MULTIVARIATE ANALYSES

A PCA of linear postcranial measurements resulted in two factors with eigenvalues > 1.0, together accounting for 88.2% of the variance. PC1 described postcranial variation primarily attributable body size; all variables except Ulna Diameter (UDML; see Table 2 for abbreviations) had very high positive loadings on PC1 (> 0.700, most > 0.850; Table 3). PC2 described postcranial variation primarily attributable to locomotor habit; UDML and HWD had high positive loadings (arboreal and semifossorial mammals typically have robust forelimb bones) whereas Metatarsal III Length (MT3L) had a high negative loading (distal limb elongation is typical of cursorial and some bounding mammals).

A plot of these first two PCs generally supports the interpretations above (Figure 6). On PC1, some of the smallest genera plot furthest to the left (e.g., Cavia, Herpestes, Myoprocta) and some of the largest plot furthest to the right (e.g., Canis, Mazama, Ourebia, Muntiacus). Most cursorial mammals have low values on PC2 (e.g., most artiodactyls, Lepus, Dolichotis) and semifossorial and arboreal mammals have high values on this axis (e.g., Mellivora, Arctictis, Dinomys, Erethizon). Significant variation exists in both of these patterns, however. Conspicuous exceptions include the cursorial canids, which plot high on PC2 instead of low, and some small semifossorial carnivorans (i.e., Mephitis, Herpestes), which have values close to zero on PC2. In the case of canids, their position on PC2 may reflect their more generalized morphology (relative to artiodactyls) and/or their tendency to dig on occasion. For the small carnivorans, this apparently reflects an overall trend for smaller-bodied mammals (i.e., those with low PC1values) to approach zero on PC2. In functional terms, this suggests that smaller mammals tend to have less specialized postcrania for a given locomotor habit than larger bodied ones; in other words, postcranial morphologies correlated with force production (i.e., diggers and climbers) and speed (i.e., runners and bounders) tend to be more pronounced in the larger mammals included in this dataset. Phylogenetic factors also influence the positions of taxa in the PC morphospace; all three hyraxes plot very close on PC2, despite the more arboreal habits of Dendrohyrax. This is no doubt partly attributable to the less specialized morphology of Dendrohyrax relative to most other arboreal taxa (e.g., Arctictis, Erethizon).

Protypotherium plots near the center of both PC axes, extremely close to the cursorial rodent Dasyprocta. Other nearby taxa include two semifossorial rodents (Myocastor and Lagostomus) and an arboreal one (Capromys). All of these except Dasyprocta have body masses of 6-9 kg, within the range inferred for Protypotherium. The position of Protypotherium in PC morphospace suggests that it is postcranially more similar to medium-sized caviomorph rodents than to comparably-sized carnivorans, lagomorphs, hyracoids, or artiodactyls; it says little about its locomotor habit other than that its postcrania are not as modified for force production as Erethizon or Dinomys nor as modified for speed as Dolichotis.

A DFA of postcranial measurements classified 87.2% (34/39) of extant mammals correctly by locomotor habit; all variables were significantly different among groups except Femur Diameter (FDML), and the first two functions demonstrated significant differences among groups. DF1 primarily separated cursorial and bounding mammals from arboreal and semifossorial ones (Figure 7); MT3L and Tibia Length (TL) had the greatest positive correlations with DF1 and UDML and Distal Humerus Width (HWD) had the only negative correlations (Table 3). DF2 separated bounding mammals from all others; HWD and MT3L had the highest correlations (> 0.400), but nearly all other variables had correlations between 0.300 and 0.400, with the exception of UDML (0.123).

Misclassified mammals included: Dendrohyrax (semifossorial instead of arboreal), Herpestes (generalized instead of semifossorial), Mephitis (arboreal instead of semifossorial), Myoprocta (generalized instead of cursorial), and Oreotragus (bounding instead of cursorial). Except for Dendrohyrax, the true locomotor group of each of these mammals had the next highest posterior probability; in Dendrohyrax, it had the second lowest posterior probability (higher only than cursorial). It is notable that in no case was an arboreal or semifossorial mammal misclassified as bounding or cursorial (or vice versa).

Protypotherium was classified as arboreal with a high posterior probability (0.984) but an extremely low conditional probability (0.000). In other words, based on these data, Protypotherium resembles arboreal mammals more than those in other locomotor groups, but it is quite different from the arboreal mammals considered here. This is apparent in Figure 7; Protypotherium plots closest to the arboreal group centroid but is much farther from that centroid than any member of that group. Semifossorial was the second most probable classification for Protypotherium. Based strictly on the DFA of linear postcranial measurements, it is unlikely that Protypotherium was a highly cursorial or bounding mammal.

A PCA of functional indices resulted in two factors with eigenvalues > 1; the first two accounted for 63.6% of the variance (Table 4). PC1 primarily described variation due to locomotor habit and therefore resembled PC2 of the linear postcranial measurements analysis; arboreal and semifossorial mammals generally had high values on this axis whereas cursorial and bounding mammals generally had low values. Epicondylar Index (EI) and Femur Robustness Index (FRI) had the highest positive loadings (Index of Fossorial Ability, IFA, and Humeral Robustness Index, HRI, are slightly lower) and Crural Index (CI), Metatarsal/Femur Index (MFI), and Brachial Index (BI) all had high negative loadings; the former are expected to be greater in arboreal and semifossorial mammals whereas the latter are expected to be greater in cursorial and bounding mammals. Only a few exceptions to the locomotor distributions on PC1 are evident (Figure 8). Ailurus, an arboreal carnivoran, is positioned close to the cursorial side of PC1; it has the lowest EI of any arboreal or semifossorial mammal and the lowest FRI of any mammal (along with Canis and Vulpes). Hyemoschus, a cursorial artiodactyl, is positioned on the edge of the semifossorial/arboreal region; among cursorial mammals, it has the lowest CI and relatively low MFI and BI. Given that Hyemoschus inhabits dense woods and is a member of a more basal family of artiodactyls, this is not particularly surprising; more notable is its distance from the other tragulid in this analysis (Tragulus), which plots among other cursorial artiodactyls. All three hyraxes plot close together in the semifossorial/arboreal region, despite two of these (Procavia and Heterohyrax) being classified as bounding mammals. Hyraxes are somewhat generalized in their morphology, and one might therefore have expected them to plot closer to the center of PC1 (if not with other bounding mammals). The two mammals classified as generalists in this analysis, Felis and Cavia, are split between the two regions; Felis (which is relatively long-legged) plots on the left side of PC2, and Cavia (which has relatively shorter limbs) falls to the right.

PC2 was primarily a phylogenetic axis, with mammals more or less distributed based on their ordinal affiliation, irrespective of locomotor habit; these groupings had much overlap, however. HRI, BI, and Gluteal Index (GI) had the highest positive loadings on this axis and no variable loaded negatively.

Protypotherium plots closest in the PCA to Herpestes and Mephitis, two small (2-3 kg), semifossorial carnivorans. Other closely positioned mammals include Hyemoschus (discussed above) and Potos (an arboreal carnivoran similar in size to Herpestes and Mephitis). Given that Hyemoschus plots distant from other cursorial artiodactyls, Protypotherium can best be described as plotting in the generalized end of the semifossorial/arboreal region of PC1 and among rodents and carnivorans on PC2. It is quite distant from most cursorial and bounding mammals.

A DFA of functional indices classified 87.2% (34/39) of known cases correctly; all variables were significantly different among groups except IFA, and the first two functions were significant. DF1 primarily separated cursorial and bounding mammals from arboreal and semifossorial ones (Figure 9); EI had the greatest positive correlation with DF1 whereas MFI and CI had the greatest negative correlations (Table 4). DF2 primarily separated bounding mammals from all others; MFI had a high positive correlation whereas FRI and IFA had smaller negative correlations.

Two of the mammals misclassified by the DFA of functional indices were also misclassified by the DFA of linear postcranial measurements: Mephitis was classified as arboreal instead of semifossorial (as in the DFA of linear measurements), and Dendrohyrax was classified as generalized instead of arboreal (it was classified as semifossorial in the DFA of linear measurements). The three other misclassified mammals included Hyemoschus (generalized instead of cursorial), Cuniculus (arboreal instead of semifossorial), and Dasyprocta (generalized instead of cursorial). As in the other DFA, except for Dendrohyrax, the true locomotor group of each of these mammals was second most probable. In no case was an arboreal or semifossorial mammal misclassified as bounding or cursorial (or vice versa).

Protypotherium was classified as semifossorial with a moderately low posterior probability (0.493) and a higher conditional probability (0.654). The second most likely classification for Protypotherium was arboreal, with a posterior probability of 0.441. In contrast to its position in the DFA of postcranial measurements, Protypotherium falls well within the morphospace occupied by extant mammals, close to Myocastor (Figure 9). Concordant with the DFA of postcranial measurements, the DFA of functional indices suggests it is unlikely that Protypotherium was a highly cursorial or bounding mammal.