Introduction

Isolation on islands is a factor well-known to induce evolutionary changes, and a rather frequent event is the origination of giant forms in small mammals like rodents. The Pleistocene deposits of Mallorca and Menorca (Balearic Islands, Spain) provide a good case study. They yielded a peculiar mammalian fauna including the giant dormouse Hypnomys (e.g., Bate 1919; De Bruijn 1966; Mills 1976), a rodent of the family Gliridae, the closest relative of which is considered to be the extant garden dormouse Eliomys (McKenna and Bell 1997). A recent review of the vertebrate fauna from the Balearic Islands (Bover et al. 2008) sheds new light on the context of Hypnomys evolutionary history. The genus Hypnomys includes three species in Mallorca: Hypnomys waldreni Reumer, 1979 from the late Pliocene, H. onicensis Reumer, 1994 from the late Pliocene—early Pleistocene boundary and H. morpheus Bate, 1919 from the middle and late Pleistocene and Holocene. In Menorca, H. morpheus has been recorded, and two other species of unclear status (Reumer 1982, 1994) have also been described, H. mahonensis Bate, 1919 and H. eliomyoides Agusti, 1980. A fourth species has been considered in Mallorca from the early Pliocene, but with no specific assignment (Hypnomys sp.). In Eivissa a species of Hypnomys was found in the late Pliocene deposit of Cova de ca na Reia, but as with the fourth species of Mallorca, no specific assignment has been proposed. Another species, H. gollcheri de Bruijn, 1966 from Malta deposits, was at one time considered. For Storch (1974), H. gollcheri and Leithia cartei Adams, 1867 are probably synonymous taxa, but Zammit-Maempel and De Bruijn (1982) included it in Maltamys. Hypnomys sp. was also cited from Nuraghe Su Casteddu in Sardinia (Esu and Kotsakis 1980), but later emended to Tyrrhenoglis (e.g., Kotsakis 2003). So Hypnomys is restricted to the Balearic Islands and only H. morpheus has a very rich fossil record. In the late Pleistocene, this species was only found associated with two other endemic species: Nesiotites hidalgo Bate, 1945 (Soricidae; Bate 1945) and Myotragus balearicus Bate, 1909 (Caprinae; Bate 1909), a poorly diversified fauna as frequently found on islands. Man colonized Mallorca probably between 2350 and 2150 cal BC (Alcover 2008; Bover et al. 2007), the last documentation of the three endemic mammals postdating 3650 cal BC (Myotragus), 3030 cal BC (Nesiotites) and 4840 cal BC (Hypnomys) (Bover and Alcover 2003; Bover and Alcover 2007). Since the endemic fauna is considered to have become extinct before 2350 cal BC, it can be assumed that humans cohabited for some time with these three endemic species and that the latter rapidly became extinct (Bover and Alcover 2007). However, humans were not the sole invaders at the end of the Pleistocene: Eliomys and Apodemus, then Mus and Rattus, also settled on the Balearic Islands.

Fig. 1.

Zygomasseteric construction in Balearic dormice. A. Skull of extant Eliomys quercinus ophiusae (MNHN1983-832) in lateral (A₁) and anterior (A₂) views. B. Skull of Hypnomys morpheus in lateral (B₁) and anterior (B₂) views. Arrows show the origin and the insertion of the lateral portions of the masseter. The skull of Hypnomys morpheus (B) corresponds to a reconstruction. Eliomys and Hypnomys are represented at the same scale. The map summarizes the evolutionary history of Balearic glirids—Hypnomys is a lineage derived from an Eliomys species isolated by the sea level rise that followed the Messinian salinity crisis, then Eliomys quercinus ophiusae followed the first human colonization (dashed arrow represents a hypothetical pathway of colonization).

Since Hypnomys and Eliomys are phylogenetically very close, Hypnomys represents an insular lineage derived from an Eliomys species isolated on the Balearic Islands by the sea level rise that followed the Messinian salinity crisis (Fig. 1; Bover et al. 2008). According to Mills (1976: 36), Hypnomys exhibits many “characters in common with less advanced glirids” such as a U-shaped coronal suture, lingulate interparietal, diffuse supra-occipital process, converging jugals, diverging upper cheek teeth alveoli, the form of the zygosquamosal process and its anterior extension, the form of the bulla, position of the “ectopterygoid” foramina, extension of lateral pterygoid process, a small or occluded maxillary foramen, the relative size of the iliac surface, a large terminal phalanx of the first toe and the overall robustness of the mandible. Considering the dental pattern, Zammit-Maempel and de Bruijn (1982) and Holden (2005) suggested to include Hypnomys within Eliomys. However, H. morpheus depicts an original zygomasseteric construction (Fig. 1), in which the zygomatic arch was strong indicating that the lateral masseter muscle was highly developed and anchored on a wide zygomatic plate. Among traits to be considered is the relatively small-sized infraorbital foramen, through which the medial masseter passes. Such an arrangement is unique among extant and extinct glirids, the lateral masseter evolving in Hypnomys toward a sciuromorphous-like condition (Fig. 1B₁). These diagnostic traits justify referring the insular species to a distinct genus derived from Eliomys. Regardless, H. morpheus appears distinct from any Eliomys species and a comparison of this insular form with the latters may shed light on the evolution under isolation. Moreover, Eliomys is unique among extant glirids by being clearly partially insectivorous and a predator of molluscs, insect larvae and occasionally even of small mammals (Ognev 1963). Considering that interspecific interactions, primarily competition, might be the driving force behind this evolutionary specialisation, we investigate whether an absence of congeners would lead to the evolution of morphology. It can be consequently ex- pected that any departure of H. morpheus from such a diet must be easy to recognize, using extant glirids as reference.

The precise comparison given by Mills (1976) will be considered using the mandible as a proxy, the hypothesis being that there must be changes in mandible shape correlative to the changes in morphology of the zygomatic plate (Hautier et al. 2008). As a matter of fact, the study of the mandible is easier compared to the study of the skull due to the two dimensional structure of the dentary bone and the number of complete lower jaws of Hypnomys at our disposal. In addition, many studies have shown that the mandible provides good information both on change of size as well as changes in shape (Angerbjörn 1986; Renaud and Millien 2001; Michaux et al. 2002; Renaud and Michaux 2003). The present study also aims at testing the conclusion of the mandible shape analysis using data provided by a microwear analysis of the molars, with each component of the masticatory apparatus showing characteristics that may reflect one parameter of the species niche, i.e., the diet. We investigated whether Hypnomys occurring on Balearic Islands in an impoverished insular fauna reversed the evolutionary direction of Eliomys lineages. For Hypnomys that evolved with few or no competitors, we predict that changes in habitat use would lead to the evolution of a more generalist morphology. This case will not only show how a lineage could modify its habitat use under isolation, but it may also provide clues to its extinction.

Material and methods

For morphometric analyses, studied specimens come from the collection of the MNHN (collection Vertébrés supérieurs Mammifères et Oiseaux), the collection of the Department of Mammalogy of the AMNH, and the collection of the IMEDEA. We analysed 124 glirid mandibles of both sexes, representing 4 species of 3 different genera: Hypnomys morpheus (fossil specimens from holocene deposits of Balearic islands), Eliomys quercinus (fossil specimens from holocene deposits of Balearic islands and extant insular and continental representatives), E. melanurus and Dryomys nitedula. In order to reduce the intraspecific effects related to allometric changes, only adult specimens showing the third molar erupted were considered in the analysis.

The material used for the microwear analysis belongs to the collection of the IMEDEA. Two species and 37 specimens were included: 24 extinct Hypnomys morpheus from Mallorca and 13 extant Eliomys quercinus ophiusae from Formentera. Many altered specimens were rejected from analysis because of post-mortem deterioration (Andrews 1990; King et al. 1999). Dryomys nitedula was not included in this analysis because the current analysis techniques cannot deal with very small teeth. The list of measured individuals is given in the Appendix 5.

Mandible outline.—Two morphometric methods are commonly used for describing the morphology of rodent mandibles: landmarks and outline analyses (Bookstein 1991; Foote 1989). However, the simple morphology of the mammalian mandible requires the digitalization of type 2 landmarks (i.e., points of maximum curvature along the outline), which are more sensitive to certain errors in their location than points defined with bone sutures. In this context, we decided to quantify the shape of mandibles using outline analyses based on the Fourier method (Renaud and Michaux 2003; Michaux et al. 2007), because outlines give access to features involved in the insertion for masticatory muscles. The outline corresponds to a two-dimensional projection of the vestibular side of the mandible (Renaud and Michaux 2003). As teeth were often missing, only the outline of the dentary was considered. Left mandibles only were measured. If the left mandibles were broken, mirror images of the right ones were computed. The starting point of the outline was chosen at the anterodorsal edge of the incisor alveolus and for each mandible 64 points at equally spaced intervals along the outline were recorded (Renaud and Michaux 2003).

Two Fourier methods are commonly used: the Radial Fourier Transform (RFT) and the Elliptic Fourier Transform (EFT). Here, we applied the Elliptic Fourier Transform, a method allowing a description of complex outlines (Kuhl and Giardina 1982). This method is based on the separate Fourier decompositions of the incremental changes of the x- and y- coordinates as a function of the cumulative length along the outline (Kuhl and Giardina 1982). Thus, the outline is approximated by a sum of trigonometric functions of decreasing wavelength (i.e., the harmonics). Any harmonic corresponds to four coefficients: An, Bn for x, and Cn, Dn for y, defining an ellipse in the xy-plane. The coefficients of the first harmonic, describing the best-fitting ellipse of any outline, are used to standardize both the size and orientation of the mandible. After standardization, these coefficients correspond to the residuals, and are not considered in the following statistical analyses (Crampton 1995; Renaud et al. 1996).

An advantage of the EFT method is that if the wavelength of the harmonic is low, more substantial details of the morphology of the mandibles can be considered. Given that the measurement noise increases with the rank of the harmonics, the rank of the last one was empirically determined as the coefficient of variation of the harmonic amplitude (i.e., the square root of the sum of the squared Fourier coefficients; Renaud et al. 1996) of repeated measurements of five specimens. As shown in previous works (e.g., Renaud and Michaux 2003), the first seven harmonics offer a good compromise between measurement error, information content and numbers of variables to consider. Following inverse processes [Inverse Fourier transform method (Rohlf and Archie 1984)] the coefficients of Fourier allow us to reconstruct the mandible outline and to visualize the shape changes.

Statistical procedures were performed with R1.5.0 (Ihaka and Gentleman 1996). For each outline, 24 coefficients comprising seven harmonics (EFT7) were considered. The intraspecific shape variation was compared with the interspecific one with Multivariate analyses of variance (Manova) on these Fourier coefficients, using the species as a factor (Claude et al. 2003). A multivariate regression of the Fourier coefficients on size, estimated by the square root of the outline area (Michaux et al. 2007), permitted us to assess the effect of allometric variation on the overall shape of the mandible. The morphological variability of extant glirids was quantified with a principal component analyses (PCA). Extinct taxa were then added as supplementary data. Because our data set consisted of a relatively large number of variables, the shape space was simplified to the first Principal Components (the number of PCs was defined in order to keep 98% of the interspecific shape variance). Manova in association with a test of significance (Wilk's Lambda test) was performed on these PCs in order to assess the effects of phylogeny (i.e., subspecies assignments) and geographical range. A Factorial Discriminant Analysis (FDA) of shape coordinates was performed to assess if there was a discrimination of the mandible outline with the geographic range.

Microwear analysis.—Microwear features are generated on enamel dental facets during the course of mastication (Walker et al. 1978; Solounias and Hayek 1993). Recent studies (e.g., Solounias and Semprebon 2002; Merceron et al. 2004a; Nelson et al. 2005) showed that the diet of extinct species might be deduced from the dental microwear pattern, especially the number of pits (semi-circular scars) and scratches (elongated scars). Although tooth microwear has been widely studied in fossil and extant ungulate species (e.g., Solounias and Semprebon 2002; Merceron et al. 2004a), it has been also considered as a good indicator of diet in other taxa such as rodents (Nelson et al. 2005). The orientation of the scratches is also usually considered to identify the direction of jaw movement during chewing (Charles et al. 2007). Forty molars were selected for this microwear analysis. Only teeth of Hypnomys in association with a mandible were taken into consideration. Microwear was measured on the protoconid and the hypoconid of lower second molars (M₂) for both modern glirid species and Hypnomys. We decided to choose the protoconid and the hypoconid because they provided a large and flat occlusal surface which contrasts with the morphology of all other cuspids of the glirid tooth that merge into crests.

Fig. 2.

Distribution of the area (A) and circularity (B) of the mandibles of Hypnomys and Eliomys.

The teeth were carefully cleaned using acetone and cotton swabs. Microwear was measured on translucid casts made using polyvinysiloxane (Coltene President Microsystem®) and transparent epoxin resin (In Epox®, ADAM Montparnasse) left to cure for one day. For this study, dental facets were digitized using a ZEISS® SV11 M2B with the × 115 objective and transmitted-light through the stereomicroscope. This method is cheaper, simpler and totally non-invasive (no need for coating) compared to methods using a scanning electron microscope (SEM; Hayek et al. 1992; Solounias et al. 1988). Then, a 0.01mm² area was delimited on each facet to take variations in dental microwear patterns along molar facets into consideration (Gordon 1982). Dental microwear is quantified using Optimas® (v.6.5.2) software (Media Cybernetics®).

Fig. 3.

Shape differentiation of the mandible on the first two axes of the Principal Components Analysis (PCA) performed on Fourier coefficients of the mandibles. Outlines are reconstructed on the first two canonical axes, the light grey outline represents the maximum values of the axes, and the dark grey outline corresponds to extreme reconstruction.

We documented four microwear variables: the total number of scratches (Ns), the total number of pits (Np), the number of wide scratches (Nws) and the number of large pits (Nlp). Two supplementary variables can be deduced from them: the number of fine scratches (Nfs) and the number of fine pits (Nfp). The value of 5 urn for scratch width or pit diameter discriminates fine scratches or pits from wide ones. Scratches were distinguished from pits using the minor/major axis ratio. The pits have a ratio higher than 1/4, whereas scratches have a lower one (Grine 1986). Four variables were integrated in the multivariate analyses: the total number of scratches (Ns), the number of wide scratches (Nws), the total number of pits (Np) and the number of large pits (Nlp). Statistical procedures were performed with R1.5.0 (Ihaka and Gentleman 1996). A probability level of 0.05 was assumed for all tests.

Results

Mandible outline.—The mandible of Hypnomys is always bigger than the mandible of all other glirids here studied (Fig. 2A). The circularity (Fig. 2B) corresponds to the ratio of perimeter length squared by the area, as calculated with Optimas®. In fact, the circularity of the mandible gives an estimation of its compactness or roundness that is related to the degree of relative differentiation of its processes. Compared to extant glirids, Hypnomys shows low values of circularity and is characterized by a mandible with a high and large ascending ramus and more weakly differentiated processes. In our dataset, Hypnomys is clearly individualized by the overall massivity of its mandible, as it was observed by Mills (1976).

The intraspecific variation of the shape is obviously less important than the interspecific variation (Wilk' s Lambda test: Value = 0.0634, F = 10.73, p < 0.001) and does not significantly contribute to the shape differentiation. Manovas on PCs_1–12 indicated a significant morphological differentiation of the mandible outline within the dataset involving phylogeny (i.e., subspecies, Wilk's Lambda test: Value = 0.0134, F = 7.43, p< 0.001). Respectively, 46.8% and 21% of among-group variance is explained by PC₁ and PC₂ (Fig. 3). The mandibles of Hypnomys are distinct from the mandibles of Eliomys regarding the first component, which appears highly correlated to the size of the coronoid and condylar processes, the individualization of the angular one, and an overall robustness of the ascending ramus. The second axis includes components related to the orientation of the coronoid and condylar processes. However, these axes are weakly informative regarding the shape variation in relation with the phylogeny (Fig. 3), and the genera Hypnomys and Dryomys remain very close to Eliomys in the morphological space.

Fig. 4.

Plot of the discriminant analysis of the shape coordinates (the first twelve PCS, i.e., 98% of the interspecific shape variance) versus geographic range.

A factorial discriminant analysis allowed a complete discrimination for the geographical range (Fig. 4). Manovas on PCs_1–12 indicated a significant morphological differentiation of the mandible outline within the dataset involving geographical range (Wilk's Lambda test: Value = 0.6501, F = 3.77, p < 0.001). Mandible morphologies related to clades are significantly different and the mandible of Hypnomys and Dryomys are discriminated from the mandibles of Eliomys on the first discriminant axis. Mahalanobis distances (Appendix 1) indicated that the morphology of the mandible of Hypnomys is more similar to that of Dryomys (d [Hypnomys—Dryomys] = 75.5) than to that of Eliomys (d [Hypnomys—Eliomys] = 90.7). Although high variability is known to be an important feature in fossil island mammals, the mandibles of Eliomys present a high morphological variation compared to that of Hypnomys and Dryomys. This variability could be explained by the heterogeneity of the Eliomys sample, which includes continental and insular populations. The outline of the mandible allowed us to distinguish the continental representatives of the genus Eliomys from the insular ones (i.e., Balearic Islands, Lipari, Corsica and Sardinia; Fig. 4). The isolation of the latter in the shape space confirmed an insular divergence involving the morphology of the mandible. Nevertheless, they remained close to the representatives from Spain, North Africa, Israel and Saudi Arabia on the first and second discriminant axes. E. quercinus and E. q. ophiusae of Balearic Islands were associated in the shape space with E. quercinus of North Africa and E. quercinus lusitanicus of Spain, respectively.

The multivariate regression of the Fourier coefficients on size, estimated by the square root of the outline area, was highly significant (p < 0.001). Thus, allometry plays an important role for determining the pattern of morphological diversification. The size was compared to the main morphological differentiation displayed on PC1 to visualize this allometric effect (Fig. 5). Among the extant glirid sample, size and shape of the mandible are highly correlated (R = 0.46, p < 0.001). It appears that H. morpheus does not lie on the same allometric trajectory.

Microwear analysis.—The numbers of scratches and pits observed in all studied specimens are listed in Appendix 2. The Student tests (Appendix 3) on microwear features of both protoconid and hypoconid are not significant (p > 0.05), and these facets present similar microwear patterns. The total number of scratches and pits, associated to percentiles of each feature, yield information about the main characteristics of the microwear patterns. Manovas were performed on the four microwear variables (Nfs, Nws, Nfp, Nlp) measured on the protoconid and the hypoconid of the M₂ of Hypnomys and Eliomys (Table 1). They indicate significant differences (p < 0.05) between microwear patterns exhibited by the two genera. In terms of the average number of scratches and pits (Appendix 2), Hypnomys exhibits a greater range of differences than Eliomys. Considering the standard deviation (Appendix 4), it appears that intraspecific variation is more important in Hypnomys than in Eliomys. The latter is always characterized by a low number of scratches (Ns < 15). On the contrary, the number of scratches, especially fine ones, is very variable in Hypnomys. It results from univariate tests (Table 2) that the differentiation of Hypnomys is principally due to differences in the number of fine scratches (Nfs).

Fig. 5.

Allometric relationship between the size (estimated from the square root of outline area) and the main shape signal (scores on the first principal components). The dashed line represents the linear regression between both variables for all extant glirids.

Table 1.

Multivariate analyses of variance (Manova) with effect species on four microwear variables: the number of fine scratches (Nfs), the number of wide scratches (Nws), the number of fine pits (Nfp) and the number of large pits (Nlp).

Table 2.

Univariate analyses of variance (Anova) on four microwear variables (Nfs, Nws, Nfp, Nlp) with effect species for the hypoconid and the protoconid.

Discussion

The mandible.—Compared to Eliomys, Hypnomys morpheus is clearly a giant dormouse (Fig. 2A). An increase in size also affected insular Eliomys. With the exception of the continental Spanish specimens, mainland specimens of Eliomys of our sample always have smaller mandibles than insular ones. Although some specimens from Mallorca are obviously larger than mainland ones, E. quercinus individuals living on the biggest islands (i.e., Corsica and Sardinia) are similar in size to mainland forms. These results are consistent with previous works (Angerbjörn 1986; Michaux et al. 2002; Renaud and Michaux 2003) indicating a much more complex determinism of size in insular species that depends on several combined factors like the area of the island and the presence of predators (Michaux et al. 2002; Millien 2004). In their detailed study on adaptive trends in the mandible of Apodemus wood mice, Renaud and Michaux (2003) identified a random shape differentiation on some islands. In our analysis, the genus Eliomys depicts a different case of differentiation, all the insular forms of the genus being associated on the first and second discriminant axes (Fig. 4). Concerning shape differentiation in the evolution of the Balearic glirids (i.e., H. morpheus and E. quercinus ophiusae), the the first two principal components show two distinct shifts (Fig. 3). A dorsoventral expansion of the ascending ramus and a reduced coronoid process characterize the mandible of Hypnomys whereas all insular forms of Eliomys mandibles are dorsoventrally compressed and have a well developed coronoid process distally positioned. For Satoh (1997), an increasing weight of the mandible implies an increase of the area of insertion of the masticatory muscles (especially the masseter). Thus, differences in size could be responsible for the shape differentiation. However, we showed (Fig. 5) that H. morpheus did not lie on the same allometric trajectory than extant Eliomys. If this divergence in mandibular shape cannot be explained by a simple allometric relationship, other factors (e.g., ecological factors) may play an important role in shaping the mandible.

The shape and size differentiation of the mandibles of H. morpheus could be related to a way of life different from that of Eliomys. The dorso-ventrally compressed mandible of Eliomys does not contradict a partially insectivorous diet (Ognev 1963; Storch 1978). Shearing meat or crushing insects requires less occlusal pressure than grinding plants or seeds (Satoh and Iwaku 2006). On the contrary, a strong incisai bite is required to kill insects or small vertebrates and the development of mandibular processes in Eliomys could provide a high lever advantage. Dental characters, like the high concavity of the occlusal surface of the cheek teeth (Freudenthal and Martín-Suárez 2007) underscore an adaptation towards a more insectivorous diet. In the shape space, the morphology of the outline of the mandibles of H. morpheus is clearly distinct from that of Eliomys and remains close to that of Dryomys nitedula a glirid more omnivorous than Eliomys (Grzimek 1975). The diet of Dryomys consists of seeds, acorns, buds, fruits, and occasionally insects, eggs and small vertebrates (Grzimek 1975). The more massive mandible of H. morpheus likely indicates an omnivorous species that included also very hard items (i.e., acorns, nuts, fruit) in its diet as a function of habitat and season. The masseteric arrangement (see Fig. 1), associated with the overall robustness of the skull and the mandible of H. morpheus as underlined by Mills (1976), is also consistent with our hypothesis. The differences observed between the shape differentiation of the mandible of Eliomys and Hypnomys might be related to their different evolutionary history. H. morpheus evolved without competition prior to the arrival of humans and their domestic livestock. Its size increased and it adapted to a widened niche including harder food. On the contrary, E. quercinus ophiusae cohabited with other introduced rodents (like Apodemus, Mus, Rattus) and kept its initial adaptation. This explains both morphological divergences observed on the Balearic Islands.

Fig. 6.

Digitized photographs of the protoconid of the second molars. A. Eliomys quercinus ophiusae (IMEDEA 7357), Formentera, Balearic Islands; extant specimen. B. Hypnomys morpheus (IMEDEA 63839), Cova Estreta, Pollença, Mallorca, Holocene. Note the higher number of fine scratches in Hypnomys.

The cheek teeth.—The microwear analysis may allow an independent assessment of this interpretation of the diet of H. morpheus. The dental microwear patterns of extant species depend on the food items that were consumed just before the death of the animals (e.g., Solounias et al. 1988; Teaford and Oyen 1989; Fortelius and Solounias 2000; Merceron et al. 2004a, b, 2005). The grazing species present a lower percentage of pits than the browsing ones. The statistical analysis indicates that extant Eliomys quercinus ophiusae and Hypnomys morpheus differ significantly in the number of scratches. Like other glirids, E. quercinus ophiusae is largely omnivorous but with a great tendency to eat eggs, insects and small vertebrates. In fact, Eliomys is the most carnivorous glirid currently known (Ognev 1963). The dental microwear pattern of the omnivorous-insectivorous E. quercinus ophiusae (Fig. 6A) is in accordance with a diet dominated by hard food items. The specimens analysed display a large percentage of pits, especially wide ones, and large scratches. Such a microwear pattern could be explained by a diet mainly composed of insects. Kahmann and Thorns (1972) recorded a highly insectivorous diet for E. q. ophiusae during some seasons. A similar microwear pattern, characterized by a higher number of pits and a high frequency of large pits and scratches, is also found in ground squirrels (Nelson et al. 2005). This microwear pattern was related to an abrasive diet composed by seeds and insects and the presence of more grit in the food items. For Strait (1993), fruit-eaters could present a similar dental microwear pattern and we must take this into account for the dietary reconstruction of extinct species. However, Nelson et al. (2005) showed that the omnivorous ground squirrels differed from the frugivorous tree squirrels by a higher number of pits.

The microwear pattern of H. morpheus (Fig. 6B) is more variable than that of E. quercinus ophiusae. The important intraspecific variation of the microwear pattern of Hypnomys could attest that it was able to adapt its diet to the seasons. This suggests a more omnivorous diet than that of E. quercinus ophiusae. The specimens always display larger percentages of scratches, especially fine ones, than Recent glirids do. The abundance of fine scratches suggests that H. morpheus could be able to eat graminoids, despite the fact that the presence of earthworms (and snails) in the diet could also explain some narrow scratches (Silcox and Tedford 2002). Fine scratches likely have an ambiguous meaning. All the specimens of H. morpheus show a high number of pits and large pits, which indicates the intake of hard particles (insects, worms, fruit, and graminoids). An incorporation of grit into the diet could also explain a high frequency of coarse features (Solounias and Semprebon 2002). These features were observed in species living in dry habitats (camel, pronghorn) and in species feeding on roots (bush pig). This could attest that H. morpheus was more terrestrial than arboreal, supporting Mills (1976). Finally, it seems that H. morpheus had a mixed diet. By extension, we can suppose that, being the sole omnivorous mammal of its size on the island it was able to adapt to a great number of habitats.

Conclusions

Geometric morphometrics of the mandible, as well as microwear analysis of the teeth bring new insights on the extinct dormouse Hypnomys from Mallorca and Menorca (Balearic Islands). The special features of Hypnomys already described by Mills (1976) are correlated to a massive mandible and a tooth microwear pattern characterized by a high number of fine scratches. No insular populations of Eliomys display such trends in their mandible morphology and tooth microwear pattern. By comparison with the European garden dormouse Eliomys, these traits can be interpreted for Hypnomys as indicating an omnivorous diet. Hypnomys was likely a large omnivorous glirid able to eat hard food items that may have even included graminoids. These results are also consistent with the interpretation of Mills (1976) who suggested that H. morpheus was terrestrial rather than arboreal. The evolutionary divergence of Hypnomys from Eliomys could be due to the fact that the Hypnomys lineage evolved within a highly impoverished fauna among which it was the sole rodent. Hypnomys that had access to diverse food resources rapidly became an endangered species when its natural environment was altered by the settlement of the first human populations and their domestic livestock.