Dental specimens selected for this study had been inventoried by Jehenne (1985) and housed in the CVCU of Poitiers (Centre de Valorisation des Collections Universitaires). Eighteen specimens belonging to D. guthi and sixteen to D. quercyi were used for analyses. Only four specimens of B. milloquensis were included in this study due to the lack of available material. The material chosen included both lower molars (M2), directly selected on mandibles, and isolated upper molars (M1, M2 and M3) (Appendix A). The microwear signature is not significantly different between lower and upper molars, even if the positions of these last could not be determined (Teaford and Walker, 1984). The microwears were analysed on the lingual blade of the paracone and on the buccal blade of the protoconid (Merceron et al., 2005). In cases where these facets were not preserved, their analogous facets were chosen on the metacone or hypoconid.

The protocol established by Merceron et al. (2004b) was applied for the casting procedures and quantification of the microwear patterns on the enamel surfaces. The dental facets were digitized using a Leica DC 300 digital microscope camera and the image acquisition software Leica IM50 connected to a light stereomicroscope (Leica MZ125) at low magnification (×30). The dental microwear was quantified in a 0.09 mm² area using Optimas v. 6.2. image analysis software (Media Cybernetics^®).

Microwear scars are defined as pits or scratches as follows: pits are circular to sub-circular scars in enamel whereas scratches correspond to lengthened straight scars with parallel edges. A scratch has a ratio between its minor and major axes lower than ¼, whereas a pit has a higher ratio (Grine, 1986).

Microwear quantification was made by a single observer (AN). The number of scratches (N _s ) and the number of pits (N _p ) were counted. The total number of microwear scars (N _tot = N _s  + N _p ) and the percentage of pits (%p = N _p /N _{tot x100}) were calculated for each species.

The results obtained were compared with those acquired by Merceron et al., 2004a, Merceron et al., 2004b and Merceron et al., 2005 for 25 extant ungulate species with known varied diets extracted from the ‘UNGULATES’ database. These species are classified under five different feeding preferences, each reflected by a particular microwear pattern: leaf browsers, fruit/leaf browsers, grazers, mixed-feeders and generalists (Table 1).

Statistical analyses were undertaken to test whether the observed intergroup differences of microwear patterns were significant or not. All variables were rank transformed to ensure normality and homeoscedasticity. Univariate analyses of variance (ANOVA) were applied to test the significance of the observed differences between samples (Table 2). The combination of two pairwise tests (Fisher's least significant difference and Tukey's honestly significant difference) determined the sources of significant variation for each dental facet (Conover and Iman, 1981 and Sokal and Rohlf, 1998) (Table 3).

The majority of scratches observed on the enamel surface of fossil ruminants are very fine (Fig. 1). However, the microwear pattern observed on D. quercyi teeth tended to be qualitatively coarser than that observed for D. guthi, with the presence of wider scratches and more marked pits.

D. quercyi had a pit percentage (%p) of 56.1% whereas D. guthi possessed one of 42.3% (Table 1; Fig. 2). Furthermore, D. quercyi had a lower number of scratches (N _s ) than that obtained for D. guthi. These observed differences were statistically significant (Table 2 and Table 3) and decisive for the assignation of each Dremotherium species to a particular diet.

D. guthi possessed mean N _s , N _p and %p values which were not significantly different from those observed for the grazers. This result is clearly illustrated in Fig. 3, where D. guthi can be seen to fall within the range of grazing species (Table 1). In contrast, D. quercyi was statistically closer to both browsers and mixed-feeders (Table 3). This result is also well illustrated in Fig. 3, where D. quercyi is clearly comparable with the mixed-feeders in addition to possessing a mean N _s value which falls within the range of browsers.

Despite the small sample size, B. milloquensis presents a N _p (11.0) much lower than those obtained for the two other fossil species and a %p sharply lower than 50% (37.4%) (Table 1). Even if this species was graphically outside of the distribution of dietary categories (Fig. 3), it was close to the grazers in having very low N _p and %p values which were not significantly different from those of the grazing extant species (Table 3).

All of the analyses indicated that both Dremotherium species had different feeding preferences. From the variables tested, D. quercyi presented important similarities with the extant browsers and mixed-feeders. In particular, it is very close to the extant mixed-feeder Tragelaphus scriptus (Fig. 3). This species adopts different feeding strategies during the year. In the wet season, it preferentially feeds on young items of grasses and non-graminoid herbaceous plants, and throughout rest of the year it principally consumes dicotyledon leaves (about 80%). Thus, it is not surprising that its microwear profile is very close to the extant browsers and some studies classify T. scriptus in this category (Gagnon and Chew, 2000). Nevertheless, D. quercyi also possessed a high number of large pits probably reflecting its occasional consumption of fruits such as Acacia pods (Jacobsen, 1974). Considering these facts, we assumed that D. quercyi principally consumed leaves and also occasionally consumed some fruits such as pods, thus explaining its more marked pits and wide scratches.

The results indicated that D. guthi was very close to extant grazers such as the topi, ourebi and zebra, indicating that it had to have consumed grasses (Gagnon and Chew, 2000). Nevertheless, D. guthi possessed very fine scratches compared to these extant species which would indicate a consumption of C₃ grasses (Merceron et al., 2004a, Merceron et al., 2005 and Solounias and Semprebon, 2002). Indeed, such fine scratches are clearly observable in an extant species of Equidae, Equus przewalskii, which exclusively consumes C₃ grasses. On the contrary, the microwear pattern was different from those observed in extant species which only feed on C₄ grasses, such as the zebra Equus burchelli.

To a lesser degree, B. milloquensis also displayed similarities with grazers. However, this extinct species possessed a very low number of scratches (mean N _s  = 18.5) compared to D. guthi (mean N _s  = 22.8) and all of the extant grazers studied (N _s ranging from 22.6 to 29.8). Consequently, it is possible that B. milloquensis had a diet principally composed of grasses, but maybe different to that of D. guhi. Nevertheless, this result was only supported by four specimens of B. milloquensis and has to be tested with supplementary material.

The existence of grasses at the end of the Oligocene is not surprising since their presence had been already attested in North American deposits dated back to the same period (Strömberg, 2002). Considering that the C₃/C₄ transition took place in the Late Miocene, Cerling et al. (1997) supposed that only C₃ grasses were widespread in Europe at the period studied. This assumption was notably supported by the presence of very fine scratches in the fossil species.

The consumption of grasses by two brachydont ruminants is unexpected. Indeed, the presence of low-crowned cheek teeth in an ungulate is often perceived as a browsing adaptation (Janis, 1990 and Janis et al., 2000). Thus, Janis et al. (2002) linked the expansion of grasses to the appearance of hypsodonty in ungulates. High-crowned cheek teeth are considered to be an adaptation to an abrasive diet such as grasses, which are richer in phytoliths than other types of plants. Moreover, the evolution of hypsodonty seems to be closely connected to the diversification of C₄ grasses (richer in phytoliths than C₃ grasses) (Bouchenak-Khelladi et al., 2009 and Hodson et al., 2005). To sum up, the hypsodonty/grasses co-evolution credibly became more marked during the Miocene, in spite of the previous establishment of C₃-grasslands during the Oligocene. Therefore, it is likely that brachydont D. guthi and B. milloquensis were forerunners in grass consumption at the end of the Oligocene. Additionally, it is necessary to keep in mind that the grass type supposedly consumed in our study was the C₃ type, which possessed a less important degree of abrasion than the C₄ type.

This feeding behavior probably enabled the avoidance of more intense interspecific competition during this period and may have been necessary for the survival of these moschids. Moreover, the appearance of a grazing adaptation early in the evolution of ruminants has already been demonstrated in the Early Miocene brachydont Procervidae and some Miocene Giraffidae (DeMiguel et al., 2008, Solounias et al., 1988 and Solounias et al., 2000).

The diversity of resources consumed by the ruminants of La Milloque justifies their body size differences and so implicates the presence of a relatively heterogeneous environment at the end of Oligocene. The smallest species, D. quercyi, had to live in rather closed environments by having a diet mainly composed of leaves. Notably, studies concerning the post-cranial skeleton of this species revealed that it must have been adapted to rather light forests (Blondel, 1998). The presence of rodents such as Eomyidae, Gliridae, and anthracotheriids such as Microbunodon minimum at La Milloque also confirms the existence of close forested areas (Brunet, 1979, Comte, 2000, Lihoreau, 2003 and Vianey-Liaud, 1991). D. guthi and B. milloquensis, whose diet consisted of grasses, probably fed in more open areas. Such environments were notably attested by the presence of the cursorial rhinoceros Mesaceratherium aff. paulhiacense and the Dipodidae (Brunet et al., 1981 and Comte, 2000). Moreover, the La Milloque cenogram shows few large and medium mammals and a high number of small mammals, which characterizes an open and slightly arid environment in this locality (Legendre, 1987).

The dental microwear analysis revealed different feeding behavior for the moschids of La Milloque. The smallest species, D. quercyi, possessed a dental microwear pattern close to extant leaf browsers, and to the mixed-feeder T. scriptus. The largest species, D. guthi and B. milloquensis, were close to extant grazers. Thus, they constitute the oldest brachyodont ruminants to have included a significant proportion of grass in their diets. Notably, during this period, only C₃ grasses were likely to be widespread. Considering these results, we assert the existence of a mixed-environment at La Milloque, which probably ranged from light forest to more open areas composed of C₃ grasses. This conclusion is in agreement with previous studies. At last, DeMiguel et al. (2010) suggested that dietary specialization on leaves or grass occurred independently in several lineages of ruminants and that mixed feeding may have been the primitive dietary state of this group. In order to check this hypothesis, it would be interesting in future works to take into account the other Oligocene ruminant species and to combine dental microwear and mesowear analyses.