New material of Epiaceratherium and a new species of Mesaceratherium clear up the phylogeny of early Rhinocerotidae (Perissodactyla)

Reduction of the anterior dentition (i.e. incisors and canines) is a major adaptative trait of the Rhinocerotidae among Perissodactyla. However, the corresponding evolutionary sequence was lacking a robust phylogenetic frame to support it thus far. Here, we describe a new Oligocene species of Rhinocerotinae, Mesaceratherium sp. nov. from the Swiss locality of Bumbach (MP25 reference level). In addition, we identify the only known complete mandible of Epiaceratherium magnum, an early-branching rhinocerotid, as well as one of the earliest European rhinoceroses. We also compute a parsimony analysis based on morpho-anatomical characters to assess their phylogenetic position and elucidate the early evolution of the Rhinocerotidae. Our results allow to propose a new scenario for the reduction of the anterior dentition in which upper and lower dentitions would have undergone distinct evolutionary trajectories.

We modified characters 2 and 3 from the original matrix of Antoine [27] as follows: -2: Maxilla: foramen infraorbitalis: 0, above P1-2; 1, above P3; 2, above P4; 3, above molars -3: Nasal notch: 0, above P1-2; 1, above P3; 2, above P4-M1 Several analyses were performed to assess the identification of the mandible NMB.O.B.928 from Rheinbetts and the material from Bumbach. We tested different sets of terminals, adding taxa by incrementation, or by merging some specimens into a single terminal. If, in the resulting trees, several specimens were grouped into a single clade with the holotype specimen, we merged them together. When merging the scores of these terminals, the differences in character states are considered as polymorphism in the new coding. During the first analysis, the taxonomic sample included two nonrhinocerotid perissodactyls as outgroups (Tapirus terrestris and Hyrachyus eximius) and the ingroup consisted of Epiaceratherium naduongense, Epiaceratherium bolcense and Molassitherium delemontense, as well as specimens attributed to M. albigense, specimens from Bumbach (scored in a single terminal) and the mandibular specimen from Rheinbetts (NMB.O.B.928). We first selected these taxa to test our a priori referrals to these respective genera (Epiaceratherium and Molassitherium). New terminals were then added consecutively, documenting other taxa to which they could be referred, especially within Mesaceratherium and Pleuroceros. Other taxa were added to test the monophyly of Epiaceratherium and Molassitherium and, more broadly, to better understand the early evolutionary steps of Rhinocerotidae, as well as to stabilize the topology of the tree and to test its robustness. A branch-and-bound search algorithm was first used for the analyses encompassing 12 to 16 terminals, after which the heuristic algorithm was faster and almost as efficient. The addition sequence was set to 'furthest' during the branch-and-bound search. We used a random addition sequence of 1000 replicates and held 100 trees at each step during the heuristic search with a TBR swapping algorithm with no reconnection limit and swapping on all trees.

Nomenclatural act
The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature, and hence the new names contained herein are available under that Code from the electronic edition of this article. This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix http://zoobank.org/. The LSID for this publication is: urn:lsid: zoobank.org:pub:C4732CCF-996F-48D1-AC33-C1E733CBDFD9. The electronic edition of this work was published in a journal with an ISSN, and has been archived and is available from the following digital repositories: CLOCKSS, LOCKSS, Portico and PubMed Central.
Diagnosis: Stem rhinocerotine lacking i3 and a lower canine, with a wide postfossette on P2-P4, a protoloph usually constricted on M1-M2, a straight posterior half of the ectoloph on M1-M2, and a posterior valley usually closed on p2.
Stratigraphical distribution: late Middle Eocene to Early Oligocene (South Asia) and Early to early Late Oligocene (Europe).
Geographical distribution: From East to West, Northern Vietnam, Pakistan, Czech Republic, Northern Italy, Germany, Switzerland and France.
Epiaceratherium magnum Uhlig, 1999 [14] Emended diagnosis: large species of the genus with a horizontal mandibular symphysis with divergent i2, a metacone fold weak or absent on P2, a crista sometimes present on P3, a metaloph directed posterolingually on P3-4, a crochet usually present on upper molars, a crista usually absent on upper molars, a V-shaped ectolophid groove developed until the neck on lower cheek teeth, a single-rooted d1, an anteroposterior diameter/height ratio inferior to 0.65, and a nearly straight caudal border of the astragal trochlea. The skull is unknown. Stratigraphical distribution: Early to early Late Oligocene. Geographical distribution: Germany, France and Switzerland. Epiaceratherium cf. magnum was described in lower Oligocene deposits of Pakistan [1].
Description. The mandible NMB.O.B.928 (figure 1) is both well preserved (left ramus) and partly reconstructed (right ramus). The angle between the symphysis and the corpus mandibulae is particularly open, as the symphysis is almost horizontal. The symphysis is wide and without ventral keel. The dorsolateral borders of the symphysis, constricted, form sharp and acute dorsal ridges, between incisors and cheek teeth. The posterior edge of the symphysis is located at the level of p2. There are six mental foramina on each side: the two most anterior are large and ventral, whereas four others are more lateral and smaller. On the left side, the four smaller foramina are located below p1-2, whereas on the right two are located below p2-3 and other ones are anterior to p1. On the medial side of the corpus mandibulae, the groove for the mylohyoid nerves and vessels is well marked and extends from m1/2 to the ramus. The ventral edge of the corpus is completely straight in lateral view. The ramus is reconstructed on the right side, but partly preserved on the left. It is vertical, with a well-developed coronoid process. The large foramen mandibulare opens slightly below a hypothetical horizontal line formed by the teeth neck.
The lower dental formula is i1, i2, d/p1-m3. The premolar series is long compared to the molars (Lp3-4/Lm1-3 > 0.5). The cement is globally absent on cheek teeth. The first incisors are typically incisor shaped and well developed, with a clear neck and a thin lingual cingulid. They are partly worn by contact with upper teeth (and food) and almond shaped in occlusal outline. The very large second incisors are tusk-like and slightly divergent. They are also very worn, without a clear neck. The wear surface is triangular, with a much more worn medial side. There is no cingulid.
The p1 is absent on both sides, but there are small alveoli, anterior to p2, for a small single-rooted d/p1. There are very thin and smooth external rugosities on the ectolophid of p2-3. The ectolophid groove of p2-m3 is angular but vanishing above the neck. On p3, this groove is very oblique, and becomes almost horizontal above the neck. The trigonid of the cheek teeth is angular and forms an acute angle. The entoconid and metaconid are not constricted. The posterior valley of p2 is lingually closed, while on p3-4 it is very narrow and V-shaped, in lingual view. The lingual cingulid is completely absent on p2-m3. The labial cingulid is only partly present on m2-3, but very weak. The paralophid of p2 is developed, isolated and spur-like. The hypolophid of the lower molars is oblique. There is no lingual groove on the entoconid of m3 table 1. royalsocietypublishing.org/journal/rsos R. Soc. Open Sci. 7: 200633 Comparisons. The mandible NMB.O.B.928 can be unambiguously attributed to a rhinocerotoid, based on the typical shape of the lower cheek teeth (ectolophodont cheek teeth with a cristid obliqua directed towards the protoconid). An attribution to the Amynodontidae, Eggysodontidae, Hyracodontidae or Paraceratheriidae can be excluded based on its anterior dentition, due to the tuskshaped second lower incisor and the absence of third lower incisor and lower canine [32]. The mandible can be further referred to a small to middle-sized rhinocerotid, thus excluding large-sized European Oligocene rhinoceroses such as Ronzotherium or Diaceratherium [4,16,17,22].
Six genera of small to medium-sized Rhinocerotidae are known in Europe during the Oligocene-Early Miocene. Pleuroceros, Plesiaceratherium and Protaceratherium (sensu [19]) are all reported from the ?latest Oligocene to the Early Miocene [33]. Pleuroceros differs from NMB.O.B.928 in having a smooth   [17]. The lower teeth of Molassitherium delemontense are very poorly known [19,36] and very similar, but they differ in having a slightly more oblique hypolophid on m1. Molassitherium cf. delemontense from Nuceto [36] have a more constricted paralophid on p2. To sum up, the mandible can be assigned to Epiaceratherium, as supported by our phylogenetic analyses. Both entities share the absence of i3 and c, the absence of lingual cingulid on the lower premolars, as well as external rugosities on the ectolophid of the premolars and the usually closed posterior valley on p2, which are diagnostic characters shared by all species of the genus [12]. Within this genus, E. naduongense and E. bolcense differ by the biradiculate d/p1 whereas it is uniradiculate on NMB.O.B.928, as in E. magnum [12,14]. Epiaceratherium bolcense further differs by the constriction of the metaconid on the cheek teeth [13], whereas E. naduongense differs by a slightly more convex ventral border of the mandible and a slightly more upraised symphysis [12]. Therefore, we refer to the mandible NMB.O.B.928 as Epiaceratherium magnum. Stratigraphical distribution: MP25-MN3 (emended from [22]). Geographical distribution: France, Switzerland, Germany, Pakistan [16,[22][23][24][25]32,37].  Differs from M. welcommi by a posterior border of the mandibular symphysis at the level of p2, lower cheek teeth without a constricted entoconid, an ectolophid groove deep to angular and not interrupted above the neck, an angular and acute trigonid, a rounded distal border of the lunate and a magnum-facet does not reach the anterior side of the lunate.
Differs Etymology: From the last name of the original person who discovered these specimens, Gottlieb Tschan, from Merligen (Bern Canton, Switzerland).
Description: The material from Bumbach is poorly preserved, sliced and crushed. Both sides are preserved on the mandible NMBE5033614 (figure 2a-e). Only p3 and a part of m3 are preserved on the left side, whereas p1-3 and labial parts of p4-m2 are preserved on the right side. The symphysis is broken, but it is nonetheless upraised compared with the corpus mandibulae. The symphysis is quite wide, and its posterior border is located at the level of p2. As on the previously described mandible NMB.O.B.928 from Rheinbetts, there are several mental foramina on each side. On the right side, two large ones are at the level of p1/2 and p3 and a much smaller one is between p1 and i2. On the left side, three large foramina are at the level of p1/2 and a smaller one below p2. However, the symphysis is very badly preserved, and some parts are missing. The lingual groove for the mylohyoid nerves and vessels seems to be well marked on the corpus, but it could also be an artefact from the restoration. The ventral base of the corpus mandibulae is straight. On the other mandible available, NMBE5033615 (figure 2e,f ), only the right part is preserved, with p4 and a part of m1. The symphysis is broken, but the ramus is preserved, and it is inclined forward and upward. The coronoid apophysis is broken and the foramen mandibulare is located below the teeth neck. The premolar series seems long compared to the molar series, but it is difficult to estimate precisely the exact length of the latter. Cement is completely absent. Only the root of one i2 is partly preserved on NMBE5033614 (figure 2a-e). Based on the fragmentary and heavily crushed remaining part of the symphysis, i1s were probably absent. The third incisor and the canine are also absent. The p1 is present, large and single-rooted. There are very weak vertical rugosities on the anterior part of the ectolophid of p2-3. The ectolophid groove of the cheek teeth is deep to angular and does not vanish above the neck on m2 but does on premolars. The trigonid is angular and forms an acute angle in occlusal view. The opening of the posterior valley is acute on premolars, and the lingual cingulid is very weak, but present below the anterior valley. Labial cingulid is present on the trigonid, and below the ectolophid groove. The paralophid of p2 is curved, without constriction, and developed. Two distal fragments of humerus (NMBE5033619 and NMBE5033620; figure 3a-d ) are referred to this taxon. The diaphysis is slender. The humeral crest forms a right dihedron with the epicondylar crest. The fossa olecrani is narrow and high. The trochlea is poorly constricted in its median part (egg cup sensu [27]).
royalsocietypublishing.org/journal/rsos R. Soc. Open Sci. 7: 200633 The medial lip is more developed. The lateral epicondyle is rather low, moderately developed and bearing a shallow distal gutter. There is no scar (synovial fossette) on the anteroproximal part of the trochlea.
The radius is represented by two deformed specimens, but nearly complete. NMBE5033621 (figure 3i) includes also the proximal part of the ulna. The latter is not completely in anatomical connection, the radius being rotated by about counter-clockwise 90 o and the effective welding is due to taphonomical process. However, it shows clear synostosis traces resulting from a contact radius/ ulna. From the radius NMBE5033621 (figure 3j ), the anterior border of the proximal articulation is straight, the medial border of the diaphysis is rather straight, the m. extensor carpi groove is wide and deepened by the strong tuberculum dorsale lying beside it, and the posterior expansion of the scaphoidfacet is low. On NMBE5033622, the proximal ulna-facets are clearly separate and the insertion of the m. biceps brachii is shallow. From the ulna, the olecranon is thick, forming a rather closed angle with the diaphysis. The posterior tip of the olecranon (insertion of the m. triceps brachii) is salient with respect to the distal border of the process. By its dimensions, the humeral cochlea fits the distal trochlea of the humerus NMBE5033620.  The carpus is slender. Both the left scaphoid (NMBE5033628; figure 3k,l ) and lunate (NMBE5033627; figure 3m,n) are posteriorly incomplete but they fit together with the radius-ulna (NMBE5033621) and belong probably to the same individual. The posteroproximal facet of the scaphoid for the lunate is lacking, but the two bones are posteroproximally in contact. The trapezium-facet and the magnumfacet are not visible from the scaphoid. The lunate has no ulna-facet. The anterior side is smooth, with a rounded distal border. The magnum-facet does not reach the anterior side. In lateral view, the distal pyramidal-facet is elliptic and anteriorly elongated by a thin band stretching to the anterior border. The unciform-facet is sagittally elongated.
The unciform NMBE5033623 (figure 3o,p) is well preserved. The proximal facets are separated by an acute ridge. The semilunate-facet roughly outlines a quarter-circle nearly flat. The pyramidal-facet is larger, slightly concave transversally and regularly convex sagittally. There is a wide posterolateral  The anterior one is elliptic and subhorizontal, while the posterior one is circular and lower. The diaphysis is curved laterally, at the proximal third of the bone. The m. interosseus is restricted to the proximal half of the diaphysis. The diaphysis is slightly sagittally flattened. The intermediate relief is high and acute. The medial lip is lower than the lateral one. The latter is transversally concave while the former is flat. In lateral view, the proximal end bears a small articular McV-facet.
Comparisons. As for the specimen from Rheinbetts, these specimens from Bumbach can be unambiguously attributed to a rhinocerotid, based on the typical shape of the lower cheek teeth (ectolophodont cheek teeth with a cristid obliqua directed towards the protoconid) and the anterior dentition (absence of i3 and c, and tusk-shaped i2). It can also be further referred to a small to medium-sized rhinocerotid, thus excluding large-sized European Oligocene rhinoceroses such as Ronzotherium or Diaceratherium [4,16,17,22].
Among the six genera of medium-sized Rhinocerotidae known in Europe during the Oligocene-Early Miocene interval, Pleuroceros differs by a nearly horizontal mandibular symphysis (upraised on NMBE5033614), a smooth and U-shaped external groove on lower cheek teeth (deep to angular on NMBE5033615 and NMBE5033614) and a continuous lingual cingulid on lower premolars (reduced on NMBE5033615 and NMBE5033614) [25]. Plesiaceratherium differs by premolars with shallow ectolophid groove (deep to angular on NMBE5033615 and NMBE5033614) [27,34]. Protaceratherium differs by the narrower symphysis in dorsal view (wider on NMBE5033614), the smoother ectolophid groove of the cheek teeth (deep to angular on NMBE5033615 and NMBE5033614) and the stronger cingulid (usually absent on NMBE5033615 and NMBE5033614) [19,35]. Molassitherium albigense (lower dentition unknown for 'M.' delemontense) differs by the much stronger cingulid on the lower cheek teeth (usually absent on NMBE5033615 and NMBE5033614) [17,19]. Epiaceratherium differs by a posterior valley usually closed on p2 (open on NMBE5033614) and the absence of lingual cingulid on the lower   McIII from Laugnac (Mesaceratherium paulhiacense) McIV from Laugnac (Mesaceratherium paulhiacense) premolars ( present on NMBE5033614) [12]. In agreement with our phylogenetic results, we refer the material from Bumbach to Mesaceratherium. Mesaceratherium shares with these specimens a lingual cingulid on lower premolars, an angular ectolophid groove of the cheek teeth and a rather upraised mandibular symphysis [16,22,25,37]. Furthermore, the postcranial remains from Thézels assigned to M. gaimersheimense also share with the specimens from Bumbach the absence of scar on the trochlea, the egg-cup-shaped trochlea and the high fossa olecrani of the humerus, the straight anterior border of the proximal articulation and medial border of the epiphysis of the radius, the closed angle between the olecranon of the ulna and the diaphysis, the triangular facet for the unciform on the McIII, and the overall similar dimensions (tables 3 and 4) but differ by other characters, such as a possibly tetradactyl manus, the remote pyramidal and McV facets on the unciform or the keeled anterior side of the lunate [22]. Therefore, based on this unique combination of characters, we refer to these specimens from Bumbach as Mesaceratherium tschani sp. nov. (1) whether the specimens from St-Henri/St-André and Moissac (described by Lihoreau et al. [38]) all belonged to Molassitherium albigense, which our analyses have confirmed; (2) whether Molassitherium and Epiaceratherium were monophyletic in their previous acceptation; however, here the results cannot support their monophyly, but a larger taxonomic sampling is necessary to support this claim; (3) whether the specimen NMB.O.B.928 from Rheinbetts could be confidently referred to Epiaceratherium magnum, which has been confirmed by the analyses; (4) whether the specimens from Bumbach could be referred to Molassitherium (as suggested by Scherler et al., 2013: Online Resource 1), which has been partly refuted by the results.
Thus, more taxa were needed to test at least two hypotheses: monophyly of Molassitherium and Epiaceratherium and refined taxonomic assignment of the specimens from Bumbach. To test them, we added six early-branching genera of Rhinocerotidae (Uintaceras, Teletaceras, Penetrigonias, Trigonias, Amphicaenopus and Subhyracodon) in order to stabilize the tree base, through the analyses 6-11. Based on the results of these analyses, it became clear that Molassitherium as originally refined was not monophyletic. However, no referral to known species can be proposed for the specimens from Bumbach based on these analyses. We then added several more derived rhinocerotid taxa to further test the non-monophyly of Molassitherium and get a stable placement for the specimens from Bumbach, which tended to branch close to Pleuroceros and/or Mesaceratherium without clear preference. Thus, during analyses 12 to 20, nine new genera were added, comprising Oligocene to Miocene genera of Rhinocerotidae, representing two subfamilies: Elasmotheriinae and Rhinocerotinae (comprising Rhinocerotini and Aceratheriini [Aceratheriina + Teleoceratina]) as well as Ronzotherium, an early large-sized rhinocerotid. Finally, the lesser known species Mesaceratherium paulhiacense was added to further test the monophyly of Mesaceratherium, and the placement of the specimens from Bumbach. Based on these results, these specimens are attributed to Mesaceratherium tschani sp. nov. This is the placement resulting from most of our analyses, including the most comprehensive ones, and for which the topology of tree is the most stable. Thus, adding more taxa would now be unnecessary, although it could lead to new topologies. Indeed, our identifications would probably not drastically change as they are well supported by numerous characters. The definitive tree (single most parsimonious tree) resulting from the most complete taxonomical sampling is presented in figure 4.
Our new phylogeny leads to an important change concerning the monophyly of Molassitherium, which cannot be supported by our results. Indeed, the species 'M. delemontense Becker & Antoine, 2013 [19]' should be assigned to Epiaceratherium instead. Molassitherium delemontense was erected on a very well-preserved skull from Poillat (Jura Canton, Switzerland), which shared numerous similarities with the skull of Molassitherium albigense. However, based on our results, it shares more characters with Epiaceratherium than with Molassitherium, such as: a foramen infraorbitalis above P3 (char. 2), as in E. bolcense, but also M. albigense; a sharply deviating anterior tip of the zygomatic process in ventral view (char. 37), as in E. bolcense, but contrary to M. albigense; a long premolars series compared to the molars (char.  Table 5. Taxonomic sample used for each analysis, with details on the search algorithms used and a summary of corresponding results. The resulting most parsimonious trees, or consensus trees are available in electronic supplementary material, S2, where page numbers correspond to the analysis number in the first column.   Epiaceratherium that are also found in M. albigense, it also shares with Epiaceratherium 11 synapomorphies that are not found in Molassitherium. Therefore, in agreement with these results, we propose the new combination 'Epiaceratherium delemontense (Becker & Antoine, 2013 [19])'. This could imply that this species would have had, like all epiaceratheres, a complete upper dentition, including three pairs of incisors and one pair of canines. A hypothetical representation of the complete skull of Epiaceratherium is shown in figure 5 and was generated by virtually assembling the type cranium of E. delemontense comb. nov. (specimen MJSN POI007-245; three-dimensional model available in [39]) to the hypothetical gypsum reconstruction of the snout of E.

Evolution of the anterior dentition in early Rhinocerotidae
The reduction of the anterior dentition is one of the major adaptative traits of the Rhinocerotidae, which developed much earlier than the horn(s). Indeed, the first horns may have appeared in the American genus Diceratherium, from the Early Oligocene [27,40,41] whereas the reduction of the anterior dentition was initiated already in Late Eocene times, notably with Trigonias, a well-known genus from North America. The lower anterior dentition of Trigonias has been subject to some confusion. Lucas [41] originally considered the alveolus just behind the tusk-shaped i2 as the alveolus for i3 when he created this genus, but later, Gregory and Cook [42] believed that it could be in fact the alveolus for a canine. Indeed, on one mandible from a juvenile individual (No. 1027), two alveoli are present distally to i2,  which indicates the presence of both i3 and a lower canine (or di3 and dc). Based on the shape and position of these alveoli, they suggested that the single tooth retained in the adult was the canine, instead of i3. However, Russell [43] showed that in some specimens the canine was definitely lost in the subadult while the i3 was still present. Thus, the lower canine is lost in the adult, but may still be present in the juvenile. However, Prothero et al. [32] and Antoine [27] consider that the lower canine and i3 are lost in the adult Trigonias (see [32]: fig. 4, character 28 and [27]: p. 134), following Radinsky [44], who also considered the lower canine and i3 as lost in the adult, and not only in the juvenile. The confusion can perhaps be traced back up to Scott [45] who indicated the presence of three lower incisors in the dental formula of Trigonias ( p. 776) yet later indicated in the description that 'of the third incisor and the canine no vestige remains' (p. 777), perhaps meaning that these teeth were physically absent, but not their alveoli. On the contrary, Wood [46,47] considered the presence of the lower i3 as a defining generic character, whereas the presence of the lower canine would be variable. Here, we choose to follow Lucas [41] and Wood [46,47] and consider the lower i3 as present in the adult. Indeed, although the paratype mandible USNM-4815 (in Smithsonian National Museum of Natural History) belongs to an old individual with well-worn teeth, as reported by Wood [46], the alveoli for the i3 are still very large and not sealed (figure 6), which indicates that this tooth would have been present during most of the animal's life. However, there may still be some intraspecific variability concerning the presence of this i3, since some individuals reported by Russell [43] show no trace of i3 ( fig. 26 for example).  The evolution of the major traits concerning upper and lower anterior dentition is optimized on the most parsimonious tree, using the accelerated transformation parameter (ACCTRAN), favouring reversals instead of convergences ( figure 7). Thus, according to our phylogeny, the absence of a lower canine can be regarded as a synapomorphy of the Rhinocerotidae, acquired quite early during the evolution of this group ( figure 7). Yet, the presence/absence of an upper and/or lower canine remains a rather labile character in this group and may further be subject to sexual dimorphism, as in other laurasiatheres, including perissodactyls. For example, the canine can be present in the adult male domestic horse, but not in the female [50]. Even within rhinoceroses, a rudimentary canine can sometimes be present in some individuals, as observed on one skull of Dicerorhinus sumatrensis (copy in MJSN collection). On some skulls of Subhyracodon, an upper canine may be present, although only on one side and not the other [40]. We nonetheless coded it as absent for this species, as it is the most-often seen condition. In other groups of perissodactyls, there can be an important sexual dimorphism on the size of the canine (e.g. in Lophiodontidae, as reported by Vautrin et al. [51]). Thus, this might explain why the lower canine could have reappeared at least twice during the evolution of the Rhinocerotidae, in Uintaceras and Teletaceras, for example ( figure 7).
Another typical feature of the Rhinocerotidae is the tusk-shaped second lower incisor, which might have been acquired at the same time as the lower canine was lost (figure 7). It is indeed tusk-shaped in all Rhinocerotidae (when it is present), except in Uintaceras who might still have retained an incisor-shaped i2 [52]. However, the anterior dentition of this genus remains poorly known, and was not found in connection with the mandible, but the roots preserved in the lower jaw suggest nonetheless that i2 may have been smaller than i3.    [1,12,19,25,40,48,49]. Menoceras arikarense and Hispanotherium matritense have been added to illustrate convergent reduction of anterior dentition among early Elasmotheriinae, and were placed according to previous formal phylogenies [19,25].
Epiaceratherium is peculiar, as it possesses a complete upper anterior dentition, but has lost its i3 and c. This differs from the condition seen in all other rhinocerotids, in which I3 is lost at the same time as i3. Thus, i3 has been lost at least twice independently among Rhinocerotidae. I3 is also the first upper anterior tooth lost, before the successive loss of C and I2, and contrary to what occurs in the lower jaw. In Penetrigonias dakotensis, I3 is lost while C is unambiguously retained on the anterior tip of the maxilla [40]. However, Russell [43] identified I3 on the premaxilla in Penetrigonias sagittatus. Yet, Prothero [40] considers it as a canine. Thus, more investigation would be needed to determine the exact condition in this species. Finally, the loss of I2 is a distinguishing feature of Rhinocerotinae and Elasmotheriini, who retain only the typical chisel-tusk complex formed by I1/i2, as well as a non-functional i1 in many taxa.

Conclusion
Based on parsimony analyses of Rhinocerotidae, we identify an unpublished mandible (NMB.O.B.928) from the 'middle' Oligocene of Rheinbetts (Basel Canton, Switzerland) as belonging to Epiaceratherium magnum. As the first complete and well-preserved mandible available for this species, it allows for assessing the absence of i3 and c. Along with this mandible, newly prepared specimens from Bumbach (MP25, early Late Oligocene, Switzerland) are referred to a new species: Mesaceratherium tschani sp. nov. These remains document the oldest occurrence of Mesaceratherium, which was previously only known from MP28 onward.
Our new phylogenetic analyses support a new combination: Epiaceratherium delemontense (Becker & Antoine, 2013 [19]) comb. nov. replaces 'Molassitherium delemontense Becker & Antoine, 2013 [19]'. Optimization of concerned characters further allows for inferring that this species would have retained a complete upper anterior dentition, as in all other species assigned to Epiaceratherium. A new evolutionary scenario is proposed for tracking anterior dentition reduction in Rhinocerotidae. Based on this new scenario, the third lower incisor would have been lost at least twice independently, whereas I3 and C would have been lost only once, successively.
Data accessibility. All data generated or analysed during this study are included in this published article (and its electronic supplementary material files). The morphological data matrix is available in NEXUS format in electronic supplementary material, S1. Three-dimensional models are available from the MorphoMuseuM repository (https:// doi.org/10.18563/journal.m3.116) with DOIs https://doi.org/10.18563/m3.sf.534 (three-dimensional surface model of the mandible NMB.O.B.928) and https://doi.org/10.18563/m3.sf.534 (three-dimensional surface model of the archetype reconstruction of Epiaceratherium).