DATABASE

A comparative database was developed for 2200 individuals representing upper molar teeth from 64 extant ungulate species (table 1)Table 1. The material was studied in the mammalogy collections of several museums (see Acknowledgments). The variables included in table 1Table 1 are explained below and under Methods. The raw data are available on request from either author.

SAMPLES FOR MESOWEAR STABILITY TEST

Mesowear analysis is applicable only if factors related to diet have a significantly stronger influence on wear patterns than factors related to structure or wear stage. The former problem is addressed by the present study as a whole, but the latter requires special attention. It was surprisingly difficult to find large populations of individuals of well-known age to investigate the effect of wear stage. We were able, however, to study a population of 36 reindeer from Canada (a personal collection) for which age is known from annuli in the lower first molar roots (see Solounias et al., 1994, for more details) (table 2). Individuals from six additional species were selected (table 2): the plains zebra aged by the degree of incisor wear (following methods of Bone, 1964; Klingel, 1965; Slade and Godfrey, 1982);Thomson's gazelle, bohor reedbuck, bushbuck, and black-fronted duiker by number of keratinous rings on horns, size of horns, and second molar height; and African buffalo by size and curvature of horns (following a study for the springbuck, Rautenbach 1971).

UNGULATE DIETS—DERIVING table 1Table 1

Several factors can complicate the relationship of diets to morphology in extant species. Although investigators who study the evolution and adaptation of ungulates have related morphology to reported diets (e.g., Webb, 1983; Fortelius, 1985; Janis, 1988; Solounias et al., 1994), it is clear that diets reported in the literature are often based on small samples from single populations, and that diet would be best studied by extensive field observation or fecal collections. Many ungulates are also opportunistic, and their diets vary by place and by season, so that what can be observed for a few years and in certain places may not hold for the thousands to millions of years that a particular species existed. We acknowledge that dietary information for Recent species is often less than perfect but cannot suggest any immediate remedy. Furthermore, for the method presented here, even a rough classification based on physical properties of plant foods would be preferable to one based on systematics or general appearance.

General dietary classes (browser, grazer, mixed feeder) are problematic for two reasons. They lump a diversity of foods and lifestyles under three broad categories, and scientists have used different definitions for them, not often explicitly stated. However, these dietary classes are so firmly entrenched in the literature that any new approach must somehow relate back to them. Although our study emphasizes a continuum—a spectrum of diets based on their mechanical (wear) properties in terms of abrasion and attrition—we have chosen this general classification as the base against which to compare the results of this study. We use the conventional 90% cutoff points to define the classes: browsers take <10% grass, grazers <10% browse. The huge spectrum between these extremes is lumped into the obviously heterogeneous mixed feeder category.

Table 1Table 1 lists the basic dietary classifications compiled from Janis (1988) and several other sources (in particular: Tener, 1965; Schaller, 1967; Hofmann, 1973; Labâo-Tello and Van Gelder, 1975; Schaller, 1977; Sinclair, 1977; Gauthier-Pilters and Dagg, 1981; McDonald, 1981; Chapman and Feldhammer, 1982; Kingdon, 1982a; 1982b; Nowak and Paradiso, 1983; Hofmann, 1985; Schaller et al., 1986; Hofmann, 1989; Nowak, 1991). In recognition of some uncertainty and ambiguity we have used two parallel classifications of our 64 species: a “conservative” classification (cons) where doubtful cases are treated as intermediate (mixed feeder) and a “radical” classification (radi) where such cases are treated as extreme (browser or grazer). Comparison of results from the two classifications enables us to gauge the effect of such subjective choices and to evaluate their impact on the results. The variable Jad1 gives the diet as reported by Janis (1988); Jad2 gives the corresponding value translated to the simple browser-grazer-mixed feeder classification. Our conservative dietary classification differs from that of Janis (1988) for three species: Antilocapra americana, Capreolus capreolus, and Procavia capensis, but agrees with the present opinion of Dr. Janis (personal commun., July 1999). Table 1Table 1 also features the ad hoc classifying variable Class, with four values: fo (fossil), no (no particular class), mb (“mabra,” minute abraded brachydont, identified under Results, Cluster Analysis), and ty (typical of its dietary class).

TEST CASE: A KNOWN DIETARY SUCCESSION IN THE SERENGETI

We tested this method on a grazing succession from the Serengeti (Bell, 1971), involving plains zebra, topi, wildebeest, and sometimes the hartebeest and Thomson's gazelle. Bell observed that the zebra feeds on the rough higher grasses as it migrates. The removal of these grasses by the zebra enables the topi, the wildebeest, and the hartebeest to follow and feed on lower grasses. Finally, the gazelle follows, feeding on the smallest and softest vegetation. The precise sequence of this succession is essential for these species to find their required vegetation. We predict that the mesowear will replicate this succession such that the zebra and the gazelle will bracket the other three species.

FOSSIL SAMPLES

FOSSIL BOVIDS: Pachytragus laticeps and Pachytragus crassicornis are two species of early Caprini (Bovidae) from the Miocene of Samos collected by Brown in 1924 (Gentry, 1971; Solounias, 1981). P. laticeps has longer and more backwardly curved horn cores than P. crassicornis, which is also smaller and has a more posteriorly placed toothrow (Gentry, 1971). Based on general consideration of the cranial morphology and dentition, Gentry (1971) suggested that P. crassicornis was more advanced and adapted to a harsher environment than P. laticeps. In contrast, Solounias and Moelleken (1992b) determined from a tooth microwear analysis that P. laticeps was a grazer and P. crassicornis a mixed feeder. Based on masseter attachment morphology and using the morphology of the masseter muscle attachments, Solounias et al. (1995) concluded that both species were mixed feeders and suggested that P. laticeps was further removed from the browsing mode than P. crassicornis. We have collected mesowear data for these species to see whether this problem can be resolved.

FOSSIL EQUIDS: Hayek et al. (1992) used microwear to study the paleodiet of the following fossil equids: Mesohippus insignis from the Olcott Formation of Nebraska, Cormohipparion goorisi from Trinity River Pit, Flemming Formation, Cold Springs Texas, Cormohipparion quinni (formerly C. sphenodus), from Valentine Formation, Cornell Dam Member Nebraska, and Cremohipparion proboscideum Quarry 1 of Samos, Greece (Sondaar, 1971). We have collected mesowear data from the same populations in order to compare the results obtained by the two techniques.

All fossil samples were from the Division of Paleontology of the American Museum of Natural History (New York).

SCORING ROUTINES FOR MESOWEAR

Although additional mesowear features can be defined (a larger and more detailed study is in progress), the present study bases mesowear on two variables: occlusal relief and cusp shape (fig. 1). Ungulate teeth were inspected at close range, using a hand lens when appropriate. The first several hundred specimens were photographed and traced on paper, but once the standards had been set to our satisfaction the rest of the material was recorded by direct scoring. We provide several figures, simple outlines of molars, which show the buccal outlines of a selection ungulate teeth illustrating the various morphologies. Artistic shadow has not been included in order to emphasize the shape of the tooth outlines. Parallax due to photography was ignored as we only scored qualitatively three general categories. We only used specimens in which the last molar was in occlusion and the first molar retained an occlusal shape similar to the second molar. Consequently, the effect of age was minimized, as discussed further under Results. After experimenting with various choices and standards, we settled on the sharper buccal cusp of the second upper molar; that is, either the paracone or metacone. This was done for simplicity, and we would like to stress that our experience strongly indicates that the choice of molar buccal cusp is not critical. (For example, the paracone and metacone are usually identical in mesowear of a single individual.) The selection of the sharpest cusp will drive the data slightly toward sharpness. This is a conservative decision, since after the very first wear stage, sharpness is never an artifact of wear stage, whereas blunting may be so.

Occlusal relief was classified as high or low, depending on how high the cusps rise above the valley between them (fig. 1). After some practice, simple scoring is sufficient, but in borderline cases a quantitative index can be constructed as follows. The buccal profile of the tooth is projected onto a plane. The vertical distance between a line connecting two adjacent cusp tips and two adjacent valley bottoms is measured, and divided by the length of the whole tooth (fig. 1). For selenodont forms and plagiolophodont equids, the limit between high and low was arbitrarily set at 0.1, for hyracoids at 0.05, and for rhinoceroses at 0.03. These values were calibrated by the relief observed in the species included in the study, to separate the subjectively “low” from the subjectively “high.” Negative relief (cusp tip lower than sides) was sometimes seen in hypsodont equids, and was treated as low. The progressive blunting of a cusp (see below) will inevitably reduce occlusal relief. That is to say, cusp shape and relief are not entirely independent, but converge at the low and blunt (“grazer”) end of the spectrum. We included occlusal relief, anticipating its usefulness for the study of fossil forms, particularly nongrazing hypsodont plagiolophodonts, which include many hipparions. Occlusal relief is used in the analyses as a percentage and is given in table 1Table 1 as percent of high relief (perhigh).

Cusp shape was scored as sharp, rounded, or blunt (fig. 1, table 1Table 1) according to the degree of facet development. A sharp cusp has (practically) no rounded area between the mesial and distal phase I facets, a rounded cusp has a distinctly rounded tip without planar facet wear but retains facets on the lower slopes, while a blunt cusp lacks distinct facets altogether. Cusp shape is also used as a percentage and is given in table 1Table 1 as three variables: persharp, perround, and perblunt.

Dental structure and phylogenetic history obviously influence both occlusal relief and cusp shape. The blunt cusps of true Bovini and the low occlusal relief of present-day horses are at least partly the result of obvious structural modifications of the teeth. Apart from calibrating the cut-off points for high and low relief for individual groups as explained above, we have not attempted to correct for these influences here, although this may later become necessary. It is our contention that these differences are due to adaptive evolution and largely follow the pattern established by mesowear. They appear to amplify rather than disguise the mesowear signal, and are therefore not a problem unless the resolution desired is very high. A brief “how to do it” description of mesowear analysis is given within Methods.

HYPSODONTY

The goal of our study is to develop a method for the dietary interpretation of extinct species, and the currently best established gross morphological predictor of diet in ungulates is hypsodonty (Janis, 1988), the main aspect of functional durability in the face of wear (Janis and Fortelius, 1988). Hypsodonty must obviously be included in our study, but is perhaps best conceived of as a proxy for overall wear rate (Solounias et al., 1994), rather than as a morphological character as such, just as the mesowear variables are conceived as proxies for kinds of wear.

In deciphering extinct species, the degree of hypsodonty can usefully be employed to select an appropriate subset of extant species for comparison. The logic here is simply that the relationship between attrition and abrasion is studied separately for different total wear regimes. However, for the cluster and discriminant analyses, all crown heights were considered together.

Hypsodonty indices (table 1Table 1, hypind) were taken from Janis (1988): width divided by the length of third lower molar (cement has been excluded in these measurements). For convenience, the species were partitioned into the conventional categories: brachydont (b), mesodont (m) or hypsodont (h). With such a tripartite subdivision, the hypsodonty of most taxa can be readily determined by observation. It is primarily the mesodont species and those on the borderlines that need to be measured or carefully examined for correct assignment. Mesodont artiodactyls are those with indices between 2.6 and 3.4 and all perissodactyls and hyraxes with indices between 2.0 and 3.0. Brachydont artiodactyls have indices of 2.5 or lower, and hypsodonts have indices of 3.5 or lower (table 1Table 1, hyp). The camel was classified as hypsodont despite the unexpectedly low index reported by Janis (1988), but the index reported was used in the analyses. Hog deer, Grant's gazelle, nyala, and the bushbuck are all just below our limit for mesodonts. The takin, Thomson's gazelle, and the common waterbuck are all just above our limit for hypsodonts. Figure 1 shows examples of brachydont, mesodont, and hypsodont teeth.

STATISTICAL ANALYSES

The significance of differences observed in simple comparisons was tested using the Kruskal-Wallis test and the chi square test, as appropriate. Hierarchical cluster analysis was performed for several sets of species, with Euclidean distance and complete linkage (to enhance the distinctness of clusters), using three mesowear variables with and without the index of hypsodonty. Three of the four mesowear variables are percentages of cusp sharpness and add up to 100; therefore a maximum of two of those were included for any analysis). Discriminant analysis was performed for single variables and for all combinations of the mesowear variables plus the index of hypsodonty, using two dietary classifications (conservative and radical) alternately as grouping variable. We report the percentage of correct classifications from the jackknifed classification matrix (table 3). All statistical tests were performed on Systat 7.0 in a PC environment, using the default settings except where noted.

PRACTICALITIES OF MESOWEAR DATA COLLECTION AND ANALYSIS

Since this paper is intended to introduce an easy and generally applicable method of paleodiet reconstruction we agree with one of our reviewers (Dr. R. L. Bernor) that a “cookbook” section is warranted.

The first step in mesowear analysis is obviously to obtain the material. It is important to select the specimens such that teeth in very early and very late wear are excluded, and to avoid scoring the shape of cusps that are either damaged or modified by structural elements (like the paracone of a rhinoceros). Pilot studies have revealed that lower teeth consistently score more rounded than do uppers, and we suggest that the two should not be mixed. Using lower teeth requires building up a comparative sample of lower teeth. We used only uppers in this paper. A sample of less than ten specimens should be treated with caution, while more than 30 is probably excessive.

The scoring procedure itself is described in the method section of this paper and is not repeated here. There is no doubt that it can be refined, but care should be taken not to loose the generality of the method, since restricting it to a single, morphologically uniform group will severely limit the choice of recent species available for comparison. Several of our figures are intended as an aide for scoring. Once the mesowear data exist as a file they can be analyzed by appropriate statistical means (we are sure that both our choice and our application of methods can be improved upon!). The raw data allow comparisons between pairs or groups of species using (for example) the chi square test, while the discriminant and cluster analyses require summary data in the form of (for example) percentages of occlusal relief and cusp shape values.

Blind reliance on statistics is to be discouraged, for reasons exemplified by the “mabra-syndrome” identified in this paper. An important part of more detailed mesowear analyses than those reported here will be the selection of the appropriate comparative sample, much as one would do for estimating body mass by regression techniques. It would clearly not be wise to analyze a small and brachydont species in relation to a sample of large and hypsodont ones, for example. The figures of teeth in this study are offered as an aide to make a first rough assessment of the general mesowear regime of a sample.

STABILITY OF RELIEF AND CUSP SHAPE DURING WEAR

Reindeer Specimens: Ten young specimens (49–51 months) showed a high consistency with 80% sharp cusps. The 16 older individuals (61–77 months) had 87% sharp cusps and 12% rounded. The oldest 10 individuals, ranging 85–97 months, had 70% sharp and 30% rounded cusps. All specimens were high in relief. Chi square tests of the three groups gave low probabilities that the three age classes are different from each other (p values ranged from 0. 552 to 0.187). We conclude that, although there is some increase in roundedness with wear, the mesowear signature is nevertheless reasonably maintained throughout an individual's life (table 2).

Six additional species were also selected for the study of ontogenetic effects on the occlusal relief and cusp shape. All species showed that these parameters remain relatively stable from early middle wear until advanced age. In 24 Thomson's gazelles with 17 to 23 horn rings 83% of cusps were sharp. Seven younger individuals had 71% sharp cusps, while in 7 older individuals 29% of cusps were sharp. In the black-fronted duiker, the proportions of cusp shape (in this case, sharp and rounded) are reasonably stable throughout the life span of the individuals studied (table 2). Low relief and blunt cusps were found in 11 individuals of plains zebra spanning 2–13 years (aged by incisor wear stages), showing that the bluntness is genuinely present in early as well as late wear. The crown height decreased from 13.0 to 8.0 mm in the bushbuck and we found 75% and 80% sharp cusps. High and rounded morphology was found in four plains reedbucks from horn stages 1–4 with 100% high and rounded cusps. In three African buffaloes, horn stages 1–3, the crown height decreased from 22.9 to 12.8 mm and the cusps were all high and rounded (table 2). Similarly, consistent cusp shape morphology during wear was commonly encountered during the recording of the large data set used in this study.

CLUSTER ANALYSIS

Cluster analysis, using the index of hypsodonty and all mesowear variables except percent rounded cusps, polarizes the full set of 64 Recent species into a pattern that groups grazers and browsers at the extremes, with mixed feeders in between (fig. 2). The figure shows hierarchical cluster diagrams of all Recent species included in this study. A is a cluster based on index of hypsodonty, percent high occlusal relief, percent sharp cusps and percent blunt cusps. B is a cluster based on the same variables as (A) except for the index of hypsodonty. C is a cluster based on percent sharp cusps and percent blunt cusps only whereas D is based on the same variables as (C) plus index of hypsodonty. Figure 1 shows that the grazers group more clearly than browsers, which, although clumped at one end, also intersperse with mixed feeders throughout the mixed feeder range. Mixed feeders also intersperse with browsers. There are two main clusters, corresponding to attrition-dominated and abrasion-dominated wear, respectively. The attrition-dominated cluster is divided into two subclusters, one containing mostly browsers, the other mostly browse-dominated mixed feeders. The abrasion-dominated cluster also shows two subclusters, one containing “extreme” grazers (bison bb, white rhino cs, topi dl, and the zebras eb, eg), the other containing the rest of the grazers and the graze-dominated mixed feeders (the abbreviations after the names are the labels used in table 1Table 1 and the clusters). This second subcluster is also divided into (1) all the remaining grazers, a few graze-dominated mixed feeders, and one browser (the greater kudu TT) and (2) mainly graze-dominated mixed feeders and a set of small species (mainly duikers and hyraxes) that are abrasion-dominated for reasons not related to grazing, as discussed below. Somewhat unexpectedly, the effect of hypsodonty on this pattern is negligible: virtually the same tree is obtained whether hypsodonty is included (fig. 2A) or excluded (fig. 2B). This is universally observed in other sets analyzed but is not shown further here. It is fortunate for the practical reason that the index of hypsodonty is far more difficult to obtain for a species than are the mesowear variables. The omission of occlusal relief (fig. 2C) has a much more noticeable effect, most evident in the disintegration of the grazing cluster seen in the previous analyses. Adding hypsodonty recaptures some of the structure (fig. 2D), but does not produce the clear cluster of grazers seen in figures 2A, B.

In figure 2A and B, two Indian deer (barasingha Cd, chital Ax), usually regarded as mixed feeders, cluster with the second rank of grazers (hartebeest ab, wildebeest ct), and we suspect that they may be best interpreted as grazers, as indeed they are in our “radical” classification in table 1Table 1. The mixed feeders associated with the third rank of grazers are also all strongly grass-oriented and in some cases (African buffalo Cs, mountain reedbuck Rf) may be equally well classified as grazers. The main anomaly is the greater kudu TT, a browser that persistently clusters with these graze-oriented mixed feeders and grazers. The reason for this anomaly is essentially a simple calibration problem: the greater kudu shows 100% rounded cusps, like many typical grazers such as the common waterbuck (ke). The fact that the cusps of the greater kudu are considerably less rounded is not picked up by our crude analytical procedure.

Close inspection shows that dispersed browsers are principally small species with significantly rounded cusps: the water chevrotain HY; the duikers DR NA NI NG SL; the hyraxes DD, HB, and two selective, long-necked browsers; the gerenuk LW, and the dibatag EI. The water chevrotain and the duikers are well known to be highly frugivorous, and are thus not typical browsers (Gautier-Hion et al., 1980; Kingdon, 1982a; Lumpkin and Krantz, 1984; Feer, 1989; Nowak, 1991). The rounding of their cusps is probably due to the “tip crushing wear” typically associated with frugivory (Janis, 1990). The hyraxes are opportunistic feeders with varied diets including (in Heterohyrax and Dendrohyrax) a high proportion of insects (Kingdon, 1974). Why the grass dominated mixed feeder Procavia capensis Pc shows such strong attrition-dominated wear is unclear, but it seems reasonable to treat the water chevrotain, the duikers and hyraxes as a special case (“mabra,” for “minute abraded brachydont”) until more information becomes available or a more adequate dietary classification is devised. There is no good reason to exclude the dibatag and the gerenuk, although we hypothesize that their highly selective feeding results in too little attrition to mask even the small amount of abrasion involved. As with the greater kudu, this may also be a calibration problem involving the degree of rounding of the cusps.

Excluding the water chevrotain, the duikers, and the hyraxes produces a more distinct clustering (fig. 3), again with only minor differences due to inclusion (fig. 3A) or exclusion (fig. 3B) of hypsodonty. The mixing of grazers and grass-dominated mixed feeders reported above is still seen, and the greater kudu TT still clusters with these forms, but the remaining browsers are much less dispersed. The mixed feeders, Indian rhinoceros Ru and springbuck Ma, fall among the browsers and the browser bongo BE falls among the attrition-dominated mixed feeders. The springbuck is a selective mixed feeder (Bigalke, 1978) that could well have a mechanically browserlike diet and wear pattern. The Indian rhinoceros is represented by a very small sample (5 specimens), and in any case its placement in the less extreme of the two browser clusters is no great anomaly. The bongo is undoubtedly a browser but is said to dig up and ingest roots (Nowak, 1991), a habit that could well account for the extra abrasion detected.

It may be of interest to explore the subset of species for which good dietary data are available and where the interpretation seems to be uncontroversial. We have selected 27 species to form a set of such “typical” species. Their clustering pattern is essentially free of anomalies (fig. 4); there are three main clusters, one for true grazers, one for less extreme grazers and mixed feeders, and one for browsers. The grazer-mixed feeder cluster is cleanly divided into two subclusters, one for grazers and one for mixed feeders. The browser cluster is also divided into two, with the slightly more abrasion-dominated Sumatran rhinoceros DS, giraffe GC, and mule deer OH forming their own subcluster. Although such a set of “typical” species may not say much about the dietary diversity of living species we feel that it forms a good basis for comparison with fossil forms, as shown below.

The “typical” set can also illustrate the relative contribution of individual variables to the resolution of the clusters. For a more comprehensive and quantitative treatment we refer to the discriminant analysis reported below, especially table 3. Figure 5 we show the two most powerful variables and the weakest variable obtained by discriminant analysis for the typical set: percent sharp cusps (fig. 5A), hypsodonty (fig. 5B) and percent high relief (fig. 5C). It is clear that both percent sharp cusps and the index of hypsodonty alone are capable of polarizing the species along a dietary axis, although neither works as well as the full set of mesowear variables. It is also clear that hypsodonty produces a decidedly more mixed result in terms of diet than does percent sharp cusps. The result based on the highly skewed variable percent high relief, is clearly much less satisfactory than the others, but as shown above, it does significantly increase resolution within the dietary classes when used together with the other mesowear variables. Percent high relief seems to be particularly critical for the recognition of grazers and it alone does, in fact, recognize all of the extreme grazers as such (fig. 5C).

DISCRIMINANT ANALYSIS

Discriminant analysis was performed to quantify the resolution of mesowear analysis with respect to the three conventional dietary classes of browser, grazer, and mixed feeder. As already shown in the cluster analyses, mesowear analysis resolves the structure within these classes in a biologically meaningful way, which is completely missed by an analysis based on the main classes only. For example, cusp shape without occlusal relief gives higher scorings but a poorer clustering pattern than does cusp shape with occlusal relief (figs. 2B and C, table 3). These analyses should thus not be seen as a test of “how well” mesowear analysis works, only of how well it classifies diet at the most general and conventional level in the different sets of species, and especially of how the “conservative” and “radical” dietary classifications affect the pattern.

SINGLE-VARIABLE ANALYSES: For the full data set of 64 living species, the single variable that classified the species best was the index of hypsodonty (hypind), with an overall correct classification (jackknifed matrix reported throughout) of 65% for the conservative (cons) and 59% for the radical (radi) classifications (table 3). The second-best single variable was percent high relief (perhigh), which correctly classified 55% (conservative) and 56% (radical) overall. Percent blunt cusps (perblunt) alone correctly classified 58% of radical but only 48% of conservative. The mean of all single-variable analyses of the full set were 50% for conservative and 56% for radical.

For the set without the “minute abraded brachydonts” (mabra) the pattern was similar and the percentages correctly classified overall were about 5% higher on average, with hypsodonty index producing 63% for conservative and 57% for radical. Percent sharp cusps (persharp) performed distinctly better for this set, about as well as hypsodonty index, with 61% for conservative and 69% for radical. For the typical set, where conservative and radical coincide, percent sharp cusps (persharp) correctly classified 96% of the species, followed by percent rounded cusps (perround) and hypsodonty index at 81% and a mean single-variable value of 78% (table 3).

TWO-VARIABLE ANALYSES: The combination of hypsodonty index and percent blunt cusps (perblunt) performed best, at 80% correct for conservative and 72% for radical. The index of hypsodonty with persharp also classified well, at 76% for both conservative and radical. Other combinations classified less successfully, with a mean value of 61% for conservative and 63% for radical.

For the minute abraded brachydonts-free set, the pattern was essentially the same but percentages correctly classified averaged 7–8% higher. For the typical set six out of the ten combinations correctly classified 96% or more of the species, percent high relief with either persharp or perround both giving 100% correct.

THREE-VARIABLE ANALYSES: As the number of variables increase the percentage of species correctly classified mounts and the difference between the conservative and radical classifications is reversed, so that whereas the radical classification gets the highest percentages of species correctly classified for one and two variables, from three variables onward, the conservative classification gets the highest scores. For three variables and the full set of species, the combinations of hypsodonty index with any two cusp shape variables performs best (80% correct for conservative, 72% for radical), followed by combinations of hypsodonty index, percent high relief, and any cusp shape variable (76% for conservative, 70–72% for radical). The mesowear variables without hypsodonty give significantly lower values and the pattern is reversed, with higher percentages correctly classified for the radical classification. The mean percentage correctly classified are 72% for conservative and 68% for radical.

For the minute abraded brachydonts-free group, the pattern is the same but the percentages average 6% higher for conservative and 4% higher for radical. For the typical set, all combinations except one correctly classify 96% or more of the species. The exception is hypsodonty index, percent high relief, and percent blunt, at 81% correct. This is a recurring pattern: percent high relief and percent blunt together seem to perform relatively poorly in most combinations of two or three variables. Both variables are highly skewed since most taxa have high relief and few blunt cusps.

FOUR-VARIABLE ANALYSES: All combinations of hypsodonty with occlusal relief and two of the three cusp shape variables give the same result. For the full set of species, 76% (conservative) or 72% (radical) are correctly classified. For the minute abraded brachydonts-free set these values are 80% (conservative) and 73% (radical). The typical set is classified at 96% correct.

MESOWEAR AND DIET—A CLOSER LOOK

This section gives a more detailed breakdown of the general patterns found in cluster and discriminant analyses of hypsodonty, occlusal relief, and cusp shape in relation to diet.

HYPSODONTY: The index of hypsodonty is significantly different among conservative and radical dietary classes (Kruskal-Wallis test, P < 0.001) except that the difference between mixed feeders and grazers is only significant for the conservative classification (P = 0.003).

OCCLUSAL RELIEF: High occlusal relief is the common state except in grazers, which cover the full range (fig. 6). No browser has less than 80% high relief, and only two mixed feeders (the brachydont lesser kudu and the hypsodont saiga) have lower values. The only hypsodont browser, the pronghorn, has 100% high relief, but very high values are also found for most hypsodont mixed feeders. There is no evidence that occlusal relief would systematically change with hypsodonty in browsers or mixed feeders, but in grazers the most hypsodont forms have low relief. These are the “extreme grazers” identified by the cluster analyses, primarily the species of Equus and certain Alcelaphini. Grazers as a group have significantly lower relief than either browsers or mixed feeders (Kruskal-Wallis test, P < 0.001 for both the conservative and the radical classifications). The only species that show predominantly low relief are those of the zebras and the white rhinoceros, which are all plagiolophodont grazers. However, plagiolophodont forms may also show high relief (for example the hipparions included in this study), and feral horses that have lived on browse also show high occlusal relief (unpublished data). Therefore, low relief appears to be a strongly associated with grazing (or at least with highly abrasive food).

CUSP SHAPE: Cusp shape appears to be largely independent of hypsodonty, as all three cusp shapes occur in all crown height classes (fig. 7). Looking separately at the percentages of sharp, rounded, and blunt cusps allows two further distinctions: no extant grazer has more than 40% sharp cusps, and no browser or mixed feeder has more than 10% blunt cusps. Despite nearly complete overlap of ranges, all three dietary groups in all combinations are significantly different from each other in cusp shape (persharp and perblunt; Kruskal-Wallis test; P ≤ 0.001 for both conservative and radical), except that mixed feeders are not significantly different from browsers. In most cases cusp shape does not seem to change with hypsodonty, but the only hypsodont browser (the pronghorn) has a very high percentage of sharp cusps, indicating a high attrition level and arguing against any role for abrasive food or dust. In contrast, percent rounded cusps seems to decrease with increasing hypsodonty in grazers, suggesting that increased abrasion may not be the main reason for the increased wear in these forms. The plots shown in figures 6 and 7 change only in details if the radical dietary classification is used. Hierarchical cluster diagrams for fossil and Recent species are shown in figure 8.

FIGURES OF SELECTED TEETH: A sample of sharp and high relief teeth is shown in figures 9 and 10. Both perissodactyl and artiodaclyl teeth show that the apices are similarly sharp although the relief is clearly higher in artiodactyls. Dots next to each tooth are 5 mm apart; all views are buccal of adult upper molar teeth (mostly M2). Figures 11–15 show teeth of high relief and rounded apices. Again the perissodactyl overall has the lower relief (fig. 15I). We find that the condition of sharp and rounded remain the same despite the degree of relief. For example, figure 14D and 14I have similarly rounded apices but the degree of relief differs. Figure 16 shows high and low cusps which are all rounded whereas figure 17 shows high rounded (17H, K) and low blunt cusps (17F, I). Figures 18 and 19 show low and blunt cusps of perissodactyls. In some of the Equus individuals cusp are more than blunt they are convex (fig. 18F and 19D). So far we have classified such convex cusps as blunt. Figures 20 and 21 show cusps of mixed modality. For example, figure 20A shows sharp and rounded cusps on the same tooth. Figure 20I shows a blunt cusp and a rounded cusp on the same tooth. Figure 21A shows strongly blunt cusps and 21D and F mixed rounded and sharp cusps. We find that the mixed patterns within the same dentition are not common.

COMPARISON OF CUSP-SHAPE HISTOGRAMS: The cusp mesowear histograms for a selection of extant species from table 1Table 1 are given in figures 22–26. Histograms are based on the raw data and are the same with the summary values given in table 1Table 1.

Figure 22 shows the percentages of cusp tip shapes for selected browsers; the data show a large array of differences between species. Alces alces is the only species with 100% high and sharp cusps. The remaining browsers can be arranged according to a decreasing percentage of sharp cusps down to Tragelaphus strepsiceros, which has 100% rounded cusps. Note that the hypsodont browser Antilocapra has a high proportion of sharp cusps, as expected from its diet. Hyaemoschus and most duikers (Cephalophus) have a small percentage of blunt cusps, probably because of “tip-crushing” wear due to frugivory as noted above. Litocranius also shows strong rounding, possibly because of the very low overall rate of wear (attrition not enough to mask even very low abrasion). Dendrohyrax dorsalis is the only browser in our data set that features some blunt cusps, but as discussed above hyraxes are too omnivorous to be qualified as typical browsers and their dental wear is not well understood.

Figure 23 shows the percentages of cusp tip shapes for selected hypsodont mixed feeders. Ovibos, Saiga, and Capra are attrition-dominated and all species show at least 25% sharp cusps. Blunt cusps are only seen in Lama and Capra, less than 10% in both cases.

Figure 24 shows the percentages of cusp tip shapes for selected “traditional” mixed feeders, arranged from the attrition-dominated Antidorcas to the abrasion-dominated Aepyceros. Most species have about 50% sharp and 50% rounded cusps, while blunt cusps are absent except for a few in Gazella thomsoni, the more grass-oriented of the two gazelles.

Figure 25 shows the percentages of cusp tip shapes for some mixed feeders of the Indian monsoon forest. All are highly abrasion-dominated, like African fresh grass grazers (and the browsing greater kudu). The cervids (Axis axis and Cervus duvauceli) particularly show a grazerlike profile with over 20% blunt cusps, and both are indeed treated as grazers in our “radical” dietary classification (table 1)Table 1.

Figure 26 shows the percentages of cusp tip shapes for grazers. The data show a diversity that promises potential for future palaeodiet analysis. The Reduncini (Redunca and Kobus) have a very high percentage of rounded cusps, a signal apparently related to marginal or “fresh grass” grazing. All other grazers have a significant proportion of blunt cusps. The Alcelaphini (Connochaetes, Alcelaphus, and Damaliscus) are dominated by rounded cusps but also show significant proportions of sharp and blunt cusps. The zebras (Equus) stand out with almost subequal amounts of cusps of each type, while the white rhinoceros (Ceratotherium) has over 30% blunt cusps and no sharp ones at all. Bison is unique in our data set in having mostly blunt cusps (over 70%).

THE SERENGETI FEEDING SUCCESSION (TEST CASE)

Subjective inspection of the mesowear of the five species involved indicates four grazers (Equus burchelli, Alcelaphus buselaphus, Connochaetes taurinus, and Damaliscus lunatus) characterized by low relief and numerous blunt cusps, and one browser or browse-dominated mixed feeder (Gazella thomsoni) with a higher relief and higher percentage of sharp cusps. Chi square analysis of the cusp shape distribution of all five species in the succession indicates very low probability that the distributions are identical (P < 0.001). When the zebra is excluded, the difference is still highly significant (P < 0.001), but when only the grazing bovids are compared no significant difference is detected (P = 0.924). This is the predicted result because they all feed at the same vegetational level. Comparison of the four grazers also shows a marginally significant difference (P = 0.025). Damaliscus is represented by only five individuals, however. If it is excluded, the difference between Alcelaphus, Connochaetes, and Equus is more marked (P = 0.008). Thus simple analysis of cusp-shape distribution indicates a statistically significant relationship between the species that corresponds to the known one. The same result can also be seen in the cluster diagrams (figs. 2–5; table 4).

PILOT APPLICATIONS TO FOSSIL UNGULATES—TABLE 4

PACHYTRAGUS: The samples from the two Quarries at Samos were (marginally) significantly (P = 0.011) different in cusp shape distribution although their dentitions are otherwise identical (Gentry, 1971). Neither species showed low relief or blunt cusps, the hallmarks of extreme grazers. Based on microwear analysis, P. laticeps has been characterized as a grazer and P. crassicornis as a mixed feeder (Solounias and Moelleken, 1992b). The mesowear signal suggests the reverse, however; P. crassicornis being more abrasion-dominated, with a higher percentage of rounded and a lower percentage of sharp cusps.

In the cluster analysis (fig. 8A), P. laticeps is aligned with moderate mixed feeders in the attrition-dominated “browsing part” of the tree: the lesser kudu Tragelaphus imberbis, Capra ibex, Ovibos moschatus, and the gazelles. In contrast, P. crassicornis falls among more hypsodont and open-adapted mixed feeders in the abrasion-dominated “grazing half” of the tree: Camelus dromedarius, Lama glama, Aepyceros melampus, and Tragelaphus angasi. In terms of modern analogs, P. crassicornis resembles the modern mesodont mixed feeders chousingha (Tetracerus quadricornis) and nyala (Tragelaphus angasi). P. laticeps has mesowear different from all the mesodont species included in the extant data set but appears most similar to mixed feeders such as the gazelles and the lesser kudu.

FOSSIL EQUIDS. The selected four hipparions have microwear data already analyzed by Hayek et al. (1992). As predicted, mesowear clustering (fig. 8A, B) shows Cormohipparion goorisi closest to the grazing end of the tree, with the less extreme grazers and grass-dominated mixed feeders, while Merychippus insignis and Cormohipparion quinni are both placed in the next cluster, with the least extreme of the grass-dominated mixed feeders such as Tetracerus quadricornis, Camelus dromedarius, and Lama glama. Cremohipparion proboscideum clusters with the less extreme browsers, near Odocoileus hemionus and the mixed feeder Antidorcas marsupialis. No hipparions cluster near the zebras.