Bovidae (Mammalia) from the early Middle Miocene of Maboko, Kenya

ABSTRACT We describe here the Bovidae from the early Middle Miocene site of Maboko, Kenya, known for its rich and highly diverse assemblage of Primates. They are attributed to two taxa of slightly different sizes, Turcocerus africanus n.sp. and Kubanotragus pickfordi. In addition to Kenya, Turcocerus ranges through the Middle Miocene from Turkey to China; it might have originated in Africa but the lack of precise dating weakens any phylogeographic hypothesis. Kubanotragus has similar chronological and geographic ranges. Both Maboko species retain low levels of hypsodonty compared to their northern relatives and may have been less specialised for grazing than Eurasian taxa. The Maboko Bovidae illustrate an early diversification of the family in Africa that appears to pre-date the major diversification of Bovidae and appearance of the diverse clades that are characteristic of modern African pecoran faunas.


Introduction
Middle Miocene bovids from Maboko Island, Kenya, are the oldest radiometrically dated Bovidae currently known, although they are certainly younger than the earliest Eurasian representatives of the family (Ginsburg and Heintz 1968;Solounias et al. 1995;Ginsburg et al. 2001;Li et al. 2021) and post-date molecular estimates for the origin of the family by at least 7 million years (Chen et al. 2019).Together with biostratigraphically dated fossils from Jebel Zelten, Libya (Hamilton 1973), and Al-Sarrar and Ad Dabtiyah, Saudi Arabia (Thomas et al. 1982;Gentry 1987), they document the early diversification of Afro-Arabian bovids, soon after their appearance at Arrisdrift, Namibia (Morales et al. 2003).

Geological and historical context
Historically, Maboko bovids were first discovered by Archdeacon Owen during the early 1930s and sold to the British Museum of Natural History (Cole 1975;Andrews 1981;Andrews et al. 1981).Precisely where on the island Owen collected his fossils is unknown.The excavation and mapping in place of 287 bovid fossils from intact Miocene sediments in three stratigraphic layers at Maboko Main by two of us (MM and BB) from 1987 to 1997 (Benefit and McCrossin 1989;Geraads et al. 2012;Arney et al. 2022; Figure 1; Suppl.Data 1 Figures 2-13), allows us to make suggestions about the provenience of Owen's fossils.The general completeness, limited distortion and light colour of the London Natural History Museum (NHMUK) specimens are consistent with fossils excavated from undisturbed Miocene strata at both the oldest fossiliferous stratum in the Maboko sequence Bed 3 (a yellow sandy clay) and Bed 5 white clay (5 w) which occurs 2 metres higher in the stratigraphic column at the locality of Maboko Main (Suppl.Data 1 Figs.[16][17][18].Horncore NHMUK PV M 15543 in Owen's collection has the same morphology as two of the most complete horncores excavated from intact Miocene yellow/green sand Bed 3 (KNM [Kenya National Museum]-MB 21849 and 35021), and horncore NHMUK PV M 15544 closely resembles fairly complete horncores from both Bed 3 (KNM-MB 24892) and Bed 5 w (KNM-MB 21504).Provenience information for bovid fossils collected by Pickford between 1982 and1984 and by BB and MM follow Maboko locality and stratum designations established in the Kenya Palaeontology Gazetteer (Pickford 1984), with the additional distinction between upper (white clay) and lower (brown clay) levels of Bed 5 (Benefit 1999;Retallack et al. 2002).Since 1982, 40 bovid specimens were collected from the surface of Maboko Main including from topsoil overlying intact Miocene sediments, and smaller collections come from Bed 6 surface exposures at the Maboko Cliffs locality immediately east of Maboko Main (n = 1; Suppl.Data 1 Fig. 17); Bed 12 at Maboko Southeast (n = 7); and Beds 15-16 at Maboko Cliffs East (n = 1; Suppl.Data 1 Fig. 18).Incomplete locality information in the accession catalogue of the National Museums of Kenya is available for 37 additional bovid specimens collected by Pickford, and for the 6 Maboko bovid specimens collected between 1932 and 1982 by MacInnes (1933-34), Leakey (1949Leakey ( -1951) ) and Andrews and Pilbeam (1973).These expeditions are known to have focused their fossil collecting efforts at Maboko Main (Andrews et al. 1981), but it is not clear whether surface finds from other strata are included in the sample.Based on its morphology which differs from specimens excavated from Beds 3-5 at Maboko Main (see below), we suspect that at least one specimen, KNM-MB 548 (field number 1934) collected by MacInnes, may derive from a higher stratum .
Full provenience information is given in Suppl.Data 2A for the 350 Maboko bovid fossils that belong to the collections of the National Museums of Kenya.This is especially important for the excavated material because the information is only recorded in field catalogues and on accession labels but not in the accession catalogue of the KNM.
Radiometric dating of the Bed 8 tuff at Maboko Cliffs yields a 14.71 ± 0.16 Ma upper limit for the age of Bed 3-6 fossils (Feibel and Brown 1991), but no absolute dates are yet obtainable for the base of the Maboko sequence.The presence of kenyapithecine apes, Victoriapithecus monkeys, choerolophodont gomphotheres, Climacoceras giraffoids, and listriodont suids in the lowest fossiliferous stratum at Maboko (Bed 3) suggests that the fauna is biostratigraphically more recent than c.17 Ma Early Miocene deposits at Rusinga where these taxa are absent (Pickford 1981;Andrews et al. 1981;Peppe et al. 2009).Maboko Main is usually considered to be closer to 15 Ma and placed in Faunal Set III of the East African Faunal Zonation (Pickford 1981), but could theoretically be as old as 16 Ma.Bovids from Bed 12-16 are sandwiched between the Bed 8 Tuff and a phonolite, which caps the Maboko Formation.Laser fusion 40 Ar/ 39 Ar dating of alkali feldspars of the Ombo phonolite at Maboko yields a youngest age of 13.8 ± 0.04 Ma for fossils from Beds 9-20 (Feibel and Brown 1991), the latter being the highest stratigraphic layer on the island (Pickford 1984).Although Maboko Bed 12 bovids are closer in age to the 13.7 ± 0.3 Ma bovids from Fort Ternan (Gentry 1970;Pickford et al. 2006), the presence of Climacoceras africanus rather than Climacoceras gentryi in Beds 12-16 indicates their fauna is less derived and somewhat older than Faunal Set IV at Fort Ternan.
Nearly two-thirds of the excavated bovid fossils derive from Bed 3 sediment (N = 208; 5.9% of the 3,510 identifiable mammal fossils from this stratum).The small number of Bed 3 bovid individuals (minimum number of individuals [MNI] = 8 based on the number of left scaphoids, or 10 based on half of horncore sample) is represented by at least 29 different postcranial elements (Suppl.Data 2A), indicating that a small number of disarticulated skeletons are preserved in the stratum.This is consistent with other taxonomic groups from Bed 3 including kenyapithecine apes (McCrossin and Benefit 1993, 1994, 1997;McCrossin et al. 1998) and rhinocerotids (Geraads et al. 2012).The presence of ooids and oncoids within Bed 3 indicate that the fossils accumulated within a moderately low energy beach ramp or littoral environment associated with a freshwater lake of unknown depth and size (Watkins 2004).Bed 3 fossils tend to be well preserved and undistorted.The 49 bovid fossils (1.4% of the 3,494 Bed 5b fossils; bovid MNI = 3 based on left m3s) excavated from the 90 cm thick Bed 5 brown clay (Bed 5b) overlying the Bed 4 limestone at Maboko Main are less well preserved with tooth enamel that is often corroded.Bed 5b has been interpreted as representing a riparian woodland or localised wetland forest because of its association with a Dhero paleosol (Retallack et al. 2002).Bed 5b mudstone is replaced a metre higher in the sequence by Bed 5 white clay which is associated with slicken slides and Yom paleosol (Arney et al. 2022: Fig. 2) the latter of which is interpreted as a moist wooded grassland laid down during a time when rainfall was seasonal (Wynn & Retallack 2001;Retallack et al. 2002).This interpretation is supported by the discovery of a fossil grass seed (Suppl.Data 1 Fig. 19; B. Jacobs personal communication) within Bed 5 white clay matrix attached to excavated Victoriapithecus partial mandible KNM-MB 18736.At present the seed has not been identified to a specific taxon and we do not know if it is a C 3 or C 4 grass.Although herbivore enamel isotopic values do not indicate the presence of C 4 grass (Arney et al. 2022), its abundance in the diets of Maboko herbivores may be too low to detect.Of the 15 bird taxa identified in Bed 5 w, 10 are aquatic (cormorants, bitterns, pelicans, storks, flamingos, herons, lily-trotters, ducks, gulls and river kingfishers) indicating the presence of lakes, rivers, marshes or swamps (Retallack et al. 2002;Mayr 2014a); one is a hoatzin typically associated with flooded forest (Mayr 2014b); and three others (bustards, hornbills, and houbaras) suggest proximity to open woodland or savanna (Retallack et al. 2002;Mayr 2014a,b;McCrossin and Benefit 2016).The presence of the gastropod Limicolaria in the stratum reflects a seasonal pattern of rainfall during the time it was deposited (Pickford 1995;Retallack et al. 2002).The 30 bovid fossils excavated from Bed 5 w (1.8% of the 1,645 mammals identified; bovid MNI = 3 based on right P4s) include well-preserved horncores, jaws, and postcranial remains.

Material and methods
Most of the material is housed in the National Museums of Kenya, Nairobi (KNM), but a few other specimens were also examined in the Natural History Museum, London (NHMUK).Comparative material is housed in the Muséum national d'Histoire Naturelle, Paris (MNHN).Dental nomenclature follows Geraads et al. (2013).Uppercase denotes upper teeth, lowercase denotes lower teeth.

Subfamily indet
Genus Turcocerus Köhler, 1987 Type-species.Oioceros (?) grangeri Pilgrim, 1934 Diagnosis Bovidae of small size; horncores rather short, inserted directly above the orbits, moderately inclined backwards, slightly convergent in their basal parts but diverging higher up, with a weak to moderate homonymous torsion, slightly to moderately compressed transversely, their base somewhat swollen and thicker than the pedicle, which is long.Skull with weak cranio-facial flexion, broad across the orbits, which are large and deep.

Referred material
Other specimens were excavated from Bed 5 w sediments three metres higher in the sequence at the same locality.Teeth from beds  5b, 6 and 12 are tentatively attributed to this taxon (Suppl.Data 2A).

Etymology
From Africa, because the Maboko species is the first found in this continent.

Diagnosis
A species of Turcocerus of small size, with horncores distinctly compressed transversely, long premolars with transverse cristids, and brachyodont molars, with weak goat-folds and convex lingual walls.Differs from T. grangeri in its much smaller size, compressed horncores, longer premolars, less aegodont teeth; from T. gracilis in its more compressed horncores, longer premolars, more brachyodont molars; from T. halamagaiensis in its larger size, longer horncores, presence of torsion, and greater curvature; from T. kekemaidengensis in its smaller size, lesser horncore compression, shorter premolars.

Horncores
The holotype horncore KNM-MB 24892 (Figure 2A) is rather short (~8 cm long) and almost straight, and conical rather than cylindrical, as its antero-posterior diameter quickly decreases upwards; the basal part looks somewhat inflated in medial or lateral view.It is uprightly inserted, directly above the orbit.In medial view, its posterior border makes an angle of about 80° with the frontoparietal plane, its anterior border an angle of about 65°.In front view, the horncore first approaches its counterpart for about 5 cm, then diverges from it, so that the upper half of the horncores were slightly divergent.The horncore also has a slight forward curve, and a slight anticlockwise torsion (so-called homonymous, because the right horncore describes a dextral spiral).The basal crosssection is oval, with a main axis slightly oblique in respect to the sagittal plane and distinctly compressed transversely (30 × 22.1; Suppl.Data 2 F), with stronger curvatures of the cross-section outline anteriorly and posteriorly; these almost become keels higher up.The pedicle is rather long relative to the length of the horncore proper.This is the only specimen that unambiguously includes the mid-frontal suture; it shows that the frontal is rather flat between the horncores, and that the orbit extends far medially, its medial wall being at most 10 mm from the sagittal plane, if not closer still.There is no postcornual fossa.
Most other specimens resemble KNM-MB 24892.KNM-MB 21504 (Figure 2B) is ~9 cm long, has no hint of keels, is more inclined backwards than the holotype, and its pedicle is longer.A posterior longitudinal groove occurs between two well-demarcated ridges, similar to that of T. halamagaiensis (see Li et al. 2021, fig.2) but appears to be more medially positioned on the African specimen.There is an inconspicuous, shallow postcornual fossa.The supra-orbital foramen is small, located at the surface of the frontal bone, medially to the ridge that extends in front of the horncore.The orbit is as large as that of KNM-MB 24892.KNM-MB 27599 (Figure 3) has a distinct torsion and elongated conical shape; the section is compressed but without keels.The horncore is deeply grooved with the same medially positioned posterior groove as seen in KNM-MB 24892.Its base is much thicker than the pedicle, which is very long.KNM-MB 27603 is very stout, distinctly inclined backwards, and has a quickly decreasing anteroposterior diameter; the pedicle is short, and not much less thick than the horncore.KNM-MB 20347 cannot be precisely oriented but has an oval crosssection and is compressed transversally.KNM-MB 29562 is a small horncore, rather strongly inclined backwards; it might be from a juvenile individual, but the moderate compression and lack of keels differ from the tip of adult horncores, in contrast to what would be expected in this case.KNM-MB 33211 much resembles the holotype; it shows additionally that the frontal bone is hollowed by a small sinus anteriorly.A few other, incomplete specimens are of the same type.On average, horncores excavated from Bed 3 are larger than those from Bed 5 w, although specimens from both strata overlap in size (Suppl.Data 2 F-G).
We also assign to this species NHMUK PV M 15544 (Suppl.Data 4; Gentry 2010, fig. 38.8) that includes only the dorsal part of the braincase and the incomplete right horncore.Behind the frontals, on each side of the parietal, a pair of temporal lines first converge towards the midline and then run more and more parallel to each other backwards; however, the distance between the two lines of each pair slightly increases posteriorly, so that the more medial lines meet before reaching the (missing) occiput, while the lateral ones do not.The horncore is obliquely inserted above the orbit, and slightly divergent from its counterpart.It is compressed transversely, was certainly quite short, and is not curved backwards, as the outline of its anterior and posterior borders suggest instead that its tip might have been slightly curved forwards.Thus, it resembles the previous specimens except for its absence of torsion or increase of divergence upwards.However, these differences might well result from a slight distortion of the fossil (the frontal and parietal bones are heavily cracked) and we regard them as insufficient for species distinction.Thomas (1979) assigned this specimen to an Eotragus close to E. sansaniensis (recte: E. clavatus), but Gentry (2010) tentatively identified it as Homoiodorcas, a genus based upon H. tugenium Thomas, 1981, whose holotype horncore, from the pre-hipparion levels of Ngorora, is strongly curved and inclined backwards, and does not show the quick decrease of antero-posterior diameter of NHMUK PV M 15544.We therefore reject Gentry's identification.

Dentognathic material
Although horncores are not found in direct association with dentognathic material at Maboko, we feel confident assigning mandibles and teeth from Bed 5 w to T. africanus because only horncores of this taxon are known from this stratum.In turn, specimens from Beds 3, 5b, 6 and 12 with similar molar dimensions (Suppl.Data 2B-E) and morphology are attributed to this species.At Maboko Main, horncores of T. africanus are the most common, as are the teeth attributed here to the same species.Since no taxonomically diagnostic horncores are known from Beds 5b, 6 or 12, attribution of teeth from these beds to T. africanus remains tentative.
The Bed 5 w T. africanus sample includes two of the most complete bovid mandibles known from Maboko.They are of similar size, and differ only in subtle details of dental morphology.KNM-MB 21459 (Figure 4B) with left p4-m2, preserves the symphysis to corpus below m2.Anteriorly, the part of the corpus between p2 and canine root is mesio-distally long (33 mm; 143% as long as the premolar series) and dorso-ventrally shallow (9.4 mm; 43% as deep as the ramus below m2).The mental foramen is long (7.5 mm).The corpus deepens posteriorly below p4 (6.4 mm), m1 (18.3 mm) and m2 (21.9 mm), making it different from Fort Ternan specimens assigned initially to 'Gazella' sp.(Gentry 1970: plate 15 KNM-FT 64361), but similar to the larger 'Oioceros' tanyceras in shape (Gentry 1970: plate 9) from the same site.The premolar row of KNM-MB 21459 (c.22.4 mm) is also long relative to estimated molar row length (35 mm), its size and relative dimensions being close to those observed in KNM-MB 20350 and 9204 (see below; Suppl.Data 2D).Dimensions of KNM-MB 25148 from Bed 12, which we tentatively assign to K. pickfordi (see below) are larger.Premolar row length is relatively shorter (at least slightly) in other species attributed to Turcocerus and in other bovids from Fort Ternan.KNM-MB 21459 is also similar to the larger T. grangeri (Pilgrim 1934: fig. 7) in having a corpus that deepens posteriorly and is long and supero-inferiorly shallow anterior to p2.The p4 of KNM-MB 21459 is minimally worn but its anterior border appears to be damaged, obscuring details of the anteparacristid, which is visible from the lingual perspective where it is separated from the paraconid.The p4ʹs metaconid is tall with a steep anterior wall that extends close to the cervix, and is equal in height and crown area to the protoconid.The lingual side of the protoconid does not expand into the large mesosinusid gap between paraconid and metaconid.The metasinusid is narrow between the slightly oblique transverse cristid and more oblique entocristid.The telosinusid is enclosed by the intersection of lingual ends of the entocristid and hypocristid.The entocristid is shorter than the metaconid and of equal height to the hypocristid which is more expanded than the former.The slight to moderately worn m1 and very slightly worn m2 are brachyodont with a hypsodonty index (H/L) of 0.7 for m1 and 0.75 for m2.Buccally, the m1 and m2 of KNM-MB 21459 have well-developed basal pillars (ectostylids), somewhat wrinkled buccal enamel, and a hint of a goat fold.Lingually, the crowns are slightly convex.An unworn right m1 in mandible fragment, KNM-MB 7882 (Figure 4D) is similar in having a basal pillar, buccalward expansion of the hypoconid, and similar buccal and lingual contours as KNM-MB 21459.Its lingual cuspids rise high (~3 mm) above crown height at intersection of the postmetacristid and preentocristid (5 mm).
KNM-MB 20350 from Bed 5 w (Figure 4A) is a left mandible preserving m1-m3, premolar alveoli, the corpus between the mental foramen and the back of m3.Only the upper half of the corpus below m2-3 is intact.The specimen is minimally larger than KNM-MB 21459 with a similarly long premolar relative to molar row length (Suppl.Data 2D).The corpus is shallow anteriorly (9 mm) but deepens below m1 (21 mm).Molars of KNM-MB 20350 differ from KNM-MB 21459 in being somewhat narrower bucco-lingually relative to length (Suppl.Data 2C; L/W = 1.52 for m1 and 2) compared to KNM-MB 21459 (L/W = 1.41 for m1 and 1.42 for m2); having buccal cusps that are more columnar and vertically oriented with slightly less basal expansion in width compared to KNM-MB 21459; absence of m1 and m2 basal pillars though present on m3; smoother enamel buccally; and in being slightly more hypsodont (m2 H/L = 0.84; m3 H/W = 1.26).We interpret these differences as reflecting intraspecific variation.The relatively small m3 hypoconulid is positioned on the buccal side and separated from the entoconid by a strong entostylid.In occlusal view, the hypoconulid is C-shaped, with a lingual concavity forming a vertical furrow, partly blocked antero-lingually by the entostylid, a morphology also found on isolated m3 KNM-MB 20349.A similar m3 hypoconulid configuration is seen in several Miocene taxa including T. grangeri (Pilgrim 1934: fig.7) and Fort Ternan 'Gazella' (Gentry 1970: plate 16-3) but not 'O'.tanyceras (Gentry 1970: plate 9-3).
KNM-MB 9204 from Bed 6 at Maboko Cliffs (East) is tentatively assigned to T. africanus based on its size and morphology.It is the most complete of the T. africanus dentognathic specimen (Figure 4C) with both a right partial maxilla with P4-M3 and an associated partial left mandible with p3-m3.Only the upper half of the corpus is preserved.Upper and lower first molars are heavily worn but third molars show moderate wear.The KNM-MB 9204 P4 is about as bucco-lingually broad as M1, and has a flat lingual surface between robust styles.Premolars are rather large relative to the molars, with an estimated lower Pm/M index of c. 66.The upper molars have rather strong styles and a flat metacone, but a distinctly inflated paracone buccal wall.Mandibular premolar to molar lengths ratio is close to those of KNM-MB 20350 and 21,459 (Suppl.Data 2D).Lower third and fourth premolars of KNM-MB 9204 (Figure 4C) are not very different from each other, with distinct paracristid, anteparacristid and entocristid.They differ from the p4 of KNM-MB 21459 in having an entocristid and hypocristulid that are unfused lingually leading to lingual opening of the telosinusid, as well as in having an unexpanded metaconid.From occlusal and lingual views, it is clear that the lingual side of the p4 protoconid cusp is inflated and extends into the mesosinusid, a feature almost absent in KNM-MB 21459.Lower molars of KNM-MB 9204 have convex lingual walls, almost no metastylid, a minute ectostylid, basal pillars, and only a hint of a goat fold; the metaconid connects the hypoconid in early wear; the m3 talonid consists mostly of a conical hypoconulid showing only a hint of the lingual concavity of KNM-MB 20350 and 20349.
The mandibular molars of T. africanus are brachyodont with hypsodonty indices ranging from 0.75-1.04based on m1-m2 metaconid height/crown length (Hall and Cote 2021 and our own measurements) and from 1.06 to 1.26 based on m3 metaconid height/crown width, with goat folds that are at most incipient and slightly convex lingual walls.Isolated molars attributed to T. africanus are listed in Suppl.Data 2.
Horncores similar to those of Maboko were first described by Pilgrim (1934) from the Middle Miocene (c. 14 Ma) of Tunggur (China), as 'Oioceros (?) grangeri' and 'O.(?) noverca'.These species differ mostly in size and have short, stout horncores that resemble those of Turcocerus africanus, but are more circular in cross section.Associated teeth are distinctly high-crowned (especially so for Middle Miocene forms), the premolars are short (Suppl.Data 2B, D), upper teeth have flat labial walls and prominent styles, lower molars have a goat fold and flat lingual walls.Köhler (1987) erected the genus name Turcocerus for O.(?) grangeri.
From the Middle Miocene site of Çandır in Turkey, Köhler (1987), followed by Geraads (2003), described a few horncores similar to those of Turcocerus africanus.She called them T. gracilis.These horncores are less compressed than those from Maboko (Suppl.Data 2 G), and we regard this as a valid difference at species level.In addition, at Çandır the premolar row is much shorter than the molar row and the molars are distinctly hypsodont, as at Tunggur.
The horncores of Turcocerus halamagaiensis (Ye, 1989) from the Middle Miocene of Botamoyin, China, whose teeth are also hypsodont, do not much differ from the Çandır ones (Li et al. 2021a).The species is also present in the Tabenbuluk area at Xishuigou, a site said to be of lower Miocene age (but see below).The horncores of Turcocerus kekemaidengensis (Ye et al. 1999) from the late Middle Miocene are also similar, but less compressed (Suppl.Data 2 G; Dmitrieva 2007;Li et al. 2021a).Li et al. (2021a, table 3) stated that these species differ from the Middle Miocene Eurasian genus Eotragus Pilgrim, 1939, in the presence of homonymous torsion, but the difference is far from clear, as torsion is absent in T. halamagaiensis (Ye 1989, pl. 1;Li et al. 2021a: fig.2), while E. clavatus (Gervais, 1850) from Sansan, France, the type-locality of MN 6 biostratigraphic zone (Made 2012), may rarely show a hint of incipient torsion (MNHN-Sa10799).Other similar horncores have been described from Israel (Tchernov et al. 1987) and Thailand (Suraprasit et al. 2015).Pseudoeotragus seegrabensis Made, 1989 is similar to Eotragus clavatus, but the single horncore is more slender and the dentition more hypsodont.
Eotragus clavatus from Sansan (MNHN collections) further shares with Turcocerus africanus a skull that is broad across the orbits (as deduced from the frontal bone of the latter species), salient orbital rims, a weak frontal flexure, temporal lines with an identical pattern, horncores inserted directly above the orbits, moderately inclined backwards, without backward curvature or torsion but with incipient outward and forward curvature.Most horncores from Sansan have no keel, but they may be present anteriorly (MNHN-Sa2494) or posteriorly (MNHN-Sa10804), so that the difference with Maboko is not great.Other differences are that E. clavatus is slightly smaller on the average, and that a sinus is probably always present in the frontal, as best shown by MNHN-Sa1183 and MNHN-Sa10799.Thus, although they are at similar evolutionary grades, E. clavatus shows a mixture of primitive and derived features compared to T. africanus; perhaps the clearest derived feature of the latter is the slight torsion.Eotragus clavatus horncores are smaller on average than those of T. africanus, but its mandibular premolar and molar row lengths are larger than T. africanus.
Earlier forms of Eotragus, E. noyei Solounias et al., 1995, and E. minus;Ginsburg et al., 2001, both from Pakistan, as well as Namacerus gariepensis Morales et al., 2003, from the early Middle Miocene of Namibia, have smaller and shorter horns (Suppl.Data 2 F-G) and the latter smaller teeth than T. africanus.The braincase of N. gariepensis resembles that of NHMUK PV M 15544 in the course of the temporal lines.

Diagnosis
A genus of small Bovidae with protruding large orbits, temporal lines split into an upper and a lower branch, horncores inserted vertically above the middle of the orbits, long and slender, remarkably straight, almost parallel, with slight to moderate transverse compression, and with at most an incipient homonymous torsion.

Type-locality
Maboko; the specimen was purchased from Archdeacon Owen, and likely came from Bed 3, the only stratum in which similar horncores are found.

Diagnosis
Braincase rounded, basioccipital lacking paired tuberosities, but with a central ridge flanked by depressions on either side.Tentatively associated teeth are larger than would be expected based on horn core size.Cheek-teeth brachyodont, but less so than those of T. africanus.Premolars absolutely and relatively about as long as those of E. clavatus.Lower premolars with oblique talonid cristids, and metastylid restricted to the top of the crown in the molars.Differs from K. sokolovi in its slightly larger size, longer horncores, absence of outward curvature; differs from K. gaopoensis in the transverse compression of the horncores, shorter premolars, and less hypsodont molars.

Horncores
The holotype was described by Thomas (1984a) as a member of the Giraffoidea, but as previously noted by McCrossin et al. (1998: p. 379), Morales et al. (2003) and Gentry (2010), it is in fact bovid as a clear distinction can be observed between the pedicle and the horncore proper.Besides this, not much can be added to Thomas' description.The horncores are slightly divergent in front view, and the partly preserved left one is virtually straight and apparently lacks torsion but close examination shows that it is partly reconstructed.The temporal lines are identical to those of NHMUK PV M 15544, assigned above to Turcocerus africanus.
The almost complete right horncore KNM-MB 21849 (Figure 5A) resembles the holotype in being uprightly inserted above the orbit, slightly compressed transversely, and long and slender; it fully lacks torsion.The length of the broken horncore is 12 cm, but is estimated to have been close to 20 cm long by projecting its anterior and posterior profile vertically.It differs in being slightly larger, and in that the frontal is narrower between the orbits and between the horncores, so that the orbits certainly approached each other more closely near the midline.KNM-MB 35021 (Figure 5B) is still more complete, and shares the same characters including great length (~16.2 cm).
These specimens resemble Turcocerus africanus in being of similar size, and in having a long pedicle, a faint postcornual fossa, a flat frontal, a large, deep orbit, paired temporal lines of similar course, and horncores that are slightly divergent upwards.They differ in having a less transversely compressed basal crosssection fully lacking keels, in being only slightly inclined backwards (raising almost vertically), in the absence of torsion, and in being long, slender, and almost cylindrical rather than conical (the diameters only slowly decrease upwards).Thomas (1984a) assigned mandible KNM-NC 7816 from Nyakach to N. pickfordi, a specimen that might seem too large to go with horncore KNM-NC 7973 of this species.A few molars of similar size to those of KNM-NC 7816 were excavated from Maboko Bed 3 (Suppl.Data 2A).As for the Nyakach specimens, these Bed 3 molars seem too large to belong to the horncores of K. pickfordi recovered from that stratum; they are larger than those of Turcocerus.In spite of this inconsistency, we tentatively assign these isolated molars to K. pickfordi because only two bovid taxa appear to be present in Bed 3 based on horncores.Since these Bed 3 teeth are similar in size and morphology to those found in partial mandible KNM-MB 25148 collected from the surface of Bed 12 at Maboko Southeast, we tentatively assign the latter specimen to K. pickfordi although no diagnostic horncores can confirm the presence of the species in this stratum.

Dental remains
KNM-MB 25148 from Bed 12 (Figure 4G) is a left partial mandible with p3-m2 and part of m3.Its molar row length is larger by 8 mm than any other Maboko specimen (Suppl.Data 2), as is its estimated premolar length, falling close to E. clavatus, but several isolated lower molars from Maboko, such as KNM-MB 29588, KNM-MB 24041, and KNM-MB 21912, are only slightly smaller, and we assign them all to K. pickfordi.Although worn and weathered, the p4 of KNM-MB 25148 is more elongated and complex anteriorly than those of T. africanus specimens KNM-MB 21459 and 9204.The paraconid is expanded, partly filling the mesosinusid.The premolar talonid cristids are oblique and close off the telosinusid as in KNM-MB 21459, with a worn metaconid that appears to be smaller than the protoconid.On molars the buccal sinusid is deep and the entoconid remains isolated from the hypoconid until late wear; the connection occurs later than in teeth assigned to Turcocerus.
The precise provenience on Maboko is not known for a third partial mandible that appears to have been collected in 1934, KNM-MB 548 (Figure 4E;MB 1934in Whitworth 1958and Gentry 1970).It preserves corpus depth below a highly worn and weathered dp4 and less worn m1-2 (Whitworth 1958;Gentry 1970).Both m1 and m2 lack basal pillars, and differ from most excavated specimens in having goat folds, and less convex lingual crowns.Molar enamel is also more finely wrinkled than that of KNM-MB 21459.The m2 of KNM-MB 548 is quite hypsodont (H/L = 1.04,Hall and Cote 2021; H/W = 3.58) its height to width index being comparable to hypsodonty indices calculated for m3s of extant Gazella dorcas, Nanger granti and Gazella thomsoni (Damouth & Janis 2011;C. Janis pers. comm).The unworn lower molar KNM-MB 29588 has a similarly high crown; in general, the largest Maboko teeth are rather hypsodont.In comparison, hypsodonty indices for T. africanus m2s from Maboko Main range from 0.75 to 0.92 (H/L, Hall and Cote 2021, this study).
Of the isolated upper molars, upper M3 KNM-MB 21536 (Figure 4H) from Bed 3 is larger than molars of KNM-MB 9204, has more accentuated metastyle, mesostyle and parastyle; and has a slight hint of a basal pillar which are not apparent on molars of KNM-MB 9204.

Comparisons
The very slender, upright, and virtually straight horncores from Maboko are so similar to those described as Kubanotragus in Eurasia (see below) that their generic assignment is hardly questionable.This horncore shape is reminiscent of some bovid female horncores, but the wide size range of the abundant dental material from Maboko clearly shows that we are not dealing with female specimens of Turcocerus.
The shape of the basioccipital of Kubanotragus pickfordi is unlike that of most bovids, in which there are well-marked anterior tuberosities, replaced here by a pair of depressions on either side of a sagittal crest, but such a primitive morphology is also close to the one found in Namacerus (Morales et al. 2003) and probably Eotragus (Made 2012, and our observations), and it can be expected in early bovids.From the slightly younger site of Nyakach, Thomas (1984a) correctly assigned to the same species a horncore KNM-NC 7973.However, the mandible KNM-NC 7816 that he also assigned to N. pickfordi is much too large to go with this horncore; the hypoconid and entoconid remain isolated until late wear, as in the similar-sized Bed 12 bovid KNM-MB 25148.
From the Middle Miocene of Belometchetskaya in Georgia, Sokolov (1949) described as Eotragus (?) martinianus (Lartet, 1851) three horncores, of which he figured one (Sokolov 1949: Fig. 2) obviously of the same species as the holotype horncore of Kubanotragus sokolovi Gabunia, 1973, later described from the same locality.These horncores are so similar, in their remarkable straight, cylindrical shape to KNM-MB 21849 and KNM-MB 35021 from Maboko that genus identity is hardly questionable.Thomas (1984b) also reported K. sokolovi from Cheskewala in the Chinji of Pakistan.Hamilton (1973) assigned to Eotragus the horncores NHMUK-M26688 and NHMUK-M26689 from Jebel Zelten; they resemble Kubanotragus but are more inclined backwards.We assign them to Kubanotragus ?sp.
The horncores of Kubanotragus gaopoensis Chen, 1990, from Gaopo, Lantian, are almost identical with those of K. sokolovi (the apparent difference in size might be due to the fact that Gabunia's horncore measurements were taken on the pedicle: Thomas 1984b), but associated dentitions much differ from KNM-MB 9204 and KNM-MB 20350 in their short premolars and hypsodont molars.From Thymiana on the Greek island of Chios, Bonis et al. (1998) described as Hypsodontus cf.gaopense [sic] a similar frontlet; as at Gaopo, the molars are hypsodont and the premolars are short.
No horncore from Çandır closely resembles Kubanotragus.By contrast, from the nearby site of Inönü that they regarded as somewhat younger than Çandır, Geraads et al. (1995) reported two types of horncores: larger, more conical, and spiralled (similar to Turcocerus), smaller, more cylindrical, and straighter (similar to Kubanotragus), showing the co-existence of the two genera.

Postcranial material
Bovid postcrania from Maboko are not part of articulated specimens and therefore are more appropriate for ecomorphological than the taxonomic analysis that is the focus of this paper.At least 105 postcranial remains were excavated from Bed 3, consisting of 41 bones from 9 different elements of the upper limb, 43 bones from 11 different elements of the lower limb, 3 types of phalanges and 3 vertebras.In comparison, 13 bovid postcrania were excavated from Bed 5b and 15 from Bed 5 w, and an additional 22 specimens were collected from the surface of Maboko Main.Of these, vertebral axis KNM-MB 27496 (Bed 3) is definitely more like the Fort Ternan second type of Gentry (1970 pl. 16, fig. 7, bottom), but he did not identify it to taxon.
A femur KNM-MB 24906 (Bed 3) lacks the cursorial characters (Gentry 1970) of a notch between trochanter and articular head, very unequal lips of the distal trochlea, and deep supra-condylar fossa.Twenty-three astragali from Maboko are attributed to Bovidae with 13 from Bed 3, 2 from Bed 5b, 3 from Bed 5 w, and 5 from the surface of Maboko Main.KNM-MB 21587 (Bed 3) is similar in size to those of the tragulid Dorcatherium chappuisi.The majority are small and homogeneous in size, but their size range (Suppl.Data 2 H) would also be compatible with two species of slightly different sizes, in agreement with the small size difference between T. africanus and K. pickfordi.However, KNM-MB 25378 from the surface Maboko Main is the size of a large gazelle astragalus, as are some additional specimens in the NHMUK.KNM-MB 22077 (Bed 5 w), a left MC III/IV, has length (180 mm) and proximal medio-lateral breadth (21 mm) dimensions similar to those of the Eritrean gazelle Gazella tilonura, and relative breadth to length proportions (11.7) most similar to Gazella gazella, Antidorcas marsupialis, and Pelea capreolus (Scott 1985).Although the MC III/ IVs of Aepyceros melampus is slightly broader relative to length, it shares with KNM-MB 22077 a proximal end with the greatest antero-posterior thickness positioned asymmetrically and just medial to the sagittal midline; a magnum-trapezoid facet that is antero-posteriorly intermediate in length.Other partial metapodials, such as KNM-MB 21661 (Bed 3) are somewhat smaller.
Bovids are important markers distinguishing Early from Middle Miocene faunas in Africa.Based on present evidence, their numbers and diversity in terms of body size and horncore morphology increase from the earliest to latest Middle Miocene in Africa.Bovids are far less abundant and diverse at Maboko (Suppl.Data 2I) than at Fort Ternan (Gentry 1970).This difference is probably linked with the global cooling documented towards the end of the Middle Miocene, which led to an expansion of grasslands favoured by bovids (review in Steinthorsdottir et al. 2021), but the depositional environments of both sites increase the contrast.The Maboko environment ranged from open canopy forest to open woodland; in contrast, Fort Ternan depositional environments are strongly tied to volcanic activity along the flanks of the Tinderet volcano and are thought to be associated with more open woodland and grassland environments that supported far fewer primates (3% of the total fauna based on MNI) than Maboko and no cercopithecoid monkeys (Shipman et al. 1981;Arney et al. 2022: table 4).At Fort Ternan, tragulids and giraffoids are as rare as the primates, but at Maboko the tragulids always make up a high proportion of the mammalian fauna.Tragulid teeth are adapted to consume softer, less fibrous items than those of bovids and giraffoids, and their limbs for movement in more densely wooded or forested environments (Geraads 2010: fig. 36.1).
Although bovids are rare throughout the Maboko sequence from Bed 3 to Bed 15, their abundance fluctuates in the various beds (Suppl.Data 2I).Together with that of the giraffoid Climacoceras, it is negatively correlated with the abundance of tragulids and primates.).They are more common in the deposits of Bed 3, associated with large expanses of open water including lakes, but rarest in the densely wooded wetlands of Bed 5b, and still rare in the more open woodland associated with seasonal rainfall and the presence of grass of Bed 5 w (Retallack et al. 2002;Arney et al. 2022).Together tragulids and primates make up a combined 78% of the mammalian fauna in Bed 5b based on NISP, but only 41% of the fauna in Bed 3 and 49% in Bed 5 w, strongly supporting interpretations of Bed 5b as more densely wooded or forested than other strata in the Maboko sequence.Compared to tragulids, Maboko bovid and giraffoid molars are better adapted to the consumption of fibrous browse, and their limbs (including fused and elongated metacarpals) to movement in more open canopy woodland.Clearly, at Maboko bovids and giraffoids are least abundant where primates and tragulids are most common (Bed 5b), and most abundant where primates and tragulids are the least common, a pattern similar to rhinocerotids (Geraads et al. 2012).

Conclusions
The Bovidae material from Maboko consists mostly of isolated teeth and postcranial remains, but includes no taxonomically significant cranial part besides horncores, and only a few frontals, mandibles and tooth-rows.Horncores allow for their unambiguous identification to Turcocerus and Kubanotragus, genera originally named on the basis of Miocene material from Eurasia.This has important biogeographic implications because each genus is dispersed over three continents (Fig. 6; Suppl.Data 2 J).Whether the Maboko bovids evolved from precursors in Africa or Eurasia is an open question, as primitive bovids are known from both regions.Miocene evidence for incipient horncores in ruminants is slender.Li et al. (2021b) described as Amphimoschus a partial cranium from Xishuigou showing a short supraorbital process.The Xishuigou fauna is said to be in the range 17-19.7 Ma (Wang et al. 2013), an age that would fit the evolutionary stage of this cranium, but it also has Turcocerus halamagaiensis, a species defined in the significantly younger Halamagai fauna (Ye 1989), which makes such an early age unlikely (we note that its Platybelodon dangheensis would also be the earliest record of its genus).Thus, it could in fact be similar in age to Jiulongkou from where Chen and Wu (1976) described as Oioceros (?) jiulongkouensis an adult cranium with very small supraorbital protuberances and a strong sagittal crest.It is conceivable that forms at this evolutionary stage gave rise to proto-Eotragus forms ('Eotragus' noyei, 'Eotragus' minus, and Namacerus), whose horncores are larger but remain smaller than in European primitive Eotragus, reported from the Spanish sites of Buñol, Córcoles, and Tarazona (Alférez 1981;Moyá-Solá 1983;Astibia 1987;Mazo et al. 1998).Further increase in size led to Eotragus clavatus, well known from Sansan (Made 2012).
Potentially, minor changes could easily have led to Turcocerus: some thickening of the horncore base, homonymous twisting, and broadening of the anterior part of the basioccipital, although this part of the skull is not preserved before the age of Tunggur.We may surmise that the sagittal crest, present at Xishuigou and Jiulongkou was lost in derived forms (Sansan, Seegraben), but reappeared at Tunggur in T. grangeri, perhaps because of its large size.
Alternatively, proto-Eotragus forms, rare but widespread from Namibia to China, could have given rise by vicariance in Eurasia to Eotragus (first definitely known from Spain), and in Africa to Turcocerus (first represented there by Turcocerus africanus) before its dispersal to Eurasia.This hypothesis looks more satisfactory because no definite Eotragus is known in Africa, but is weakened by the lack of evidence of Turcocerus in Africa earlier than in Eurasia.
The affinities of Kubanotragus are unclear, because it remains poorly known; it is perhaps more closely related to Hypsodontus, best illustrated by H. pronaticornis from Çandır, Paşalar (Gentry 1990), and Prebreza (Pavlovic 1969), than to Turcocerus, even though Hypsodontus is larger.These three genera are often regarded as members of a subfamily Hypsodontinae, but its monophyly is questionable, because the shared characters of these early bovids: upright horncore insertion, absence of backward curvature or heteronymous torsion, absence of cranial flexure, shape of the basioccipital, etc., are probably just primitive.Thus, we prefer to leave these genera near the base of the Bovidae, without assigning them to subfamily.
The dental remains of these forms tell an interesting story.All Eurasian dentitions associated with Turcocerus, Kubanotragus, and Hypsodontus are hypsodont for their geological age, and have flat lingual walls, goat folds, and shortened premolars, and are thus probably adapted for grazing.The single mandible from Jebel Zelten (NHMUK M26685) displays similar features.By contrast, in dentitions from Maboko, the molars are lower-crowned, their lingual walls are more convex, the central valleys disappear later in wear, and the premolars are relatively longer.However, because the characteristic horncores unambiguously show that they belong to this informal group, we regard these tooth characters as of lower taxonomic significance.It may be that in the Middle Miocene of Kenya, Kubanotragus and Turcocerus filled the niches that were occupied in Eurasia by taxa absent from Maboko, especially Tethytragus, unknown in Africa, and Gentrytragus, which only appears at Fort Ternan (Gentry 1970(Gentry , p. 2010;;Azanza and Morales 1994).

Figure 1 .
Figure 1.Stratigraphic profile of Maboko Main Beds 2-5 based on the side walls of excavation units.Photographs from top to bottom: eastern wall of Pits 1 and 3 taken in 1987; southern wall of Pit.Note that the structure of the Bed 4 limestone changes laterally, being thinner and less solidified at Leakey's Trench with slumping and convolutions due to softer sediments and exposure to stronger hydraulic activity that occurred after Bed 3 was laid down (Retallack et al., 2002; see Suppl.Data 1).That Bed 5 w white clay is superimposed over Bed 5b brown clay is shown in Suppl.Data 1.