Lower topological resolution is obtained with partitioned analysis restricted to skull features which indicates that 1) there are some conflicts among the skull features, 2) dental traits provide the stronger phylogenetic signal in our analysis. However, it should be stressed that some of the partitioned analyses restricted to skull morphology recover the clade Paenungulata and a clade corresponding to Afroinsectiphilia minus Macroscelidea (Text S1, III, Cladograms 7-8), although they do not resolve the position of Ocepeia.
The Retention Indices of resulting trees are between 50 and 60, indicating strong homoplasy. For instance, the shortest tree (all features unordered, Fig. 21) includes only 13 non homoplasic characters, i.e. 8% of all analyzed informative features (75 exclusive synapomorphic states, 2% of all sates).
The exclusive sypapomorphies are indicated by an asterisk (*) before the state number.
Common nodes and topological patterns. There are several recurrent nodes that are shared among all resulting trees from our different analyses. We discuss these successively from the base to the higher part of the trees.
Todralestes always has the basalmost position, as sister-group of all other taxa except the eutherian outgroup. It is indeed excluded from placentals, and should be considered as a stem eutherian as for the Zhelestidae [71]. However, it should be noted that Todralestes variabilis remains poorly known (45% of characters unknown in matrix), being documented only by cheek teeth.
Zhelestidae, Protungulatum, and Arctocyonidae have a basal position in the trees that is not fixed relative to each other. However, the Zhelestidae is generally the most basal clade.
Ocepeia has a basal position in the trees, outside the Paenungulata, ''Altungulata'' (lophodont ungulates) and ''Taxeopoda''. It most frequently belongs to a basal afrotherian clade together with Potamogale, Ptolemaia, Orycteropus, corresponding to the molecular-based clade Afroinsectiphilia minus Macroscelidea, although in some case (e.g., unordered analysis) it is a more derived branch (sister group of all other taxa). Relationships with Afroinsectiphilia afrotherians result from the analysis of the whole matrix and from partitioned analysis restricted to dental features, which indicates that the phylogenetic signal is mostly dental. Bremer support for this node (Afroinsectiphilia and Ocepeia) is rather low. Within Afroinsectiphilia, Ocepeia joins preferentially Potamogale (e.g., 148-1, postorbital process reduced), and Ptolemaia groups with Orycteropus in many of our analyses (ordered, unordered, and step matrix; Text S1, part III). The clade (Ptolemaia, Orycteropus), that is well supported by 3 exclusive synapomorphies (*27-2, *50-2, *116-2; ordered analysis), agrees with the conclusions of Simons & Gingerich [72] and Seiffert [73].
In the higher part of the MPTs obtained in this work, the main clades are stable and robust; in nearly all trees, we recover the following clades: (Phenacodonta (Teilhardimys, Macroscelidea, (Perissodactyla, Paenungulata))). This node is supported by 3 exclusive synapomorphies of which the most noticeable are the upper molars with lingual root enlarged and bearing a sulcus, and with developed interloph. Upper molar with incipient lophodonty (89-1) appears at this node or at the next node (Teilhardimys, Macroscelidea, (Perissodactyla, Paenungulata)).
The relative position of Teilhardimys and Macroscelidea to each other (i.e., same clade or not) changes in several trees. In the partitioned analysis restricted to skull features (Text S1, part III, Table 7. Ocepeia daouiensis, main derived features of the dentition (with presumed rank of apomorphy; * anthropoid-like feature). Numbers refer to those of the matrix that is analyzed in this work (see Text S1, part II). doi:10.1371/journal.pone.0089739.t007 Cladograms 7-8), the Macroscelidea groups with basal afrotherians, as in the molecular-based Afroinsectiphilia hypothesis. Following Gheerbrant et al. [37] and Gheerbrant [69], we recover the sister-group relationships of Perissodactyla (usually allied to Radinskya) and Paenungulata, that corresponds to the taxon ''Altungulata''. This is a very robust node in our analysis (ordered analysis: 5 exclusive synapomorphies: *25-2, *35-1, *42-2, *49-2; 4 homoplastic synapomorphies: 11-0, 16-1, 52-1, 90-2). Consequently, it would refute the monophyly of the clades Afrotheria and Laurasiatheria. It is supported especially by the lophodonty (*25-2, *35-1, *42-2) and the enlarged M 3 (*49-2). However, the latter feature implies unlikely reversal in Eritherium, and the lophodonty is most likely convergent in Perissodactyla and Paenungulata, as indicated by detailed structural traits (see below).We suggest, following Gheerbrant [69], that our cladistic analysis remains unable to detect convergence of lophodonty in Laurasian and African ''ungulates'' because of fossil gaps, i.e. because of unknown early structural steps in its evolution. We lack data especially on the dental ancestral morphotype of perissodactyls, but also on some questionable paenungulates such as Phenacolophus, Minchenella, and anthracobunids. Phenacodonts are plesiomorphic with a low stem position and do not help to solve this question in the analysis. Only the partitioned analysis restricted to cranial features (ordered analysis; Text S1, part III, Cladograms 7-8) does not recover sister-group relationships of Perissodactyla and Paenungulata. This confirms that the signal for ''Altungulata'' here is mostly dental (e.g., lophodonty).
The clade Tethytheria is generally recovered at least for the crown groups Sirenia and Proboscidea, but the position of stem taxa such as Embrithopoda, Anthracobunia, Desmostylia is variable, and the Bremer supports are low. Tethytheria is not recovered only in partitioned analysis with step matrix restricted to dental features (Text S1, part III, Cladogram 11), in which Proboscidea joins the Hyracoidea.
The clade Proboscidea is one of the most stable in all our analyses. It is supported by a strong Bremer Index.
Embrithopoda. A clade (Minchenella (Phenacolophus (Embrithopoda))) is very stable (several analyses, see Text S1). It is supported by 10 homoplastic features in the ordered analysis (Fig. 22). However, it should be noted that a recent analysis of dental enamel microstructure argues against close relationships of Phenacolophus and Embrithopoda [77]. We concur, and suggest that the cladistic relationships of Phenacolophus recovered here results from unresolved significant dental convergences, in the same way as between perissodactyls and paenungulates. This is probably related to our poor knowledge of Phenacolophus (59% of its characters unknown in matrix) and Minchenella (50% of its characters unknown in matrix) which are known only by dental specimens and are among the less known analysed taxa. Actually, it should be emphasized that even the described dental material of Phenacolophus [78] is very poorly preserved -a point which is widely underestimated in current phylogenetic discussions; for instance the structural dental homology remains widely to check. Embrithopods are usually sister-group of Sirenia as recovered by Seiffert [73], except in partitioned analysis (ordered) restricted to dental traits in which they are stem tethytheres (Text S1, part III, Cladograms 5-6).
The different topologies obtained.
N Unordered analysis ( Fig. 21; Text S1, part III, Cladograms 1-2). The analysis of the matrix with all features unordered provides a single resolved tree, which is the shortest tree (length     our discussion of the relationships of Ocepeia. Ocepeia is the sister group of Potamogale, and both are sister group of the clade (Ptolemaia, Orycteropus), corresponding to the clade Afroinsectiphilia (minus Macroscelidea). Analysis with standard ''implied weighting'' command provides a single tree that fixes relative position of Protungulatum and Arctocyonidae (Text S1, part III, Cladogram 4). In this analysis Ocepeia is strongly autapomorphic, as it is in the unordered analysis (36 homoplastic states, 5 exclusive autapomorphies).
N The partitioned ordered analysis (Text S1, part III, Cladogram 5) restricted to dental features recovers similar topology than in the unpartitioned analysis ( Fig. 22; e.g. Ocepeia = basal afrotherian) except in the derived position of Orycteropus that clusters with Proboscidea, probably as the result of the optimization of many features (50% of unknown data in Orycteropus; see Table S1). The partitioned ordered analysis restricted to skull features and to taxa only documented by skull features (Text S1, III, Cladogram 8) recovers two main clades: basal afrotherians (Potamogale, Orycteropus and Macroscelidea, i.e. clade Afroinsectiphilia) and Paenungulates (but including Zhelestidae). The position of Ocepeia is surprisingly not resolved in the partitioned ordered analysis restricted to the skull. Similar results are obtained with the TNT ''implied weighting'' command.
N Ordered analysis with step matrix (Table S4 and Text S1, part III, . In this analysis we ordered the state transformation according to the most likely successive structural steps, and we excluded unlikely reversals (e.g., reappearance of teeth; see Table S3). The analysis recovers 8 trees of which the consensus agrees (Text S1, part III, Cladogram 9) with the ordered analysis ( Fig. 22; Text S1, part III, Cladogram 3). Length, retention and consistency indices of trees from step matrix analysis (Text S1, part III, Cladogram 9) are very close to the ordered analysis. In this analysis, Proboscidea are supported by a strong Bremer Index, and Ocepeia clusters with basal afrotherians. ''Implied weighting'' command provides a single tree (Text S1, part III, Cladogram 10) which differs in the interchanged position of Hyopsodus and Ocepeia, the latter which is excluded from basal afrotherians. In this analysis Ocepeia is much less autapomorphic, with 28 autapomophic traits.
N The partitioned analysis including ordered features and step matrix (Text S1, part III,  shows that dental features yield the most significant phylogenetic signal: 1) dental features provide one single tree very similar to the ordered analysis, except for the position of Orycteropus that surprisingly joins Proboscidea (as in analysis without step matrix); 2) skull features provide an unresolved basal polytomy (Text S1, part III, Cladogram 12).
N Constrained analysis for paenungulates relationships of Ocepeia ( Fig. 23; Text S1, part III, Cladogram 13). We constrained the analysis to cluster Ocepeia with Paenungulata by increasing the weight of features shared with paenungulates (Tables 6, 7, 12). The most important shared characters with paenungulates that impacts topology in the cladistic analysis are characters related to the selenodont pattern of upper molars (ectoloph dilambdodont and joined to a developed mesostyle: characters 98, 99, 101). Other shared features (see Table 12), such as the occurrence of an entolophid (34-1), do not support cladistic relationships with paenungulates at similar weight cost. Constrained analysis with weighted upper molar selenodonty results in only two trees, the consensus of which is well resolved ( Fig. 23; Text S1, part III, Cladogram 13) and significantly longer than in other (unweighted) analyses (RI = 53.2, CI = 37.5, Length = 767). Bremer supports in resulting trees are higher than in other analyses, including for nodes among Paenungulata (e.g., Proboscidea, Tethytheria). In this topology the node Paenungulata (node 23) is supported by 5 exclusive synapomorphies (*40-3, *42-3, *61-1, *126-1, *161-1 and 4 homoplastic features (28-1, 37-0, 130-1, 152-1). However, the node is paradoxically not supported by features related to the selenodonty, following our hypothesis; these features support in fact the more internal paeungulate node 31 ((Phenacolophus, Embrithopoda) (Hyracoidea (Ocepeia, Orycteropus))). In these trees Ocepeia is noticeably less autapomophic (16 homoplasies; 2 exclusive states, including: 118-2, extensive pneumatization; 122-1, palatine posterior to M 1 ). Within Paenungulata, Ocepeia has a nested derived position rather than a stem position; it  Table 8. Ocepeia grandis n. sp., dimensions of lower dentition (mm): lower teeth. N Finally, we analyzed the position of Ocepeia in the exclusive context of the afrotherian relationships (i.e., relationships within the monophyletic clade Afrotheria); in this way we excluded from the analysis putative laurasiatherian ungulates such as Perissodactyla, Radinskya, Phenacodonta, and even Hyopsodus. In all obtained trees, Ocepeia keeps the same outgroup position with respect to Macroscelidea and Paenungulata, in the same clade with Potamogale and Orycteropus (Afroinsectiphilia). This is especially related to the absence of hypocone and lophodonty in Ocepeia.

Discussion: Relationships of Ocepeia, morphological support and significance
Our cladistic analysis confirms that Ocepeia daouiensis is plesiomorphic among placentals, as was indicated in previous studies of more limited material [69]. This is most noticeably shown by the skull construction that is eutherian-like in many respects (Table 5), and this is confirmed in the analysis restricted to cranial features (Text S1, part III, . Some characters of the dentition, such as the absence of a hypocone, the presence of a paraconid, and the large canine, are also remarkable plesiomorphies. As a matter of fact, the several and noticeable ungulate-like features of Ocepeia daouiensis are not optimized in the phylogenetic analysis that excludes it from African and Laurasian ''ungulates'' (including primitive taxa), but instead relates it to basal afrotherians.
Relationships with Placentalia. Whereas Todralestes is a stem eutherian, several features indicate unequivocal relationships of Ocepeia with crown placentals. This is especially true of the inner ear morphology (e.g., cochlear canal with two full turns), as identified from the CT scan study.
Relationships with Afrotheria. The most striking result of the cladistic analysis of Ocepeia is its relationships with Afrotheria, and especially its basal position within Afrotheria. The primitive morphology of the skull of Ocepeia agrees with its basal position in Afrotheria, and also with the basal position of Afrotheria among placentals.
However, the relationship with an exclusive clade of basal afrotherians such as Potamogale, Ptolemaia and Orycteropus, i.e. the clade Afroinsectiphilia minus Macroscelidea, remains poorly supported morphologically. The single identified exclusive synapomorphy with afroinsectiphilian implies an unlikely character transformation in Ocepeia (120-1.3: rostrum very narrow to very wide). Six homoplastic synapomorphies (ordered analysis) also support the node, most with low RI , and of which the best is the loss of P 1 1 (11-3 RI = 41; 72-1 RI = 75). Within basal afrotherians, the reduced postorbital process (148-1) is interpreted as an unambiguous synapomorphy of Ocepeia and Potamogale.
A peculiar relationship of Ocepeia with basal afrotherian taxa such as Potamogale is questionable in regard to its many remarkable ungulate-like features, for instance. We interpret the basal afrotherian placement of Ocepeia in our trees as a rather more general phylogenetic signal for its afrotherian relationships. However, our analysis does not recognize a monophyletic clade Afrotheria and, hence, any exclusive afrotherian synapomorphy in the Moroccan genus. This is related to the more general problem that the monophyly of the supercohort Afrotheria remains poorly characterized morphologically on the basis of available fossil and neontological morphological data (but see [73,76,79]). In this regard, Ocepeia should help to characterize the ancestral morphotype of the afrotherians.
The sister-or stem-group position of Ocepeia to the whole Afrotheria cannot be indeed rejected, especially if we agree with the hypothesis of an ungulate-like ancestral afrotherian morphotype, instead of an insectivore-like ancestral afrotherian morphotype, as raised by Seiffert [73,75,76]. However, this does not fit well with 1) the early diversity of the insectivorous-grade mammals from the Paleocene of Morocco [32,33,34], and 2) several striking paenungulate-like traits of Ocepeia, such as the selenodonty (see below) that remain unknown in extant and fossil insectivoran-like afrotherians (Afroinsectiphilia). Moreover, the synapomorphies that support in our cladistic analysis a relationship of Ocepeia to insectivoran-like afrotherians (Afroinsectiphilia) are weak (see above).
Relationships with Paenungulata. Within Afrotheria, our cladistic analysis places Ocepeia outside of the Paenungulata. Even the constrained analysis that clusters Ocepeia with paenungulates results in a hypothesis of nested paenungulate position that is   [68] (Text S1, parts I-II). All features are processed as unordered and unweighted in this tree (see Text S1 for character distribution). Single most parsimonious tree obtained from the ''traditional search'' command (see Text S1, part III, Cladogram 1). Tree length: 596. Retention index: 53.6. Consistency Index: 40.8. In Figure 21b, the black and open white circles represent respectively strict and homoplasic synapomorphies. This is the shortest tree obtained in all our analyses. The implied weighting analysis does not change the topology. In this analysis Ocepeia is by far the most autapomorphic (see text). doi:10.1371/journal.pone.0089739.g021 refuted by unlikely implied characters transformations such as a secondary loss of the hypocone. Whatever the cladistic results, the new material of Ocepeia described here evidences a combination of primitive eutherian features, placental traits and ungulate-like traits. The ungulategrade traits include proto-ungulate-like and paenungulate-like features that are summarized in Tables 6, 7 and 12. These traits have a noticeable heterogeneous distribution in our trees. The generalized ungulate-grade traits of Ocepeia are the low and bunodont teeth (e.g., 25-1, 88-1), and the posteriorly projecting broad angular process of the dentary (62-1). The most remarkable paenungulate-like features of Ocepeia are dental traits such as especially the presence of an incipient entolophid (34-1, see also [26]), and the ectololoph selenodont (W-shaped centrocrista, 101-2) and linked to an inflated mesostyle (98-1, 99-2 or 99-1), and lower molars with large inflated lingual cusps (28-1). The ectoloph is secondarily lost within paenungulates (e.g., Proboscidea), and the entolophid is known in perissodactyls. Some noticeable paenungulate features of the skull such as the periotic amastoidy (161-1), the developed zygomatic process of the maxilla, and lower molars with enlarged lingual cusps are known in Ocepeia. Amastoidy in the only trait restricted to paenungulates in our analysis. Several other traits of Ocepeia are known in paenungulates (Table 12), but not in all of them for several reasons such as reversals, convergences, or poor knowledge. The most remarkable ones are the high (deep) zygomatic arch with ventral process below the orbit (145-1), the short process of the maxilla in the orbit rim (136-1), presence of a coronoid fossa (63-1), and wide nasal cavity (126-1). Some paenungulate-like traits of Ocepeia seem even restricted to proboscideans (Table 12: 7-1, vestigial I 3 ; 67-1, enlarged I 2 ; 162-1, large pars mastoidea). The ancestral condition of several of these traits remains poorly known in several paenungulate orders such as hyracoids (e.g., features 126, 145), sirenians (63,98,99,101,136) and embrithopods (63,136). Table 12 summarizes the interpretations of the paenungulatelike traits of Ocepeia in the trees (ordered analysis). The shortest trees (unconstrained parsimonious analyses, Figs. 21-22) admit noticeable parallelisms of Ocepeia and Paenungulata in their ungulate-like traits (Table 12), which instead are autapomorphic in both taxa. Resemblances with Proboscidea are also interpreted as convergences in the cladistic analysis. As a result Ocepeia is strongly autapomorphic in our trees (31 autapomorphies in ordered analysis, 41 in unordered analysis).
In fact, the shared traits of Ocepeia and Paenungulata conflict with the shared lophodonty of Paenungulata and Perissodactyla which is interpreted in our parsimony analyses as a synapomorphy for a questionable clade ''Altungulata'', and which challenges the monophyly of Afrotheria in our trees. The morphotypic bunodont incipiently lophodont pattern of primitive proboscideans and paenungulates (89-1, also shared by Teilhardimys and primitive macroscelideans) is well distinct from the advanced lophodont pattern (89-2) of primitive perissodactyls, supporting the probable convergence of Laurasian and African ungulate-like mammals [69]. Together with remarkable shared features of Ocepeia and  [68] (Text S1, parts I-II). This tree results from an analysis in which Ocepeia is constrained to cluster with paenungulates by increasing the weight of characters shared with paenungulates (Tables 6, 7, 12), i.e., characters 98, 99, 101 that are directly related to the selenodonty. Consensus of two trees (Text S1, part III, Cladogram 13). This tree is well resolved but significantly longer than trees resulting from other analyses (Figs. 21, 22 paenungulates such as the selenodont ectoloph, that is unknown in primitive perissodactyls, this challenges such a clade ''Altungulata'', and agrees with the alternative hypothesis of the relationships of Ocepeia and Paenungulata. In this hypothesis, the proto-ungulate features of Ocepeia best fit with its primitive stem position with respect to Paenungulata. The most noticeable plesiomorphic features that exclude Ocepeia from the crown group of extant paenungulate orders are the presence of paraconid, the absence of hypocone and related absence of bilophodonty, and the large canines. Crown paenungulates also depart by the M 3 with two lingual roots (115-1), the molar trigonid compressed with reduced paracristid (29-1), presence of coronoid foramen (61-1), orbit more anterior (above P 4 -M 1 , 135-2) and possibly relatedly the tuber maxillae more posterior to orbit (143-1), and a more developed posttympanic process (158-1). Such hypothesis of a stem paenungulate position of Ocepeia would actually be more parsimonious for several features at the nodes of stem and crown Paenungulata.
In our view, the selenodont morphology of the molar ectoloph is a structural landmark of the monophyly of the paenungulates including stem taxa such as Ocepeia, i.e. as a key feature of their ancestral morphotype. The incipient entolophid is another, but more derived, significant shared derived trait. Together with the relationships of Ocepeia with basal afrotherians (afroinsectiphilians) obtained in the parsimony study, this would make it the first known fossil link between insectivorelike and ungulate-like afrotherians.
Recent Eocene discoveries evidenced intriguing paenungulate dental and postcranial affinities of the European louisinid Teilhardimys and some primitive afrotherians such as early macroscelideans (e.g., [79]). In contrast to Ocepeia, these taxa share with paenungulates especially a large hypocone and the primitive bunodont-bilophodont pattern. However, they lack the selenodont ectoloph linked to a well-developed mesostyle and other paenungulate features seen in Ocepeia such as the incipient entolophid, implying unsuspected noticeable convergent evolution of ungulate-like features within Afrotheria (see Table 12).
The autapomorphic and endemic nature of Ocepeia. Our cladistic analysis emphasizes the autapomorphic nature of Ocepeia, that is here distinguished at least at familial level in the new family Ocepeiidae (see diagnosis). The ordered analysis (Fig. 22) records 31 autapomorphies; 5 are exclusive autapomorphies (68-1, 118-2, 120-3, 122-1, 154-1) and others are mostly homoplastic convergences with paenungulates (see above and Table 12). In this topology, Ocepeia is by far as the most autapomorphic taxon compared in this work. The constrained analysis however shows that Ocepeia is much less autapomorphic when it clusters with paenungulates; in this topology Ocepeia has only 18 autapomorphies (including 2 exclusive states), and Orycteropus is consistently much more autapomorphic. The most remarkable specialized features of Ocepeia are the anthropoid-like traits of the skull face and dentition (Tables 6, 7), such as the rostrum short and robust and the shortened anterior dentition. Other important specialized features of Ocepeia are the Table 12. Cladistic interpretation of shared characters of Ocepeia with ungulates and paenungulates (ordered analysis, Fig. 21).

Characters
Description Convergence with Remarks

28-1 (RI = 88)
Lower molars with lingual cusps larger than labial cusps skull with elongated meso-and postero-cranial region as illustrated especially by the very long parietal and related short frontal. The noticeable pneumatization of the skull bones (a feature that evolves parallely in large proboscideans) is also remarkable; its adaptive significance remains unknown (structural pattern specialized for enhanced sound transmitting?). These autapomorphic traits are evidences of the strong endemic nature of Ocepeia, and the long evolutionary history of its lineage in Africa.

Conclusions
The new material reported here from the Paleocene (Selandian) of the Ouled Abdoun Basin considerably enhances our knowledge of early afrotherians and African condylarth-like mammals. It enlightens the skull morphology of Ocepeia daouiensis that is reconstructed (Figs. 11-13) and studied here. O. daouiensis is today the best known Paleocene mammal from Africa. Another new species of the same genus, O. grandis n. sp., is described from the same basin in a higher Paleocene level of Thanetian age. It differs mostly by its larger size, and it is a probable direct descendant of O. daouiensis. O. daouiensis and O. grandis belong to the new placental family Ocepeiidae, that is only known from the Paleocene of the Ouled Abdoun Basin (Morocco) at that time.
The cladistic analysis of Ocepeia supports its placental and afrotherian relationship. However, it does not fix its ordinal and supra-ordinal position within Afrotheria. The most parsimonious trees (MPTs) support a relationship of Ocepeia with basal insectivore-like afrotherians (i.e., afroinsectiphilians), in contrast to its many paenungulate-like traits. These most parsimonious trees involve 1) noticeable parallelisms of Ocepeia and Paenungulata in their ungulate-like traits, 2) a mostly dental relationships of Paenungulata and laurasiatherian ungulates such as Perissodactyla. The Paenungulata -Perissodactyla sister group relationship ( = ''Altungulata'') recovered in our most parsimonious trees is considered here as the result of a convergence obscured by fossil gaps that make relationships of Ocepeia uncertain within Afrotheria. The shared derived features of Ocepeia and paenungulates are not optimized as exclusive synapomorphies because they conflict with undetected convergences, in our most parsimonious trees, of Paenungulata and Perissodactyla such as the bilophodonty. A recent study of the enamel microstructure of Phenacolophus [77] confirms for instance the convergence of this bilophodont Laurasian genus with paenungulates. These convergences of Laurasian and African ungulates also conflict with a monophyletic clade Afrotheria in our most parsimonious trees.
Within Afrotheria, the remarkable paenungulate-like features of Ocepeia are much more significant than the few and weak synapomorphies recovered in our MPTs with afroinsectiphilians. The selenodont pattern is one of the most remarkable shared derived features of Ocepeia and paenungulates. However, the skull and dentition of Ocepeia show a much more primitive construction with respect to extant paenungulate orders. This excludes unambiguously Ocepeia from the crown clade Paenungulata that departs for instance in the bilophodont molar pattern. Such a combination of primitive and derived features in Ocepeia actually best fits with its stem position to Paenungulata. The sister-group relationship of Ocepeia to Paenungulata needs however to be further and more formally established, hence our referral with a question mark.
Ocepeia actually shows a remarkable mosaic of autapomorphic, ungulate-like and plesiomorphic features. In particular, the combination of plesiomorphic eutherian-like and afrotherian-like skull features, and of proto-ungulate and paenungulate-like dental traits, makes Ocepeia the first and best known ''transitional fossil'' between insectivore-like and ungulate-like afrotherians [80,81].
The family Ocepeiidae is representative of a specialized endemic African lineage characterized by unexpected anthropoid-like features of the dentition and muzzle, and by other skull traits such as elongated meso-and postero-cranial region (e.g., long parietal) and short frontal, and remarkably pneumatized bones (e.g., occipital). It indicates an old history in Africa, since the very beginning of the Tertiary or earlier. This supports the old endemic evolution of Afrotheria and placentals in Africa, in contrast to hypotheses supposing their origin from Paleogene Laurasian stem ''condylarths'' such as louisinids and apheliscids [82]. Table S1 Matrix of Ocepeia: Data completeness. We distinguish here absence of data (character marked by a question mark) from unapplicable characters (''-'' = gaps in TNT); unapplicable characters correspond to features that cannot be homologized. Undocumented and unapplicable features are however processed similarly in TNT. The larger number of unapplicable traits occurs in Orycteropus that is the most specialized taxon compared in our cladistic analyses. (DOC)