Patterns of morphological integration in the appendicular skeleton of mammalian carnivores

DATA

The dataset comprised the following appendicular bones: scapula, humerus, radius, ulna, pelvis, femur, and tibia from 46 living species (Table 1). It is worth noting that only one side of the pelvic girdle (the innominate bone) was analyzed. Although special attention was paid to families with large representatives (e.g., felids, canids, hyaenids, and ursids), several species of mustelids, procyonids, and ailurids were also included to cover a wider range of carnivoran ecomorphological diversity. As indicated by complete epiphyseal fusion to diaphysis, only adult specimens were collected to avoid the effects of ontogenetic variation. Furthermore, only bones from the same side of an individual were collected to avoid the effects of left–right asymmetry. All the specimens analyzed (Table S1) were housed at the American Museum of Natural History (AMNH, New York) and the Natural History Museum (NHM, London).

GEOMETRIC MORPHOMETRICS

A set of 3D homologous landmarks was directly digitized onto each bone using a Microscribe G2X digitizer (Immersion Corporation). The landmarks (Fig. 2) were selected according to different anatomical criteria (Table S2). The 3D coordinates (x, y, z) of all the landmarks were imported into Excel using the Immersion software package (Immersion Corporation, http://www.3d-microscribe.com). The surface of each appendicular bone of a jaguar (Panthera onca [AMNH-139959]) and the pelvis of a snow leopard (Uncia uncia [AMNH-100110]) housed at the AMNH (Mammalogy Department) were scanned using a 3D-mobile surface scanner (Nextengine HD) and the ScanStudio Pro I software package. Our aim was to locate the landmarks digitized onto the sample specimens in the 3D models to transform them into each of the models derived from statistical analyses using a morphing procedure (Wiley et al. 2005; Drake and Klingenberg 2008; Schoenebeck et al. 2012; Singleton 2012; Martín-Serra et al. 2014a). The morphing procedure was performed with the Landmark software package from the Institute of Data Analysis and Visualization (IDAV 2002–2006).

A Procrustes fit (Dryden and Mardia 1998) was individually performed on the raw coordinates of the landmarks digitized on each bone using the MorphoJ software package (Klingenberg 2011) that removes the effects of rotation, translation, and scaling (Rohlf and Marcus 1993). Once the specimens were aligned, Procrustes coordinates and centroid size were both averaged by species. We analyzed species averages instead of specimens for the following reasons: (1) our main aim was to study the influence of functional adaptations in the patterns of morphological integration at an evolutionary level; for this purpose, interspecific covariation patterns are more likely to reflect these effects than intraspecific ones, which may also reflect developmental constraints in the appendicular skeleton (Klingenberg 2008); (2) furthermore, it is more appropriate to analyze species averages than specimens to avoid mixing the effects of intra- and interspecific covariation and allometry (e.g., see Fabre et al. 2014); and (3) the independent contrasts used to correct for phylogenetic effects on the patterns of covariation were performed using species averages.

ACCOUNTING FOR THE EFFECTS OF SIZE AND PHYLOGENY

We investigated the effects of both size and phylogeny on bone shape variation given that they can influence the degree of covariation between pairs of morphological structures.

A phylogenetic tree was assembled (Fig. 3) using the Mesquite software package (Maddison and Maddison, 2011) to test for the presence of a phylogenetic signal in the sample. The tree was based on the topology and branch lengths (million years before present) of previously published phylogenies (Koepfli et al. 2007; Nyakatura and Bininda-Emonds 2012). The tree topology was based on the phylogeny proposed by Nyakatura and Bininda-Emonds (2012). However, the well-developed phylogeny created by Koepfli et al. (2007) for the family Procyonidae was used to avoid polytomies.

A multivariate regression (Monteiro 1999) of Procrustes coordinates on centroid size was performed in each bone to investigate the influence of size on bone shape. A permutation test (10,000 permutations) was used to assess the statistical significance of all regressions (Drake and Klingenberg 2008). The shape of those bones that showed significant interspecific allometry was decomposed into residual and predicted components (Drake and Klingenberg 2008). These residuals were used in subsequent multivariate analyses to explore the morphological covariation that accounts for size effects (hereafter called size-free species averages data).

The MorphoJ software package (Klingenberg 2011) was used to perform a permutation test developed for multivariate analyses (Klingenberg and Gidaszewski 2010) to quantify the presence of phylogenetic signals in limb bone shape (Laurin 2004). This test compares the tree length calculated from shape data with the length of 10,000 random trees generated by permutations of the mean species shapes to the tips of the phylogenetic tree. The P-value obtained indicates the proportion of trees that result in a tree length equal to or less than the reference tree and also indicates the statistical significance of the phylogenetic signal (Gidaszewski et al. 2009; Klingenberg and Gidaszewski 2010).

These analyses provided a new dataset derived from the phylogenetic independent contrasts of the principal component (PC) scores (Felsenstein 1985), hereafter called size-free independent contrast data. We performed the independent contrast on the PC scores of all PCs because, while these are independent variables, Procrustes coordinates are intercorrelated and remove the same phylogenetic variance multiple times when they are used for independent contrast (P. D. Polly, pers. comm.). The phylogenetic signal of bone shape was also calculated on the PC scores instead of on the size-free species averages data for the same reason exposed above.

The SDs of the standardized contrasts were also obtained using the Mesquite software package (Maddison and Maddison 2011), allowing us to perform a multivariate regression (Monteiro 1999) of the contrasted PCs (through the origin) versus the SDs of the contrasts for each bone (following Díaz-Uriarte and Garland 1998). These regressions were performed with MorphoJ to determine whether there was a correlation between the contrasts of shape in relation to its SDs. A lack of correlation would indicate that the data satisfied the assumptions required by phylogenetic independent contrasts (Díaz-Uriarte and Garland 1998).

COVARIATION STUDY

Covariation between bones was evaluated using two-block partial least-squares (2B-PLS) analysis (Rohlf and Corti 2000), which is a multivariate method used to calculate covariation between two blocks of variables that are treated symmetrically without assuming a causal relationship (Zelditch et al. 2004). Specifically, it is used to calculate the linear combinations of the original variables (i.e., PLS or singular axes) that maximize covariance between the two blocks (Rohlf and Corti 2000; Zelditch et al. 2004).

The degree of covariation between each pair of bone shapes was quantified using the RV coefficient described by Escoufier (1973), extended in the context of morphological integration and modularity by Klingenberg (2009), and applied to morphometrics by several authors (e.g., see Goswami and Polly 2010b; Chamero et al. 2013; Figueirido et al. 2013; Klingenberg and Marugán-Lobón 2013; Piras et al. 2013; Fabre et al. 2014; Goswami et al. 2014). This coefficient measures the degree of covariation between two blocks of variables in relation to the total amount of variance of each block (Klingenberg 2009) and is calculated as:

where tr is the trace (i.e., the sum of the diagonal elements of a squared matrix), S₁₂ is the matrix of covariance between the two blocks (S₂₁ is the transpose), and S₁ and S₂ are the matrices of covariance within each of the two blocks, respectively (Klingenberg 2009).

A permutation test with 10,000 permutations was performed to assess the statistical significance of this coefficient against the null hypothesis of the complete absence of covariation between the two blocks of variables. This test randomly shuffles the observations of the subsets and recalculates the RV coefficients. The proportion of random RV coefficients equal to or higher than the actual RV coefficient yields a P-value that indicates the presence of statistically significant covariation. The MorphoJ software package (Klingenberg 2011) was used to perform these analyses.

We used 2B-PLS analyses and the RV coefficient to study the degree of covariation between pairs of bones within limbs and between limbs (Fig. 1A). The analysis of within-limbs covariation was restricted to comparisons of anatomically connected bones, leading to a total of six pairs: four for the forelimb (i.e., scapula-humerus, humerus-radius, humerus-ulna, and radius-ulna) and two for the hindlimb (pelvis-femur, femur-tibia). Within-limb integration was used as a “control” for comparison with the between-limbs integration patterns. Between-limbs covariation was analyzed by comparing pairs of serially homologous bones and functionally equivalent bones in the fore- and hindlimb. In the case of serially homologous bones, girdles (scapula-pelvis), both stylopods (humerus-femur), and two zeugopods (radius-tibia) were compared. The fibula was not included in the analysis because of its small size in carnivorans and thus the ulna-fibula comparison was not considered in the serial homology analysis. In line with Raich and Casinos (1991), Gasc (2001), and Schmidt and Fischer (2009), three pairs of functionally equivalent bones were also compared: ulna–tibia, scapula–femur, and humerus–tibia.

The set of 2B-PLS analyses and the RV coefficients were calculated using three different datasets: (1) from the average of the Procrustes coordinates by species, to explore shape covariation without accounting for the effects of phylogeny and size; (2) from the size-free species average data (i.e., residuals that only account for interspecific allometry); and (3) from the size-free independent contrast data (i.e., PCs that account for both interspecifc allometry and phylogenetic relationships). The previous section provides a detailed explanation of how these datasets were obtained.

An angular comparison between the respective PLS axes (Cheverud 1982; Klingenberg and Zimmermann 1992; Klingenberg and McIntyre 1998; Klingenberg and Zaklan 2000) was performed to test whether the main morphological changes that associate a given bone with each of the other bones are the same. Angular comparison was used to calculate the angle between the directions of two vectors—in our case, two PLS axes—and to compare them to the distribution of angles obtained by simulating 100,000 random vectors within an n-dimensional morphospace (n depends on the number of landmarks). The proportion of angles equal to or less than the one obtained for the actual vectors indicates the P-value that shows whether the shape changes explained by one PLS axis are statistically different from those explained by other axes (e.g., it shows whether the first PLS axis obtained for the radius in the humerus–radius comparison is similar to the one obtained in the radius–ulna comparison). The MorphoJ software package (Klingenberg 2011) was used for the PLS and angular comparisons.

TESTING THE INFLUENCE OF FUNCTIONAL SPECIALIZATION ON LIMB BONE INTEGRATION

The effect of functional specialization on the patterns of covariation in the appendicular skeleton of carnivorans was explored by classifying the species into two functional categories: cursors and noncursors (Table 1). Species with a functional similarity in their fore- and hindlimbs during locomotion were classified as cursors and species with forelimbs with more functions than their hindlimb, thus having a functional dissociation between both limbs, were classified as noncursors (see Introduction).

A bootstrapping approach was used to test differences in RV coefficients (see Fruciano et al. 2013 for a similar approach) between both ecological groups for all pairs of bones (i.e., comparisons within limbs, and between serially homologous and functionally equivalent bones) to investigate whether the limb bones of cursors and noncursors have different degrees of morphological integration. Given that the RV coefficient is directly comparable to a squared correlation coefficient (Josse et al. 2008; Klingenberg 2009), confidence intervals were obtained for the differences between any pairs of RV coefficients using the procedure developed by Chan (2009) for squared multiple correlation coefficients. The bootstrap samples were generated by randomly drawing N1 and N2 observations with replacement from cursors and noncursors, respectively. In each step, differences were calculated between the RV coefficients for cursors and noncursors. Obviously, differences close to zero suggest similarity in the RV coefficients of both groups. Using a sample of 10,000 differences, 90 and 95% bootstrap standard confidence intervals and bootstrap percentile intervals were constructed. The null hypothesis of equality between the two populations of RV coefficients was rejected if the zero value was outside these intervals (see Manly 2007: p. 62). The Mathematica software package (Wolfram, 2005) was used to perform the bootstrapping procedures.

Different Procrustes alignments for each functional subsample were calculated and the influence of size and phylogeny was again tested for each individual subsample. Similarly, the 2B-PLS analyses were performed and their respective RV coefficients were also calculated for both subsamples independently, using the size-free species averages data and the size-free independent contrast data.

INFLUENCE OF BODY SIZE AND PHYLOGENY ON THE SHAPE OF APPENDICULAR BONES

Effects of interspecific allometry were significant for all bones in the whole sample, with the only exception of the tibia (Table 2). Furthermore, the phylogenetic signal in shape (PC scores) was statistically significant for all bones (Table 2). The regressions between the contrasts of shape on their respective SDs were not significant (all P values > 0.1). Therefore, the data satisfy the assumptions required for phylogenetic independent contrasts.

PATTERNS OF LIMB COVARIATION

Covariation of bones within limbs

The RV coefficients obtained from all comparisons of bones within both limbs were highly significant (P < 0.01). The RV coefficients obtained with the Procrustes coordinates (Fig. 4A, left) and the size-free coordinates for the species averages (Fig. 4B, left) were higher than those obtained with the size-free independent contrasts (Fig. 4C, left). Furthermore, forelimb bones had a greater degree of covariance than hindlimb bones and this difference increased when allometric and phylogenetic effects were taken into account (Fig. 4A–C; left column). Another important aspect of the integration patterns of the appendicular skeleton was that covariance between long bones was greater than between long and girdle bones of both limbs (Fig. 4A–C; left column).

Similarly, the percentages of covariance explained by the first PLS axes were slightly lower in all comparisons when the size-free independent contrasts were used (Table 3). In addition, these percentages were also lower for the girdle bones compared to the percentages obtained for the long bones (Table 3). Figure 5 shows the scatter plots of the first PLS axes of both blocks of variables and their associated morphological changes obtained from the analyses performed with the size-free species averages. The morphological change associated with these PLS axes was the degree of robustness in all bone comparisons (Fig. 5A–D, F), except for the pelvis in the case of the pelvis–femur comparison (Fig. 5E). In fact, as the femur becomes more robust, the pelvis develops a narrower iliac crest and both the ischium and the pubic symphysis become shorter (Fig. 5E).

Table 3. Percentages of covariance explained by the first axis obtained from each PLS analyses. Three series of PLS were performed: (1) with the Procrustes coordinates; (2) with the size-free species averages; and (3) with the size-free independent contrasts. S, scapula; H, humerus; R, radius; U, ulna; P, pelvis; F, femur; T, tibia

		Procrustes coordinates	Size-free species	Size-free contrasts
Within limbs	S-H	74.96	70.1	53.27
	H-R	90.56	91.49	85.41
	H-U	88.85	91.83	90.93
	R-U	88.2	95.33	92.34
	P-F	66.27	68.63	37.55
	F-T	93.38	93.29	72.73
Between serial homologues	S-P	60	64.51	54.14
	H-F	82.6	81.86	80.13
	R-T	96.15	97.69	84.18
Between functional equivalents	U-T	94.46	97.57	88.83
	S-F	66.68	62.29	54.57
	H-T	95.77	95.09	80.86

The distribution of taxa along all the PLS axes is very similar: most living canids (especially the maned wolf, Chrysocyon brachyurus) and some felids, such as the cheetah (A. jubatus), the bobcat (Lynx rufus), and the serval (Leptailurus serval), occupy the morphospace region that corresponds to slender bones (Fig. 5A–F). In contrast, the morphospace region for robust bones is occupied by species, such as the Northern river otter (Lontra canadensis), the European badger (Meles meles), the giant panda (Ailuropoda melanoleuca), and the sloth bear (Melursus ursinus; Fig. 5A–F). The only exception to this pattern is that the scapula of ursids is extremely robust—and that of felids is only moderately robust—in relation to the humerus, which leads to a slightly different distribution of taxa in the scapula–humerus comparison (see Fig. 5A).

Covariation of bones between serial homologues

All the RV coefficients obtained from the comparisons of serially homologous appendicular elements (i.e., girdles, stylopods, and zeugopods) were statistically significant (P < 0.01). Similar to the case of the within-limb analyses, the RV coefficients and the percentages of covariance explained by the first PLS axes (Table 3) obtained from the Procrustes coordinates (Fig. 4A, center) and the size-free species averages (Fig. 4B, center) were higher than those obtained with the size-free independent contrasts (Fig. 4C, center). Furthermore, the RV coefficients obtained from each comparison gradually increase from girdles to zeugopods (Fig. 4A–C, center).

Figure 6A–C shows the scatter plots of the first PLS axes obtained with the size-free species averages and the morphological changes associated with them. In the case of the scapula–pelvis comparison (Fig. 6A), the morphological change in the scapula is the degree of robustness. Regarding the pelvis, as the scapula becomes more robust, the iliac crest widens, the pubic symphysis and ischium both shorten, and the ischial tuberosity undergoes increased dorsolateral projection (Fig. 6A). In this axis, felids have slender scapulae and pelvises with narrow iliac crests. At the opposite extreme, ursids have robust scapulae and pelvises with wide iliac crests.

The analyses of the stylopod (humerus-femur) and the zeugopod (radius-tibia) were very similar (Fig. 6B, C). In both cases, the morphological change related to these axes was the degree of bone robustness. Thus, a more robust stylopod/zeugopod in the forelimb is associated with a more robust stylopod/zeugopod in the hindlimb, and vice versa. The distribution of species along these axes is very similar to the distributions obtained for the within-limbs comparisons: most canids occupy the morphospace region characterized by slender morphologies, whereas several ursids (especially A. melanoleuca and M. ursinus) and mustelids (M. meles and L. canadensis) occupy the region characterized by robust shapes.

INFLUENCE OF FUNCTIONAL SPECIALIZATION ON THE PATTERNS OF LIMB BONE SHAPE INTEGRATION

Table 2 shows the results of the regressions of shape on size for each functional subsample (cursorial and noncursorial taxa), and the results obtained for the presence of phylogenetic signal in the size-free shape variation. The graphic models shown in Figure 7 depict the RV coefficients obtained for all comparisons. The bootstrap test confirmed that there were significant differences in the RV coefficients between cursors and noncursors in some of the comparisons. Figure 8 depicts the results of this bootstrap test and shows the random distribution of the RV differences obtained from the size-free species averages as well as the 90 and 95% bootstrap percentile intervals for the anatomically connected bones (Fig. 8A–F), between serial homologues (Fig. 8 G–I) and functional equivalents (Fig. 8J–L). Table 4 also shows the bootstrap percentile and the confidence intervals at the 90 and 95% levels for the size-free independent contrasts.

None of the within-limb comparisons using the size-free species averages were statistically different from zero, as the null difference was within the 90 and 95% intervals in all cases (Fig. 8A–F). However, when the size-free independent contrasts data were used, the differences in RV values for the humerus–ulna, radius–ulna, pelvis–femur, and femur–tibia comparisons were significantly different from zero at the 90% confidence interval (Table 4). Although this is not the usual level of statistical significance, it may suggest the presence of a general trend.

Regarding the between-limb comparisons, the differences in the RV coefficient between the comparisons of the serially homologous scapula–pelvis, humerus–femur, and radius–tibia (Fig. 8H, 8) were not significant for the size-free species averages. However, in the case of the radius-tibia comparison, the null values for the differences in the RV coefficients were outside the 95% confidence intervals for size-free independent contrasts (Table 4). The differences in the RV coefficients between cursors and noncursors were not significant in the comparisons between functionally equivalent proximal bones (scapula–femur; Fig. 8K; Table 4). In contrast, the functional ulna–tibia comparison yielded significant results at the 90% interval when the species averages data were used (Fig. 8J), and at the 95% confidence interval when the independent contrasts data were employed (Table 4). Similarly, the humerus–tibia comparison was significant at the 95 percentile interval (Fig 8L) for the size-free species averages and at the 95% confidence interval for the size-free independent contrasts (Table 4).

PATTERNS OF LIMB COVARIATION REVEAL A HIGHLY INTEGRATED APPENDICULAR SKELETON IN CARNIVORANS

The results obtained from the PLS analyses and the RV coefficients show that the carnivoran appendicular skeleton is highly integrated to function as a whole. Furthermore, the angular comparison test indicated that shape covariance is related to the degree of bone robustness in all the comparisons, with the sole exception of the pelvis, which has a more complex shape variation. Thus, the axis of maximum shape covariance is strikingly similar to the main axes of shape variability that were obtained in previous studies of both limbs using principal components analyses for each appendicular bone (Martín-Serra et al. 2014a,2014b). As expected, the distribution of species along the PLS axes generally resembles the distributions found in our previous studies (Martín-Serra et al. 2014a,2014b). The high correspondence between the maximum shape covariance (between bones) and the maximum shape variance (for each bone) confirms the remarkably tight integration of the carnivoran appendicular skeleton. In addition, although both allometry and phylogeny are significant sources of bone shape variation, they do not seem to strongly influence the patterns of morphological integration. In fact, although the strength of the association between elements was slightly reduced when the effects of phylogeny were taken into account, integration remained significant.

We hypothesize that this remarkable integration pattern could be partly due to the strong biomechanical constraints that terrestrial locomotion place on the major limb bones. This biomechanical constraint could be related to a trade-off between resistance to weight-bearing stresses (achieved by robust bones) and the maintenance of energetic efficiency during locomotion (achieved by slender bones), given that adaptations for economical locomotion are counterbalanced by those for increased strength (Pasi and Carrier 2003; Kemp et al. 2005; Martín-Serra et al. 2014a,2014b). However, it is noteworthy that the data do not allow this trade-off to be rigorously tested, as the data are neither kinetic nor kinematic. Thus, given that all the species included in this study are quadrupedal, the bones of the fore- and hindlimbs vary in a coordinated manner as a function of their robustness, because both limbs are affected by similar biomechanical needs. The most noteworthy exceptions are the northern river otter (a semiaquatic species) and the European badger (a digging species), which is probably due to the fact that the functions of the fore- and hindlimbs differ more in these species than in other carnivorans (Tarasoff 1972; Fish 1994).

DIFFERENCES IN THE WITHIN- AND BETWEEN-LIMBS PATTERNS OF COVARIATION

Despite the high degree of morphological integration between bones, the strength of this covariation is not constant throughout the appendicular skeleton. As indicated by the RV coefficients of the comparisons between the serially homologous elements, the two girdles seem to be the least integrated elements that could be related to their different evolutionary history compared to the long bones (Young 2004; Kardong 2006). However, although the shape covariance of the scapula with all other limb bones remains more or less constant and is associated with changes in robustness (Figs. 5A and 6A, E), the shape covariance of the pelvis with other bones is more variable (e.g., see Fig. 5E for the shape covariances of the pelvis and femur and Fig. 6A for those of the pelvis and scapula). This difference in the pattern of integration between both girdles is also supported by the low percentages of shape covariation explained by the first PLS axes in all comparisons (Table 3). A possible explanation for this result could be related to their structural differences relative to the axial skeleton: whereas the scapula is a mobile element, the pelvis is fixed to the vertebral column. Thus, the scapula contributes to stride length during locomotion (as do other long bones) and the pelvis anchors important hindlimb muscles and transmits their movements to the trunk (Raich and Casinos 1991; Gasc 2001; Polly 2007; Schmidt and Fischer 2009; Williams et al. 2009; Hudson et al. 2011). According to this morphofunctional interpretation, the pelvis should not be expected to be as constrained as the scapula by the same type of biomechanical demands that primarily affect other limb bones. Alternatively, the pattern of covariance could be influenced by the fact that the pelvis is the result of the fusion of three different elements (ilium, ischium, and pubis). Although it is beyond the aims of this article to investigate the nature of the differences between both girdles, this topic deserves further in-depth study.

Another clear difference in the degree of shape covariation between serially homologous bones is that zeugopods (radius–tibia) appear to be more integrated than stylopods (humerus–femur). Developmental and functional reasons could underlie this difference: due to the fact that as a limb bone is more distally positioned in the appendicular skeleton, its shape and dimensions are more subject to functional adaptations (Figueirido et al. 2011), and are less limited by developmental constraints because distal growth plates close later (Weisbecker et al. 2008; Weisbecker 2011; Geiger et al. 2014).

Similar results were obtained by the comparisons between functionally equivalent bones and between serially homologous bones. The strong covariation between the ulna and the tibia (Fig. 4) suggests that they may have undergone a certain degree of convergence in their functions. The results appear to partially support the alternative hypothesis concerning the displacement of functional equivalence in relation to serial homology (Raich and Casinos 1991, Gasc 2001, Schmidt and Fischer 2009). In fact, the humerus and tibia show high covariation (see Fig. 6, right column), which suggests that their functional equivalence may have a real effect on the morphological changes of these bones. In contrast, covariation between the scapula and the femur is weak (see Fig. 4, right column), which is probably due to their different developmental processes and timing (Geiger et al. 2014). Thus, the presumed functional equivalence between both bones does not have a substantial effect on their morphological covariation.

FUNCTIONAL SPECIALIZATION INCREASES INTEGRATION WITHIN AND BETWEEN LIMBS

As indicated by the bootstrap tests, the results suggest that the degree of integration in the appendicular skeleton seems to be higher in cursorial species than in noncursorial ones. Furthermore, when phylogeny is taken into account, the appendicular skeleton of cursorial taxa remains more integrated than that of noncursorial taxa, although to a lesser degree than when phylogeny is not taken into account (see Table 4). Although a more in-depth research is needed to find the processes that underlie this pattern (perhaps comparing different levels of integration), following our results we hypothesize that it may be due to the fact that cursors have a forelimb specialized for locomotion, whereas noncursors also use the forelimb for grappling prey, swimming, digging, or manipulating food. Thus, whereas both limbs of cursors converge in their functions, the limb functions of noncursors diverge, which appears to be related to the lower degree of integration between the appendicular elements. Furthermore, in cursorial taxa the comparisons between distal functional equivalents (ulna–tibia and humerus–tibia; Fig. 8) are more integrated and are more susceptible than other elements to modification by functional causes. This finding is also supported when phylogeny is taken into account, because these comparisons plus the radius–tibia appear as still more integrated in cursors than in noncursors (Table 4). The results indicate that the appendicular bones are more integrated in species that specialize in locomotion and that such specialization particularly affects the distal elements.

Young and Hallgrímsson (2005) and Bell et al. (2011) demonstrated a similar process but in the opposite direction regarding the tight association between functional specialization and morphological integration. They showed that the highly integrated pattern between serially homologous elements decreases in specialized taxa with extremely different limb functions (e.g., flying vertebrates). Thus, in this case, functional specialization entails a divergence of functions between both limbs. The particular developmental strategies of marsupial mammals also reduce limb integration (Bennett and Goswami 2011; Kelly and Sears 2011). Marsupials are altricial and need well-developed forelimbs to climb toward the mother's teat (Sears 2004). This underlies the difference in developmental timing between the fore- and hindlimb in marsupials (Weisbecker et al. 2008; Goswami et al. 2009; Sears 2009; Geiger et al. 2014) and, as a result, a reduced level of morphological integration (Bennett and Goswami 2011; Kelly and Sears 2011).

Strikingly, the case regarding the appendicular skeleton of carnivorans seems to be the opposite, as the ancestral condition of basal carnivorans is a less integrated appendicular skeleton and generalized morphofunctional limbs (Heinrich and Rose 1995, 1997; Heinrich and Houde 2006; Spaulding and Flynn 2009; Tomiya 2011; Samuels et al. 2013). Therefore, taxa specializing in cursorial locomotion lose the “nonlocomotory” skills of their forelimb and increase the degree of integration between their limbs. More interestingly, the results also show that the pattern of integration may not only change due to large differences in function or development among distantly related clades (e.g., bats, birds, and marsupials), but could also change due to subtle differences in function or development within closely related species at an evolutionary level.