New data on Amynodontidae (Mammalia, Perissodactyla) from Eastern Europe: Phylogenetic and palaeobiogeographic implications around the Eocene-Oligocene transition

Amynodontidae is a family of Rhinocerotoidea (Mammalia, Perissodactyla) known from the late Early Eocene to the latest Oligocene, in North America and Eurasia. European Amynodontidae are very rare, and all remains belong almost exclusively to a single post—Grande Coupure genus from the Oligocene, Cadurcotherium. The “Grande Coupure” defines an extinctions and dispersal-generated originations event in Europe that is nearly contemporaneous with the Eocene-Oligocene transition. Perissodactyls are one of the major groups affected by this event: Palaeotheriidae went almost extinct during this crisis, whereas Rhinocerotidae appeared for the first time in Europe. Study of fossiliferous Eastern-European localities from this age is crucial for the understanding of this crisis. We report here three new localities of Amynodontidae in Eastern Europe. Two of them are dated from the Eocene (Morlaca, Romania; Dorog, Hungary), whereas the other is either Late Eocene or Early Oligocene (Dobârca, Romania). The skull from this latter locality belongs unexpectedly to the same individual as a previously described mandible attributed to “Cadurcodon” zimborensis. As a result, this specimen can be allocated to its proper locality, Dobârca, and is assigned to a new genus, Sellamynodon gen. nov. It is characterised by an extraordinary growth of the nuchal crest, a unique character among amynodontids. Along with this remarkable material from Dobârca, two specimens from another Romanian locality, Morlaca, have been recently discovered and are dated from the Late Eocene. They belong, as well as new material from Dorog (Middle Eocene, Hungary), to the genus Amynodontopsis, also found in North America. The new Hungarian material represents the earliest occurrence of Amynodontidae in Europe. New phylogenetic hypotheses of Rhinocerotoidea are proposed, including the new material presented here, and show that Amynodontidae may be closer to the polyphyletic family ʽHyracodontidaeʼ than to Rhinocerotidae. Amynodontidae, with their deep preorbital fossa and extremely reduced premolars, display in fact a very derived condition, compared to rhinocerotids.


Geology
The specimen from Hungary was given to Miklós Kretzoi during his lifetime, though he never mentioned or described it. It is labelled "Dorog substratum" and therefore we assume that it comes from the Dorog coal Formation in Dorog Basin (Hungary). This is confirmed by the especially dark and carbonaceous preservation of the specimen. This formation is either latest Lutetian or Bartonian in age (late Middle Eocene; [25,26]).
The mandible from Dobârca was originally described as coming from the Zimbor Strata (Sălaj County, Romania) and assigned to "Cadurcodon" zimborensis [30], but its allocation was corrected as coming from the Dobârca locality (Sibiu County) [31]. Indeed, the recent discovery of the matrix counter-impression of the mandible in the collections of the UBB revealed that it was labelled "Dobârca". Furthermore, it permitted the association with a previously unpublished skull, which was properly labelled "Dobârca", and articulates perfectly with the mandible. On the basis of these new data, we assume that both belong to the same individual and thus come from that same locality. However, its age remains uncertain. Dobârca is located at the boundary between the Transylvanian Basin and Southern Carpathians. In Dobârca Valley, fluvial red beds are in contact with the metamorphic rocks of the Cândrelului Mountains and overlaid by grey-bluish sandstone interbeddings [32]. The amynodontid remains were probably unearthed from this layer, but no other fossils are recorded in it. The succession can be compared with the red clay recorded eastward in Rodului Valley, at Apoldu de Sus. There, Mészáros et al. [33], mentioned a pile of "red clay, gray-greenish clayish and mica sandstone" with Chara, considered by these geologists as "Burdigalian-Helvetian" (i.e. Lower Miocene), laying on Upper Eocene (Priabonian) deposits. On the southern border of the Transylvanian Basin, Eocene lithostratigraphic units were coined by Mészáros [34] at Turnu Rosu-Porcești, near Sibiu. The youngest one, named Valea Nișului Formation is Priabonian-Lower Oligocene. The Oligocene rocks refer to "sandstone with fish scales", bearing also plant remains. Visibly the rock succession of Dobârca is not identical neither with the one of Turnu Roșu-Porcesti, nor with the Ighiu Formation (Late Eocene-Early Oligocene; [35]) from the southwestern area of the basin. Considering the rock succession from Apoldu de Sus as the most convenient for a comparison with the one of Dobârca, one may consider the sandstone bearing the amynodont remains as late Eocene or Early Oligocene in age, which can be included for instance in Valea Nișului Formation. The similar rocks of Apoldu are herein considered as Oligocene, not Miocene [32,33].

Nomenclatural acts
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:6BBF9472-2A60-4C51-BD82-3156FD9A4827. 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: PubMed Central, LOCKSS.

Anatomical terminology and characters
Cranial and dental terminology is from Antoine [36]. Anatomical features described follow the same sequence as Antoine [36,37], completed by characters newly observed in this work. Capital letters are used for upper teeth (I, C, P, M), and lower-case letters for lower teeth (i, c, p, m). Italicized numbers between parentheses refer to the character number in the data matrix (see S1 File). Dental and cranial measurements are taken according to Uhlig [38] and Antoine [36], respectively. Dimensions are given in mm.

Phylogenetic analysis
The morphological characters matrix is based on a published matrix [36,37], and is available in Supporting information (S1 File) in NEXUS and TNT format. Fifteen characters were added to better resolve the relationships within Amynodontidae (for which the original data matrix was not created specifically). Eight of them (290-297) are taken from another matrix [4]: characters 8,9,16,23,25,39,41  Characters were scored based on specimen observations and publications, including descriptions, pictures and illustrations ( Table 1). The cladistic analysis was performed with the heuristic search in PAUP Ã version 4.0b10 and 4.0a159 [39] and with the traditional search in TNT version 1.1 [40]. In PAUP, 1000 replicates of random-addition-sequence were performed, with a TBR branch-swapping algorithm and MulTrees option in effect. Starting trees were obtained by stepwise addition and 100 trees were held at each replicate. In TNT, starting trees were obtained from Wagner trees. 1000 replicates of addition-sequence were performed using TBR algorithm and saving 100 trees per replication. Seven characters do not form a morphocline and were considered as unordered: characters (72), (94), (102), (187), (190), (292) and (293); all the others were ordered. Four characters have been modified from the original matrix [36,37] to form morphoclines: (36), (60), (103) and (140) (see data matrix in S1 File). Uninformative and constant characters were deleted prior to the analyses ("exclude uninf" and "exclude constant" commands in PAUP and "xinact" command in TNT); thus, 184 characters were kept during the analysis for 38 taxa. Decay index (Bremer support) was calculated with β Ronzotherium filholi (Osborn 1900) the use of TreeRot.v3 software [41] and PAUP Ã version 4.0b10 (TreeRot script was not compatible with PAUP Ã version 4.0a159). The taxonomic sampling includes five taxa previously coded [37]: Tapirus terrestris (considered as outgroup for this analysis), Hyrachyus eximius and three unambiguous Rhinocerotidae (considered part of the branching-group [36,37]): Trigonias osborni, Ronzotherium filholi and Aceratherium incisivum. Thirty-three new terminals were added to the matrix, including the new amynodontid remains (two taxa), seventeen amymodontids, one tapiroid (Protapirus simplex [= Protapirus validus]; also considered as outgroup), and the branching-group. Terminals in the branching-group have been chosen to test the monophyly of Amynodontidae as well as their position inside the Rhinocerotoidea. Therefore, they have to represent all major groups of rhinocerotodoids (Rhinocerotidae, Eggysodontidae etc.). It includes another species of Hyrachyus (H. princeps; though it was considered a junior synonym of H. eximius [90], we consider it as valid based on several anatomical differences), two equivocal early diverging rhinocerotids (Uintaceras radinskyi and Teletaceras radinskyi) and 10 hypothetical "hyracodontid" species. However, the goal of this analysis is not to resolve the phylogeny of Rhinocerotoidea, which is beyond the scope of this study. The branching-group should only lead to a better polarization of characters, which is necessary here because the two outgroups are both tapiroids and thus act as a single monophyletic outgroup. Terminals in the ingroup (i.e. our group of interest) have been chosen to represent nearly all amynodontids, to allow the placement of the new material. The taxonomic sampling and coding sources are provided in Table 1.

Remarks on taxonomy
The taxonomy of Amynodontidae was recently revised [4], and we follow these suggestions on most of the taxonomic considerations. Therefore, we consider Lushiamynodon Chow & Xu, 1965 a synonym of Sharamynodon Kretzoi, 1942 andSianodon Xu, 1965 (with type species S.  [37] β Teletaceras radinskyi Hanson 1989 Hanson 1989 [84] β Trigonias osborni Lucas 1900 Antoine et al. 2010 [37] β Triplopus obliquidens ( β symbol before taxon name indicates that it belongs to the "branching-group". https://doi.org/10.1371/journal.pone.0193774.t001 New Amynodontidae from Eastern Europe bahoensis) as a junior synonym of Cadurcodon Kretzoi, 1942. We also agree not to recognize "Sianodon" ulausuensis and "Sianodon" gaowangouensis as belonging to Cadurcodon, since both have three upper incisors, whereas Cadurcodon only has two. "Sianodon" ulausuensis Xu, 1966 has not been included in the analysis because it acted as a "wildcard" taxon, probably because the referred cranial specimen is much distorted and most teeth are not preserved or incomplete. "Sianodon" gaowangouensis however has been included, which was represented by a palate specimen bearing complete dental series [8]. However, we follow Lucas (2006; [3]) and consider Zaisanamynodon protheroi as valid but Procadurcodon orientalis as a nomen dubium (nomen vanum). Metamynodon bugtiensis Forster-Cooper, 1922, from the Bugti deposits in Baluchistan, was described as a gigantic amynodontid [91], based on the belief that the last erupting tooth (still far in alveoli) was a M3, mostly because there was no space behind it for another tooth. Yet, dissecting the specimen more than 10 years later (Figs 20A and 22 in [73]), it was revealed that what was considered as P4 and M1 were in fact DP3 and DP4. The first preserved tooth is then a DP1, and not a P2 or DP2. This is also congruent with the fact that the M1 is usually the most worn tooth (e.g. Fig 19.5 in [75]). This specimen should therefore be referred to an aberrant form of Paraceratherium bugtiense lacking the M3 [73,92].

Emended diagnosis
A medium-sized amynodontid (Lm1-3 = 114.0), distinguished from all other members of the family by the huge development of nuchal crests, a very high zygomatic arch, a deeply concave dorsal profile of the skull, the absence of a sagittal crest, the absence of a lingual groove on the body of mandible and a ramus of mandible inclined forward. Within Metamynodontini, differs from Paramynodon birmanicus by the presence of a postorbital process on zygomatic arch, a right-angled anterior tip of the zygomatic process of maxilla in ventral view, an upraised mandibular symphysis, a continuous labial cingulum of lower premolars and a labial cingulum of lower molars always present. Differs from Megalamynodon regalis in having a right-angled anterior tip of the zygomatic process of maxilla in ventral view, a well-developed paraoccipital process, a mental foramen at the level of p2-4, a well-developed coronoid process of the mandible, a low condylar process of the mandible and a talonid of m3 enlarged. Differs from Metamynodon planifrons in having a high anterior base of the zygomatic process of maxilla, an upraised mandibular symphysis, a mental foramen at the level of p2-4, the presence of labial cingulum on lower premolars and molars, an enlarged talonid of m3 and a strong paracone fold on M3.
Remarks. Sellamynodon zimborensis also differs from early diverging Amynodontidae (Rostriamynodon grangeri, Amynodon advenus and Sharamynodon mongoliensis) by the absence of p2 and in having a smooth external groove of lower cheek teeth, an enlarged talonid on m3 and a high orbit on the skull. Further differs from Rostriamynodon grangeri in having an anterior border of the orbit above P4-M2, a zygomatic index (= zygomatic width / frontal width) superior to 1.5 and a mental foramen at the level of p2-4. Further differs from Amynodon advenus in having continuous labial cingulum on lower premolars and molars, a strong paracone fold on M3 and a well-developed paraoccipital process. Further differs from Sharamynodon mongoliensis by the presence of a postorbital process on zygomatic arch, an upraised mandibular symphysis and a posterior margin of the mandibular symphysis at the level of p2-4.
Differs from Cadurcodontini, in having a flat postglenoid process and a low condylar process of the mandible. Further differs from Zaisanamynodon in having a postorbital process on the zygomatic arch, a developed nuchal tubercle, a zygomatic index superior to 1.5, a rightangled anterior tip of the zygomatic process of maxilla in ventral view, a subtriangular foramen magnum, an upraised mandibular symphysis, a mental foramen at the level of p2-4, in having cement on upper cheek teeth, continuous labial cingulum on lower premolars, and a wide space between condylar and coronoid processes of the mandible. Further differs from Amynodontopsis and Cadurcotherium in having a zygomatic index superior to 1.5, weak or variable cement on cheek teeth, labial cingulum of lower molars always present and a high orbit on the skull. Further differs from Cadurcodon in having a well-developed coronoid process of the mandible, an upraised mandibular symphysis, weak or variable cement on cheek teeth, an oblique hypolophid of lower molars and a high orbit on the skull.

Description
Skull (Figs 2A-2D and 3A-3D). In dorsal view (Figs 2A and 3A), the sutures for nasals (if present) and parietals are not distinguishable but there is a distinct mesial line that could be the suture line separating the two frontals. Frontals are wide and supraorbital foramen are present. The left anterior side of the skull is more complete than the right side. On its anterolateral extremity, the skull is slightly dorsally elevated and the bone surface is coarse. The postorbital process of the frontal does not enclose the orbit, which is open into the wide temporal fossa. There is no frontoparietal or sagittal crest, only a very low ridge at the posteriormost preserved part of the skull (that could potentially give birth to a larger posterior crest) and no sign of any horn attachment. The braincase is distinguishable by the constriction of the skull. It is completely bordered posteriorly by a large bony "frill" that is formed by the extraordinary growth of the nuchal crest. It is a rather thin, almost circular, bony edge, lacking ornamentation, which arises from the posterior end of the mastoid or temporal crest. It extends at its widest part up to around eight cm from the rest of the skull, but it is broken on the left and posterior sides. The zygomatic arches are preserved, they are wide and high. The maximum width at the zygomatic (Zw) is 317.0, whereas it is 196.0 at the frontals (Fw): the zygomatic index (= Zw/Fw) is therefore of 1.6.
In ventral view (Figs 2B and 3B), the only preserved tooth is the left M3, but there are still parts of the roots for the right M3. The skull is not preserved anteriorly to this teeth. The retromolar space is small. The left zygomatic arch is complete and diverges quickly and strongly from the skull on the left side. It starts approximately above the M2 (not preserved) and is separated from the skull above the middle of the M3. The right side of the skull is slightly less complete; the anterior part of the zygomatic arch is broken posterior to its maxilla suture. The choanae open approximately at the level of M3. The vomer is rounded. The basioccipital is well-preserved and there is a sagittal crest on the basilar process. It is bordered laterally by the piriform fenestra (= lacerum foramen) and the jugular foramen. This fenestra is widely open and shaped like a 'C'. The hypoglossal foramen is in the middle of the ventral condyloid fossa, posterior to the jugular. The caudal alar foramen is present anteriorly to the oval foramen and to the piriform fenestra. The retroarticular process (post-glenoid apophysis) is wide and flat. The articular tubercle of the squamosal is straight and rather high. The posttympanic process is poorly developed whereas the paraoccipital process is more developed and they are both fused at their base.
Viewed laterally (Figs 2C, 2E and 3C), the dorsal profile of the skull is deeply concave. The nuchal crest is oriented posterodorsally. Posteriorly, the occipital condyles are well-expanded and oriented posteroventrally. The zygomatic arch is very high and deeply convex dorsally. It is very thick dorsoventrally: approximately 6 cm at its broadest point, from ventral to dorsal side. The postorbital process is present on the jugal. The orbit is large (around 6 cm in diameter) and well-defined, and its anterior border was anterior to the M3, possibly above premolars or M1. The suture between the jugal bone and squamosal is straight and oblique. The anterior margin of the orbit is much more anterior than the M3, it may have been approximately above the M1 (not preserved). The external acoustic meatus is closed.
In occipital view (Figs 2D and 3D), the "frill" is very large (even though it is only preserved on the right side). The foramen magnum is triangular. There is a small nuchal tubercle (or occipital protuberance) on the occipital. The stylomastoid foramen is visible between the posttemporal and paraoccipital processes. The condyles are smooth, without ridge and slightly expanded ventrally. They are directed almost perpendicularly to the frill.
Mandible (Figs 3E and 4). The mandible is almost complete, some parts have been reconstructed from the counterprint in the sediment. Only the anterior part of the symphysis is missing and p3-4 are incomplete on the right side. The basic dimensions are given in Codrea & Şuraru [30]. The symphysis was either slightly upraised or horizontal, and not very massive. Its posterior margin stops at the level of p4. There are several mental foramina and they are mostly located at the level of p3. The lower base of the body of mandible is straight and there is no lingual groove medially. The height of the body is almost constant. The ramus of mandible is stout and inclined forward. The coronoid process is well developed longitudinally, but very thin, and the condylar process is low. There is a wide gap between these two processes. There is no post-cotyloid process. The mandibular foramen is large and below teeth neck. The angle of the mandible is very strong, and acute. The masseteric fossa is extremely marked, forming a deep pocket in the upper part of the ramus.
The dental formula is classical for Amynodontidae, with a reduction of premolars to p3-p4, and three lower molars. The anterior dentition is unknown, though it is possible to distinguish small root remains, but it is not possible to determine if those were for canines or incisors. There is however a rather long diastema anterior to the p3, and no alveoli for p2 or p1. Dentition (Fig 5; Table 2). The M3 (Fig 5A) is not much worn and enamel is wrinkled. The roots are not visible. The ectoloph wall is covered in part by a very thin layer of cement. Labial and lingual cingulum are absent. The posterior cingulum is continuous. The tooth is very simple and shows no enamel folding. There is no crochet, antecrochet or crista. It is quadrangular: protoloph, metaloph and ectoloph are clearly distinct. The ectoloph is long and nearly longitudinal. Protoloph and metaloph are short and transverse. The metaloph is slightly broken posteriorly. There is no constriction of protocone or hypocone. Parastyle and metastyle are present whereas the mesostyle is absent. The paracone fold is strong and there is no Concerning lower cheek teeth (Fig 5B and 5C), p3 is distinctly shorter than p4 and premolariform, with very little developed hypoconid and entoconid, whereas p4 is almost molarifom, but still retains a developed paralophid. Talonid is much wider than trigonid on p4. Trigonid and talonid of lower molars are rounded and obtuse and the talonid of m3 is longer than trigonid. Metaconid and entoconid are never constricted. The hypolophid of lower molars is oblique, but not as sagittal as in Cadurcotherium. There is no lingual groove of the entoconid. The labial cingulum of lower cheek teeth (Fig 4B) is strong and continuous. The lingual cingulum ( Fig 5C) is interrupted under the entoconid and metaconid of lower molars but continuous on premolars. The external groove is absent and there is no cement.
Remarks. The presence of a large nuchal crest on the posterolateral border of the skull seems to be a unique character among Amynodontidae. It may be homologous to the lambdoidal crest of Zaisanamynodon [3,89], but it is much more developed here. No other observed taxon shares such a deeply concave skull. In most species, the skull is in fact quite flat or even convex and the nuchal crest is never as developed (for instance Metamynodon planifrons [68] or Amynodon advenus [46]). This morphology is more common in Brontotheriidae, such as Diplacodon gigan [99] for example.
It also differs from all other Amynodontidae by the absence of a sagittal crest, or at least by its very strong reduction (it is possible that a very low sagittal crest was present on the posteriormost part of the skull, but this part is broken). The presence of a large sagittal crest was in fact considered as a diagnostic character of amynodontids [2,15]. Also, contrary to all other rhinocerotoids, the dorsolateral surface of the braincase is completely smooth because there is no parasagittal ridge. Parasagittal ridge and sagittal crest are normally used as muscle supports, which means that for Sellamynodon the muscles had to connect differently on the skull (on the nuchal crest for example), or were much reduced. The absence of sagittal crest could also mean that the specimen was a juvenile [21], but this hypothesis is not compatible with the presence of a fully grown and slightly worn M3. It could also be linked to sexual dimorphism, but we lack data for comparison.
The M3 though, is typically amynodontid-like. Its quadrangular shape, with its long and flat ectoloph and the presence of a metastyle are diagnostic characters of Amynodontidae [2]. In Rhinocerotidae such as Trigonias or Ronzotherium for example, the metacone is lost or fused with the paracone and the ectoloph is very short. The amynodontid M3 is more similar to the M3 of some hyracodontids such as Hyracodon, Triplopus or Forstercooperia, with which it shares a metastyle, though the ectoloph is shorter and more lingually deflected in these taxa. The absence of crista, crochet and antecrochet is also diagnostic of Amynodontidae [2]. Therefore, according to this unique combination of characters (absence of sagittal crest, great development of the nuchal crest, quadratic M3, reduction of lower premolars to two and elongation of talonid) this specimen represents a new genus of amynodontid, as was suspected when the mandible of Sellamynodon zimborensis was first described [30].
The body mass of Sellamynodon zimborensis was estimated following the regression equation [m = exp(1,5133 x ln(L x W) + 3,6515] of Perissodactyla [100]. The body mass (m) is in grams, L and W are length and width of m1 in mm, respectively. The first lower molar (m1) was proposed as the most suitable for this estimation, due to its low variability. With L = 28.5mm and W = 19.6mm (see Table 2 in [30]), the body mass of S. zimborensis is estimated to be around 550kg. Recently, the same equation was used for 18 amynodontid species [4] and it appears that the bodymass of S. zimborensis is very close to the most primitive amynodontid, Rostriamynodon grangeri (535kg). It is also similar to some specimens of Amynodon advenus and Amynodontopsis bodei (S2 in [4]). We also used a similar equation based on the ectoloph length of M3, which gives an even smaller body mass, of approximately 410kg.

Description
Maxillary (Figs 6A-6C and 7). The zygomatic arch (Fig 6A) is slender and does not deviate strongly from the maxilla. It starts above M2 and is high above the teeth crowns. There is a ventral groove between the zygomatic and the teeth on the specimen from Morlaca, not visible on the specimen from Dorog.
Dentition (Figs 6 and 7; Table 3). The roots of upper molars (clearly visible on UBB MPS V545; Fig 6A and 6C) are fused on the lingual side and separated on the labial side. There is no cement. Lingual cingulum is only slightly present under the protocone of M1 and the hypocone of M3. Labial cingulum is present and continuous but very thin and weak. The median valley is narrow, acute and V-shaped. The protocone is slightly anteriorly constricted on HNHM PAL 2017.54.1. There is a distinctive oblique lingual blade joining the base of the hypocone to the top of the protocone (Figs 6B, 7B and 7C). Because of this blade, the posterolingual edge of the protocone is sharpened and not rounded. There is no crochet, antecrochet or crista. The medifossette is always absent. Metaloph and protoloph are transverse. The paracone fold is present and it is strong on the M3 (Figs 6A, 6B, 7B and 7C). There is no metacone fold. A metastyle is present on the M3 and it is large on the M2. The M1 is the smallest tooth of the row. The hypocone of the quadratic M3 is much more labial than on the other molars, and the metaloph much shorter.
The lower molar (probably m1 or m2; Fig 6D-6F) belongs to a young individual, it is almost not worn. The ectolophid groove separating trigonid and talonid is present and vanishing before the neck. The talonid is much wider than the trigonid. The trigonid is rounded and obtuse and the hypolophid is oblique. Metaconid and entoconid are not constricted. Lingual cingulum is present and reduced and labial cingulum is present anteriorly and may have been continuous, but only a small part is visible since the tooth is not fully erupted.
Remarks. The upper molars of the specimens from Morlaca and Dorog share all the diagnostic characteristics of the Amynodontidae: absence of crista, crochet and antecrochet, parastyle reduced and quadratic M3 with large metastyle and labial metacone [2].  New Amynodontidae from Eastern Europe Several characteristics differ from those of the skull of Sellamynodon zimborensis. Cement is completely absent on upper molars, the metastyle is directed postero-labially and the lingual blade joining the protocone to the base of the hypocone is present on the specimens from Morlaca and Dorog. Furthermore, even though it is incomplete, the zygomatic arch is not as strong as in S. zimborensis. In ventral view, it is also less deviated from the maxilla than in S. zimborensis and a ventral groove is present. The lower molar also differs from S. zimborensis by a greater development of the external groove of the ectolophid and by a greater width of the talonid compared to the trigonid.
Within Cadurcodontini, it differs from Cadurcodon by a weak paracone fold on M1-2 and from Cadurcotherium by the presence of lingual and labial cingulum on upper molars and an oblique hypolophid on lower molars. It also differs from Zaisanamynodon by a shorter metastyle on M1-2 and the absence of a postfossette on M1. It shares with Amynodontopsis a moderate angle of the zygomatic process of the maxilla in ventral view, the presence of an external groove on lower cheek teeth, a weak paracone fold on M3 and the absence of cement on upper teeth, but differs from A. bodei by a higher anterior base of the zygomatic process of the maxilla and a metasyle of M3 directed more posterolingually. It is therefore tentatively assigned to Amynodontopsis aff. bodei. Its body weight is estimated to be around 500kg with the equation based on the length of M3, which is close to the previously estimated body weight of Amynodontopsis bodei (S2 in [4]).

Phylogenetic relationships
One most-parsimonious tree of 825 steps (excluding uninformative characters; CI = 0.30, HI = 0.70, RI = 0.51 and RC = 0.15; Fig 8) was retrieved during the heuristic search in PAUP Ã version 4.0a159 and PAUP Ã version 4.0b10. The same topology was found with the traditional search in TNT version 1.1. It should be noted that PAUP Ã version 4.0b10 was unable to show unambiguous synapomorphies in ACCTRAN optimization, whereas PAUP Ã version 4.0a159 depicted them accurately. This version was then used to visualize transformations, as well as TNT and Mesquite version 3.2 [102]. The complete list of transformations is provided in S2 File.
Suprageneric relationships within Rhinocerotoidea are divergent from other recent phylogenies, especially concerning the Hyracodontidae, which is clearly paraphyletic (following the "classical" definition of the family), as also recovered by recent study [23]. The results presented here are disputable for several reasons. Firstly, the taxonomic sampling is unequal depending of the families (it is almost exhaustive for Amynodontidae, but poor for Rhinocerotidae and Hyracodontidae). Secondly, our matrix was not designed to resolve the phylogeny of early rhinocerotoids. Phylogenetic results outside Amynodontidae presented here are thus preliminary, but this analysis is necessary to discuss their position within Rhinocerotoidea.   Table 1. Thick bars represent temporal and spatial distribution of taxa and thick bars with dashed lines have uncertain ages. Geological time scale produced with TSCreator [103]. Numbers at nodes are Bremer support values.
Indeed, it was previously investigated only through the use of two basal amynodontid taxa [20][21][22][23], which is clearly insufficient to represent the variability of this group.
Most nodes have very low support values, i.e. very few clades are supported with a Bremer support value superior to 1. The best supported clades are basal to Amynodontidae, with Bremer values of 5: the clade Aceratherium-Ronzotherium and the clade Uintaceras-Teletaceras. Amynodontidae have a Bremer support value of 3 and within Amynodontidae, only two clades have support values higher than 1: Zaisanamynodon and the clade Amynodontopsis aff. bodei-Amynodontopsis bodei. These low support values indicate that taxonomic inferences should be taken cautiously. They are probably due to a small taxonomic sampling of Rhinocerotidae, for which the original characters matrix was designed [36,37], compared to an almost exhaustive sampling of Amynodontidae.

Rhinocerotidae
Rhinocerotidae sensu stricto (Trigonias, Ronzotherium and Aceratherium) is supported by 11 unambiguous synapomorphies, including the presence of tusk-like i2 (79), the absence of i3 (81) and of lower canine (82). They are supported by a Bremer value of 4. Contrary to most previous studies [20][21][22][23]104], the monophyletic Hyrachyus is here closer to Rhinocerotidae than to all other rhinocerotoids. However, our taxonomic sampling of Rhinocerotidae is very small, so the proposed relationships are only tentative and should be further investigated in future studies. Yet, Hyrachyus and Rhinocerotidae indeed share seven synapomorphies whose placement is unambiguous: hypocone and metacone of P4 separated (108), lingual cingulum of lower premolars usually absent (147), mesostyle of D2 present (165), lateral vertebral foramen of the atlas absent (185), indentation on the medial side of the magnum present (218), antero-distal groove of the tibia present (242) and latero-distal gutter of the fibula shallow (250).

"Hyracodontidae"
Seven unambiguous synapomorphies support the smallest clade including Amynodon and Eggysodon (Bremer support = 3): the mandibular symphysis is massive (42), the constriction of the metaloph is present on P2-4 (86), the metaloph of P2 is transverse (95), the metastyle of M1-2 is long (120), the hypolophid of lower molars is oblique (161), the ectolophid fold of d2-3 is absent (177) and the metapodials intermediate relief is low and smooth (280). According to our results, "Hyracodontidae" are basal to Amynodontidae and are clearly paraphyletic. The monophyly of Hyracodontidae has long been debated, but was hardly ever found by computed phylogenetic analyses [20,21], and it can in fact be considered as a wastebasket taxon [105]. Recently, this group has been redefined as the most basal clade of rhinocerotoids that includes Triplopus and Hyracodon [23], but we did not find this clade either. In our tree, Triplopus is found as sister-group of Proeggysodon and Hyracodon as sister-group of Paraceratheriidae (Bremer support value of 2). In our analysis, "Hyracodontidae" can be divided into three clades: Eggysodontidae, Paraceratheriidae and an unnamed clade. Eggysodontidae is the most basal of them and shows two dichotomies: Eggysodon-Allacerops and Proeggysodon-Triplopus. This clade is supported by three unambiguous synapomorphies: P3-4 protocone and hypocone connected by a lingual bridge (102), lower molars lingual cingulum always present (157) and proximal border of the anterior side of magnum concave (217). Within Eggysodontidae, Eggysodon and Allacerops share seven unambiguous synapomorphies and are supported by a Bremer value of 2.
The next most basal clade ("unnamed clade") includes Pappaceras confluens and two equivocal rhinocerotids: Uintaceras radinskyi and Teletaceras radinskyi. It is quite strongly supported by a Bremer value of 4. Teletaceras was classically considered as a primitive Rhinocerotidae while Uintaceras was often considered as the closest sister-group to this family [22,106,107]. Yet, according to our results they both branch outside of the Rhinocerotidae. Though it is not quite surprising for Uintaceras (a similar position far from Rhinocerotidae was also recently recovered [23]), it is more startling for Teletaceras, who shares with Rhinocerotidae the I1/i2 tusk complex, though it is less developed than in true rhinocerotids [107]. However, they do not share the absence of lower canine. Again, this may be an effect of our small taxonomic sampling of Rhinocerotidae. As previously explained, taxonomical results discussed here are tentative. Nonetheless, the union of Uintaceras and Teletaceras is strongly supported, with a Bremer support value of 5, and they share four unambiguous synapomorphies, including the reduction of lower canine (284). They both share six unambiguous synapomorphies with Pappaceras confluens: lateral apophysis of nasal present (1), P2-4 postfossette narrow (89), upper molars labial cingulum always present (109), lower molars labial cingulum reduced (150), p2 paralophid isolated (154) and lower molars labial cingulum usually present (159).
Furthermore, we did not find any close relationship between Pappaceras (P. confluens or "P." meiomenus) and Paraceratheriidae. We did find a monophyletic Paraceratheriidae clade with a Bremer support value of 3, including Paraceratherium and Urtinotherium, supported by three unambiguous synapomorphies: mandibular symphysis nearly horizontal (53), ectoloph and metaloph of M3 fused [= ectometaloph; (133)] and Cc 2 and Cc 3 facets of the astragal always independent (263). This clade was found as sister-group of Hyracodon nebraskensis, with whom it shares six unambiguous synapomorphies: sphenorbital foramen and foramen rotundum are fused (16), contrary to "P." meiomenus where they are "probably separated by a thin plate" [23], protocone and hypocone of P2 are separated (94), P3-4 protocone and hypocone are connected by a lingual bridge (102), upper molars antecrochet is usually present (110), upper molars lingual cingulum is always absent (114) (it is always present on "P." meiomenus) and lower canine is reduced (284) (even absent in Paraceratherium). It reveals that Forstercooperiinae and Paraceratheriidae (sensu [23]) may not be monophyletic: "P." meiomenus is closer to amynodonts than Paraceratheriidae, whereas P. confluens is closer to Uintaceras and Teletaceras than to Paraceratheriidae. However, our results must be compared with caution considering that the species used for our analysis differ from their analysis. A comprehensive revision of this group, which is beyond the scope of our study, is needed.
"Pappaceras" meiomenus was recently described as a forstercooperine and the "earliest known unequivocal rhinocerotoid" [23]. Interestingly, it was stated that Forstercooperiinae was the sister clade of Paraceratheriinae, and that both formed a clade (Paraceratheriidae), sister to the Amynodontidae. The Forstercooperiinae clade was united by two synapomorphies: nasal notch above canines and M1-2 metaloph short (S1 in [23]), but we did not find the monophyly of this clade with our matrix. However, based on their results (Fig 4 in [23]), the genus Pappaceras is not monophyletic, as it includes Forstercooperia totadentata. "P." meiomenus should rather be included in a new genus, or alternatively, Pappaceras should be a junior synonym of Forstercooperia. In any case, with an Arshantan age Pappaceras remains the earliest known unequivocal rhinocerotoid.
In contrast, our results suggest that "P." meiomenus could be considered as one of the first amynodontid. We have found indeed that it shares with them three unambiguous synapomorphies: the premolar/molar series is very short (length of P3-4 / length of M1-3 < 0.42, (63); it is here equal to 0.38), upper P1 is sometimes absent (91) and M3 metastyle is directed posterolabially (297). The close resemblance of "P." meiomenus with Rostriamynodon or other Amynodontidae was also previously noticed [23], as they share "a wide frontal between the orbital and followed by a flange overhanging the postorbital cavity", "incisors arranged in straight lines and weakly converging anteriorly", "ridge-like nuchal crest", "a relatively large canine, distinctly reduced P1, a short postcanine diastema, P4 transversely wide with high and strong protoloph, weak and short metaloph (metaconules), and M3 quadrate in outline" [23]. In particular, the reduction of P1 and the distinctly quadrate M3 in addition to its labially deflected metacone strongly suggest an amynodontid, and very much differ from other Pappaceras or Paraceratheridae. In Pappaceras confluens for example (Fig 1 in [69]), the upper P1 is always present and two-rooted (whereas if present, it is one-rooted in "P." meiomenus), a primitive state, and the ectoloph and metaloph of the triangular M3 are almost fused (i.e. the metacone is not lingually deflected). Finally, it also shares a "distinct preorbital fossa" [23], which is a characteristic of Amynodontidae, even though it may also be present in Forstercooperia or Paraceratherium. In Rostriamynodon as in "P." meiomenus the maxilla is excluded from the border of the nares, due to the broad extension of the premaxilla. The relatively large size of the canines is also characteristic of Amynodontidae, as well as the long metastyle of the M3.
The monophyly of Amynodontidae is retrieved and Rostriamynodon grangeri, Amynodon avenus and Sharamynodon mongoliensis are placed as early diverging amynodontids, but not Amynodontopsis, contrary to recent results [4]. The clade Amynodontidae is supported by a Bremer support value of 3. The monophyly of Amynodontidae (defined here as the smallest clade including Rostriamynodon and Cadurcodon) is supported by six synapomorphies whose placement is unambiguous: lacrimal process is absent (8), anterior base of the zygomatic process of the maxilla is high (10), rostral end of nasal bones is broad (24), P1 in adults is always absent (91), upper canine is strong (283) and upper molars parastyle is reduced (296).
Two clades previously defined [4] are identified: Metamynodontini and Cadurcodontini. The recent definition of Metamynodontini is the "stem-based taxon that includes Metamynodon planifrons and all amynodontids closer to it than to Cadurcodon ardynensis". Following this definition, Sellamynodon zimborensis is included in this clade, as well as Paramynodon and Megalamynodon. S. zimborensis is found as sister-group of Metamynodon planifrons, based on the sharing of one unambiguous synapomorphy: the external auditory pseudomeatus is closed (18). Megalamynodon, Metamynodon and Sellamynodon zimborensis are in a same clade and share three unambiguous synapomorphies: protocone and hypocone of P3 are separated (102), upper molars crista is usually absent (112) and upper postcanine diastema is short (291). Metamynodontini is supported by four unambiguous synapomorphies: zygomatic arch is high (11), postglenoid process of squamosal is flat (42), upper molars antecrochet is always absent (110) and the humerus distal articulation has a deep median constriction (diabolo-shaped; 194).
The other clade, Cadurcodontini, was recently defined as the "stem-based taxon that includes Cadurcodon ardynensis and all amynodontids closer to it than to Metamynodon planifrons" [4]. It is here supported by six unambiguous synapomorphies: the nasal notch is above P4-M1 (3), the postglenoid process of the squamosal is dihedral (42), the M1-2 paracone fold is weak (118), the olecranon fossa of the humerus is low (193), the talonid of m3 is longer than trigonid (288) and the diastema between upper incisors and canine is absent (290). Cadurcodon, Cadurcotherium and Zaisanamynodon are all monophyletic. Amynodon reedi Stock, 1939 was described based on a skull fragment with P4-M2, from the Late Eocene of California [48]. According to our phylogenetic results, it is not found close to Amynodon advenus (the type species of Amynodon) at all, but instead as sister-group of Zaisanamyodon. Indeed, it shares one unambiguous synapomorphy with this genus (upper molars lingual cingulum usually absent; 114) and only differs from it by one autapomorphy (the hypocone is anterior to metacone on P3-4; 103). Except in size difference, the morphology of the teeth is very similar to this genus, much more than to Amynodon. For instance, contrary to A. advenus, the lingual cingulum of P4 is complete, the M1 paracone fold is also poorly marked, and the M3 parastyle is more pronounced, as in Zaisanamynodon. The M1 metastyle is much shorter, as in Z. borisovi and Z. protheroi and the metaloph of P4 does not connect to the protocone as in Z. borisovi. Therefore, we suggest to assign this species to "Amynodon" reedi until its material is re-examined. Sianodon gaowangouensis is found as sister group to the clade including Amynodontopsis, Cadurcodon and Cadurcotherium. Here, within Cadurcodon, C. bahoensis is more basal than C. maomingensis, contrary to recent results [4]. The sister group of Cadurcodon includes Cadurcotherium, Amynodontopsis bodei and the specimens from Morlaca and Dorog (scored under one same taxon: Amynodontopsis aff. bodei), that are found in our phylogenetic analysis as sister-group of Amynodontopsis bodei. They are united with a Bremer support value of 2 and by four unambiguous synapomorphies: the anterior tip of the zygomatic process does not strongly deviate from the maxilla (37), the upper molars labial cingulum is always present (109), the external groove of lower cheek teeth is developed (140) and the M3 paracone fold is weak (289). Amynodontopsis aff. bodei has one autapomorphy: the M3 metastyle is directed posterolingually (297). Here, Amynodontopsis and Amynodontopsis aff. bodei are placed as sister-group of Cadurcotherium, with which they share four unambiguous synapomorphies: the rostral end of nasal bones is narrow (24), the mandibular ramus is inclined backward (60), the upper molars antecrochet is always absent (110) and the lower molars labial cingulum is always absent (159).

Phylogeny
From a phylogenetic point of view, and with the current taxonomic sampling, the monophyly of "Hyracodontidae" is seriously questioned. We confirm indeed that they could be divided in three monophyletic clades: Eggysodontidae, Paraceratheriidae and another unnamed clade including Pappaceras confluens and equivocal rhinocerotids like Uintaceras and Teletaceras. It is interesting to note that in these three clades, some forms show strong convergence with Rhinocerotidae. For instance, Teletaceras radinskyi possesses the I1/i2 chisel-tusk complex (sensu [108]) of rhinocerotids, though not associated to the loss of I3/i3 and canines [107], but its position in our tree may be an artifact of our taxonomic sampling and is thus tentative. Similarly, Paraceratherium shows the characteristic reduction of the anterior dentition by the loss of I3-C and i3-c, but without sharing the strictly I1/i2 complex, materialized instead by I1/i1 [88]. However, comparing our results with those of Wang et al. [23], we did not find the monophyletic clade including Hyracodon and Triplopus, that they named Hyracodontidae.
Furthermore, Wang et al. [23] recently described Pappaceras meiomenus as a forstercooperiine paraceratheriid, whereas our results suggest that it could be in fact the closest sister group of Amynodontidae. In any case, the authors had also reported the close resemblance of this form with Rostriamynodon, an early diverging Amynodontidae. Amynodontidae are without much doubt monophyletic and can be easily diagnosed by a number of synapomorphies (reduction of premolars row length compared to molars [also shared with P. meiomenus], loss of upper P1, M3 metacone labially deflected and strong canines). The association of these characters with a deep preorbital fossa is also possibly diagnostic. All these characters are very derived compared to those of "Hyracodontidae".
Here, our results clearly differ from the most recent and comprehensive phylogenetic study on Amynodontidae [4] by the position of the genus Amynodontopsis. We show that it is much more derived than previously recovered [4], as it is indeed sister-group of Cadurcotherium (Fig 8). Cadurcotherium probably shows the most derived characters in Amynodontidae with cemented, hypsodont and very laterally-compressed cheek teeth. Although Amynodontopsis clearly does not show the same degree of derived character states as Cadurcotherium, it notably shares with it a very similar development of the preorbital fossa [15,49]. This fossa, associated to very short nasal bones and retracted narial notch, is a strong evidence for the existence of a proboscis [109], such as found in Tapirus, which is used to manipulate food. The similarities of this facial structure between Cadurcodon and Amynodontopsis (which were grouped together in a group named "cadurcodonts") was already noted [109]. However, it was suggested that Cadurcotherium could belong to a different group [109], the metamynodonts, but its skull was unknown at this time. Contrary to Wall [2,109], and as previously suggested [15], Cadurcotherium belongs in fact to the same group as Cadurcodon and Amynodontopsis. It probably possessed a proboscis, like the other members of this group. Furthermore, the clade Cadurcotherium-Amynodontopsis shows a very similar development of the proboscis, as suggested by the reduction of nasal incision to the front of M1, the extension of the preorbital fossa medial to the orbit in dorsal view, a short diastema, and a reduced preorbital portion of the skull [15]. These characters indicate the presence of a well-developed proboscis, but not as developed as in Cadurcodon [15,109]. However Amynodontopsis differs from Cadurcotherium by its anterior dental formula: it is complete in Amynodontopsis (3I/3i) (though the incisors were very small), but reduced in Cadurcotherium cayluxi (2I/1i), which may indicate two different feeding habits, as also indicated by the differences in hypsodonty and cementation of cheek teeth. Amynodontopsis, with its complete incisors formula shows an identical condition to Tapirus for example.

Palaeobiogeography
From a European context of mammalian history, the record of Amynodontidae in the Eocene of South Alpine-Carpathian Europe, as opposed to their absence in the rest of Europe, allows to propose a new picture of mammalian interchanges in Eurasia and improves the understanding of their diversification during the Eocene-Oligocene transition. Due to high sea-level stand of the Middle and Late Eocene, Europe was disconnected from the main continental landmasses, producing its own endemic fauna [110]. Europe had become an archipelago in a warm, tropical sea, isolated from North America by the complete opening of the Atlantic Ocean, and from Asia by the Turgaï Strait, which connected Arctic Ocean to the Tethys Sea [111,112]. After the Eocene-Oligocene transition, a fall in sea-level related to the "Oi-1 glaciation" (ca. 32 Ma; [111]) had notably caused the drying-up of the Turgaï Strait, thereby establishing a land connection with Asia. The effects in Western Europe are assumed to be extinctions and dispersal-generated originations of the Grande Coupure [17,113,114]. Besides, numerous authors have suggested that climate change due to the Oi-1 glaciation combined with competition produced this faunal turnover (e.g. [18,100,115]). The latter marked a change from the endemic European faunas to ones with major components of Asian origin, such as Rhinocerotoidea (e.g. [18,100,114,[116][117][118]). The scarce records of Eocene European Rhinocerotoids were limited to the Early-Middle Eocene cosmopolitan Hyrachyus [90,119], and two Amynodontidae, assumed to be Late Eocene, Amynodon hungaricus (Tápiószele?, Hungary; [12]) and Cadurcodon ardynensis (Nikolaevo and Kameno, Bulgaria; [11]). Concerning Hyrachyus, its presence in Europe seems to be related to the Early Eocene palaeogeographical context during which Europe was still connected to North-America by the DeGeer Route [120,121]. Moreover, the inclusion of this genus within Rhinocerotoidea is still debated [23,122], though according to the results of our phylogenetic analysis, Hyrachyus could be considered as sister group of Rhinocerotidae (Fig 8). Consequently, the report here of Amynodontidae in the Eocene of Hungary and Romania is truly exceptional, considering the scarcity of Eocene rhinocerotoids in Europe. The discovery of Sellamynodon zimborensis in Dobârca (Romania) is also of great importance for the study of European faunas in the context of the Grande Coupure event. It represents indeed the only European species of the Metamynodontini tribe, which was previously exclusively restricted to North America (Megalamynodon and Metamynodon) and Asia (Paramynodon birmanicus, in Myanmar). However, the age of S. zimborensis still needs to be clarified: it could either be Late Eocene or Early Oligocene. If the age of Dobârca is Early Oligocene, then this new taxon represents the only occurrence at that time in Europe of an amynodontid other than the well-known Cadurcotherium, whose European distribution is restricted to France (e.g. [14,15]) and Switzerland [16] during the Early Oligocene, but extended to Spain [123] and Bosnia and Herzegovina [124] during the Late Oligocene. It is also known in Pakistan during the entire Oligocene [9]. Otherwise, if the age of the Dobârca locality is Late Eocene, then it documents a new early occurrence of Amynodontidae in Europe, which means that they were regionally more diversified than what was previously suggested by the fossil record. Furthermore, with the report of Amynodontopsis aff. bodei in Dorog (late Middle Eocene, Hungary) and in Morlaca (Late Eocene, Romania), we document the first record of the widespread Amynodontopsis out of Asia and North America, and also  Table 4 for symbols and colours explanations as well as localities and occurrences references. Palaeogeographical map reconstructions during the Late Eocene (C) and Early Oligocene (D) modified from Ron Blakey (jan.ucc.nau.edu/rcb7/ mollglobe.html).
https://doi.org/10.1371/journal.pone.0193774.g009 Based on these new data and according to the distribution maps of Amynodontidae from the late Middle Eocene to Oligocene in Eurasia (Fig 9, Table 4), their presence is evidenced in South Alpine-Carpathian Europe before any other representative of the family is known in Western Europe.
The palaeogeographic setting explains this situation, because from the late Middle Eocene to the earliest Oligocene (i.e. to MP20) Western Europe was isolated from Northern Europe by the North Sea and Northern Europe from Asia by the Turgaï Strait [16,18,112,[139][140][141]. Regarding South Alpine-Carpathian Europe, it was mostly disconnected from the rest of Europe by the Perialpine and Paratethys seas-they may have been in connection at some point [142,143]-but connected to Asia [112,140,141]. Perissodactyl faunas in this part of Europe seem more similar to those found in Asia (e.g. the reported amynodontid Amynodontopsis, as well as Amynodon hungaricus [12] and Cadurcodon ardynensis [11], but also Brontotheridae such as Brachydiastematherium transylvanicum in Hungary [144] or Sivatitanops in Bulgaria [11], and "Hyracodontidae" like Prohyracodon orientale [31,145] than to those found in North and Western Europe [11]. Excluding a few sporadic and enigmatic reports [11,142], typical late Eocene Western European taxa such as Palaeotherium or Plagiolophus are indeed absent from South Alpine-Carpathian Europe. The extension phase of the distribution of Amynodontidae to Western Europe occurred only after the Eocene-Oligocene transition ( Fig  9B). However, this new distribution is strictly limited to the genus Cadurcotherium, unknown before the Oligocene. The occurrence of this genus in Western Europe, along with Epiaceratherium, Eggysodon or Ronzotherium (e.g. [16,38,116,146]), represents the typical record of the post-Grande Coupure fauna (sensu [17]). The dispersal routes used by these new immigrants in Western Europe through North Alpine (due to the drying up of the North and Turgaï Seas) and/or South Alpine dispersal routes (due to the closure of the Paratethys and Perialpine seas)-is still debated [16,142,147]. Considering that none of the typically Eastern-European Eocene perissodactyls has ever been found in Western Europe, even after the Grande Coupure, they should not be regarded as "precursors" of the post-Grande Coupure faunas, but rather as representatives of a palaeobiogeographical extension in Eastern Europe of North Americano-Asiatic faunas during the Late Eocene. These observations suggest that, at least for rhinocerotoids, an arrival at the very beginning of the Oligocene through the northern route is more likely, but should be quantitatively investigated. A similar case is evidenced for the family Anthracotheriidae (Artiodactyla) whose early dispersal in the Balkan area and the Italian peninsula, during the Late Eocene, is recorded by taxa that differ from the newly arrived taxa in Western Europe after the Grande Coupure [148,149], as well as for ruminants, for which a South-Alpine route is suggested during the "Bachitherium Event" [150].
Supporting information S1 File. Morphological data matrix used for the phylogenetic analysis in NEXUS format and TNT format. (ZIP) S2 File. List of synapomorphies retrieved during the phylogenetic analysis with PAUP, in Excel format. Changes in blue with double-arrows are unambiguous. Number of steps and CI of characters are also included. Numbers on the tree are node numbers. Refer to the included tree for node numbers and to the morphological data matrix for characters description (S1 File). (XLSX)