OUP user menu

Limb specialization in living marsupial and eutherian mammals: constraints on mammalian limb evolution

E. McKenna Kelly , Karen E. Sears
DOI: http://dx.doi.org/10.1644/10-MAMM-A-425.1 1038-1049 First published online: 14 October 2011


Evolution of the marsupial forelimb complex is argued to be constrained by the functional requirements of the newborn's crawl to the teat. This constraint is often used to explain why marsupials are limited in their diversity relative to eutherians. However, despite the constraint's importance to mammalian evolution, one of its fundamental corollaries, that marsupials exhibit less forelimb specialization than eutherians, has not been tested explicitly. We used morphometric analyses of mammalian limbs to test 2 hypotheses concerning their specialization: marsupials have less-specialized forelimbs than eutherians, and marsupials tend to specialize their hind limbs rather than forelimbs and eutherians their forelimbs rather than hind limbs. In support of the 1st, marsupial forelimbs are more similar among functional groups, display smaller morphospaces, and have lower morphologic divergence from the average mammalian state than eutherian forelimbs. In support of the 2nd, marsupial hind limbs are more dissimilar than forelimbs and display higher morphological divergence, but the opposite is true for eutherians. Results suggest fundamental differences in the mechanistic underpinnings of limb specialization in marsupials and eutherians and support a constraint on the marsupial forelimb complex, and possibly the mammalian hind limb.

Key words
  • development
  • functional constraints
  • limb
  • macroevolution
  • mammalian evolution
  • marsupials
  • morphological evolution
  • morphometrics
  • reproductive strategies
  • specialization

The form of the limb governs a mammal's range of locomotor, social, and feeding behaviors (Polly 2007). Therefore, the evolution of the mammalian limb complex (i.e., girdles and the limb proper) has played an integral role in the diversification of the group (Blob and Biewener 1999; Ji et al. 1999; Oxnard 1963; Polly 2007; Schmidt and Fischer 2009; Sears 2004; Tardieu 2010; Wayne 1986). This is most evident in the dichotomy between the 2 main groups of extant mammals, marsupials and eutherians.

Despite arising at approximately the same time in evolutionary history, marsupials and eutherians have experienced very different levels of success, with marsupials currently comprising only 6% of modern mammalian species and exhibiting a limited range of girdle and limb morphologies (Cooper and Steppan 2010; Lillegraven 1975; Sears 2004; Springer 1997). In previous research the relatively reduced morphologic diversity of the marsupial shoulder girdle (Sears 2004) and reduced rate of morphological evolution among the distal elements of the forelimb proper (Cooper and Steppan 2010) were linked to constraints imposed by the unique functional requirements of the marsupial newborn. These functional requirements result from marsupials, in contrast to eutherians, giving birth after extremely short gestation times to immature neonates that immediately must complete a crawl to the teat where they attach and continue to develop (Gemmell et al. 2002; Hughes and Hall 1988; Lee and Cockburn 1985; Sharman 1973). This crawl is powered entirely by the forelimb complex, and, consequently, the forelimb complex of the marsupial newborn is advanced in its development (Sears 2009a) and displays a highly modified shoulder girdle to provide the structural support and areas of muscle attachment necessary for the crawl (Cheng 1955; Klima 1987; Sears 2004). The constraint argument is that by having to form the specific crawl morphology at a specific time in ontogeny (i.e., at birth), the marsupial forelimb complex is relatively less free to vary its development, evolve new forelimb morphologies, and thereby specialize the forelimb for divergent functional behaviors (Lillegraven 1975; Sears 2004, 2009a). In contrast to the forelimb, the marsupial hind limb, which plays no role in the crawl, is expected to be freer to evolve and specialize.

This predicted pattern of marsupial limb specialization, with the hind limbs more specialized than the forelimbs, is opposite of that recently proposed by Schmidt and Fischer (2009) for mammals. Specifically, they proposed that because the mammalian hind limb is more important for propulsion than the forelimb, the mammalian forelimb should exhibit a higher level of evolutionary flexibility and be more capable of specialization for diverse functions. To test their hypothesis they examined limb proportions across mammals and found that, in general, mammalian forelimbs have greater variability in the proportions of their limb elements than do hind limbs (Schmidt and Fischer 2009). However, although their ideas are intriguing, their data set was dominated by eutherians, and they did not address explicitly differences in limb specialization between eutherians and marsupials. Therefore, despite the importance of the marsupial constraint to our understanding of mammalian evolutionary history, one of its fundamental corollaries, that marsupials should exhibit less forelimb specialization than eutherians, has never been expressly tested. Furthermore, the relative specialization of the forelimbs and hind limbs in marsupials and eutherians has not been investigated.

Traditionally, specialization of the limbs has been assumed from measures of morphological diversity (Cooper and Steppan 2010; Schmidt and Fischer 2009), but morphological diversity and limb specialization are not necessarily the same. For the purposes of this paper limb specialization is defined as the morphological evolution of a limb away from the generalized ancestral mammalian limb state (e.g., the morphological adaptation of the bat forelimb for flight). In contrast, morphological diversity (Foote 1994) reflects the overall range of morphologies achieved by a group. To fill these important gaps in our knowledge we quantified forelimb and hind-limb specialization among mammals. We use these data to test the hypotheses that marsupials have less-specialized forelimbs than eutherians, and that marsupials tend to specialize their hind limbs rather than forelimbs, and eutherians their forelimbs rather than hind limbs.

Materials and Methods

Data collection.—We measured the maximum length and width (measured on the shaft midway along the anterior-posterior axis) of skeletal elements of the girdles (scapula and pelvis), stylopod (humerus and femur), zeugopod (ulna, radius, tibia, and fibula), and autopod (metacarpals, metatarsals, and phalanges) from 196 therian taxa (82 marsupials and 114 eutherians). However, the autopod ultimately was excluded from analyses, because it was present in less than one-half of examined specimens (see Appendix I for list of taxa). We restricted our analyses to limb length and width to maximize homology and thereby comparability among skeletal elements and species. In general, for marsupials 1 species per genus was examined, and for eutherians 1 species per family was examined, following the classification of Wilson and Reeder (2005). We sampled marsupials at a higher intensity because of their lower taxonomic richness relative to eutherians and to insure that we accurately quantified their full range of limb specialization. More than 1 species per family or genus were examined if the family or genus contained members of multiple targeted functional groups. In total, we investigated representatives from 21 marsupial families (95% of total marsupial familial diversity) and 97 placental families (84% of total placental familial diversity). The sampled families represent the majority of the morphological and functional variety found within Theria.

Most adult quantitative data were obtained from osteological specimens housed at the Field Museum of Natural History in Chicago, Illinois. However, we also examined specimens housed at the Smithsonian National Museum of Natural History (Washington, D.C.) and in the University of California Museum of Paleontology (Berkeley, California), and in Australia at the South Australian Museum (Adelaide, Australia), the Australian Museum (Sydney, Australia), and the University of New South Wales (Sydney, Australia).

Adult quantitative measurements < 150 mm were sampled with digimatic calipers (Mitutoyo, Aurora, Illinois), those from 150 mm to 30 cm with vernier calipers (Fowler, Des Plaines, Illinois), and those >30 cm with a metric tape measure. Three measurements were taken for every length and width and averaged to minimize measurement error.

Functional groups.—We targeted 3 functional groups: arboreal quadrupeds, terrestrial quadrupeds, and bipedal saltators. These particular groups were selected because they have representatives from multiple mammalian orders, providing sufficient sampling of taxa for statistical analysis. We assigned species to functional groups based on published behavioral descriptions of their primary mode of locomotion (Nowak 1999). A result of this was that species that were described as locomoting equally in a terrestrial and arboreal quadrupedal manner were excluded from the functional groups analyses. However, these and all taxa were included in the analyses of all marsupials and all eutherians. For each functional group only 1 representative from any given genus was included. In total, arboreal quadrupeds were represented by 21 marsupial and 27 eutherian species, terrestrial quadrupeds by 17 marsupials and 33 eutherians, and bipedal saltators by 11 marsupials and 7 eutherians (see Appendix I for specific taxa in functional groups).

Data and analyses.—Quantitative measurements were log-transformed prior to analysis to standardize variance (Sokal and Rohlf 1995). A principal component analysis, including all measurements, was performed for marsupials and another for eutherians. In both analyses principal component 1 was correlated strongly and positively with all variables, suggesting that it is a good proxy for body size. Therefore, to minimize the effect of body size each length and width measurement (e.g., humerus length) was regressed against the appropriate principal component 1 (either the marsupial or eutherian), and the resulting residuals were used in further analyses unless otherwise noted (Sears 2004; Sears et al. 2007). Principal component analysis was performed in JMP version 8.0.2 (SAS Institute, Cary, North Carolina).

Correlation matrices were generated in JMP 8.0.2 for all measured variables for all functional groups (arboreal quadrupeds, terrestrial quadrupeds, and saltators). To assess the impact of sampling on the matrices the original data set was resampled with replacement, and the correlation matrices reestimated 1,000 times. The resulting matrices then were compared with the original observed matrix using the mean matrix correlation as an estimate of matrix repeatability, t (Marriog and Cheverud 2001). To test whether matrices were significantly similar to one another we applied the Mantel's test (9,999 replicates). Two matrices were considered significantly similar when the matrix correlation exceeded 95% of randomly generated correlations (Goswami and Polly 2010; Marriog and Cheverud 2001). Matrix correlations were performed using both Pearson (parametric) and Spearman (nonparametric) algorithms in R using the Vegan program (http://vegan.r-forge.r-project.org). These analyses allowed us to test our hypothesis that marsupials have less-specialized forelimbs than eutherians. For data to support this hypothesis marsupials should display higher correlations between functional groups (e.g., arboreal and terrestrial quadrupeds) than eutherians for forelimb-only data sets, consistent with a lower degree of limb specialization in marsupials for specific functional behaviors. These analyses also allowed us to test our hypothesis that marsupials tend to specialize their hind limbs rather than forelimbs and eutherians their forelimbs rather than hind limbs. For the data to support this hypothesis marsupials should display higher correlations between functional groups for their forelimbs than hind limbs, and eutherians should display higher correlations between functional groups for their hind limbs than forelimbs.

To provide a visual assessment of limb specialization we performed 2 additional principal component analyses on the log-transformed (but not body-size-adjusted) data—1 for the forelimb and 1 for the hind-limb data sets—that included all marsupials and eutherians. We then plotted the resulting principal component 2 and principal component 3 for all functional groups to visualize the relative morphological space occupied by marsupials and eutherians (Foote 1994, 1995; Sears 2004); principal component 1 was excluded because its high correlation with all variables suggests that it reflects size rather than shape. For our 1st hypothesis to be supported marsupial forelimbs should plot closer to the morphospace origin (which is taken to represent the limb morphology of an average, generalized mammal) than those of eutherians, both when all taxa are included and for each functional group. For our 2nd hypothesis to be supported marsupial hind limbs should plot further from the morphospace origin than marsupial forelimbs, and the opposite should be true for eutherians. However, note that principal component analysis represents a data visualization tool rather than a statistical test. We used the absolute distance from the centroid for any given taxon as a representation of its degree of limb specialization. To calculate this metric we calculated the average value for each measurement (e.g., scapula length) for marsupials and eutherians independently. We then subtracted the resulting average value for any given measurement from that of each taxon and took the absolute value of the result. The resulting value represents the distance from the average value, or centroid, for a given measurement for a given taxon. For each taxon we then summed the distances from centroid for all measurements. The resulting absolute distance from centroid (hereafter referred to as distance from centroid) represents the morphological divergence of the limbs of the taxon in question from the average mammalian limb, which in this study represents the degree of limb specialization. Distances from centroid then were pooled to allow statistical comparisons of the forelimbs and hind limbs among marsupials and eutherians as a whole, and among functional groups, using the Kruskal-Wallis test (χ2Sokal and Rohlf 1995). For each marsupial and eutherian comparison (e.g., marsupial compared to eutherian forelimbs) we also performed a combination of bootstrapping and rarefaction to assess the effect that sampling might have on the results (Kowaleski and Novack-Gottshall 2010). To do this we 1st subsampled the more-speciose group (usually eutherians) at the level of the smaller (usually marsupials). Then, for both the subsampled larger group and the smaller group we resampled the taxa with replacement 1,000 times and calculated the mean distance from centroid for every repetition. The resulting mean distance was compared between marsupials and eutherians for each repetition, and a P-value was obtained by dividing the number of repetitions in which the original result was upheld by the total number of repetitions (1,000). For the data to support our 1st hypothesis marsupial forelimbs should have significantly lower mean distances from centroid than those of eutherians for each functional group. For the data to support our 2nd hypothesis the marsupial forelimb should display a significantly lower distance from centroid than the hind limb, and in eutherians the opposite pattern should be observed.


Patterns of correlation for marsupial and eutherian limbs.— All matrices displayed relatively high repeatability indices (t), with values ranging from 0.84 to 0.97, with an average repeatability of 0.93. Because results of Spearman and Pearson matrix correlations are similar, only Spearman (rs) results are discussed here in detail. The average correlation between the forelimb matrices of the marsupial functional groups (n = 3, rs = 0.685) is higher that that of eutherians (n = 3, rs = 0.233; Table 1). Furthermore, every functional group comparison displays a higher correlation in marsupial than eutherian forelimbs, and all 3 correlations are significant in marsupials, whereas only 1 is for eutherians. The average correlation between marsupial forelimb matrices (n = 3, rs = 0.685) is also higher than that of marsupial hind limbs (n = 3, rs = 0.366). However, the average correlation between the marsupial (n = 3, rs = 0.366) and eutherian (n — 3, rs = 0.357) hind-limb matrices is very similar. Every marsupial functional group comparison displays a higher correlation in forelimbs than hind limbs, and although all 3 of these comparisons are significant in marsupial forelimbs, only 2 are significant for the hind limbs. In contrast, eutherians display an opposite pattern in which the average correlation among eutherian forelimb matrices (n = 3, rs = 0.233) is less than that of the eutherian hind limbs (n = 3, rs = 0.357). Moreover, every functional group comparison displays a higher correlation in eutherian hind limbs than for forelimbs, and although 2 of these comparisons are significant in eutherian hind limbs, only 1 is for the forelimb.

View this table:
Table 1

Correlations between functional group matrices for marsupial and eutherian forelimbs and hind limbs. P-values are located above and to the right of the diagonals and Spearman correlation coefficients (rs) are below and to the left. P-values marked with an asterisk (*) are statistically significant at the 0.05 level. Arboreal = arboreal quadupeds, Bipedal = bipedal saltators, Terrestrial = terrestrial quadrupeds.

Marsupial forelimb
Arboreal (n = 21)0.005*0.005*
Bipedal (n = 11)0.599<0.001*
Terrestrial (n = 17)0.6940.762
Marsupial hind limb
Eutherian forelimb
Arboreal (n = 27)0.1020.008*
Bipedal (n = 7)0.3130.543
Terrestrial (n = 33)0.407−0.022
Eutherian hind limb

Morphospace of eutherian and marsupial limbs.—For every functional group, the morphospaces occupied by eutherian forelimbs and hind limbs extend farther from the origin than those of the comparable limbs of marsupials (Fig. 1). This is the case even for functional groups in which the morphospace occupied by marsupials is larger than that of eutherians (e.g., bipedal saltators; Figs. 1C and 1D). However, no difference in the overall range of morphospace occupation (i.e., overall size of the morphospace cloud) of the forelimbs and hind limbs of eutherians, or of marsupials, is obvious.

Fig. 1

Morphospace occupation (plot of principal component 2 [PC2] on the x axis versus PC3 on the y axis) of marsupial and eutherian forelimbs (arboreal quadrupeds = A, bipedal saltators = C, and terrestrial quadrupeds = E) and hind limbs (arboreal quadrupeds = B, bipedal saltators = D, and terrestrial quadrupeds = F) by functional group. Light circles and ellipses represent marsupials, and dark squares and ellipses represent eutherians.

Specialization of eutherian and marsupial limbs.—For all functional groups, and a comparison of all marsupials and eutherians, the forelimbs (all taxa: χ21 = 160.98, P < 0.0001 arboreal quadrupeds: χ21 = 56.27, P < 0.0001 terrestrial quadrupeds: χ21 = 32.77, P < 0.0001 bipedal saltators: χ21 = 52.35, P < 0.0001) and hind limbs (all taxa: χ21 = 17.51, P < 0.0001 arboreal quadrupeds: χ21 = 10.42, P < 0.0001 terrestrial quadrupeds: χ21 = 17.05, P < 0.0001 bipedal saltators:) of eutherians display significantly greater distances from centroid than comparable marsupial limbs (Fig. 2A).

Fig. 2

Comparisons of the distance from centroid (see “Materials and Methods” for definition and calculation) of the A) forelimbs and hind limbs within marsupial or eutherian functional groups, and of the B) forelimbs and hind limbs of marsupials and eutherians by functional group. The bottom and top of each box represent the 25th and 75th percentile for the data set, respectively, with the horizontal line within each box representing the 50th percentile. The black squares represent means, and the whiskers the minimum and maximum values for the data set. Starred P-values are statistically significant at the <0.05 level.

Forelimbs of eutherians display a greater distance from centroid than their hind limbs for all functional groups, and the analysis of all eutherians (Fig. 2B). Of these differences in distance from centroid, 2 are significant (all eutherians, χ21 = 31.39, P < 0.0001 eutherian arboreal quadrupeds, χ21 = 16.50, P < 0.0001) and 2 are not (eutherian saltators, χ21 = 2.31, P < 0.13 eutherian terrestrial quadrupeds, χ21 = 2.95, P = 0.0001). In contrast, distance from the centroid is greater for the marsupial hind limb than forelimb for every functional group and within all marsupials. However, these differences are relatively slight when compared to the differences displayed by eutherian limbs (all marsupials, χ21 = 0.96, P = 0.0001 marsupial arboreal quadrupeds, χ21 = 0.04, P = 0.0001 marsupial terrestrial quadrupeds, χ21 = 1.71, P = 0.001 and are accordingly only significant for 1 functional group (marsupial saltators, χ21 = 6.83, P = 0.0001). All centroid results remained intact after resampling (P-values ranged from <0.001 to 0.008 for the forelimb comparisons, and <0.001 to 0.036 for the hind limb).


Although evolutionary constraints have long been proposed as important mediators of morphological diversification (Alberch 1982; Arthur 2001; Brakefield 2006; Breuker et al. 2006; Galis et al. 2001; Lillegraven 1975; Salazar-Ciudad 2006), documenting their existence has been difficult (Beldade and Brakefield 2003; Klingenberg 2003; Resnick 1995). As a result, relatively few rigorous studies of constraints have been performed (Beldade et al. 2002; Ciampaglio 2002; Cooper and Steppan 2010; Domazet-Loso and Tautz 2010; Frankino et al. 2005; Kalinka et al. 2010; Polly 1998; Sears 2004; Wagner 1995; Zelditch et al. 1993). In this study we tested one of the main corollaries of the marsupial constraint hypothesis (Cooper and Steppan 2010; Lillegraven 1975; Sears 2004), that marsupial forelimbs should be less free to specialize than marsupial hind limbs and the forelimbs of eutherians. To do this we used morphometric data from a broad suite of mammals to test the hypotheses that marsupials have less-specialized forelimbs than eutherians, and that marsupials tend to specialize their hind limbs rather than forelimbs, and eutherians their forelimbs rather than hind limbs.

We found morphologies of marsupial forelimbs to be more highly correlated between functional groups than those of eutherian forelimbs, which suggests that marsupial forelimbs are not as specialized for any given functional behavior as eutherian forelimbs. Consistent with this, we found that the morphospace range occupied by marsupial forelimbs is closer to the origin than that occupied by eutherian forelimbs in all comparisons. Furthermore, we also found that eutherian forelimbs display significantly greater morphologic divergence relative to the average mammalian state (as measured by distance from centroid) than marsupial forelimbs, consistent with the forelimb morphology of eutherians being more specialized. Taken together, these results provide strong support for the 1st of our hypotheses—that marsupials are less likely to specialize their forelimbs than eutherians.

The 2nd hypothesis is easiest to analyze by breaking it into its 2 components, and addressing separately whether marsupials tend to specialize their hind limbs than forelimbs, and eutherians tend to specialize their forelimbs than hind limbs. Relevant to the marsupial component, we found that forelimb morphologies of marsupials from different functional groups are more highly correlated than those of their hind limbs. This suggests that morphology varies more between functional groups in marsupial hind limbs than forelimbs, which is consistent with the hind limb being more specialized for different functional behaviors. Consistent with this, we found that the morphologic divergence from the average mammalian state was higher for marsupial hind limbs than forelimbs, but not significantly so. In contrast to the marsupial pattern we found that hind-limb morphologies of eutherians from different functional groups are more correlated than those of their forelimbs and that eutherian hind limbs also display a significantly greater morphological distance from the average mammalian condition than eutherian forelimbs. These data are consistent with eutherian forelimbs being more functionally specialized than eutherian hind limbs, and thereby support the 2nd component of the 2nd hypothesis.

When taken together, our results suggest that fundamental differences exist in the mechanistic underpinnings of limb specialization in marsupials and eutherians. This is in line with previous studies that found fundamental differences between marsupial and eutherian limbs in the sequence of limb developmental events (Bininda-Emonds et al. 2007; Sears 2009b; Weisbecker et al. 2008), developmental growth rates (Sears 2009a), and integration (Bennett and Goswami 2010; Goswami et al. 2009; Kelly and Sears 2011). The results of our study suggest that when marsupials specialize, they tend to do so in their hind limb rather than forelimb, likely because of the constraint on their forelimb from the functional requirements of their newborn's crawl. In contrast, eutherian specialization tends to occur more in the forelimb, possibly because of the hind limb's functional role in locomotor propulsion, as proposed by Schmidt and Fischer (2009). This suggests that different functional constraints (Alberch 1982; Maynard Smith et al. 1985; Richardson and Chipman 2003; Schwenk 1995), one acting very early in life (marsupials) and one much later (eutherians), could have channeled the evolution of the limb differently in the 2 main mammalian groups.

The average correlation of hind-limb morphologies between functional groups, however, is similar in marsupials and eutherians. This suggests that the differences in forelimb and hind-limb specialization observed in marsupials and eutherians could be driven largely by differences in the capability of the forelimb to specialize in these groups. If this is the case, it would be consistent with a situation in which the evolution of all mammalian hind limbs is constrained because of the integral role of the hind limb in locomotor propulsion (Schmidt and Fischer 2009). For marsupials this would mean that the evolution of both their forelimbs and hind limbs is functionally constrained, although as a result of different locomotor requirements. Furthermore, if the mammalian hind limb is inherently constrained, the marsupial forelimb constraint becomes that much more limiting to the evolutionary history of the group. Further testing is needed to tease apart these intriguing scenarios. In conclusion, our results support the hypothesis that the evolution of the marsupial forelimb complex has been constrained relative to that of eutherians and the marsupial hind-limb complex, consistent with studies by previous researchers (Cooper and Steppan 2010; Lillegraven 1975; Sears 2004). Furthermore, our results are consistent with the existence of an additional functional constraint on hind-limb evolution across all mammals, and thereby with the results of Schmidt and Fischer (2009). Results of this study also are consistent with suggestions of previous authors (Brakefield 2006; Breuker et al. 2006) that study of constraints has the potential to bridge developmental systems and evolutionary processes. We advocate future inquiry into this system to identify the important roles that functional constraints have had on the evolution of the mammalian limb and thereby mammalian evolutionary history.


We thank J. Marcot for his programming assistance and constructive suggestions on how to improve this manuscript. We also thank staff at the Field Museum, National Museum of Natural History, Australian Museum, South Australian Museum, University of California Museum of Paleontology, and the University of New South Wales for allowing access to specimens. We especially thank B. Stanley of the Field Museum for his outstanding collections management and research support. Research was supported by the National Science Foundation (0104927) to KES.

Appendix I

View this table:

Taxa (following Wilson and Reeder 2005) included in the analyses. In cases where multiple taxa per family are listed those indicated with * were excluded from the all marsupial/all eutherian analyses. Functional groups: Arboreal = arboreal quadruped, Terrestrial = terrestrial quadruped, Bipedal = bipedal saltator. Functional groups are indicated only for taxa included in the functional groups analyses. Species were assigned to locomotor functional groups based on behavioral descriptions provided by Nowak (1999).

GroupOrderFamilyGenusSpeciesFunctional Group


  • Associate Editor was Elizabeth R. Dumont.

Literature Cited

View Abstract