Sobrarbesiren cardieli

Díaz-Berenguer, Ester, Moreno-Azanza, Miguel, Badiola, Ainara & Canudo, José Ignacio, 2022, Neurocranial bones are key to untangling the sea cow evolutionary tree: osteology of the skull of Sobrarbesiren cardieli (Mammalia: Pan-Sirenia), Zoological Journal of the Linnean Society 196 (4), pp. 1671-1703 : 1674-1691

SOBRARBESIREN CARDIELI DÍAZ-BERENGUER ET AL., 2018

Holotype: Complete skull of an adult individual ( MPZ 2017 /1) ( Figs 1-12 ).

Paratype skull elements: An almost complete skull of a subadult individual ( MPZ 2017 /2), a nasal premaxillary process ( MPZ 2017 /3), an isolated?I 1 ( MPZ 2017 /4) and an isolated?I 3 ( MPZ 2017 /5) .

Referred specimens: MPZ 2020/591, a subadult–adult skull, which lacks the supraoccipital, exoccipital, basicranium and dentition; MPZ 2020/592–594, juvenile isolated left frontals; MPZ 2020/595, juvenile isolated right frontal; MPZ 2020/596, fragment of the posterior midpart of a juvenile left frontal; MPZ 2020/597, anterior part of a juvenile left frontal; MPZ 2020/598 and MPZ 2020/599, juvenile supraoccipitals; MPZ 2020/600–602, left occipital condyles; MPZ 2020/603, incomplete right squamosal; MPZ 2020/604, juvenile right squamosal that lacks the dorsal cranial portion; MPZ 2020/605, fragments of a juvenile left squamosal; MPZ 2020/606, incomplete left jugal; MPZ 2020/607, right tegmen tympani; MPZ 2020/608, left ascending ramus of mandible; MPZ 2020/610, right?P 2; MPZ 2020/611–612, left?P 3–4; MPZ 2020/613, undetermined premolar; MPZ 2020/614, deciduous premolar.

Type locality and distribution: Known only from the Castejón de Sobrarbe-41 (CS-41) fossil site (Comarca de Sobrarbe, Huesca Province, Spain), which is located in the upper part of the Sobrarbe Fm (Ainsa Basin, South Pyrenean Basin) ( Díaz-Berenguer et al., 2018). Fossils come from a marly, silty level interpreted as the filling of a distributary channel in a deltaic plain.

Diagnosis sensu Díaz-Berenguer et al. (2018): Pansirenian, based on the following synapomorphies: retracted and enlarged external nares [Ch.8 (1)]; premaxilla contacts the frontals [Ch.9 (1)]; and a P 1–5 –M 1–3 post-canine dental formula [Ch.155 (1)]; and characterized by the following combination of characters (autapomorphies indicated with an asterisk): upper dental formula 2.1.5.3 (I 2 absent); pterygoid fossa present, extending above the level of the roof of the internal nares [Ch.102 (1)]; alisphenoid canal present [Ch.101 (0)]; basioccipital rectangular and elongated* [Ch.226 (1)] ( Fig. 2C ); acromion process of the scapula hook-shaped*, which extends slightly to the dorsal border of the glenoid fossa*; posterior iliac spine of the innominate bone present*; lateral iliac spine that appears abruptly on the lateral surface of the ilium with a flattened ventrolateral surface*.

Description and comparison

The description of the skull and dental elements of Sobrarbesiren is based on eight individuals at different ontogenetic stages. The holotype skull MPZ 2017/1 is considered an adult because the basisphenoid– basioccipital suture is obliterated (erroneously interpreted by Díaz-Berenguer et al., 2018), although M 3 exhibits only slight wear. It is therefore likely to represent a young adult. In the paratype skull MPZ 2017/2, the basioccipital is not preserved; however, M 3 is erupted but not worn and the specimen is thus considered a subadult individual. In contrast, the posterior border of the zygomatic process is at the level of M 2, which is considered a feature of adults ( Mitchell, 1973; Sagne, 2001a). Therefore, this skull probably pertains to an individual at the limit between the subadult and adult stages. In addition, MPZ 2017/2 is slightly smaller than MPZ 2017/1 and MPZ 2020/591 (for measurements, see Supporting Information, Table S2). The skull MPZ 2020/591 is larger than the type skulls (Supporting Information, Table S2). Criteria based on the fusion of the skull bones are difficult to apply in this case owing to the preservation of the specimen. Moreover, criteria based on the dentition cannot be applied either. Nevertheless, on the basis of its size (Supporting Information, Table S2), it can be considered a subadult–adult individual.

Premaxilla: Both premaxillae are preserved in the type skulls MPZ 2017 /1 (holotype) and MPZ 2017 /2 (paratype) and in the skull MPZ 2020 /591 ( Fig. 1 ; Supporting Information, Table S 2). Furthermore, an isolated right nasal process of the premaxilla ( MPZ 2017 /3) has been recovered ( Fig. 3 ; Supporting Information, Table S 2). Specimen MPZ 2017 /1 preserves the most complete premaxillae, which are 178 mm (left premaxilla) and 184 mm (right premaxilla) in length, measured from the anterior edge of the symphysis to the posterior end of the nasal process. Each premaxilla is 64 mm in height at the level of I 3 and contains two incisor alveoli, for I 1 and I 3 [Ch.143 (0)]. The inferior border of the I 1 alveolus is broken in the type skulls, but this alveolus seems to be small. It is located at the anterior tip of the premaxilla and is less than half the length of the premaxillary symphysis [Ch.140 (0)]. The alveolus of I 3 lies at the posterior end of the premaxilla, slightly anterior to the premaxilla– maxilla suture. Its maximum diameter is 10 mm, but it is deformed by taphonomic lateromedial compression. In the paratype skull ( Figs 2A, B , 3A, B ), the alveolus of I 1 is 6 mm × 6 mm, but it is broken. The diameters of the I 3 and C 1 alveoli are 6 and 5 mm, respectively. The deflection of the masticating surface of the rostrum is ~32° from the occlusal plane, as measured in the holotype skull ( MPZ 2017 /1). The ventral part of the premaxilla is not preserved in the other skulls. The anterior edge of the rostrum is broad and flattened in anterior view and becomes narrower along the gently flattened dorsal border of the premaxillary symphysis to finish in a convex summit immediately anterior to the mesorostral fossa ( Fig. 1 : mf). The premaxillary symphysis is 105 mm in length in the most complete skull ( MPZ 2017 /1) and is enlarged relative to the cranium [Ch.3 (2)] and laterally compressed [Ch.10 (1)]. The ratio of the symphyseal length to the total length of the skull is ~3.5:1, and the rostral symphysis is more than half the total length of the premaxilla. The anteroventralmost end of the suture with the maxilla is on the vertical level of the symphyseal summit. The nasal process of the premaxilla is long, thin and tapering at the posterior end [Ch.6 (0)]. It is 79–82 mm long in MPZ 2017 /1, 82 mm in MPZ 2017 /2 and 90 mm in MPZ 2020 /591, as measured from the anterior edge of the mesorostral fossa to the posterior edge of the nasal process, and it extends posteriorly less than onethird of the anteroposterior length of the supraorbital process. It is long (distance from rear of symphysis to posterior end of premaxilla more than half the length of the symphysis) [Ch.7 (0)]. The nasal process of the premaxilla is curved in lateral view, and its posterior edge expands slightly lateromedially. It is 24 mm wide in MPZ 2017 /1, 19 mm in MPZ 2017 /2 and 25 mm in MPZ 2020 /591. The nasal process contacts the maxilla ventrally, the nasals posteromedially and the frontals posterolaterally. There is no contact with the lacrimal. The external nares are enlarged and retracted, as in all pan-sirenians [Ch.8 (1)]. The mesorostral fossa is long and narrow but is distorted by compression in all the skulls. Its anterior edge is rounded, and it is bounded anteriorly and laterally by the premaxillae and posteriorly by the nasals (at the level of the anterior orbital margin), with no contribution from the frontals.

The ventral/palatal surface of the premaxilla is deformed by compression in the skull MPZ 2017/1; it is not preserved in MPZ 2020/591, but it is well preserved in the skull MPZ 2017/2 ( Fig. 2A, B ). The incisive foramen is diamond-shaped and is bounded by the premaxillae laterally, anteriorly and dorsally, and by the maxillae posteriorly. It is ~ 20 mm in width and 21 mm long, and it opens downward. The rostral masticating surface is trapezoidal, and its maximum breadth is 48 mm at the level of I 3.

The isolated right nasal process of the premaxilla MPZ 2017/3 is 85 mm long ( Fig. 3 ). The nasal branch is oval in cross-section and 9 mm in maximum diameter. The posterior end is diamond-shaped and dorsoventrally flattened, and it has a maximum mediolateral width of 19 mm. It is similar in size to the nasal process of the subadult–adult skulls; hence, it is considered to pertain to a subadult–adult specimen.

Nasal: Both nasals are preserved in the three studied skulls, but they are incomplete in MPZ 2020/591 ( Fig. 4 : n; Supporting Information, Table S2). The nasals are large and anteroposteriorly elongated, and they meet along their dorsomedial borders [Ch.31 (0)], as in ‘prorastomids’ ( Savage et al., 1994; Domning, 2001a), protosirenids ( Sickenberg, 1934; Bajpai et al., 2009; Domning et al., 2017) except Protosiren smithae ( Domning & Gingerich, 1994) and some other Eocene pan-sirenians, such as Prototherium veronense and Eotheroides spp. ( Sickenberg, 1934; Samonds et al., 2009; Zalmout & Gingerich, 2012). They are set in sockets in the anteromedial margin of the frontals. The nasals extend further than the anterior border of the orbit and are exposed dorsally on the skull roof for a total length of 70 mm in MPZ 2017/1 and 61 mm in MPZ 2017/2. The anterior borders of the nasals are separated by a nasal incisure. Together, they form a V-shaped extension beyond the posterior edge of the mesorostral fossa [Ch.37 (0)]. Their dorsal surface is arched upward to reach a level higher than both frontals and parietals in MPZ 2017/1, but they are at the same level in MPZ 2017/2 and MPZ 2020/591 ( Fig. 1 ). The elevation observed in MPZ 2017/1 might therefore be attributable to the taphonomic lateromedial compression of this skull. The ratio of the maximum breadth of both nasals to their anteroposterior length is approximately one. The lateral side of the nasal shows an anterolateral constriction where it contacts the premaxillary nasal process, and its posterior edge is rounded. The nasals are shallowly separated by the frontals. The suture with the frontals is evidenced by interdigitations in MPZ 2017/2 and MPZ 2020/591 ( Fig. 4A, B ).

Vomer: The vomer ( Fig. 4A, B : v) is well preserved only in the skull MPZ 2017/2. It is exposed in the mesorostral fossa. The vomer is elongated and lies above the dorsal surface of the maxillae. Dorsally, two thin laminae contact anteriorly, forming a narrow V, which extends to the anterior border of the mesorostral fossa. They delimit a U-shaped valley. Posteriorly, the bone narrows, and the laminae become closer.

Lacrimal: The lacrimals can be observed in the three skulls ( Fig. 1 : l), but they are only preserved in part. The lacrimal is a relatively large bone (≥ 16 mm in anteroposterior width on the left side of the holotype skull MPZ 2017 /1). Its anterior edge is rounded, but the posterior side is poorly preserved. The lacrimal faces posterolaterally and is surrounded by the maxillae ventrally and anterodorsally, forming the anteromedial border of the orbit, and by the supraorbital process of the frontal dorsally [Ch.93 (0)]. The lacrimal foramen (nasolacrimal canal; Fig.1 : lf) is small and opens laterally [Ch.91 (0)]. Tubercles or protuberances are absent.

Frontal: Both frontals are preserved in the type skulls and in the skull MPZ 2020/591 ( Fig. 4 : fr; Supporting Information, Table S2). Furthermore, four complete, isolated frontals are recovered ( Fig. 5 ): MPZ 2020/592, MPZ 2020/593 and MPZ 2020/594 are left frontals, and MPZ 2020/595 is a right frontal. In addition, MPZ 2020/596 is a fragment of the posterior midpart of a left frontal, and MPZ 2020/597 is the anterior part of a left frontal. All these isolated specimens are assigned to perinatal to juvenile individuals because of their small size. The frontals form the anterior midpart of the skull roof and contact the premaxilla anteromedially and the nasal medially ( Fig. 4 ). The frontal roof of Sobrarbesiren is short relative to the parietals (i.e. interfrontal suture vs. length of parietals), as in ‘prorastomids’, protosirenids and Eotheroides aegyptiacum , Prototherium ausetanum Balaguer & Alba, 2016 and ‘ Halitherium ’ taulannense ( Domning & Gingerich, 1994; Savage et al., 1994; Sagne, 2001a; Bajpai et al., 2009; Balaguer & Alba, 2016; Domning et al., 2017), and in contrast to some other Eocene pan-sirenians, such as Prototherium veronense , Eotheroides clavigerum Zalmout & Gingerich, 2012 , Eotheroides sandersi Zalmout & Gingerich, 2012 and Eosiren stromeri Abel, 1913 (Domning, 1994; Zalmout & Gingerich, 2012: table 3). The frontals are flat between the temporal crests [Ch.42 (0)] ( Fig. 4 : tc) and do not bear knoblike bosses [Ch.45 (0)]. Their lateral walls are narrowed posteriorly at the anterior edge of the frontoparietal suture. The supraorbital processes ( Fig. 4 : sop) of the frontals are stout, dense and dorsoventrally flattened, 10–15 mm thick and show a prominent posterolateral corner [Ch.36 (0)]. Their lateral borders are not divided [Ch.44 (0)]. They reach the posterior end of the nasals. The almost straight posterior border of the supraorbital process shows a blunt bulge, which is present in the type skulls and in the skull MPZ 2020/591. In addition, the holotype and paratype skulls show several cord-like protuberances on the dorsal edge of the posterolateral corner of the frontal, immediately before the beginning of the frontoparietal suture. These cords are directed anteromedially to posterolaterally and increase in length distally. They are more marked in the holotype skull than in the paratype. The anterolateral border of the supraorbital process shows a rugose texture. The dorsolateral surface of the frontals is distinctly demarcated by low and blunt temporal crests, which are oriented anterolaterally to posteromedially. The beginning of the frontoparietal suture is marked by a dorsolateral prominence. Posteriorly, the frontoparietal suture is V-shaped. The ratio of the maximum breadth across the supraorbital processes to the maximum length of the frontals is 1.7. The lateral edges of the frontals are sharp and overhanging. The lamina orbitalis of the frontal is covered by sediment in the holotype skull MPZ 2017/1 and is not visible in the paratype skull MPZ 2017/2 because the orbits are collapsed. However, it is well preserved in the left frontal of the skull MPZ 2020/591 and is preserved, in part, in the isolated juvenile frontal MPZ 2020/593 ( Fig. 5B, E ). This lamina forms a thin wall (6 mm in thickness in MPZ 2020/591 and 3 mm in MPZ 2020/593). An orbitotemporal crest is absent. The frontals contact the parietals posteriorly and the alisphenoid ventrally.

Juvenile frontals: The general morphology of the isolated juvenile frontals ( Fig. 5 ) is similar to that of the subadult skulls. The left and right frontals MPZ 2020/593 and MPZ 2020/595 are similar in size, whereas MPZ 2020/592 and MPZ 2020/594 are smaller ( Fig. 5 ; Supporting Information, Table S2). The most anterior edge of the frontals shows a slightly concave surface inclined medially for the contact with the premaxilla ( Fig. 5D ). The specimens MPZ 2020/593 and MPZ 2020/595 show well-developed supraorbital processes, whereas these processes are small, and almost indistinguishable in dorsal view, in the younger specimens MPZ 2020/592 and MPZ 2020/594, and the lateral side of the frontal is almost flat, without the usual lateral deviation of the frontal ( Fig. 5G–L ). The medial side of the frontal, where it contacts the nasal bone, is a thin lamina, which overhangs the nasal cavity. Posteriorly, the medial side shows a thick, sickle-shaped, rugose vertical surface where the two frontals contact. In specimens MPZ 2020/592 and MPZ 2020/594, this surface is reduced in size and smoother than in the specimens MPZ 2020/593 and MPZ 2020/595.

The ventral surface of the frontals is occupied by several deep and elongated concavities. These structures seem to be absent or poorly developed in other sirenians, such as Metaxytherium krahuletzi Depéret, 1895 ( Domning & Pervesler, 2001: plates 1, 2b), Metaxytherium serresii Gervais, 1847 ( Carone & Domning, 2007: plates 1b, 3), and Eotheroides sandersi ( Zalmout & Gingerich, 2012: fig. 43), in which the ventral side of the frontal is a plane surface. The frontals reach their maximum thickness at their posterior midpart (17–22 mm), whose dorsal surface is inclined posterolaterally; this sutural surface is irregular and covered with longitudinal grooves, where the parietals contact this bone. The posteroventral side of the frontals (i.e. the anterior border of the cranial cavity) is a suboval, smooth and concave surface. Both the contact surface with the parietals and the concavities on the ventral surface of the frontals are less marked and subtler in MPZ 2020/592 and MPZ 2020/594 than in the other juvenile frontals. Both MPZ 2020/592 and MPZ 2020/594 are proportionally shorter anteroposteriorly than the other specimens, although thick. Moreover, there is a deep transverse groove between the lateral wall of the frontal and the rest of the bone ( Fig. 5K, L ), which turns medially towards the medial wall of the frontal. Both MPZ 2020/592 and MPZ 2020/594 are thus interpreted as perinatal specimens because of their size and the fact that they are less developed.

Parietal: The parietals are well preserved in the type skulls (MPZ 2017/1 and MPZ 2017/2), and the dorsal part of the parietals is preserved in the skull MPZ 2020/591 ( Fig. 4B : p; Supporting Information, Table S2). The parietals form the posterior midpart of the skull roof. They cover a length of 88 mm of the MPZ 2017/1 skull roof, 57 mm in the skull MPZ 2017/2 and> 74 mm in the skull MPZ 2020/591, less than the nasal–frontal portion of the skull (135 mm in MPZ 2017/ 1, 115 mm in MPZ 2017/2 and 120 mm in MPZ 2020/591). The parietal portion is slightly longer than the frontal portion (length of interfrontal suture: 58 mm in MPZ 2017/1, 57 mm in MPZ 2017/2 and 59 mm in MPZ 2020/591). The parietals are elongated and convex bones. The frontal processes exceed half of the interfrontal length in the type skulls and in the skull MPZ 2020/591 (estimated). Thus, they are long, as in Prototherium veronense and ‘ Halitherium ’ taulannense, unlike other Eocene pan-sirenians ( Sagne, 2001a: Ch.20). The interparietal suture and the parieto-occipital sutures are completely closed and cannot be distinguished in any of the subadult– adult specimens studied. The parietals show smooth temporal crests that begin in the anterior and lateral contacts of the frontoparietal suture; thus, these crests are marked in both the frontal and the parietal, as in other Eocene pan-sirenians except Prorastomus , with temporal crests limited to the frontal bone ( Savage et al., 1994), and Libysiren sickenbergi Domning et al., 2017 , with temporal crests limited to the posterior part of the parietals. The temporal crests show variability between the different specimens. In the holotype, they are thin and join immediately behind the frontoparietal suture for a total of 52 mm in anteroposterior length. They join immediately behind the frontoparietal suture in the skull MPZ 2020/591 for a total length of 40 mm, but in this specimen the crests are wider, resulting in a more flattened vault in its more dorsal part. In the paratype, the temporal crests are wider than in the holotype and separated 3 mm from one another ( Fig. 4A, B ). The smooth temporal crests of Sobrarbesiren differ from all other Eocene pan-sirenians, except the ‘prorastomid’ Pezosiren , in being joined, or nearly so, in the parietals. Nevertheless, the temporal crests of Sobrarbesiren are not raised, unlike in Pezosiren . Thus, a true sagittal crest is absent in all the specimens [Ch.51 (1)]. Posteriorly, the temporal crests separate before reaching the nuchal crest and delimit a shallow triangular depression. Three emissary foramina forming a triangle are situated in this depression in the holotype skull, but they are absent in the paratype skull. The parietal bone has a dorsoventral thickness of 29 mm in the skull MPZ 2020/591.

Endocranial structures: The endocranial structures can be observed in the paratype skull MPZ 2017/2, in the skull MPZ 2020/591 and in the isolated juvenile skullcaps MPZ 2020/598 and MPZ 2020/599 ( Fig. 6 ). The bony falx cerebri ( Fig. 6 : fce) is absent in all the studied specimens [Ch.223 (1)]. Instead of it, there is a shallow groove along the interparietal suture ( Fig. 6A, E ), which transforms into a deep groove between the frontals. A low tentorium osseum ( Fig. 6 : to) and shallow transverse sulci directed posterolaterally ( Fig. 6 : tsul) are present in MPZ 2017/2 and in MPZ 2020/598 [Ch.224 (1)], but are less marked in the latter. These structures are less marked in the skull MPZ 2020/591. In addition, the ventral surface of the supraoccipital is covered by small pits in the specimen MPZ 2020/598 ( Fig. 6C–F ). A blunt internal occipital protuberance ( Fig. 6 : iop) is present in all the specimens [Ch.225 (1)]. Thus, Sobrarbesiren lacks a bony falx cerebri, like protosirenids and Pezosiren (see Fig. 13B , fce), but possesses the tentorium osseum and the internal occipital protuberance, unlike ‘prorastomids’ and protosirenids, although a low tentorium osseum is present in some specimens ( Sickenberg, 1934; Domning & Gingerich, 1994; Gingerich et al., 1994; Domning et al., 2017).

Supraoccipital: The supraoccipital is complete but distorted in the holotype skull MPZ 2017/1 ( Fig. 7 : so; Supporting Information, Table S2). In the paratype skull MPZ 2017/2, it is preserved only in its dorsal midpart ( Fig. 7A, B ). Furthermore, in the juvenile skullcaps MPZ 2020/598 and 599, the supraoccipital is almost complete ( Fig. 6C–F ; Supporting Information, Table S2). The supraoccipital contributes to the most posterior part of the skull roof. The general morphology and the relationship between the dorsal and ventral width of this bone are not clear in the holotype skull.The nuchal planum is bipartite owing to a weak and short median ridge, evident in the holotype and paratype skulls. This ridge separates the oval concavities for the rectus capitis dorsalis muscle insertions ( Fig. 7 : rcd), situated near the dorsal border of the bone. Lateral to these concavities there are two other paired concavities, bigger and deeper, surrounded dorsally by the nuchal crest ( Fig. 6 : nc), interpreted as the points of insertion of the semispinalis capitis muscles ( Fig. 7 : ssc). The external occipital protuberance ( Fig. 7 : eop) is a blunt triangular boss situated on the dorsal edge of the supraoccipital. The supraoccipital shows a pronounced nuchal crest positioned at the same level as the skull roof, semicircular in shape, curved laterally and lacking rugosities, but less massive than in protosirenids ( Domning & Gingerich, 1994; Bajpai et al., 2009; Domning et al., 2017). The supraoccipital forms an angle of 105° with the parietal in the holotype skull MPZ 2017/1 and 100° in the paratype skull MPZ 2017/2, but both are distorted by fossil-diagenetic compression.

Juvenile parietal–supraoccipital skullcaps: The general morphology of the specimens MPZ 2020/598 and 599 is similar to that of the holotype and paratype skulls, but it is better preserved ( Fig. 6C–F ; Supporting Information, Table S2). The supraoccipital–parietal suture is almost obliterated but still discernible. The supraoccipital is wider in its dorsal half than at the ventral extremities of the lateral borders [Ch.64 (0)], as in the other Eocene pan-sirenians ( Domning et al., 2017; Springer et al., 2015: Ch.64). Its outline is hexagonal, and it has a relatively flat posterior surface. The external occipital protuberance is a small blunt boss on the dorsal part of the supraoccipital and, immediately below it, there is a short median ridge limited to the dorsal half of the bone. The concavities for the insertion of the semispinalis capitis muscles (the lateral, bigger ones) and the rectus capitis dorsalis muscles (the medial, smaller ones) are also present. The posterior surface of the supraoccipital forms an angle with the dorsal plane of the skull roof of 138° in MPZ 2020/598 and 130° in MPZ 2020/599. In ventral view, the ventral border of the supraoccipital is V-shaped and divided by a narrow furrow into two suboval surfaces for the fusion with the exoccipital.

Exoccipitals: The exoccipitals are preserved only in the holotype skull MPZ 2017/1 ( Fig. 7 : eo; Supporting Information, Table S2). The lateromedial compression of the skull has obscured the original morphology. Three isolated left occipital condyles, MPZ 2020/600– 602, are also preserved. The exoccipitals are connected along a midline suture, 11 mm in length, dorsal to the foramen magnum [Ch.66 (0)], as in all the Eocene pan-sirenians except Protosiren smithae ( Domning & Gingerich, 1994) . The ratio of the maximum exoccipital height to the maximum exoccipital breadth is 0.58, and the breadth of the exoccipitals is 87 mm. The foramen magnum ( Fig. 7 : fm) is oval, with its dorsal border pointed, probably distorted by the lateromedial compression, with a height that exceeds its width. The dorsolateral borders of the exoccipital are rounded and smooth [Ch.70 (1)]. The supraoccipital–exoccipital suture is not well preserved in the holotype skull. Nevertheless, the ventral border of the complete juvenile supraoccipital MPZ 2020/598 shows that this suture was V-shaped ( Fig. 6C, D ). The presence of supracondylar fossae above the condyles is not clear, but if they were present, they would have been shallow and located directly above the dorsal part of the condyle [Ch.67 (0,1)], as in ‘prorastomids’ and protosirenids ( Domning & Gingerich, 1994; Savage et al., 1994; Domning et al., 2017), except Ashokia Bajpai et al., 2009 .

The occipital condyles are kidney-shaped, wide dorsally and narrowing ventrally. They are 25 mm apart at their bases in MPZ 2017/1 and curved almost vertically ( Fig. 7C ), as in the other Eocene pan-sirenians, except Pezosiren (specimen USNM 553590). The paraoccipital processes are almost at the same ventral level as the condyles and are separated by 11 mm from them. These processes have blunted, curved ends to accommodate the periotic along with the squamosal. The presence of hypoglossal foramina is not clearly discernible [Ch.72 (?)]. The exoccipitals contact the supraoccipital dorsally, the basioccipital ventrally and the squamosal laterally.

Basioccipital: The basioccipital is preserved only in the holotype skull MPZ 2017/1. The basioccipital– basisphenoid suture is completely closed. The basioccipital is 38 mm long (from the line of fusion with the basisphenoid to the base of the foramen magnum) and 17.5 mm wide. It is elongated and almost rectangular in ventral view, maintaining the same width along its body [Ch.226 (1)] ( Fig. 2C ; see Phylogenetic analysis section), unlike the basioccipital of the rest of the pan-sirenians. The basioccipital is transversely wider posteriorly in ‘prorastomids’ ( Savage et al., 1994; basicranium USNM 553590 of Pezosiren portelli , and it is wasted in protosirenids and other pan-sirenians (see Phylogenetic analysis section). The anterior edge of the basioccipital of Sobrarbesiren is slightly more elevated than the posterior one. In ventral view, there are two symmetrical, shallow, elongated and smooth depressions for the insertion of the longus capitis muscles [Ch.227 (0)], as in Prorastomus . These depressions are deeper in other Eocene pan-sirenians, such as Pezosiren (basicranium USNM 553590), Eotheroides aegyptiacum ( Abel, 1913: table (II) XXXI, 2) and ‘ Halitherium ’ taulannense (see Phylogenetic analysis section). Protosirenids and most dugongids show convex rugosities on the basioccipital, instead of concavities, for the attachment of the longus capitis muscles (e.g. Domning, 1978, 1988; Bajpai et al., 2009; Vélez-Juarbe & Domning, 2015; Voss & Hampe, 2017; Domning et al., 2017). The depressions for the insertions of the longus capitis muscles of Sobrarbesiren are separated by a median keel, as in Pezosiren , Prorastomus and Eotheroides aegyptiacum ( Savage et al., 1994; basicranium USNM 553590 of Pezosiren portelli ; Abel, 1913: table (II) XXXI, 2). This keel is shorter in Eotheroides aegyptiacum than in Sobrarbesiren and the ‘prorastomids’, but it bifurcates before the basioccipital–basisphenoid contact, as in Sobrarbesiren . Moreover, the ventral surface of the basioccipital is depressed between the rami of the keel in both taxa. Posteriorly, the keel extends almost to the foramen magnum [Ch.228 (0)].

Basisphenoid: The basisphenoid is preserved in the holotype skull MPZ 2017 /1 and partly preserved in the paratype skull MPZ 2017 /2 ( Fig. 2A–C : bs). The basisphenoid is a flattened and short bone in ventral view (46 mm in length in the paratype skull). It is inclined anterodorsally towards its contact with the presphenoid ( Fig. 2A, B : pre). Its posterior part, immediately anterior to the basisphenoid–basioccipital suture, shows two low keels that delimit a shallow depressed area in the holotype skull, but these structures are not preserved in the paratype skull. Lateral to these rami, there are two concavities on the left and right sides, which occupy the most lateral posterior corners of the basisphenoid above the pterygoid fossa. The sutures with the pterygoid and the surrounding sphenoid bones are not clear in the type skulls. In the paratype skull the posterior contact surface with the basioccipital is exposed. It is pentagonal and rugose.

Presphenoid: The presphenoid is partly exposed between the broken palatines and the basisphenoid in the paratype skull MPZ 2017 /2 ( Fig. 2A–C ), and it is partly preserved in the skull MPZ 2020 /591. The presphenoid forms an elongated median crest. This crest gets higher towards the nasal cavity and reaches its maximum height (14 mm) in its anterior part, although it is incomplete. The sutures between the presphenoid and the surrounding bones are not visible, but it clearly contacts the basisphenoid posteriorly.

Orbitosphenoid: Not preserved.

Alisphenoid: The alisphenoids are well preserved in the holotype skull MPZ 2017/1 ( Fig. 1 : as). In the paratype skull MPZ 2017/2, only the dorsal part is still present. In MPZ 2020/591, a fragment of the alisphenoid is articulated with a left squamosal, both associated with the skull MPZ 2020/591. The alisphenoids form the lateral side of the pterygoid processes, which are broken and inclined inward. Their bases are separated by an estimated distance of 45 mm measured between their lateral sides. The alisphenoid contacts the basisphenoid dorsomedially and the pterygoids medially. The dorsolateral contact with the squamosal is straight and located at the base of the medial wall of the temporal fossa. This contact is clearly visible on the left squamosal of the skull MPZ 2020/591. An alisphenoid canal ( Fig. 2A, B : alc) is present [Ch.101 (0)], as in ‘prorastomids’ and protosirenids ( Domning & Gingerich, 1994; Gingerich et al., 1994; Savage et al., 1994; Domning, 2001a; Domning et al., 2017). It is exposed in the lateral surface of the alisphenoid in the holotype skull because the alisphenoids are broken and out of place, and in the ventral view of the paratype skull because the wings of the alisphenoid are not preserved. This canal is large (≥ 16 mm in length) in MPZ 2017/2. Its posterior opening is located in the lateral side of the alisphenoid, but its anterior opening is not clear in this skull. The foramen ovale is converted into an open notch [Ch.103 (1)].

Pterygoid: Like all non-‘prorastomid’ sirenians, Sobrarbesiren has a short sphenopalatine region, with an enlarged pterygoid process ( Domning, 2001a). The pterygoids are preserved in the specimen MPZ 2017/1, but they are broken at their roots and inclined inward. Their dorsal part is preserved in the paratype skull MPZ 2017/2 ( Fig. 2 : pt). The medial walls of the pterygoids are separated by an estimated distance of 36 mm at their base. These medial walls project further posteriorly than the lateral walls and converge dorsally at the posterior corner of the basisphenoid. The lateral side of the pterygoid presents a shallow, longitudinal groove immediately anterior to the posterior edge. The pterygoid constitutes the medial lamina and the posterior facet of the pterygoid process, which is enlarged, thickened and dorsoventrally projected [Ch.105 (1)]. The pterygoid fossa is shallow but clearly present, because the posterior edge of the pterygoid has two well-differentiated medial and lateral borders. The fossa extends above the roof of the internal nares [Ch.102 (1)], unlike in protosirenids ( Domning & Gingerich, 1994; Bajpai et al., 2009; Domning et al., 2017). In the medial corner of the posteroventral extremity of the pterygoid process, there is a small posteromedial projection of the bone, which is here interpreted as the hamular trochlea, where the tensor veli palatini muscle or the pterygopharyngeus muscle is attached ( Domning, 1977a; Domning & Gingerich, 1994). The pterygoids contact the palatines anteriorly, the basisphenoid dorsally and the alisphenoid laterally, although the sutures of the latter two are not clearly discernible.

Palatine: The palatines ( Fig. 2 : pal) and their contacts are well preserved in the paratype skull MPZ 2017/2, except for their posterior edges. The palatines are thin, flat bones that contact the maxillae anteriorly and laterally. Their anterior width is 23 mm. The two palatines are joined by a median suture from their anterior border, which extends at least as far as the midpart of P 5, and thus beyond the posterior edge of the zygomatic–orbital bridge [Ch.99 (0)], as in ‘prorastomids’, protosirenids and some other pansirenians, such as Eotheroides spp. and Prototherium ausetanum ( Samonds et al., 2009; Zalmout & Gingerich, 2012; Balaguer & Alba, 2016). The posterior border of the palatines is broken, and its morphology and limits are not clear, but it lies at least behind the anterior border of M 3 [Ch.97 (0,1)]. The remnant bone can be seen to be <1 cm thick at the level of the penultimate cheek tooth [Ch.16 (0)]. There is no sign of the two foramina that mark the maxillopalatine suture in other sirenians. The expected posterior contact with the pterygoid is broken in the paratype skull, but it is preserved in the holotype skull and forms the anterior part of the pterygoid process.

Maxilla: The maxillae are well preserved laterally in theholotypeskull MPZ 2017/1, buttheirventralsurface is better preserved in the paratype skull MPZ 2017/2, although the maxillae are fractured and compressed in this specimen ( Fig. 1B : mx). The maxillae form a lyre-shaped palate. The medial contact between them is marked by a low keel that extends almost the total length of the bone in ventral view. The deflection of the maxilla begins at the level of the P 4 alveolus in MPZ 2017/1. The dorsoventral compression of MPZ 2017/2 disguises the rostral deflection of this specimen. The maxilla encloses a small infraorbital foramen ( Fig. 1 : fio) in the holotype skull [height, 16 mm; width, 8.3 mm; Ch.13 (0)], as in ‘prorastomids’, protosirenids and Eotheroides aegyptiacum ( Abel, 1913; Domning & Gingerich, 1994; Savage et al., 1994; Bajpai et al., 2009). The infraorbital foramen is directed anteroventrally and is unobstructed [Ch.20 (0)]. Its outline is oval, but it is affected by the lateromedial compression of the skull. The infraorbital canal is short (13 mm in length) and extends from the anterior end of P 2 to the anterior end of P 4. The infraorbital foramina are placed 56 mm away from each other across the rostrum in the holotype skull and 51 mm in the paratype skull. The zygomatic– orbital bridges of the maxillae, which surround the infraorbital foramina, are long anteroposteriorly [60–65 mm in length; Ch.14 (0)]. Their anterior edges are thin and sharp, and the posterior ones are thicker and rounded [Ch.22 (0)]. The height of the zygomatic– orbital bridge above the alveolar shelf of M 1 is 20 mm, meaning that it is elevated above the palate [Ch.11 (1)], as in ‘prorastomids’ ( Savage et al., 1994) and the protosirenid Ashokia ( Bajpai et al., 2009) . Anteriorly, the maxillae form a narrow palatal gutter [Ch.23 (0)] following the palate midline, unlike in protosirenids ( Domning & Gingerich, 1994; Domning et al., 2017), with tall but thin edges bordering the tooth roots in MPZ 2017/1. The maxillary dental arcade is concave medially, and the palate is narrowest at the level of P 1, from where it broadens and attains its maximum breadth across the anterior edge of M 1 in MPZ 2017/2 (44 mm).

The maxillary dental alveoli are evident. A single canine alveolus is situated immediately behind the premaxillary–maxillary suture [Ch.144 (1)]. There is no diastema between the C 1 alveolus and the I 3 alveolus in MPZ 2017/1 ( Fig. 2C ). There is a small gap of 10 mm in MPZ 2017/2, but in this skull the ventral side of the maxilla is not preserved, meaning that this distance probably represents the spacing between the roots of the teeth and not a diastema between their crowns ( Fig. 2A, B ). Posterior to the canine alveolus, there are five alveoli for single-rooted P 1 –P 5 on the right side [Ch.145 (0); Ch.157 (1)]. In the left dental arcade of MPZ 2017/1, the permanent P 5 is preserved [Ch.146 (0)]. In this skull, the diastema between the canine and the P 1 alveolus is 14 mm and that between the P 1 and P 2 alveoli is 14 mm on the left side and 22 mm on the right side. The molar teeth (M 1 –M 3) are preserved on both sides in MPZ 2017/1 and MPZ 2017/2 ( Fig. 2A–C ). The space posterior to the maxillary dental battery is 12 mm in MPZ 2017/1. The maxilla contacts the premaxilla anterodorsally, the palatines ventromedially, the jugal dorsolaterally and the lacrimal posteriorly.

Squamosal: The squamosals are preserved in MPZ 2017/1 (the holotype), but broken and displaced ( Fig. 1 : sq; Supporting Information, Table S2). The zygomatic arches ( Fig. 1 : zps) are lateromedially compressed and shifted against the cranial roof. The paratype MPZ 2017/2 preserves the zygomatic processes of the squamosal and part of their cranial portion ( Fig. 2A, B ). The skull MPZ 2020/591 preserves the left zygomatic process but isolated from the skull. Specimen MPZ 2020/603 is an isolated right squamosal ( Fig. 8A–E ), judged to belong to a mature individual owing to its size, which preserves the zygomatic process and the cranial portion. It is undeformed and preserves the original morphology of the squamosal in posterior view, which is distorted in the skulls. In addition, a right juvenile squamosal that lacks the dorsal cranial portion (MPZ 2020/604) and a dorsal portion and the distal part of a left juvenile squamosal (MPZ 2020/605) are preserved. These specimens are assigned to juvenile individuals on account of their small size ( Fig. 8F–K ; Supporting Information, Table S2).

In the paratype skull MPZ 2017/2 ( Fig. 1A, B ), the squamosal bones are out of place, but their position is more similar to the original position than in the holotype skull. The zygomatic breadth of this skull is 150 mm. The dorsal end or cranial portion of the squamosal contacts the parietal and reaches the nuchal crest but not the temporal crests, because these are joined and thus situated in a more dorsal position on the skull roof [Ch.76 (1)]. However, in lateral view there is not a significant distance between the squamosal and the top of the cranial roof as occurs in protosirenids ( Domning & Gingerich, 1994; Bajpai et al., 2009; Domning et al., 2017), and the condition observed in Sobrarbesiren is more like that of the rest of the pan-sirenians.

The suture line with the parietal is interdigitated. The zygomatic process is robust, tapered anteriorly, and triangular in cross-section [Ch.81 (0)]. The left zygomatic process of MPZ 2017/1 is 109 mm long anteroposteriorly and 35 mm wide dorsoventrally, with its maximum height in the middle part of the bone. The concave medial side of the zygomatic process is inclined inward dorsally [Ch.84 (0)]. Its ventral side shows an inflexion from the middle of the bone to its anterior edge; this inflexion marks the contact with the jugal ( Fig. 8G, J ). This contact surface is narrow and slightly concave. The root of the zygomatic process of the squamosal is 46 mm long anteroposteriorly in MPZ 2017/1, 42 mm in the specimen MPZ 2017/2 and 21 mm in the juvenile squamosal MPZ 2020/604. The rear edge of the zygomatic process is broken and out of place in the type skulls, but a moderately inflected processus retroversus (posterior end of the zygomatic process) can be observed on the right side of the skull MPZ 2017/2 [Ch.77 (1)]. This processus is absent in ‘prorastomids’ and protosirenids except Libysiren ( Savage et al., 1994; Domning, 2001a; Domning et al., 2017).

The external auditory meatus in MPZ 2017/1 ( Fig. 1 : eam) is narrow, 6 mm in anteroposterior length, 18 mm in dorsoventral height and 6 mm (right)/ 7 mm (left) in width between the postglenoid process ( Fig. 2 : pgp) and the post-tympanic process. Thus, it is higher dorsoventrally than anteroposteriorly [Ch.82 (0)]. The post-tympanic process projects anteroventrally and is enlarged, unlike in protosirenids except Ashokia ( Domning & Gingerich, 1994; Bajpai et al., 2009; Domning et al., 2017), and the facet for insertion of the sternomastoid muscle is present [Ch.73 (0)].

The sigmoidal ridge ( Fig. 8C : sr) is present and prominent in posterior view [Ch.74 (1)]. It projects laterally (~ 15 mm) in the specimen MPZ 2020/603 ( Fig. 8A–E ). Its dorsal part is lost, but the ventral part extends to the ventral tip of the post-tympanic process. The temporal condyle (or tuberculum; Fig. 2 : tco) is prominent. The mandibular fossa ( Fig. 2 : maf) is a shallow and narrow depression 10 mm in anteroposterior breadth in MPZ 2017/2, and the postglenoid process ( Fig. 2 : pgp) is low and elongated. All of them are transversely directed. The squamosal– alisphenoid suture is not clear in the type skulls MPZ 2017/1 and MPZ 2017/2, but it is well marked in the ventral view of the partial left squamosal of the skull MPZ 2020/591 ( Fig. 2D ). It is interdigitated and is located at the level of the medial border of the temporal fossa ( Fig. 2 : tf). The contact surface of the squamosal with the alisphenoid is also preserved in the specimen MPZ 2020/604 ( Fig. 8H, K ). Posteroventrally, the post-tympanic process of the squamosal contacts the paraoccipital process of the exoccipital.

Juvenile squamosals: The general morphology of the isolated juvenile squamosals MPZ 2020/604 and MPZ 2020/605 is like that of the adult and subadult skulls, although structures such as the sigmoidal ridge, the temporal condyle and the postglenoid process are less prominent, and the mandibular fossa is almost undefined ( Fig. 8F–K ). The external auditory meatus measures 9 mm in anteroposterior length, 12 mm in dorsoventral height and 4 mm in width between the postglenoid and the post-tympanic process in the specimen MPZ 2020/604.

Jugal: The jugals are broken and distorted in all the skulls ( Fig. 1 ; Supporting Information, Table S2). Specimen MPZ 2017/1 preserves an almost complete left jugal, and MPZ 2017/2 has partly preserved jugals, which lack the ventral and postorbital processes. In addition, an isolated and incomplete left jugal (MPZ 2020/606) has been recovered ( Fig. 8L–O ). The jugal forms the ventrolateral margin of the orbit. The total length of the left jugal of the holotype skull MPZ 2017/1, the most complete one, is estimated at 149 mm, with an approximate maximum dorsoventral height of 44 mm. In MPZ 2017/1, the preorbital process ( Fig. 8 : jpre) is a thin and short plate [Ch.88 (0)] that overlaps the maxilla but does not contact either the premaxilla or the lacrimal [Ch.87 (0)]. The jugal does not extend to the anterior margin of the orbit, which is formed by the lacrimal and the maxilla. The ventral process of the jugal is broken and shifted, and its shape is not clear. Its ventral border is rounded; the lateral surface is smooth, and it is lateromedially flattened. The position of this process with respect to the postorbital process is not clear in the type skulls. The postorbital process ( Fig. 8 : jpostp) is a rounded summit that lies against the zygomatic process of the squamosal. The zygomatic process of the jugal is long and narrow, and it thins towards its rounded end. This process is lateromedially flattened and longer than the anteroposterior diameter of the orbit [Ch.89 (0)]. It reaches the anterior edge of the temporal condyle. The isolated jugal MPZ 2020/606 has a total length of 70 mm and is 32 mm in height.It lacks the ventral and postorbital processess but preserves part of the preorbital process and part of the zygomatic process. The jugal is flattened lateromedially, with a maximum thickness of 11 mm. The ventral part of the preorbital process has a shallow step parallel to the process, where the maxilla contacts the jugal, the latter overlapping the former. The ventral rim of the orbit does not distinctly overhang the lateral surface of the jugal [Ch.90 (0)]. Although the ventral process of the jugal is not preserved, in accordance with the position of the fracture surface in this specimen, the ventral extremity of the jugal would be positioned posterior to the orbit or under the posterior edge of the orbit, but anterior to the postorbital process of the jugal [Ch.85 (0,1)].

Periotic: The right periotic is preserved, in part, in the paratype skull MPZ 2017/2 ( Fig. 9A : pe), and only the pars petrosa can be described ( Fig. 9A : pp). Only the ventral view can be observed because the bone is not isolated from the skull. In addition, an isolated right tegmen tympani (MPZ 2020/607) has been recovered ( Fig. 9B, C ). The periotic is situated in a socket formed by the squamosal and the exoccipital. Its anteroposterior length is 40 mm, and its estimated width is 43 mm. The pars petrosa is formed by swollen anddensebone.Itissomewhatflatteneddorsoventrally, and its posteromedial side is rounded. On its ventral surface, a blunt promontorium ( Fig. 9 : pm) and the perilymphatic foramen ( Fig. 9 : pf) are preserved. The endolymphatic foramen is not visible. The isolated tegmen tympani is elongated and is gradually tapered towards its pointed anteromedial end ( Fig. 9B, C ). It is not fused to the alisphenoid [Ch.115 (1)], as in all pan-sirenians except Prorastomus ( Savage et al., 1994) . The anterior and ventral surfaces are rounded. The dorsal surface is also rounded but marked by a longitudinal groove. The posterior surface is covered by anteroposterior striations.

Tympanic: Three tympanic bones are recovered: the right and left tympanic of the holotype skull MPZ 2017/1 and another isolated and incomplete tympanic associated with the skull MPZ 2020/591. The tympanic bone of Sobrarbesiren is swollen, horseshoe-shaped and asymmetrical, as is usual in sirenians (e.g. Robineau, 1969; Thewissen & Bajpai, 2009). The right tympanic of the holotype skull is the most complete recovered ( Fig. 9D, E ). Its anteroposterior length is 18 mm, and its dorsoventral height is 24 mm. Its anterior side is straight but shows a small anteromedially directed protuberance. The posterior side is straight in its dorsal half and inflected anteroventrally in its ventral half. The anterior and posterior sides join in a pointed apex, which is directed anteroventrally when situated in its original position in the skull. The anterior arm ( Fig. 9 : aa) is directed dorsomedially and is longer and more robust than the posterior one ( Fig. 9 : poa), which is directed dorsolaterally. The sulcus tympanicus is slightly marked at the base of the posterior arm. The lateral side of the tympanic bone is convex, and the medial side is concave and shows a shallow, elongated dorsomedial furrow. The tympanic arch ( Fig. 9 : ta; or lumen of the tympanic arch, following Voss & Hampe, 2017) has an internal diameter of 5 mm, which is relatively narrow compared with other sirenians. The tympanic bone of Sobrarbesiren is anteroposteriorly broad and more robust than the tympanic bone of Prorastomus , although the two taxa share a relatively narrow arch (Supporting Information, Table S3; Savage et al., 1994). The tympanic bone of Eotheroides lambondrano is similar in size to that of Sobrarbesiren , but its internal diameter is twice as great ( Fig. 10C ; Supporting Information, Table S3). Other Eocene pan-sirenians also possess proportionally broader tympanic arches than Sobrarbesiren . These include Protosiren smithae ( Domning & Gingerich, 1994: fig. 3), ‘ Halitherium ’ taulannense (holotype skull RGHP D040), ‘ Prototherium ’ intermedium (skull MGP-PD 28998) and Eotheroides clavigerum ( Zalmout & Gingerich, 2012: fig. 29A). The condition of a tympanic bone with a broad tympanic arch is maintained in postEocene taxa, such as Metaxytherium albifontanum Vélez-Juarbe & Domning, 2014 ( Fig. 10D ; Supporting Information, Table S3) and extant sirenians ( Robineau, 1969: figs 8, 10E).

Malleus , incus and stapes: Not preserved.

Mandible: Mandibles are represented by only a partial left ascending ramus (MPZ 2020/608), which preserves the condyle ( Fig. 11A, B : con) and part of the coronoid process ( Fig. 11A, B : cp). This mandibular fragment is 47 mm in length and 50 mm in height, and its ventral and posterior borders are broken. The articular surface of the mandibular condyle is suboval in dorsal view, convex and inclined medially. Its lateral side is rounded, and the medial side is pointed and slightly narrower. It is 11 mm in anteroposterior length and 18 mm in lateromedial width. It is as extended medially as laterally, as in some Eocene pan-sirenians ( Sagne, 2001a: Ch.53) and unlike the clearly extended and medially projected articular surface of Prorastomus ( Savage et al., 1994: fig. 6), Pezosiren (holotype, USNM 511925) and Protosiren smithae (cast CGM 4229 of the holotype 94810). The mandibular notch is preserved, and it is narrow. The coronoid process clearly rises above the level of the dorsal edge of the condyle, despite being broken. The preserved lamina of the coronoid process is thin, 4 mm in thickness. In medial view, there is an elongated groove below the condyle, oriented dorsoposteriorly to anteroventrally, which Domning (1977a) interprets as the insertion area for the pterygoideus externus muscle.

Dentition: The type skulls MPZ 2017/1 and MPZ 2017/2 preserve the M 1 –M 3 molar series on both sides and alveoli for incisors, canines and the P 1 –P 5 premolar series, with the left P 5 being preserved in the holotype skull ( Figs 1C , 2C ). On the basis of the preserved alveoli and their position, the upper dental formula of Sobrarbesiren has been interpreted as 2.1.5.3 [Ch.139 (0); Ch.143 (0); Ch.145 (0); Ch.146 (0); Ch.150 (0); Ch.151 (0); Ch.155 (1)] ( Díaz-Berenguer et al., 2018). Therefore, Sobrarbesiren retains a singlerooted permanent P 5, like the ‘prorastomid’ Pezosiren and protosirenids ( Abel, 1907; Domning & Gingerich, 1994; Domning, 2001b) and unlike the three-rooted permanent premolar of Prorastomu s ( Savage et al., 1994). However, it shows a reduction in the number of incisors by the loss of I 2, like Prototherium spp. and the Eotheroides spp. from Egypt, although the condition for Eotheroides aegyptiacum is unknown ( Bizzotto, 1983; Pilleri et al., 1989; Zalmout & Gingerich, 2012).

In addition, two isolated incisors were tentatively identified as an?I 1 (MPZ 2017/4) and an?I 3 (MPZ 2017/5) ( Fig. 12A–D ). The description of the incisors, the P 5 and the molar series was included in the paper by Díaz-Berenguer et al. (2018). The alveoli of the canines and premolars in both skulls indicate that these teeth were single rooted [Ch.144 (1); Ch.157 (1); Fig. 2 ]. Subsequently, several isolated premolars have been recovered. Preliminarily, they are considered upper premolars on the basis of the scarcity of mandibles in the CS-41 fossil site. Specimen MPZ 2020/610 ( Fig. 12E–G ) is identified as a right upper premolar (?P 2) and MPZ 2020/611–612 as left?P 3–4 ( Fig. 12H–J ). Specimen MPZ 2020/613 is an undetermined worn upper premolar, and MPZ 2020/614 ( Fig. 12K–M ) is a deciduous premolar (left?DP 3). For the dental measurements, see the Supporting Information (Table S4).

Specimen MPZ 2020/610 is considered a?P 2 because it shows only one contact with an adjacent tooth, in this case probably P 3. The crown is composed of a main central cusp, partly worn and exposing the dentine, and a posterolingual low accessory cusp. A cingulum with numerous cuspules covers the anterior, lingual and posterior perimeter of the tooth. The root is absent.

Specimens MPZ 2020 /611–612 are considered?P 3–4 because their crowns possess two contact surfaces with adjacent teeth, and their morphology is different from that of the P 5 of the holotype skull ( Fig. 2C ). The crown is worn, but it preserves a main central cusp located labially and a lower secondary cusp located lingually. There is a strong and short anterolingual cingulum with three small cuspules. The main cusp and the cingulum are connected by two parallel cristae, and posterior to the main cusp there is another crista.

Specimen MPZ 2020/613 is a worn uniradiculate premolar and is partly broken. The main cusp has almost disappeared owing to wear, but this premolar possesses a cingulum with cusplets and an accessory cusp.

Specimen MPZ 2020/614 is a biradiculate deciduous premolar. It is preliminarily interpreted as a left DP 3 based on the number of roots for the deciduous premolars proposed for other fossil sirenians, such as Prototherium veronense , ‘ Halitherium ’ taulannense and Dusisiren spp. ( Sickenberg, 1934; Domning, 1978; Sagne, 2001a). This interpretation is also supported by the lack of contact surfaces with other premolars, which is in accordance with the sequence of dental eruption proposed by Sagne (2001a) for the Eocene pan-sirenian ‘ Halitherium ’ taulannense. The crown of the tooth has a main labial cusp and two slightly lower anterior and posterior cusps. There is also a small lingual cusp. A small cingulum with one accessory cusp is located anteriorly. A strong posterior cingulum, labiolingually narrower than the rest of the crown and almost quadrangular in shape, is divided into three accessory cusps. Anterior and posterior roots are complete, separated and divergent.

Figures

Figure 1. Skulls of Sobrarbesiren cardieli in lateral view. A, paratype skull MPZ 2017/2. B, interpretative sketch of the paratype skull. C, holotype skull MPZ 2017/1. D, skull MPZ 2020/591. Sutures are marked with continuous lines. Dashed lines indicate incomplete bones, and dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 2. Skulls of Sobrarbesiren cardieli in ventral view. A, paratype skull MPZ 2017/2. B, interpretative sketch of the paratype skull. C, holotype skull MPZ 2017/1. D, skull MPZ 2020/591. Sutures are marked with continuous lines. Dashed lines indicate incomplete bones, and dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 3. Isolated right nasal process of the premaxilla of Sobrarbesiren cardieli (MPZ 2017/3) in dorsal (A) and medial (B) views. For abbreviations, see the Material and Methods section.

Figure 4. Skulls of Sobrarbesiren cardieli in dorsal view. A, paratype skull MPZ 2017/2. B, interpretative sketch of the paratype skull. C, holotype skull MPZ 2017/1. D, skull MPZ 2020/591. Sutures are marked with continuous lines. Dashed lines indicate incomplete bones, and dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 5. Frontal bones assigned to immature individuals of Sobrarbesiren cardieli. A–F, MPZ 2020/593, left frontal in dorsal (A), ventral (B) and lateral (C) views, and interpretative sketches (D–F, respectively). G–I, MPZ 2020/592, left frontal in dorsal (G), ventral (H) and lateral (I) views. J–L, MPZ 2020/594, left frontal in dorsal (J), ventral (K) and lateral (L) views. Dashed lines indicate incomplete parts. For abbreviations, see the Material and Methods section.

Figure 6. Endocranial structures of Sobrarbesiren cardieli. Skull roof of skull MPZ 2020/591 (A) and MPZ 2017/2 (B) in ventral view. C–F, MPZ 2020/598, parietal–supraoccipital skullcap of a juvenile individual in posterior (C) and ventral (E) views, and interpretative sketches (D, F, respectively). For abbreviations, see the Material and Methods section.

Figure 7. Skulls of Sobrarbesiren cardieli in posterior view. A, paratype skull MPZ 2017/2. B, interpretative sketch of the paratype skull. C, holotype skull MPZ 2017/1. Sutures are marked with continuous lines. Dashed lines indicate incomplete bones, and dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 8. Isolated squamosals and jugal of Sobrarbesiren cardieli. A–E, MPZ 2020/603, right isolated squamosal; cranial portion in lateral view (A), interpretive sketch (B), and in posterior view (C); and zygomatic process in lateral (D) and ventral (E) views. F–K, MPZ 2020/604, right isolated juvenile squamosal in lateral (F), ventral (G) and medial (H) views, and interpretative sketches (I–K, respectively). L–O, MPZ 2020/606, left partial jugal in lateral (L) and medial (N) views, and interpretative sketches (M, O, respectively). Dashed lines indicate incomplete bones, and dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 9. Bones of the ear region of Sobrarbesiren cardieli. A, right periotic of the paratype skull MPZ 2017/2 in ventral view. B, C, MPZ 2020/607, right pars temporalis (= tegmen tympani) in ventral (B) and dorsal (C) views. D, E, right tympanic of the holotype skull MPZ 2017/1 in lateral (D) and medial (E) views. Dashed areas denote broken surfaces. For abbreviations, see the Material and Methods section.

Figure 10. Schematic drawings of tympanic bones of various sirenian taxa. A, Sobrarbesiren cardieli, right tympanic of the holotype skull MPZ 2017/1. B, Prorastomus sirenoides, right tympanic BMNH 44897. C, Eotheroides lambondrano, left tympanic (from Samonds et al., 2009: fig. 6B). D, Metaxytherium albifontanum, right tympanic (from Vélez-Juarbe & Domning, 2014: fig. 8A). E, Trichechus senegalensis, right tympanic (from Robineau, 1969: fig. 11, not to scale). For abbreviations, see the Material and Methods section. Measurements: APL, Anteroposterior length; DVH, Dorsoventral height; ID, Internal diameter.

Figure 11. Mandibular fragment of Sobrarbesiren cardieli (MPZ 2020/608) in dorsal (A) and medial (B) views. For abbreviations, see the Material and Methods section.

Figure 12. Isolated teeth of Sobrarbesiren cardieli. A, B,?I1 (MPZ 2017/4) in occlusal (A) and labial (B) views. C, D,?I3 (MPZ 2017/5) in occlusal (C) and lateral (D) views. E–G, P2 (MPZ 2020/610) in occlusal (E), labial (F) and lingual (G) views. H–J,?P3–4 (MPZ 2020/611) in occlusal (H), labial (I) and lingual (J) views. K–M, deciduous premolar (MPZ 2020/614) in occlusal (K), labial (L) and lingual (M) views.

Figure 13. Endocranial surfaces of two sirenian skulls. A, Kaupitherium bronni (sketch based on Voss & Hampe, 2017: figs 4, 11). B, Protosiren fraasi (sketch based on Sickenberg, 1934: plate I, fig. 4). For abbreviations, see the Material and Methods section.