The mammalian order Carnivora encompasses many species with a diverse array of locomotor habits, including terrestrial generalists, cursorial, scansorial, arboreal, semifossorial, and semiaquatic forms (Ewer,1973; Van Valkenburgh,1985; Nowak,2005). Although some are more common within families, these locomotor modes span across the entire order, suggesting that repeated convergent or parallel evolution has occurred across the Carnivora. Species with different locomotor habits show obvious differences in limb proportions. Behavioral specializations have led to the greater development of certain muscles relative to others, depending on which muscles are more important in a given locomotor mode. This influence on postcranial musculature should, therefore, affect the morphologies of the bones to which the muscles attach.
Ecomorphological analyses of the postcranial skeleton have been successfully applied to examine the locomotor modes of extinct animals using their extant counterparts (e.g., Van Valkenburgh,1985,1987; Munthe,1989; Anyonge,1996; Heinrich and Rose,1997; Lewis,1997; Argot,2001,2002,2004; Gingerich,2003; Christiansen and Adolfssen,2007; Schutz and Guralnick,2007; Hopkins and Davis, 2009; Hunt, 2009; Lewis and Lague,2011; Meloro,2011a; Polly,2011). These studies aim to characterize the morphological space occupied by extant taxa with differing locomotor habits, and then infer the ecology of their extinct counterparts based on their morphology. A seminal paper by Van Valkenburgh (1987) demonstrated that extinct carnivorans could be relatively well classified into locomotor categories by a comparison with extant taxa. Later studies on rodents (Elissamburu and Vizcaíno,2004; Samuels and Van Valkenburgh,2008) also found taxa could be classified well with ecomorphological analyses using functional indices.
To evaluate similarities between mammalian orders, Samuels and Van Valkenburgh (2008) classified nine extant carnivoran species using ecomorphological comparisons to rodents. These classifications were not entirely successful because of the dissimilar evolutionary histories of these two groups. Carnivorans were only correctly classified in three out of nine cases based on locomotor mode and morphology. The results of this earlier study suggest that to correctly classify the locomotor habits of fossil organisms, extant species that share a relatively recent evolutionary history must be used (Samuels and Van Valkenburgh,2008). Studying locomotor groups within families is probably the most conservative way to understand extinct organisms; however, many locomotor styles that existed in the past are no longer observed within taxonomic families, evidenced by organisms like the saber-toothed cats. Additionally, some carnivoran families are no longer extant, such as amphicyonids and nimravids. For these groups, extant members of the same order can be used as comparators. This is not to say that the behavior of extinct organisms most nearly resembles that of its closest living relative, but that when animals have evolutionarily similar starting points it is easier to see convergent or divergent morphological trajectories (i.e., when rodents and carnivorans have similar locomotor habits, it does not mean that they necessarily have similar morphologies due to their disparate evolutionary starting points).
Like Van Valkenburgh's (1987) study, we also use morphology to examine the locomotor proclivities of extinct carnivoran species to better understand their locomotor habits and their ecological roles in past communities. Here, we use a different set of morphological indices, many more species of living and extinct carnivorans, and include some forms, like saber-tooth morphs, which have some very different morphological features from living carnivorans. This study includes 107 carnivoran species from 13 living families that show terrestrial, cursorial, scansorial, arboreal, semifossorial, and semiaquatic locomotor habits (Table A1). Based on the results of previous studies, we predict carnivorans with similar locomotor habits will display similar morphological features as a result of convergent or parallel evolution. Selection pressures acting on functional features should lead to similarities between taxa regardless of their phylogenetic histories. Comparisons of postcranial material from extinct carnivorans to living species should allow us to infer their ecologies, improving understanding of the evolution of locomotor adaptations through time. Here, we apply these methods to some extinct members of the Amphicyonidae, Barbourofelidae, Canidae, Felidae, Miacidae, Mustelidae, Nimravidae, and Ursidae. Comparing these extinct species to their closest extant relatives allows us to get a more accurate picture of the niches that these extinct species filled. Taken as a whole, these inferred ecologies can help us to better reconstruct paleoenvironments and inform us of the intricate interplay between carnivorans, their prey, and their complete environments (Louys et al.,2011).
MATERIALS AND METHODS
For this study, we examined the postcranial skeletons of 333 individuals from 107 extant carnivoran species in 13 families (Table A1). We collected data for species from all living fissiped (terrestrial) carnivoran families to sample the widest range of locomotor variation possible. We preferentially chose adult, wild-caught individuals, and used members of both sexes. The species used represent nearly all of the body size range of extant carnivorans, from the < 100 g ermine (Mustela erminea) to the > 300 kg polar bear (Ursus maritimus). We also measured specimens of 25 extinct carnivorans, some represented by complete skeletons, others composites of multiple individuals. We gathered measurements of skeletal specimens of extant and extinct species from the following museum collections: Donald R. Dickey UCLA collection (UCLA), Natural History Museum of Los Angeles County (LACM), Page Museum La Brea Tar Pits Hancock Collection (LACMHC), Idaho Museum of Natural History (IMNH), the Smithsonian Institution National Museum of Natural History (USNM), University of Washington Burke Museum (UWBM), Carnegie Museum of Natural History (CM), and John Day Fossil Beds National Monument (JODA). Additional measurements were compiled from literature sources, including Scott and Jepsen (1936), Hendey (1980), Wang (1994), and Wang et al. (1999).
We measured 20 postcranial osteological characteristics (Table 1 and Fig. 1) to the nearest 0.01 mm using digital calipers; measurements included bone lengths, diameters, and the sizes of muscle attachments. These measurements were used to calculate 16 morphological indices (Table 2), as in Samuels and Van Valkenburgh (2008), which represent limb proportions and the mechanical properties of some primary locomotor muscles. These indices were used to examine the relationship between anatomy and locomotor ecology of extant species, and to reconstruct the ecology of extinct carnivoran species.
Table 1. Osteological measurements used in the analyses
Measurements used are illustrated in Figure 1.
Greatest length of the humerus
Midshaft mediolateral diameter of the humerus
Length of the deltopectoral crest
Epicondylar breadth of the distal humerus
Greatest length of the radius
Functional length of the ulna
Midshaft mediolateral diameter of the ulna
Length of the olecranon process of the ulna
Greatest length of metacarpal 3
Greatest length of proximal phalanx of digit 3 of manus
Greatest length of terminal phalanx of digit 3 of manus
Greatest length of the femur
Midshaft anteroposterior diameter of the femur
Height of the greater trochanter of the femur
Epicondylar breadth of the distal femur
Greatest length of the tibia
Midshaft mediolateral diameter of the tibia
Length of tibial tuberosity
Greatest length of metatarsal 3
Greatest length of terminal phalanx of digit 3 of pes
Table 2. Morphological indices used in the analyses, their definitions, and their inferred functional significance
Measurements indicated above are illustrated in Figure 1 and described in Table 1.
Shoulder moment index (SMI)
Deltopectoral crest length divided by humerus length (DPCL/HL). Indicates mechanical advantage of the deltoid and pectoral muscles acting across the shoulder joint.
Brachial index (BI)
Radius length divided by humerus length (RL/HL). Indicates relative proportions of proximal and distal elements of the forelimb.
Humeral robustness index (HRI)
Mediolateral diameter of humerus divided by humerus length (HMLD/HL). Indicates robustness of the humerus and its ability to resist bending and shearing stresses.
Humeral epicondylar index (HEI)
Epicondylar breadth of humerus divided by humerus length (HEB/HL). Indicates relative area available for the origins of the forearm flexors, pronators, and supinators.
Olecranon length index (OLI)
Olecranon process length divided by functional length of the ulna (ULOL/FUL). Indicates relative mechanical advantage of the triceps brachii and dorsoepitrochlearis muscles used in elbow extension. This is identical to the index of fossorial ability used by Hildebrand (1985a,b).
Ulnar robustness index (URI)
Mediolateral diameter of ulna divided by functional length of the ulna (UMLD/FUL). Indicates robustness of the ulna and its ability to resist bending and shearing stresses, and relative area available for the origin and insertion of forearm and manus flexors, pronators, and supinators.
Manus proportions index (MANUS)
Manus digit 3 proximal phalanx length divided by metacarpal 3 length (Mph3p / MC3L). Indicates relative proportions of proximal and distal elements of the manus and size of the palmar surface.
Claw length index (CLAW)
Manus digit three terminal phalanx length divided by pes digit three terminal phalanx length (Mph3t/Pph3t). Indicates the relative size of manual and pedal claws.
Crural index (CI)
Tibia length divided by femur length (TL/FL). Indicates relative proportions of proximal and distal elements of the hind limb.
Femoral robustness index (FRI)
Anteroposterior diameter of femur divided by femur length (FAPD/FL). Indicates robustness of the femur and its ability to resist bending and shearing stresses (AP diameter is used due to transverse expansion of the femora in some semiaquatic mammals).
Gluteal index (GI)
Length of distal extension of the greater trochanter of the femur divided by femur length (FGT/FL). Indicates relative mechanical advantage of the gluteal muscles used in retraction of the femur.
Femoral epicondylar index (FEI)
Epicondylar breadth of femur divided by femur length (FEB/FL). Indicates relative area available for the origins of the gastrocnemius and soleus muscles used in extension of the knee and plantar-flexion of the pes.
Tibial robustness index (TRI)
Mediolateral diameter of tibia divided by tibia length (TMLD/TL). Indicates robustness of the tibia and its ability to resist bending and shearing stresses.
Tibial spine index (TSI)
Length of distal extension of the tibial tuberosity (spine) divided by tibia length (TSL/TL). Indicates relative mechanical advantage of the hamstrings and biceps femoris muscles acting across the knee and hip joints.
Pes length index (PES)
Metatarsal 3 length divided by femur length (MT3L/FL). Indicates relative proportions of proximal and distal elements of the hind limb, and relative size of the hind foot.
Intermembral index (IM)
Lengths of the humerus and radius divided by lengths of the femur and tibia [(HL + RL) / (FL + TL)]. Indicates the length of the forelimb relative to the hind limb.
Extant species were classified into groups according to their locomotor habits (Table 3). We consulted a wide variety of literature sources to classify the ecology of each species (see Appendix Table A1). Many carnivorans are capable of performing a wide variety of locomotor behaviors as needed; thus, each species was assigned to the most appropriate group based on the proportion of time spent using different locomotor modes. Each of the locomotor groups used includes species from multiple families.
Table 3. Locomotor categories used in this study and their definitions
Complete lists of extant species included and their classifications are included in Appendix 1. Species were assigned to categories deemed most appropriate, but given that there is some gradation between categories, individual species may be capable of placement within more than one category.
Rarely swims or climbs, may dig to make a burrow (but not extensively) (e.g., ermine, grisons).
Regularly displays rapid locomotion with bounding characterized by unsupported intervals (e.g., cheetah, gray wolf).
Capable of climbing, usually for escape, does not forage in trees (e.g., pumas, black bears).
Capable of and regularly seen climbing for escape, shelter; forages actively in trees, and may rarely come down to the ground (e.g., red panda, palm civets).
Regularly digs to build burrows for shelter, may dig to forage underground (e.g., badgers).
Regularly swims for dispersal, escape, or foraging (e.g., river otters, American mink).
We used several statistical tests to examine the relationships between limb morphology and locomotor ecology, including analysis of variance (ANOVA), stepwise discriminant function analysis (DFA), and linear regression. Multivariate ANOVA (MANOVA) was used to test for significant differences among locomotor groups and families, differences in individual indices were assessed by univariate ANOVAs with Scheffe's F and Tamhane's T2 procedures used for post hoc comparisons. Stepwise DFA was used to determine which variables best separated a priori locomotor groups, and to then classify extinct species into these groups. A second DFA was run using families as groups, to examine how well families (phylogeny) can be classified using these same indices. As a selection criterion, the stepwise model included variables with F probability < 0.05 and excluded variables with F probability > 0.1. Discriminant scores were also mapped onto a phylogeny of the carnivoran species studied, allowing interfamilial and intrafamilial visualization of ecomorphological changes. We built a composite phylogeny based on a combination of molecular and morphological studies (Wayne et al.,1989; Wozencraft, 1993; Wang,1994; Baskin, 1998; Wang et al.,1999; Gaubert and Veron,2003; Yoder et al.,2003; Flynn and Wesley-Hunt, 2005; Flynn et al.,2005; Lindblad-Toh et al.,2005; Wesley-Hunt and Flynn,2005; Gaubert and Cordeiro-Estrela,2006; Johnson et al.,2006; Koepfli et al.,2006;2007;2008; Tedford et al.,2009; Eizirik et al.,2010). To visualize the functional indices used in the analyses, we used linear regressions and bivariate plots of log transformed variables (available in Supporting Information); these also facilitated the interpretation of extinct taxa, many of which were represented by incomplete specimens. The number of individuals sampled for each species studied were disparate, thus averages were used in all statistical analyses. All analyses were performed using IBM SPSS Statistics 19.
Analysis of Variance
MANOVA found significantly different morphometric indices among both locomotor categories and families (Table 4). The partial η2 value was higher for locomotor category than for family, indicating greater variance in morphometric indices is attributable to locomotor group than to phylogeny. The mean values and standard deviations of each index for each of the locomotor groups studied are summarized in Table 5. We assessed differences in the individual indices by univariate ANOVAs with Scheffe's F post hoc procedure for comparison of individual groups (Table 5).
Table 4. Results of MANOVA performed using morphological indices, with locomotor categories and families as factors
Table 5. Mean values and standard deviations of indices for each locomotor group
Terrestrial (n = 37)
Cursorial (n = 16)
Scansorial (n = 19)
Arboreal (n = 13)
Semifossorial (n = 11)
Semiaquatic (n = 11)
Indices are defined in Table 2. The numbers of species in the analysis are listed below each group. Significant differences between groups in univariate ANOVA tests at the P < 0.05 level using Scheffe's F post hoc procedure are indicated by the following: T = terrestrial, C = cursorial, Sc = scansorial, A = arboreal, SF = Semifossorial, SA = Semiaquatic.
0.424 (0.079) Sc,SA
0.524 (0.091) C
0.563 (0.081) C
0.870 (0.101) C,SA
1.005 (0.069) T,Sc,A,SF,SA
0.873 (0.067) C, SA
0.796 (0.036) C
0.828 (0.035) C
0.774 (0.066) T, C,Sc
0.077 (0.011) SA
0.070 (0.008) SF,SA
0.090 (0.014) C
0.091 (0.020) C, T
0.226 (0.041) SF,SA
0.187 (0.026) Sc,SF,SA
0.234 (0.042) C,SF,SA
0.228 (0.027) SF,SA
0.290 (0.048) T,C,Sc,A
0.298 (0.042) T,C,Sc,A
0.160 (0.040) SA
0.139 (0.031) SF,SA
0.147 (0.034) SF,SA
0.150 (0.030) SA
0.205 (0.054) C,Sc
0.216 (0.072) T,C,Sc,A
0.042 (0.010) SA,SF
0.061 (0.010) C
0.064 (0.013) C
0.475 (0.116) A
0.408 (0.076) Sc,A,SF,SA
0.530 (0.069) C
0.616 (0.080) T,C
0.522 (0.079) C
0.557 (0.080) C
1.047 (0.204) SF
0.979 (0.117) SF
1.111 (0.195) SF
1.025 (0.140) SF
1.618 (0.374) T,C,Sc,A,SA
0.982 (0.127) SF
0.071 (0.007) SA
0.064 (0.005) SF,SA
0.070 (0.005) SA
0.071 (0.006) SA
0.079 (0.006) C
0.090 (0.017) T,C,Sc,A
0.197 (0.046) SA
0.184 (0.048) SA
0.194 (0.043) SA
0.239 (0.020) T,C,Sc
0.188 (0.022) SA
0.167 (0.010) SF,SA
0.186 (0.015) SA
0.183 (0.013) SA
0.215 (0.015) C,SA
0.256 (0.053) T,C,Sc,A,SF
0.304 (0.071) SA
0.232 (0.070) SF,SA
0.303 (0.055) SA
0.307 (0.047) SA
0.375 (0.096) C
0.419 (0.089) T,C,Sc,A
0.408 (0.081) Sc,A,SF
0.446 (0.046) Sc,A,SF
0.332 (0.098) T,C,SA
0.328 (0.047) T,C,SA
0.314 (0.056) T,Sc,SA
0.464 (0.110) Sc,A,SF
0.851 (0.062) C
0.913 (0.067) T, A, SA
0.832 (0.043) C
0.825 (0.035) C
Univariate ANOVAs revealed many significant differences in these indices (Table 5); these results are summarized below and illustrated in Figure 2. Terrestrial carnivorans had generally intermediate features when compared to the other locomotor groups, but did have more elongate metatarsals (higher PES) than the scansorial, arboreal, and semifossorial groups. Cursorial carnivorans are characterized by distal elongation of the forelimb [high brachial index (BI), near 1.0, Table 5], elongate metacarpals and metatarsals (low MANUS, high PES), smaller humeral and femoral epicondyles with elongated humeri and femora [low HEI, femoral epicondylar index (FEI)], and generally gracile limbs (low HRI, URI, FRI, TRI). Scansorial and arboreal carnivorans are characterized by having primarily intermediate features, but arboreal carnivorans did show relatively elongate manual phalanges (high MANUS). Semifossorial carnivorans displayed relatively robust limbs, with enlarged humeral epicondyles and short humeri (high HEI), elongate olecranon process (high OLI), and clearly elongated front claws (high CLAW), relative to the pedal phalanges. Semiaquatic species are characterized by relatively enlarged humeral epicondyles and femoral epicondyles and short femora (high HEI and FEI), enlarged olecranon process (high OLI), enlarged tibial spine (high TSI), and elongate metatarsals (high PES).
Discriminant Function Analyses
Stepwise DFA of locomotor categories was performed using the functional indices for each extant species, as well as extinct carnivorans included as unclassified cases. Only three of the 16 indices are included in the stepwise discriminant model (Table 6). The DFA separated groups fairly well and was significant (Wilks' λ = 0.151, F(5,97) = 12.018, P < 0.001), but given the nature of the samples used, and our simplified categories representing a complex continuum of locomotor behaviors, there was some predictable overlap among groups, particularly the terrestrial, scansorial, and arboreal groups. The analysis yielded two discriminant functions with eigenvalues near 1, these accounted for 88.5% of the variance in the data set. One species, Bdeogale crassicauda (#39), was not complete enough for inclusion in the discriminant model.
Table 6. Discriminant analysis structure matrix, eigenvalues, percent variance explained, and Wilk's λ for each function
% variance explained
The first discriminant function (DF1) accounted for 52.9% of variance and primarily separated the cursorial and terrestrial from the semifossorial and semiaquatic groups (Fig. 3). The first discriminant function was positively correlated with BI and negatively correlated with FEI. All cursorial species had highly positive scores, nearly all terrestrial species had positive scores, scansorial and arboreal species had slightly positive or negative scores, and semifossorial and semiaquatic species had slightly to highly negative scores (Fig. 3). Positive DF1 scores for cursorial taxa reflect their tendency toward distal elongation of the limbs, particularly relatively elongate radii (high BI). Negative scores for semifossorial and semiaquatic taxa reflect their broader femoral epicondyles (high FEI).
The second discriminant function (DF2) accounted for 35.6% of variance and primarily separated semiaquatic and cursorial groups from all other groups (Fig. 3). The second discriminant function was positively correlated with pes length index (PES) and, to a lesser degree, FEI. All cursorial species had positive scores, semiaquatic species had slightly negative to highly positive scores, terrestrial species had positive or negative scores, and all other groups had exclusively negative scores (Fig. 3). Positive DF1 scores for cursorial taxa reflect the distal elongation of their hind limbs, including particularly elongate metatarsals (high PES). This feature is also evident in the high positive scores of semiaquatic mammals that use hind limb paddling locomotion; this locomotor mode is also facilitated by shortened femora and enlarged hind limb musculature (high FEI).
The ability of the discriminant model to separate carnivoran species into locomotor groups was assessed using the classification phase (Table 7). This classification showed 65.1% correct classification of species, with 60.4% correct classification when cross-validated (where individual species are excluded and then classified using models derived from the remaining specimens). Classification success was much better for some locomotor groups than others. More than 90% of cursorial, arboreal, and semifossorial species are correctly classified, whereas only 36.8% of scansorial species are correctly classified. Of 106 species included in the analysis, a total of 37 are misclassified (Appendix, Table A2).
Sixteen extinct species are complete enough to be included as unknowns in the classification phase of the analysis (Table 8). The amphicyonid Daphoenus vetus was classified as terrestrial. Among early canids, Hesperocyon gregarius was classified as arboreal, while Archaeocyon leptodus, Desmocyon thomsoni, Mesocyon coryphaeus, and Phlaocyon leucosteus are classified as terrestrial. The Pleistocene canids Canis armbrusteri and the dire wolf, C. dirus, are both classified as cursorial. Among felids, the American lion, Panthera atrox, and the two saber-tooth cats, Homotherium ischyrus and Smilodon fatalis, are classified as terrestrial. The Eocene carnivoramorph Procynodictis vulpiceps was classified as arboreal. The Pliocene otter Satherium piscinarium was classified as semiaquatic. The saber-toothed nimravids, Dinictis felina and Hoplophoneus primaevus, are both classified as arboreal. Nimravus brachyops, another nimravid, was classified as terrestrial. The short faced bear, Arctodus simus, was classified as semifossorial.
Table 8. Discriminant analysis classification of extinct species
Most likely group
Second most likely group
P(D|G) represents the conditional probability of the observed canonical score, given membership in the most likely group. P(G|D) represents the posterior probability that a case belongs to the predicted group, given the sample used to create the discriminant model.
The reliability of the discriminant analysis is further supported by comparison of 12 carnivorans from Rancho La Brea (RLB) to their modern counterparts (Fig. 4). Eight of these carnivoran species from RLB survive today, and all RLB samples are classified the same as their modern counterparts. However, it should be noted that the Pleistocene coyote, Canis latrans orcutti, and mountain lion, Puma concolor, diverge noticeably from extant samples. Canis latrans orcutti from RLB displays more elongate hind feet than extant samples, PES = 0.491 vs. 0.434. Mountain lions from RLB show proportionately longer radii than extant samples, BI = 0.911 vs. 0.835.
Convergence in morphology among groups with similar locomotor habits can be seen in Figure 5, which shows discriminant function scores mapped onto a composite phylogeny of the species studied. Despite being only distantly related, the cursorial and terrestrial hyaenids, canids, and felids studied have relatively elongate, gracile limb proportions, and positive DF1 scores (Fig. 5a). The semiaquatic otters and semifossorial mephitids and badgers studied all show relatively robust limbs and have negative DF1 scores (Fig. 5a). Arboreal members of eight distinct families have negative DF2 scores (Fig. 5b). Cursorial and terrestrial carnivorans from multiple families show slightly positive DF2 scores, while the most aquatic otters have strong positive scores (Fig. 5b).
Most carnivorans run, climb, dig, and swim among their regular activities, and perform these behaviors capably. Among extant families, species have independently evolved morphological specializations that allow them to more effectively perform these behaviors. Carnivoran species studied here that exhibit similar locomotor behaviors, but are distantly related, generally show similar morphological deviations from less specialized members of their families. The observed differences between locomotor groups reveal aspects of morphology associated with particular behaviors, and allow classification of group membership with some confidence. The primary goal of this study is inference of extinct species' locomotor habits; thus, the following discussion will begin by covering which morphological features can be used to discriminate locomotor groups and the accuracy of classifications. Then, we examine locomotor specializations, both in terms of how specialists deviate from their more generalized relatives and the functional performance of morphological features. Finally, we discuss the inferred locomotor habits of some extinct carnivorans, and their evolution.
These analyses revealed differences between carnivorans with different locomotor habits, and highlight relationships between skeletal form and locomotor function. The ANOVA results and box plots (Table 4, 5, and Fig. 2) indicate many significant differences between locomotor groups and suggest that relatively simple measurements can yield useful information about the locomotion of species. However, many of the features that distinguish locomotor groups are not exclusive to a single group and similar scores for an index may not indicate similar habits. For example, both cursorial and semiaquatic species have elongation of the pes, but for different functional reasons. Because many extinct species are only known from incomplete specimens, our results suggest that caution should be exercised in inference of locomotor habits from a few measurements.
The DFA results showed fairly good separation of locomotor groups (Fig. 3). The first discriminant function separated robust, power adapted semifossorial and semiaquatic carnivorans from more gracile, elongate cursorial and terrestrial taxa. This function primarily relates to BI and FEI, which are associated to distal elongation of the arm and epicondylar breadth of the femur. The second discriminant function separated the cursorial and semiaquatic taxa from the other groups. This function primarily relates to PES; both cursorial and specialized semiaquatic carnivorans are characterized by an elongate pes.
Classification of species to locomotor groups was fairly accurate, with 65.1% correct overall (Table 7). Several groups are particularly well classified, with greater than 90% accuracy for cursorial, arboreal, and semifossorial species. Terrestrial and scansorial species are not successfully classified, perhaps due to the relatively unspecialized nature of these locomotor modes (Appendix, Table A2). For example, the terrestrial arctic fox (Alopex lagopus) and maned wolf (Chrysocyon brachyurus) are both classified as cursorial; this is likely attributable, at least in part, to the relatively cursorial ancestry of extant canids. Several small and gracile felids are also classified as cursorial rather than terrestrial or scansorial. Many of the terrestrial and scansorial mustelids are misclassified as arboreal (Appendix, Table A2), possibly a function of their small body size and relatively unspecialized limb structure.
Consideration of individual families reveals the analysis performed well in classification of species from some families and not others. Appendix, Table A2 shows greater than 75% correct classification for the Canidae, Eupleridae, Herpestidae, Hyaenidae, and Mephitidae; whereas there was 50% or less correct classification for the Felidae, Mustelidae, and Ursidae. In general, families with species characterized by similar morphology and specialized locomotor habits (ex. Canidae and Mephitidae) were classified correctly, both in the original and cross-validated analyses. Families with more diverse habits, including more generalized terrestrial and scansorial species (ex. Felidae and Mustelidae) were classified very poorly. These generalized locomotor modes likely do not require morphological specialization to a degree that this would outweigh differences resulting from the separate phylogenetic histories of these families.
The overall mixed success in classifying living species means that caution should be used in applying this method to extinct species. The phylogenetic histories of species, their body sizes, and their relative degrees of specialization have a definite influence on whether or not they are classified correctly. Studies attempting to infer the habits of extinct taxa would be most reliable if they include broad taxonomic sampling and consideration of all of these factors. Application of these methods to a large sample from a single family would likely do very well in correctly classifying species' ecology, but may give misleading results for extinct species that may have had very different locomotor habits than any extant members of the family. For example, analysis of just the canid family would probably show high classification success, but would not necessarily provide realistic classification of an early canid like Hesperocyon that has dramatically different limb proportions than any living species.
Functional Analysis of Locomotor Modes
This was one of the least well-classified groups. Terrestrial carnivorans are capable of many different types of locomotion, but are not adapted for any one in particular. Many of the smaller carnivorans, including small mustelids, herpestids, and canids, fall in this group, and these species are capable of climbing, swimming, and digging without particular morphological specializations. Many of the larger cats also fall into this category, as they are built for short bursts of speed, and yet retain the ability of forelimb pronation and supination for prey capture (Gonyea,1978). Terrestrial species generally show intermediate features and had few significant differences from other groups (Table 5). This group was characterized by both positive and slightly negative DF scores. As this group was often misclassified, it is clear that the “terrestrial” morphology can fall at the edges of any of the other categories, which may be influenced by phylogeny or by function.
Previous studies of cursorial mammals, including cursorial carnivorans, have found this locomotor mode to be associated with distal elongation of the limbs, gracile limb elements, elevated foot posture, and muscle insertions close to joints (Howell,1944; Gambaryan,1974; Hildebrand,1985a; Van Valkenburgh,1987; Garland and Janis,1993; Stein and Casinos,1997; Carrano,1999). Cursorial carnivorans are characterized by distal elongation of the forelimb (high BI), elongate metacarpals and metatarsals (low MANUS, high PES), smaller humeral and femoral epicondyles (low HEI, FEI), and generally gracile limbs (low HRI, URI, FRI, TRI). Despite disparate ancestry, the cheetah, gray wolf, African hunting dog, and brown and striped hyenas show very similar limb proportions (Fig. 5). Although BI does not effectively separate all locomotor groups, cursorial taxa consistently display relatively elongate radii (BI scores near 1.0, Table 5), which increases their stride length and thus speed. The intermembral index of cursorial carnivorans is also closer to 1.0 than in other groups (Table 5), indicating more equal limb lengths; however, it must be noted that this is likely driven by the relatively elongate arms of hyaenids, all of which have IM scores at or above 1.0. As cursorial carnivorans all display digitigrade foot posture, elongation of the metacarpals and metatarsals also increases stride length, while the phalanges contribute little to effective limb length while increasing inertia (Van Valkenburgh,1987). Gracile limb elements reduce the inertia of the limb, allowing increased rotational velocity and thus stride rate (Gambaryan,1974; Hildebrand,1985a).
Scansorial and Arboreal
Many carnivorans readily climb, some forage on vegetation in trees (e.g., red pandas and palm civets) and others actively hunt in trees (e.g. clouded leopards and fossa). Previous studies have found adaptations that facilitate climbing in mammals include increased joint mobility, as well as proximal and distal elongation of the limbs (Cartmill,1985; Heinrich and Biknevicius, 1998; Argot,2001,2002; Sargis,2002a,b; Wright,2007). Scansorial and arboreal carnivorans are characterized by having primarily intermediate features, but arboreal carnivorans display relatively elongate digits (high MANUS, Table 4, and Fig. 2). Elongate digits facilitate grasping, which is important for climbing (Wood Jones,1953; Hildebrand,1978; Van Valkenburgh,1987). These species typically have equal claw lengths on the manus and pes (CLAW, Table 5), as is expected in species using both fore and hind limbs in climbing (Van Valkenburgh,1987).
Carnivorans, like most mammals, dig via scratch-digging, which requires production of large forces by the forelimbs and claws capable of breaking apart soil (Hildebrand,1985b, Lagaria and Youlatos,2006). This is accomplished by shortening the forelimb (reducing out-lever) and enlarging muscular attachments (increasing both in-lever and the area of insertion), which results in improved mechanical advantage for the digging muscles (Hildebrand,1985b). A manus with enlarged terminal phalanges and rapidly growing claws allows for effective breakdown of soil (Hildebrand,1985b; Stein,2000). Semifossorial carnivorans displayed relatively robust limbs, with enlarged humeral epicondyles (high HEI), elongate olecranon processes (high OLI), and elongate front claws (high CLAW; Table 5, Fig. 2); this is similar to the results found for digging rodents (Samuels and Van Valkenburgh,2008). Enlargement of the humeral epicondyles, particularly the medial epicondyle, increases the areas of origin for the manual and digital flexors (flexor digitorum profundus, flexor carpi ulnaris, flexor carpi radialis, and palmaris longus) and the forearm flexing pronator teres (Davis,1964). The elongate olecranon processes increase the areas of insertion as well as the in-lever of the elbow extending triceps brachii and anconeus (Davis,1964; Hildebrand,1985b; Van Valkenburgh,1987; Schutz and Guralnick,2007). Van Valkenburgh's (1987) examination of claws showed similar results, with semifossorial carnivorans characterized by elongate fore claws, while arboreal and cursorial species had equally proportioned fore and rear claws.
Among carnivorans studied, swimming is drag based, which is facilitated by 1) increasing the length of the paddling limb, 2) increasing the paddle's surface area, 3) increasing thrust produced during the power stroke, and 4) minimizing drag produced by the limb during the recovery stroke (Howell,1930; Fish,1996; Gingerich,2003). Semiaquatic species are generally characterized by relatively enlarged humeral epicondyles and femoral epicondyles (high HEI and FEI), enlarged olecranon processes (high OLI), enlarged tibial tuberosities (high TSI), and elongate metatarsals (high PES; Table 5 and Fig. 2); all of these features correspond to those modified in hindlimb paddling semiaquatic rodents (Samuels and Van Valkenburgh,2008). The large humeral epicondyles and olecranon processes of semiaquatic carnivorans may reflect enlargement of muscles used in turning and stabilizing the body while swimming (Taylor,1989; Fish,1994; Schutz and Guralnick,2007). The high FEI scores result from a combination of femoral shortening and enlargement of the areas of insertion for several muscles. Femoral shortening brings the paddling limb closer to the body, and thus reduces induced drag during the recovery stroke. The gastrocnemius muscles originate on the lateral and medial epicondyles of the femur and act in plantar flexion (Davis,1964), an important part of the power stroke. The enlarged tibial tuberosity increases the area of insertion for some of the limb retracting hamstring muscles (semitendinosus, pelvic head of biceps femoris; Davis,1964), the action of these muscles generates propulsion during the power stroke of hindlimb paddling species.
Although all carnivorans studied use drag-based swimming, there are some important variations in swimming modes and appendages used to displace water (Fish,1996). Otters, which represent seven of 11 semiaquatic species in this analysis, swim using simultaneous paddling of the pelvic limbs, supplemented by dorsoventral undulation of the spine. Mink (Neovison vison #80; Dunstone,1979) and polar bears (U. maritimus #95) use alternate paddling of the pectoral limbs. The misclassification of these two by the discriminant analysis may be linked, in part, to this difference in swimming mode. As some semiaquatic species analyzed show hindlimb dominated locomotion and others forelimb dominated, it is not surprising to find dissimilar limb structures in species with these two swimming modes (Webb and Blake,1985).
Previous researchers have inferred the locomotor habits of many extinct carnivorans, but most of these were based on qualitative assessments of morphology and environments (examples include Merriam,1912; Scott and Jepsen,1936; Kurtén,1967; Stevens,1991; Wang,1993; quantitative exceptions include Van Valkenburgh,1987; Janis and Wilhelm,1993). Most of the studied taxa from the Pleistocene are relatively closely related to extant species and their classifications are likely to represent actual ecologies fairly well; however, some of the species here are from the Eocene and Oligocene, far removed from any extant carnivorans. Many of the classifications of extinct carnivorans seem to fit what paleontologists have inferred about these species in past studies; however, a few results are puzzling.
The RLB species are interesting to examine as a group (Fig. 4) because the locomotor adaptations of these carnivorans give a clearer picture of what the Pleistocene environment at the tar pits was like. Pumas and coyotes from RLB seem to shift in the same direction in morphospace with more elongate limbs, possibly indicating a more cursorial locomotor style in the Pleistocene compared to modern species (Meachen and Samuels,2012). Pleistocene brown bears (Ursus arctos) appear to be more scansorial than recent brown bears, falling closer to extant black bears (#94). Mephitis mephitis also shows a morphospace shift, but not as pronounced as in the larger carnivorans, appearing less robust than extant striped skunks (#56). Similarly, A. simus is in a completely different morphospace than any living bear species, showing what appear to be digging adaptations. Conversely, grey wolves, foxes, badgers, bobcats, and lions are functionally indistinguishable from their extant counterparts. The species at RLB that are indistinguishable from their modern counterparts allow us to be fairly confident that the extinct species in our analysis are being accurately classified, with classification errors being attributable to real differences in limb structure.
The Eocene carnivoramorph P. vulpiceps (formerly “Miacis” gracilis) was classified as arboreal. This is congruent with a previous study that postulated early carnivorans like Procynodictis were arboreal or scansorial based on qualitative analyses of postcranial characters (Wang,1993; Heinrich and Houde,2006). The limb proportions of Procynodictis are generally similar to the extant linsang (Prionodon #84; Fig. 3, Supporting Information, Fig. S1), which is arboreal.
The amphicyonid D. vetus was classified as terrestrial, with limb proportions similar to the extant large Indian civet (#106) and Pallas cat (#33). Scott and Jepsen (1936) described the skeleton of Daphoenus as more similar to cats than dogs, suggesting it was less cursorial. Van Valkenburgh (1987) classified Daphoenus as scansorial based on the curvature of the ungual phalanx, phalanx to metacarpal ratio, and olecranon curvature. Given the generally ancestral structure of early amphicyonids, it is not surprising to find their location in morphospace where the terrestrial and scansorial/arboreal groups overlap.
The saber-toothed nimravids, D. felina and H. primaevus, are both classified as arboreal, similar to Van Valkenburgh's (1985,1987) results. These classifications may be a result of the true ecology of nimravids, but may also be attributed to their unique prey-killing behavior (Martin,1989,1998a,b). Nimravids show prey-killing characters similar to living cats and extinct saber-toothed cats, but to a greater extent. These extinct creatures most likely coupled an arboreal or scansorial lifestyle with increased forelimb strength to protect their elongated canines from damage during prey capture (Meachen-Samuels,2012), with the end-result being extreme arboreal adaptations relative to living carnivorans. Unlike the other two nimravids in the analysis, N. brachyops was classified as terrestrial, with limb proportions similar to the extant mountain lion (P. concolor, #37), which was also classified as terrestrial. The Nimravus specimens examined display more gracile features than either Dinictis or Hoplophoneus, including relatively elongate radii and metapodials, and narrow humeral epicondyles (Supporting Information, Fig. S1). These postcranial differences, coupled with relatively small sabers, potentially reflect a different hunting strategy, more like extant felids, that allowed it to avoid competition with other nimravids in the Oligocene (Meachen-Samuels,2012).
The barbourofelid, Barbourofelis loveorum, was also not complete enough for inclusion in the discriminant analysis, but its limb proportions are similar to some extant felids and some extinct saber-tooth cats and nimravids. Barbourofelis displays relatively low BI scores, broad humeral epicondyles, elongate olecranon processes, short metacarpals, and short tibial tuberosities (Supporting Information, Fig. S1). All of these forelimb characters are consistent with increased mechanical advantage and well-developed prey-grappling abilities in this species. Subduing and restraining prey with the forelimbs before delivering a precise killing bite appears to be a characteristic of saber-tooth species with thin, elongated canines (Meachen-Samuels,2012). Barbourofelis shows convergent evolution with S. fatalis with extremely elongated upper canines and robust “bear-like” forelimbs (Anyonge,1996; Meachen-Samuels,2012).
Among early canids, the hesperocyonine Hesperocyon was classified as arboreal, while Mesocyon coryphaeus and the borophagines Archaeocyon and Phlaocyon are classified as terrestrial. Scott and Jepsen (1936) described the skeletal proportions of H. gregarius as being more similar to viverrids and herpestids than extant canids. Wang (1993) analyzed the limb morphology of Hesperocyon, noting relatively primitive limb proportions and a plantigrade foot posture. Wang classified Hesperocyon as a transitory species between arboreal and cursorial locomotor styles, representing an intermediate form between the earliest carnivorans and the later cursorial canids. Similarly, Stevens' (1991) study of the early hesperocyonine Paraenhydrocyon wallovianus led to its interpretation as being scansorial. The Oligocene hesperocyonine M. coryphaeus was classified as terrestrial by the discriminant analysis and displayed limb proportions most similar to the extant raccoon dog (Nyctereutes procyonoides #14) and several herpestidae species (Supporting Information, Fig. S1). Mesocyon is significantly larger and more carnivorous than earlier canids (Van Valkenburgh,1991), its limb proportions may reflect ambush hunting on the ground in the wooded environments it inhabited. Wang et al. (1999) also considered the early borophagines Archaeocyon and Phlaocyon to be scansorial, like Hesperocyon, based on qualitative characteristics of the postcrania. The larger and later borophagine, D. thomsoni, was classified as terrestrial; in the discriminant analysis, it fell between the extant gray fox and arctic fox (Fig. 3). Munthe (1989) analyzed the postcrania of borophagine canids finding earlier members of the group to be terrestrial and not well adapted for a cursorial lifestyle; Janis and Wilhelm (1993) had similar results in an examination of metatarsal/femur ratio. The Pleistocene canids C. armbrusteri and the dire wolf, C. dirus, are both classified as cursorial. This was also suggested by several other studies (Merriam,1912; Nowak,1979; Tedford et al.,2009).
Among felids, the American lion, P. atrox, and the two saber-tooth cats, H. ischyrus and S. fatalis, are classified as terrestrial. Many of the larger cats are scansorial (e.g., jaguar, leopard), but the largest living pantherines (lions and tigers) are not known to climb frequently (Schaller,1967,1972). Therefore, it is not surprising that even larger cats would be classified as terrestrial. Although all of these felids are classified as terrestrial, they all differed slightly in their locomotor modes, likely chasing prey to different degrees. According to Anyonge (1996), the cross-sectional dimensions of the long bones place both P. atrox and S. fatalis as ambush predators. However, these species likely differed in hunting behavior, P. atrox may have chased prey for a short distance, whereas the extremely robust forelimbs (Supporting Information, Fig. S1) and the short tail of S. fatalis suggest that it did not chase prey at all, but leapt from cover onto an unsuspecting prey animal (Turner and Antón,1997; Meachen-Samuels and Van Valkenburgh, 2009;2010; Martin et al.,2011). Anyonge (1996) classified Homotherium serum as a cursorial predator, chasing prey for intermediate distances somewhere between a living lion and a wolf. H. ischyrus has longer limbs and been called more cursorial than H. serum, but still less cursorial than the living cheetah or canids (Hearst et al.,2011), though it is certainly not more specialized for running than the extant lion (Fig. 3, Supporting Information, Fig. S1). The American cheetah, Miracinonyx inexpectatus, was not complete enough for inclusion in the discriminant analysis, but shows relatively elongate and gracile limb proportions consistent with cursoriality (Supporting Information, Fig. S1), as had been previously demonstrated (Van Valkenburgh et al.1990).
The Pliocene otter S. piscinarium was classified as semiaquatic, with similar DF scores to Enhydra lutris and Pteronura brasiliensis (Fig. 3, Supporting Information, Fig. S1). This classification seems likely due to the striking morphological similarity between Satherium and Pteronura brasilensis, which has been recognized previously (Bjork,1970; Prevosti and Ferrero,2008). Satherium's limb proportions are consistent with hindlimb paddling locomotion, as is typical of extant otters. Given the phylogenetic position of Pteronura (Koepfli et al.,2008), this is likely what the ancestor of all extant otters looked like. The other extinct mustelid in our analysis, Martes nobilis (Martes americana nobilis in Anderson,1994; Hughes,2009; Lyman,2011), shows limb proportions nearly identical to the extant American pine marten (M. americana, #71), though it is slightly larger (Supporting Information, Fig. S1, see also Anderson,1970). These proportions suggest scansorial or arboreal habits like extant members of the genus. Lyman's (2011) faunal analysis found taxa that co-occurred with M. nobilis to be characteristic of open meadow or grassland environments. If this species was living in open environments, its limbs show little or no adaptation for more cursorial or fossorial habits.
The short-faced bear, A. simus, was classified as semifossorial, but had previously been described as being more cursorial than extant bears (Kurtén,1967). It is possible that the short-faced bear was more herbivorous (Sorkin,2006; Figueirido et al.,2010; Meloro,2011b) and used its forelimbs to dig for food (e.g., roots, tubers, ground squirrels). However, stable isotope studies from Beringia have found that Arctodus incorporated large quantities of meat in its diet (Bocherens et al.,1995; Matheus,1995; Fox-Dobbs et al.,2008). It may be the case that this bear was using its forelimbs to grapple with large prey, as Kurtén (1952) describes of Scandinavian brown bears. The bear may have run alongside the prey for a short time, and then used its forelimbs to immobilize it, with one paw over the shoulder and the other over the nose of the prey to expose the neck for a killing bite (Martin et al.,2011). This predatory behavior would increase the muscle masses of the flexor and extensor digitorum muscles and consequently the humeral epicondylar width, which is evident in Supporting Information, Fig. S1. Figueirido et al. (2010) recently demonstrated that Arctodus did not have particularly long legs for a bear of its size, and our results demonstrate its limb proportions are not particularly well adapted for cursoriality. This bear does not display distal elongation of the limbs and has relatively broad humeral and femoral epicondyles (Supporting Information, Fig. S1), which are characteristic of diggers. Interestingly, the large and highly predatory polar bear, U. maritimus, was also misclassified as semifossorial by the discriminant analysis, and displays overall limb proportions very similar to Arctodus. The other large bear in our analyses, Agriotherium africanum, was not complete enough for inclusion in the discriminant analysis. The elements of Agriotherium analyzed do show some interesting differences from other bears, including the presence of relatively short olecranon processes and elongate metatarsals (Supporting Information, Fig. S1). Agriotherium also has relatively elongate tibia, with a crural index value (0.889) much higher than extant bears (mean = 0.756). These features may indicate that this bear was more cursorially adapted than other ursids.
Phylogeny and Convergence
Recent studies of craniodental shape in carnivores have found morphology to be shaped by both ecology and phylogeny (Friscia et al.,2006; Figueirido et al.,2011; Meloro and O'Higgins,2011; Meloro et al.,2011). Similarly, our results show definite morphological convergence for adaptation to different locomotor modes, while the imprint of phylogenetic history remains. Limb structure does reflect phylogeny in carnivorans (Table 4); in general, closely related carnivoran species have similar morphology and locomotor habits, and in many cases this is likely a consequence of having a shared ancestor with similar ecology. Phylogenetic effects on morphology can be superceded by ecomorphological adaptations, particularly when functionally demanding ecological habits promote specialization (Meloro et al.2011). The MANOVA and discriminant analyses used here show the effects of locomotor ecology on morphology, and allow the identification of many cases of convergent evolution.
Distantly related groups with similar locomotor habits also display similar morphologies, showing clear functional convergence (Fig. 5). Cursorial and terrestrial canids, hyaenids, and felids all have more positive DF1 and DF2 scores than their relatives with other locomotor habits. This suggests convergent adaptations in these groups to deal with the functional demands of running (Garland and Janis,1993). The independently semifossorial mephitids and badgers both show negative DF1 and DF2 scores. As digging can place some extreme functional demands on an organism, it requires particular specializations of the postcranial skeleton (Hildebrand,1985b). Multiple carnivoran families studied have arboreal species, which have negative DF2 scores. Given the phylogenetic histories of these families, many must have independently evolved climbing adaptations. However, some might retain morphological characteristics of primitive carnivorans, which may have been arboreal (Heinrich and Rose,1995;1997; Heinrich and Houde,2006).
Several carnivoran families are characterized by diverse morphologies and locomotor habits, allowing assessment of convergence among species. Within the Canidae, extant and extinct species generally have positive DF1 and DF2 scores, characteristic of the terrestrial and cursorial groups (Fig. 5). The gray foxes (Urocyon cinereoargentius and U. littoralis), the raccoon dog (N. procyonoides), and the bush dog (Speothos venaticus), which all inhabit forests and have been documented climbing, have lower DF1 scores than their more terrestrial and cursorial relatives. The earliest canid in the analysis, Hesperocyon gregarious, has low DF1 and DF2 scores and was classified as arboreal. The other hesperocyonine and borophagines in the analysis were classified as terrestrial, suggesting adaptation for a terrestrial existence relatively early in the history of the family, at least by the mid Oligocene. The more scansorial, woodland adapted canids alive today likely represent a secondary return to these habits, after divergence from more terrestrial and cursorial ancestors.
Among felids, the several independently terrestrial/cursorial and arboreal species studied show similar divergence from their more generalized relatives (Fig. 5). The terrestrial serval (Caracal serval), Canadian lynx (Lynx canadensis), and lion (Panthera leo), and the cursorial cheetah (Acinonyx jubatis) have higher DF1 and DF2 scores, while the arboreal clouded leopard (Neofelis nebulosa), margay (Leopardis wiedii), and marbled cat (Pardofelis marmorata) have the most negative DF scores. The bobcat (Lynx rufus) is scansorial, but has morphology very similar to the lynx; however, fossil evidence suggests both are derived from a larger, shorter-legged lynx-like ancestor (Werdelin,1981). The extinct P. atrox, H. ischyrus, and S. fatalis were classified as terrestrial, with DF scores within the ranges of extant felids (Fig. 3). The limb proportions of P. atrox and H. ischyrus are similar to extant pantherines (Supporting Information, Fig. S1); S. fatalis, on the other hand, has limb elements that were generally more robust and less elongate than extant felid species, and some of the muscle attachments were substantially enlarged (Meachen-Samuels,2012).
The nimravids have commonly been considered ecological analogs of the felids, sharing highly similar cranial and postcranial morphology. Two of the nimravids studied (D. felina and H. primaevus) are characterized by particularly short robust legs, and were classified as arboreal. N. brachyops was classified as terrestrial and shows dramatic divergence from its arboreal relatives, with limb structure similar to that seen amongst felids (Figs. 3 and 5).
Though not particularly species rich, members of the Eupleridae from Madagascar have fairly diverse morphology and locomotor habits (Figs. 3 and 5). The highly arboreal fossa (Cryptoprocta ferox) has the lowest DF scores, very similar to the binturong (Arctictis binturong) and tayra (Eira barbara). In contrast, the terrestrial Malagasy civet (Fossa fossana) and broad-striped Malagasy mongoose (Galidictis fasciata) have DF scores more like terrestrial herpestids and viverrids. The nearest relatives of euplerids are the Herpestidae, which are all terrestrial (except for Suricata) and have very similar discriminant function scores.
Mustelids show a particularly broad range of limb morphologies and locomotor habits (Figs. 3 and 5). The four badgers analyzed, which belong to three different subfamilies most likely separate since the mid-Miocene (Koepfli et al.,2008), have similar morphology and DF scores to the distantly related Mephitidae. Semiaquatic lutrines (otters) have the lowest DF1 and DF2 scores of any taxa studied, diverging greatly from their closest relatives, the mustelines. Otters, which use hind-limb paddling locomotion, show a clear exclusive adaptation to semiaquatic life, distinct from other semiaquatic carnivorans which use other quadrupedal swimming modes (Tarasoff,1972). Limb adaptations in otters are very similar to those seen in hind-limb paddling rodents (Stein,1988; Samuels and Van Valkenburgh,2008). The close similarity of the Pliocene Satherium to extant otters suggests conservation of locomotor adaptations within the Lutrinae since their appearance in the late Miocene (Koepfli et al.,2008). The less specialized mustelid species from the Mustelinae and Guloninae are very morphologically similar, despite having long separate evolutionary histories, with gulonine mustelids commonly considered highly conservative (Anderson,1970).
In some cases, like the semiaquatic carnivorans studied here, the limbs show different solutions to the challenges presented by swimming and different degrees of specialization for aquatic life. Distinct swimming modes, and different morphological adaptations of the pelvic or pectoral limbs and spine, can yield similar swimming performance (Webb and Blake,1985). Diverse morphological adaptations can allow similar ecology (Losos,2011a); this has been demonstrated in many groups of vertebrates, with very different structural solutions to similar environmental challenges (Cartmill,1985, Hildebrand1985a;1985b; Stein,2000).
Ecologically similar carnivorans do not necessarily show identical limb morphology, but specialists tend to deviate similarly from their more generalized relatives. Overall, this may be an example of “incomplete convergence,” where similar selective conditions and adaptation may result in similar phenotype, but not necessarily to the degree that they completely outweigh preexisting differences between clades (Stayton,2006; Losos,2011a). This was also found by Meloro et al. (2011) in an examination of mandibular structure in carnivorans, where morphological variation was shaped both by phylogenetic history and ecological adaptations. Like that study, our results suggest taxonomic and ecological divergence of clades likely arose in parallel, resulting in some clades with relatively homogenous ecology (ex. Lutrinae and Mephitidae).
Some of the extinct species studied here show very different limb morphology than extant members of the same family, and are inferred to have very different locomotor ecology. Given the common occurrence of convergence in the carnivorans studied, use of phylogenies for ancestral state reconstruction of locomotor ecology is likely to give ambiguous or incorrect results. Ecological traits that evolve rapidly and often show convergence are generally not appropriate for ancestral state reconstruction (Losos,2011b). Fortunately for paleontologists, the methods used here show locomotor ecology of extinct species can be inferred with only a few measurements. Inclusion of fossil species with quantitatively inferred ecology in phylogenies should allow more detailed and accurate examinations of carnivoran evolution in future studies.
The ecomorphological techniques used here reveal how limb skeleton morphology reflects the locomotor habits of carnivorans. Carnivoran species from different families with similar locomotor habits display convergent morphological specializations. Extant members of these groups display many significant differences in limb structure, which will facilitate inference of ecology for extinct species, even from incomplete fossil specimens. Classification of locomotor groups for extant taxa was fairly accurate overall, and worked particularly well for specialized cursorial, arboreal, and semifossorial species. Phylogenetic effects and differences in locomotor modes can sometimes confound results, as is the case for semiaquatic species like hind limb dominated otters versus forelimb dominated mink and polar bear. The application of these methods to extinct carnivorans allows for a more quantitative and objective inference of their locomotor habits. This study has revealed a wide variety of locomotor habits represented in extinct species, highlights past locomotor diversity, and has shown some forms, like the robust saber-toothed nimravids, may differ from anything alive today. Future work will allow the examination of convergent evolution in different groups of carnivorans and will facilitate more accurate reconstructions of past ecosystems.
The following curators and collection managers kindly allowed access to specimens in their care: K. Molina (Donald R. Dickey Collection of the University of California, Los Angeles); J. Dines and X. Wang (Museum of Natural History of Los Angeles County); C. Shaw, A. Farrell, J. Harris (George C. Page Museum); B. Akersten and M. Thompson (Idaho Museum of Natural History); M. Brett-Surman and L. Gordon (National Museum of Natural History); A. Henrici (Carnegie Museum of Natural History); R. Eng (University of Washington Burke Museum). Discussion with B. Van Valkenburgh and comments by J. Balaban improved this paper. Comments and suggestions made by an anonymous reviewer also improved it substantially. We would like to thank M. Starck for his editorial assistance.
Table A1. Extant and extinct carnivore species used in this study, including: species numbers, common names, number of individuals measured, and locomotor group assignments.† References for locomotor groups follow below.
Table A2. Extant species misclassified by classification phase of discriminant analysis. Included are species number and name, their actual observed ecology, and ecology predicted by the discriminant analysis