Journal of Mammalian Evolution

, Volume 17, Issue 3, pp 193–209

Comparative Scaling of Humeral Cross-Sections of Felids and Canids Using Radiographic Images

Authors

    • National Evolutionary Synthesis Center (NESCent)
Original Paper

DOI: 10.1007/s10914-010-9133-y

Cite this article as:
Meachen-Samuels, J. J Mammal Evol (2010) 17: 193. doi:10.1007/s10914-010-9133-y

Abstract

The cortical thickness of long bones can be an effective indicator of locomotor modes and other stresses encountered by bone. Felids and canids are two carnivoran families that have similar levels of phylogenetic diversity and overlap in body size, but differ in their locomotor habits. Many canids and felids are cursorial, but felids also climb more frequently than canids. Felids also display a secondary use for their forelimbs not observed in any canids: they use their forelimbs to grasp and subdue prey. Large felids use their forelimbs much more extensively to subdue prey than do large canids and, therefore, should have proportionately greater forces applied to their forelimbs. This study uses a non-invasive radiographic approach to examine the differences in cortical thickness in the humerus between the Felidae and Canidae, as well as between size groups within these two families. Results show few significant differences between the two families, with a slight trend toward more positive allometry in the felids. Overall, radiographic measurements were found to be better predictors of body mass than either prey killing behavior or locomotor mode in these two carnivoran families. One canid that demonstrated exceptionally high cortical area was the bush dog, Speothos venaticus. The rarely observed bush dog has been postulated to swim and dig regularly, and it may be that the thickened cortical bone reflects these behaviors.

Keywords

Humerusx-raysCortical thicknessCarnivoreBody massAllometry

Introduction

Cortical thickness and limb bone cross-sectional area have been used to assess how the distribution of bone reflects loading in vivo. These can also be effective measures of weight-bearing properties of bones and used in body mass estimation (e.g., Ruff et al. 1989, 1991; Ruff 1990, 2003; Anyonge 1993; Demes and Jungers 1993; Biknevicius 1999; Anyonge and Roman 2006). Cortical bone thickness can also reflect locomotor modes, such as fossoriality (Biknevicius 1993, 1999; Polk et al. 2000) and arboreality, which includes brachiation (Ruff 1990; Demes and Jungers 1993; Runestad 1997; Polk et al. 2000). “Wolff’s law” or more aptly termed “bone functional adaptation” sensu Ruff et al. (2006) states that bone will remodel under the strain of mechanical loading. Bones are commonly loaded in both bending and compression, with the majority of the stresses attributed to bending (Rubin and Lanyon 1982; Bertram and Biewener 1988; Lieberman et al. 2004; Demes and Carlson 2009). Additionally, the speed and gait of locomotion can affect the deposition and strength of cortical bone (Szivek et al. 1992). Lately, this issue has been debated throughout the bone loading literature, and the validity of Wolff’s law has come into question (Bertram and Swartz 1991; Lieberman et al. 2004).

In a recent study, Lieberman et al. (2004) advocated that axial compression in conjunction with bending shifts the neutral axis away from the cross-sectional central axis. In that study, the authors caution that cross-sections of limb midshafts do not always indicate repeated loading patterns in all animals in the same way, and suggest against making any assumptions about bone loading except from in vivo data.

While recent in vivo studies have found that cross-sectional geometry of long bones does not correlate well with strain patterns (Demes et al. 1998, 2001; Lieberman et al. 2004), examining the cross-sectional geometry of cortical bone without in vivo data is not necessarily an exercise in futility. There is still evidence that strain does play a role in bone remodeling, although not as simple as was once thought (Rubin et al. 1995; Martin et al. 1998; Robling et al. 2002). Variations in bone structure are still the most effective indicators of locomotor mode among closely related species when in vivo data are absent (Ruff et al. 2006). Additionally, comparisons of bone cross-sectional properties (i.e., second moments of area) remain good estimators of mechanical ability, provided that the examination is in closely related groups that share similar body plans (Ruff et al. 2006).

Members of the families Felidae and Canidae in the mammalian order Carnivora have similar body plans, comparable levels of phylogenetic diversity (Bardeleben et al. 2005; Johnson et al. 2006), and analogous body size ranges with the exception of the largest felids, the lion (Panthera leo) and the tiger (Panthera tigris) (Smith et al. 2003; Nowak 2005). However these two families differ in their prey-killing behaviors. Felids, especially large species, rely on their forelimbs to kill prey, whereas canids do not (Ewer 1973). Some canids may use their paws while pouncing on small prey, but this behavior is functionally different from the grasping that occurs when felids catch prey (Leyhausen 1979). This ecological difference may be manifested in their cortical bone thickness.

Carbone et al. (1999) demonstrated that there is a physiological threshold that occurs around 21 kg in carnivorans. Above this threshold, carnivorans will no longer hunt small prey because of increased metabolic demands. Along with this physiological change is a concomitant morphological change in the skulls of both canids and felids. Both families show a shift to wider muzzles and more robust canine teeth to cope with the stresses of killing larger prey (Van Valkenburgh and Koepfli 1993; Meachen-Samuels and Van Valkenburgh 2009a). Additionally, felids show a morphological shift in the forelimbs, with larger sites for muscle attachment, larger muscle in-levers, and greater overall robustness for grappling with larger prey. Canids do not show a similar increase in these forelimb attributes, reflecting the fact that they do not use their forelimbs for prey capture as felids do (Meachen-Samuels and Van Valkenburgh 2009b).

Felids and canids also show differences in their locomotor modes and this might be indicated in cortical thickness. Although many canids and felids show cursorial behaviors (Sunquist and Sunquist 2002; Nowak 2005; Wang and Tedford 2008), there are many felid species that are scansorial, as well as a few arboreal species. These arboreal cats include the marbled cat (Pardofelis marmorata), the margay (Leopardus wiedii), and the clouded leopard (Neofelis nebulosa), all of which have been known to hunt prey in the trees (Sunquist and Sunquist 2002; Matsuda et al. 2008). Cheetahs (Acinonyx jubatus), on the other hand, are almost entirely cursorial and may show cortical modifications of their long bones for this purpose (Caro 1994).

As a general rule, canids do not engage in scansorial behavior (Nowak 2005). One exception is the gray fox (Urocyon cinereoargenteus), which has been found to climb trees on occasion (Trapp and Hallberg 1975). Additionally, most dogs and some cats will occasionally scratch-dig, but no living cats or dogs are known to have a fossorial lifestyle (Sunquist and Sunquist 2002; Wang and Tedford 2008). One unusual canid that is difficult to place in a locomotor category is the 5–7 kg bush dog (Speothos venaticus). S. venaticus hunts in packs and uses its forelimbs to hold down prey that may be almost as large as themselves, such as pacas (Kleiman 1972; Deutch 1983). Unlike most canids, the bush dog is not a cursorial pursuit predator (Kleiman 1972), has partially webbed feet for walking through muddy riverbanks (MacDonald 1996; Bieseigel and Zuercher 2005), and has been observed swimming and digging burrows, both in the wild and in captivity (Goldman 1920; Bates 1944).

Different locomotor modes in these species may introduce stresses that differentially load the limb bones during movement and alter their cortical thickness and distribution. For instance, during climbing more rotational and torsional forces are applied to the forelimbs (Cartmill 1985), possibly resulting in a greater cortical area or a different distribution of cortical bone in arboreal species. Fossorial habits also increase stresses and can contribute to thickening or redistribution of cortical bone (Biknevicius 1993).

This study examined different dimensions of cortical shape and thickness in canids and felids and evaluated differences between the two families. Cortical measurements may be indicative of functional differences between the two families. If these two families do not show variation in cortical bone, then it can be suggested that other properties, such as body mass may be more important.

In addition to extant canids and felids, the extinct dire wolf, Canis dirus, was also examined. C. dirus was slightly larger than any living canid, thus allowing a comparison of canids to larger felids without extrapolating far beyond the range of body sizes seen in extant canids. The dire wolf was a large, Pleistocene, North American wolf-like canid that had similar morphology to extant gray wolves. It most likely lived in packs and hunted large prey by chasing them, analogous to living gray wolves (Merriam 1912; Wang and Tedford 2008).

Materials and methods

Felids and canids were grouped first by family, and then by size (Table 1). Size category was labeled as “small” if the species average was under 20 kg or “large” if the species average was at or above 20 kg. Species masses (average values, Smith et al. 2003) and behavioral locomotor categories were compiled from the literature (Young and Goldman 1946; Schaller 1972; Ewer 1973; Leyhausen 1979; Kitchener 1991; Sunquist and Sunquist 2002; Hunter 2005; Nowak 2005; Wang and Tedford 2008). The locomotor groups used include: 1) Terrestrial—cursorial species that do not climb unless they have no other choice as a means of escape; 2) Scansorial—species that often climb for escape, eating or leisure, but do not hunt in the trees; 3) Arboreal—species that do at least some hunting in trees as well as many other activities, like escape, eating, or resting (Table 1). The bush dog was difficult to categorize, because there are accounts that is has been known to swim and burrow (Goldman 1920; Bates 1944). Because this species could not be definitively shown to be semi-fossorial or semi-aquatic, it was placed in the terrestrial category.
Table 1

Felid and canid species used in the analysis, sample size, species body size group, locomotor category, and average species masses

Sp. No.

Species

Common name

n

Family

Body size

Locomotion

Average mass in kg

1

Acinonyx jubatus

Cheetah

2

Felidae

Large

Terrestrial

50

2

Caracal aurata

African golden cat

2

Felidae

Small

Terrestrial

13

3

Caracal caracal

Caracal

3

Felidae

Small

Scansorial

16

4

Caracal serval

Serval

2

Felidae

Small

Terrestrial

12

5

Felis chaus

Jungle cat

2

Felidae

Small

Terrestrial

8

6

Felis nigripes

Black-footed cat

2

Felidae

Small

Terrestrial

1.5

7

Felis silvestris lybica

African Wildcat

3

Felidae

Small

Scansorial

5

8

Leopardus colocolo

Pampas cat

2

Felidae

Small

Scansorial

5

9

Leopardus geoffroyi

Geoffroy’s cat

2

Felidae

Small

Terrestrial

5

10

Leopardus pardalis

Ocelot

2

Felidae

Small

Scansorial

11

11

Leopardus tigrinus

Tigrina or Oncilla

1

Felidae

Small

Scansorial

2.5

12

Leopardus wiedii

Margay

2

Felidae

Small

Arboreal

3

13

Lynx canadensis

Canadian lynx

2

Felidae

Small

Terrestrial

12

14

Lynx lynx

Eurasian lynx

2

Felidae

Large

Scansorial

22

15

Lynx pardinus

Iberian lynx

1

Felidae

Small

Terrestrial

10

16

Lynx rufus

Bobcat

2

Felidae

Small

Scansorial

12

17

Neofelis nebulosa

Clouded leopard

2

Felidae

Large

Arboreal

20

18

Panthera leo

Lion

2

Felidae

Large

Terrestrial

150

19

Panthera onca

Jaguar

2

Felidae

Large

Scansorial

68

20

Panthera pardus

Leopard

4

Felidae

Large

Scansorial

60

21

Panthera tigris

Tiger

2

Felidae

Large

Terrestrial

180

22

Panthera uncia

Snow leopard

2

Felidae

Large

Scansorial

40

23

Pardofelis marmorata

Marbled cat

1

Felidae

Small

Arboreal

4

24

Pardofelis temminckii

Asian golden cat

2

Felidae

Small

Scansorial

12

25

Prionailurus bengalensis

Leopard cat

2

Felidae

Small

Scansorial

5

26

Prionailurus planiceps

Flat-headed cat

2

Felidae

Small

Terrestrial

2

27

Puma concolor

Puma or Mountain lion

2

Felidae

Large

Scansorial

55

28

Puma yaguarondi

Jaguarundi

2

Felidae

Small

Scansorial

7

29

Alopex lagopus

Arctic fox

1

Canidae

Small

Terrestrial

5.5

30

Canis adustus

Side-striped jackal

1

Canidae

Small

Terrestrial

10.25

31

Canis dirusa

Dire wolf

4

Canidae

Large

Terrestrial

63

32

Canis latrans

Coyote

2

Canidae

Small

Terrestrial

13

33

Canis lupus

Grey wolf

2

Canidae

Large

Terrestrial

34

34

Canis mesomelas

Black-backed jackal

2

Canidae

Small

Terrestrial

9.75

35

Cerdocyon thous

Crab-eating fox

1

Canidae

Small

Terrestrial

6.5

36

Chrysocyon brachyurus

Maned wolf

2

Canidae

Large

Terrestrial

23

37

Pseudalopex sp.b

South American foxes

1

Canidae

Small

Terrestrial

3.5

38

Lycaon pictus

African hunting dog

2

Canidae

Large

Terrestrial

26.5

39

Nyctereutes procyonoides

Raccoon dog

2

Canidae

Small

Terrestrial

7

40

Otocyon megalotis

Bat-eared fox

2

Canidae

Small

Terrestrial

4

41

Speothos venaticus

Bush dog

2

Canidae

Small

Terrestrial

6

42

Urocyon cinereoargenteus

Grey fox

3

Canidae

Small

Scansorial

5

43

Urocyon littoralis

Channel Island fox

2

Canidae

Small

Terrestrial

1.2

44

Vulpes chama

Cape fox

2

Canidae

Small

Terrestrial

4

45

Vulpes macrotis

Kit fox

2

Canidae

Small

Terrestrial

2.75

46

Vulpes velox

Swift fox

2

Canidae

Small

Terrestrial

2.4

47

Vulpes vulpes

Red fox

2

Canidae

Small

Terrestrial

4.75

Mass averages from Smith et al. (2003). Locomotor mode for cats compiled from Young and Goldman (1946), Schaller (1972), Ewer (1973), Leyhausen (1979), Kitchener (1991), Sunquist and Sunquist (2002), and Hunter (2005). For canids, locomotor categories were taken from Nowak (2005) and Goldman (1920) (bush dog)

aDenotes an estimated body mass from Anyonge and Roman (2006)

bDenotes the mean mass for all Pseudalopex spp.

Radiographic images are an effective way to examine the cortical thickness of limb bones in museum collections without destructive sampling. Previous studies have demonstrated that mandibular cortical thickness can be modeled as a hollow asymmetrical beam and estimated using radiographic techniques (Biknevicius and Ruff 1992a, 1992b). This technique has been successfully applied to limb bone studies by modeling the bone shaft as a hollow symmetrical beam (e.g., Anyonge 1993; Runestad et al. 1993; Runestad and Ruff 1995; Runestad 1997).

Radiographs were taken of the mediolateral and anteroposterior views of the humerus, as in Anyonge (1993) and Runestad (1997) (Fig. 1). Specimens were radiographed from the following collections: U.S. National Museum of Natural History (USNM), Natural History Museum of Los Angeles County (LACM), George C. Page Museum (LACMHC), and UCLA Donald R. Dickey collection (UCLA). For a complete list of species, see Table 1; a full list of specimens is available in Appendix I. Twenty-five of 28 extant felid species in the analyses were x-rayed on a digital x-ray machine housed at the National Museum of Natural History at the Smithsonian Institution (USNM). The remaining three felids (Puma concolor, Lynx rufus, and Lynx canadensis), 17 extant canids, and Canis dirus, were x-rayed by placing the humerus directly on a Dupont Quanta Rapid x-ray cassette containing 3M green light sensitive UVL film using a portable x-ray machine.
https://static-content.springer.com/image/art%3A10.1007%2Fs10914-010-9133-y/MediaObjects/10914_2010_9133_Fig1_HTML.gif
Fig. 1

Digital radiographic images of a Jaguar, Panthera onca, humerus USNM 49393 in mediolateral and b anteroposterior view; film radiographs of African hunting dog, Lycaon pictus, humerus LACM 30588 in c mediolateral and d anteroposterior view. Black bars indicate the measurement taken for external diameter and white bars indicate measurements taken for internal diameter. Arrows and italicized letters indicate measurements that went into equations for cortical variables.

Approximately two individuals were radiographed for each extant species, including one of each sex whenever possible to account for sexual dimorphism (Gittleman and Van Valkenburgh 1997). For the dire wolf, four humeri of various sizes were chosen to increase the chances that at least one of each sex was sampled. Juvenile individuals with unfused epiphyses were excluded for all species and wild caught specimens were preferentially chosen.

Mediolateral and anteroposterior cortical thicknesses and lengths were measured on digital x-rays in ImageJ (Rasband 2007). For film radiographs, both exterior and interior humerus diameters and lengths were measured from the x-ray film using a light box and digital calipers to the nearest 0.1 mm. In order to assure that parallax was kept constant among all specimens (Heinrich and Biknevicius 1998), the x-ray machine was placed at a constant height above the film in all radiographs taken. To be sure that parallax was not unduly biasing measurements, external measurements were also taken directly from the bone for comparison. A negligible difference (less than 1 mm) was found between the radiograph and the actual bone for Canis dirus, the largest species radiographed using this method.

Measurements were taken at the midshaft, directly distal to the deltopectoral crest. Illustrations of the planes in which the bones were radiographed as well as examples of the internal and external measurements taken from the radiographs can be found in Fig. 1. Variables calculated include cortical area (CA, mm2), second moments of area, which estimate the ability to resist cross-sectional bending in the anteroposterior (Ix) and mediolateral (Iy) planes, and polar moment of area or inertia (J) (Ruff and Hayes 1983).

Cortical area estimates the total axial tensile and compressive rigidity, and was estimated using the formula:
$$ {\hbox{CA}} = \pi \left( {{\hbox{AB}} - {\hbox{ab}}} \right)/4 $$
A = external anteroposterior diameter, B = external mediolateral diameter, a = anteroposterior diameter of the medullary cavity, and b = mediolateral diameter of the medullary cavity (Fig. 1). Since CA should not be used as an estimator of non-axial rigidity, it was used to represent the amount of cortical bone in the cross-section and as a means to estimate body mass (Ruff et al. 2006).
Second moments of area were measured to estimate anteroposterior (Ix) and mediolateral (Iy) bending rigidity of the humerus (Heinrich and Biknevicius 1998). Second moments of area were calculated using the following formulas (Roark 1965; Alexander 1968; Anyonge 1993; Runestad 1997):
$$ \begin{array}{*{20}{c}} {{\hbox{Ix}} = \pi \left( {{{\hbox{A}}^3}{\hbox{B}} - {{\hbox{a}}^3}{\hbox{b}}} \right)/64} \\{{\hbox{Iy}} = \pi \left( {{\hbox{A}}{{\hbox{B}}^3} - {\hbox{a}}{{\hbox{b}}^3}} \right)/64} \\\end{array} $$
The ability of an object to resist torsional stress was estimated as the polar moment of inertia using the formula:
$$ {\hbox{J}} = {\hbox{Ix}} + {\hbox{Iy}} $$
Together these calculations estimate the resistance to loads during activities (Runestad 1997). The polar moment of inertia, J, was divided by two as it estimates the average bending rigidity of the humerus (Polk et al. 2000).

Species averages were calculated and were used in regression analyses. Using species averages gives each species equal weight in the comparisons of families, otherwise taxa with larger sample sizes would likely have greater influence on the regression lines. To assess the differences in scaling between canids and felids, dimensionality of the measurements were adjusted by taking the square root of CA, and the fourth root of Ix, Iy, and J/2 and then by applying log10 transformation. All measurements were modified before being used for analysis, unless otherwise stated. These four adjusted measurements were regressed against log10 humerus length using ordinary least squares regression (OLS) (Smith 2009).

Because individual masses were not available for the specimens that were radiographed, another proxy for size, geometric mean (GM), was used here. The geometric mean can be defined as the nth root of the product of n measurements (Mosimann and James 1979; Madar et al. 2002). In this study the GM was calculated using the following measurements: humerus length, anteroposterior and mediolateral external humeral diameters and the mediolateral width of the distal humeral articular surface (trochlea and capitulum). Some of these measurements, such as the humeral diameters and articular widths, have been shown to be good predictors of body size in primates, ungulates, and carnivorans (Ruff 1990; Scott 1990; Anyonge 1993; Andersson 2004). Therefore, this geometric mean should give a rough overall estimate of body size for each individual, which may be more accurate than species body mass averages taken from the literature. All cortical variables were also regressed against the GM.

For cortical variables versus both humerus length and GM, differences in slope between families were tested using a likelihood ratio test in the program SMATR (Falster et al. 2006).

In addition to comparing the slopes between the two families, slopes were tested for isometric scaling (cortical variables versus bone length or GM) using reduced major axis regression (RMA). RMA was used here because it has been shown to be an effective tool when describing how size variables are related, especially when dealing with linear relationships on a logarithmic scale; it is also useful when testing the value of a slope against a known isometric constant (Warton et al. 2006).

Because of the dimensional adjustment applied to all of the data, each regression coefficient can be treated as a linear measurement. Negative allometry was indicated by slopes significantly less than one, positive allometry by slopes significantly greater than one, and isometry by slopes not significantly different from one (Schmidt-Nielsen 1993; Read and Tolley 1997). All RMA slope comparisons were calculated with SMATR using the Ho slope comparison option (Falster et al. 2006).

If these cortical measurements scale isometrically to humerus length, then cortical thickness would increase proportionally to bone length. In other words, as the bone gets longer, then the strength of the bone would lag behind the mass it is required to support. If scaling is positively allometric, then cortical bone is proportionally thickened as the animal increases in mass, helping to accommodate the larger mass. Thus, the prediction for the outcome of these slope comparisons would be positive allometric scaling in both carnivoran families, with a significantly greater positive allometry in felids to accommodate the additional use of their forelimbs for prey-killing.

To test for significant differences in cortical variables between canids and felids and between different sub-divided body size groups within families, the square root of CA and the fourth root of Ix, Iy, and J/2 were taken. Since size differences between the largest and smallest cats are of a great magnitude, an analysis of covariance (ANCOVA) was performed with GM as a covariate using SPSS v. 17.0 to test for significant differences between means and intercepts between families. To test for significant differences between the means of sub-divided size groups within and between families, an ANOVA was run using Scheffe’s F (for equal variances) and Tamhane’s T2 (for unequal variances) procedures for post hoc comparisons (p < 0.05). ANCOVA and ANOVA procedures were performed using individual specimens, to increase the sample size of each group and to reduce the variance between groups.

To test the accuracy of using cortical area as a tool for body mass estimation, log10 mean species masses of extant canids and felids (Smith et al. 2003) were plotted against dimensionally adjusted mean species CA for each family individually. More often than not, individual body masses are not available from museum collections. When this is the case, mean species body mass is often used as a substitute (e.g., Roth 1990; Ruff 1990; Van Valkenburgh 1990; Anyonge 1993; Mendoza et al. 2006; De Esteban-Trivigno et al. 2008; Sorkin 2008; Clauset et al. 2009). Additionally, geometric mean was also regressed against CA as another, more individual measurement, for body size. Since r2 can be a poor gauge of the predictive power of the independent variable (Smith 1981; Van Valkenburgh 1990), two other measures of predictive variation were considered. Percent prediction error (%PE) and the standard error of the estimate (%SEE) were both calculated as in Van Valkenburgh (1990). These values were calculated for all species collectively, and each family separately.

Although interfamilial variation, not interspecific variation, is the focus of this study, to insure that results were not being distorted or masked by phylogeny, phylogenetically independent contrasts (Felsenstein 1985) were applied to the data set. Log transformed cortical variable contrasts were regressed against log humerus length or log GM through the origin using linear regression in the PDAP module in Mesquite v. 2.6 (Maddison and Maddison 2006) to determine if phylogeny had a significant effect on scaling of any of the cortical measurements used. Phylogenies used included Johnson et al. (2006) for the Felidae and Lindblad-Toh et al. (2005) and Bardeleben et al. (2005) for the Canidae. Branch lengths were assigned from molecular data using the appropriate phylogenies.

Results

Ordinary least squares regression

The results for OLS regressions of adjusted measurements versus humerus length and GM for extant species are summarized in Tables 2 and 3, respectively. Comparisons of the slopes of OLS regressions of humerus length for felids and canids (Fig. 2a, b) show significant differences between the two families for J/2 (p = 0.53), and values that approach significance for Ix (p = 0.56). For cortical variables versus humerus length there was only one canid outlier for CA. The bush dog (#41) fell above the 95% confidence interval (CI) calculated using individual species means. Felids had lower variances than canids for all regressions (Table 2); however, there was also only one felid that fell outside of the 95% confidence interval. Caracal serval (#4) fell below the CI for all measurements, with relatively gracile humeri.
Table 2

OLS regression of log10 variables against log10 humerus length for the Felidae and the Canidae and RMA isometry assessment

Variable

Family

Slope (SE)

Intercept (SE)

r2

pa

Fa

p (isometry RMA)

CA1/2

Felidaeb

1.274 (.046)

−1.855 (.100)

0.967

0.087

2.951

0.000

Canidae

1.079 (.099)

−1.440 (.208)

0.874

0.108

Ix1/4

Felidaeb

1.324 (.043)

−2.125 (.099)

0.969

0.056

3.702

0.000

Canidae

1.106 (.091)

−1.666 (.192)

0.896

0.055

Iy1/4

Felidaeb

1.258 (.041)

−2.025 (.089)

0.973

0.066

3.173

0.000

Canidae

1.068 (.082)

−1.614 (.174)

0.907

0.135

J/21/4

Felidaeb

1.297 (.043)

−2.087 (.093)

0.972

0.053

3.801

0.000

Canidae

1.090 (.087)

−1.644 (.182)

0.902

0.056

aF and p values refer to homogeneity of slopes between families using a likelihood test

bIndicates a deviation from isometry using RMA

Table 3

OLS regression of log10 variables against log GM for the Canidae and the Felidae and RMA isometry assessment

Variable

Family

Slope (SE)

Intercept (SE)

r2

pa

Fa

p (isometry RMA)

CA1/2

Felidaeb

1.074 (.022)

−0.594 (.031)

0.989

0.510

0.420

0.000

Canidae

1.048 (.045)

−0.551 (.059)

0.969

0.156

Ix1/4

Felidaeb

1.116 (.018)

−0.815 (.026)

0.993

0.240

1.437

0.000

Canidaeb

1.071 (.029)

−0.750 (.039)

0.987

0.007

Iy1/4

Felidaeb

1.060 (.014)

−0.779 (.020)

0.995

0.476

0.500

0.001

Canidae

1.029 (.027)

−0.722 (.036)

0.988

0.217

J/21/4

Felidaeb

1.094 (.015)

−0.803 (.022)

0.995

0.194

1.686

0.000

Canidaeb

1.053 (.026)

−0.739 (.034)

0.990

0.009

aF and p values refer to homogeneity of slopes between families using a likelihood test

bIndicates a deviation from isometry using RMA

https://static-content.springer.com/image/art%3A10.1007%2Fs10914-010-9133-y/MediaObjects/10914_2010_9133_Fig2_HTML.gif
Fig. 2

OLS regression plot of a log root Ix and b log root Iy versus log humerus length for the Felidae and Canidae. Points represent species averages. Refer to Table 1 for species numbers. Refer to Table 3 for slope, intercept and r2. Dashed lines represent Felidae, solid lines Canidae.

The OLS regression of adjusted measurements, versus GM found no significant differences between canids and felids, and most species fell within the 95% confidence interval when body size (GM) is taken into account (Table 3; Fig. 3a, b). Outliers include Leopardus pardalis (#10) that fell consistently below the 95% confidence interval for CA, Ix, and J/2 with gracile humeri. Additionally, the maned wolf (#34) also fell below the CI for Iy. Falling above the CI were the Canadian lynx (#13) for J/2 and the bush dog (#41) for CA.
https://static-content.springer.com/image/art%3A10.1007%2Fs10914-010-9133-y/MediaObjects/10914_2010_9133_Fig3_HTML.gif
Fig. 3

OLS regression plot of a log root Ix and b log root Iy versus log GM for the Felidae and Canidae. Points represent species averages. Refer to Table 1 for species numbers. Refer to Table 4 for slope, intercept and r2. Dashed lines represent Felidae, solid lines Canidae.

Additionally, another set of regressions were performed where the variables were simply log transformed (no roots were taken first) and relative results between families and p-values of tests involving both rooted and non-rooted variables were not significantly different. For ease of presentation only the initially rooted variables will be discussed.

Reduced major axis regression

Reduced major axis regression was performed for each variable versus both humerus length and GM to check for the presence of allometric scaling in both families. For humerus length, felids were positively allometric for all measurements, and canids were not significantly different from isometry for any measurements (Table 2). The result was similar for all measurements versus GM, with felids showing consistent positive allometry and canids displaying positive allometry only for Ix and J/2 (Table 3). Although canids were not significantly different from isometry for humerus length, they did display a positively allometric trend for both Ix and J/2.

In order to assure that the differences in allometry between canids and felids were not being driven by the inclusion of lions and tigers, separate regressions of all variables were performed that excluded these cats. When these two largest felid species were excluded, the slope values decreased for all variables, but all felid slopes were still significantly positively allometric.

Analysis of covariance and analysis of variance

Analyses of covariance (ANCOVA) were performed using GM as a covariate in order to assess the effects of size between canids and felids. The ANCOVA performed between families showed no significant differences with regard to mean values for any measurement (Table 4). There were significant differences found between the intercepts of the two families for all measurements (p < 0.001 for all measurements).
Table 4

ANCOVA results, including mean values (and standard deviations) of rooted values CA, Ix, Iy, and J/2 (see text) using GM as a covariate for the Canidae and Felidae

Variable

Felidae (all species)

Canidae

n = 56

n = 37

CAa

9.574 (5.332)

7.755 (4.019)

Ixa

6.582 (3.670)

5.271 (2.726)

Iya

5.881 (3.142)

4.595 (1.862)

J/2a

6.265 (3.434)

4.999 (2.387)

Numbers of individuals included are listed below each category. No significant differences between canids and felids were found at the p ≤ 0.05 level. A significant difference found in the intercepts between canids and felids are indicated with a

When ANOVA post hoc tests were performed on body size categories (Table 5), small canids and small felids were not significantly different for any measurement. Large felids were significantly different from all other groups for Ix. For all other measurements, large species (canids and felids) were not significantly different from one another, but they were significantly different from small species (canids and felids). In a second analysis that excluded lions and tigers, large felids were no longer distinct from large canids for any measurement. Both Scheffe’s F and Tamhane’s T2 gave similar results. Additionally, post hoc tests in ANOVA were performed on locomotor types. No locomotor groups were significantly different from any other group.
Table 5

ANOVA results, including mean values (and standard deviations) of rooted values CA, Ix, Iy, and J/2 (see text) for the Canidae and Felidae separated by size

Variable

Small felids

Large felids

Large felids (without lions and tigers)

Small canids

Large canids

n = 37

n = 19

n = 15

n = 27

n = 10

CA

6.491 (1.818)

15.579 (4.760)

13.553 (2.548)

5.564 (1.074)

13.670 (2.832)

Ix

4.433 (1.352)

10.768 (3.085)

9.529 (1.840)

3.763 (0.703)

9.341 (1.757)

Iy

4.051 (1.136)

9.444 (2.702)

8.335 (1.571)

3.527 (0.599)

7.481 (0.402)

J/2

4.257 (1.254)

10.176 (2.908)

8.996 (1.714)

3.653 (0.652)

8.634 (1.206)

Numbers of individuals included are listed below each category. Small canids and felids were never significantly different (p ≤ 0.05). Large canids and felids were significantly different only for Ix. After lions and tigers were removed, there were no significant differences between large canids and felids

Body mass estimation

Results of the body mass estimation regression show that log √cortical area and log mean species mass are strongly correlated in both cats (r2 = 0.928, Fig. 4a) and dogs (r2 = 0.915, Fig. 4b). Percent prediction error and %SEE indicate that cortical area is a good predictor of average mass in both canids and felids and that cortical area is slightly better at estimating canid body mass based on the equation for that family, than for felids (36% SEE in canids versus 41% SEE in felids) (Table 6). There was one species that fell outside the 95% confidence interval for individual species for the log mean body mass regression, Prionailurus planiceps (#26).
https://static-content.springer.com/image/art%3A10.1007%2Fs10914-010-9133-y/MediaObjects/10914_2010_9133_Fig4_HTML.gif
Fig. 4

Linear regression plot of log mean species mass versus log cortical area (CA) for a Felidae and b Canidae, shown with a 95% confidence interval for individual species. Points represent species averages; Linear regression plot of log GM versus log cortical area (CA) for c Felidae and d Canidae, shown with 95% confidence interval for individual species. Refer to Table 1 for species numbers, and to Table 6 for line equations. Dashed lines represent extant Felidae, solid lines extant Canidae.

Table 6

Regression of log √CA versus log average species mass and log GM

Variable

Family

Slope

Intercept

r2

SE

%SEE

%PE

Mass

All

2.515

−1.216

0.928

0.142

38.69

16.742

Felidae

2.492

−1.187

0.928

0.151

41.84

16.419

Canidae

2.533

−1.242

0.915

0.133

36.09

16.958

GM

All

0.926

0.554

0.984

0.023

5.438

1.255

Felidae

0.919

0.563

0.989

0.020

4.712

1.132

Canidae

0.925

0.550

0.969

0.028

6.659

1.369

In the regression of log root CA versus log GM (Fig. 4c, d), no species are outside of the 95% confidence interval except S. venaticus, whose size is overestimated based on geometric mean (Fig. 4d). These regressions of log GM show that geometric mean is a good predictor of cortical area for both families (Table 6). However, contrary to mean species body mass estimates, the GM is slightly better at predicting the cortical area of cats (4.7% SEE) than dogs (6.6% SEE). This lower error compared to mean species mass is most likely because GM is a more realistic predictor of individual size than mean species masses from literature.

Independent contrasts

Results of independent contrasts were similar to results from the original linear regressions, with a few differences (Table 7). Once phylogeny was accounted for, the regressions of log humerus length versus the cortical variables were not significantly different between the two families for any measurement (only J/2 was significant before), and the regressions of log GM versus the cortical variables were still non-significant.
Table 7

Independent contrasts results of variable contrasts versus log humerus length (HL) or log GM regressed through the origin

Variable contrasts

Family

Slope

r2

Fa

pa

Log root CA vs Log HL

Felidae

1.164

0.938

1.009

0.297

Canidae

1.027

0.810

Log root Ix vs Log HL

Felidae

1.198

0.953

1.779

0.157

Canidae

1.046

0.869

Log root Iy vs Log HL

Felidae

1.137

0.956

0.775

0.340

Canidae

1.042

0.875

Log root J/2 vs Log HL

Felidae

1.213

0.951

1.909

0.171

Canidae

1.056

0.877

Log root CA vs Log GM

Felidae

1.036

0.971

0.270

0.591

Canidae

1.070

0.959

Log root Ix vs Log GM

Felidae

1.063

0.978

0.006

0.950

Canidae

1.066

0.984

Log root Iy vs Log GM

Felidae

1.010

0.983

1.544

0.199

Canidae

1.060

0.987

Log root J/2 vs Log GM

Felidae

1.079

0.981

0.010

0.928

Canidae

1.074

0.988

aF and p values refer to homogeneity of slopes between families

Discussion

Canids and felids showed few overall differences in cortical bone thickness. Results of the OLS regressions only showed significant differences when J/2 was regressed against log humerus length. Results of the ANCOVA suggest no differences between families. When further broken down into size groups, large felids were significantly different from other groups with regards to bending in the anteroposterior plane (Ix). However, once lions and tigers were removed from the analysis, this difference disappeared. The intercepts between the two families were distinct both including and excluding lions and tigers, the two largest felids. This may suggest that the ancestral proportions between canids and felids were different, but the overall scaling differences are very subtle. In other words, canids and felids came from different evolutionary starting points (the intercept), but are functionally on the same trajectory with regards to humerus cortical dimensions (the slope).

The only clear difference between canids and felids seemed to be in the relative allometry of cortical variables. When tested against an isometric slope value of one, canids were not significantly different from isometry for CA or Iy when regressed against GM, although felids were for all variables, even when lions and tigers were excluded from the analysis. Although there was no statistically significant difference in slopes between families, felids always had a higher slope value than canids, suggesting that there is a subtle trend toward thicker cortical bone in felids. Bertram and Biewener (1990) and Meachen-Samuels and Van Valkenburgh (2009b) found similar results when they examined the external osteology of canids and felids. Felids always displayed greater positive allometry in the humerus than did canids; however, Bertram and Biewener (1990) found that canids demonstrated positive allometry more frequently than felids with regard to the femur. It is possible that a larger sample size of individual species may reduce the risk of Type II error and elucidate significant differences between humeral cortical bone in these two families that were not found in this study (Jacquemont et al. 2009).

In this study, no differences were found in the humeral cortical thicknesses of arboreal, scansorial, and terrestrial canid and felid species. Polk et al. (2000) had similar results in their comparative study of rodents, primates, and carnivorans. They found no consistent differences in compressive strength of the femora of arboreal and terrestrial rodents, and no significant differences in compressive strength of the humeri or femora between arboreal and terrestrial primates. They did, however, find a significant difference between the cortical thicknesses of terrestrial and arboreal carnivorans, with arboreal carnivorans having relatively greater cortical cross-sectional properties. But their study only included one felid species (Felis sp.) and six canid species, all of which were classified as terrestrial. The results of the present study may have been confounded by the low number of truly arboreal felids (n = 3) and the complete lack of arboreal canids. The locomotor categories used in this study were also very general and broad; still, arboreal cats did show significant differences from other cats in external osteological characters (Meachen-Samuels and Van Valkenburgh 2009b).

Cortical thickness of long bones has been shown to be a good predictor of body mass in many species (e.g., Ruff et al. 1989, 1991; Ruff 1990, 2003; Anyonge 1993; Demes and Jungers 1993; Biknevicius 1999; Anyonge and Roman 2006). This seems to be the case in felids and canids, as body mass prediction regressions show comparatively low error relative to non-cortical measurements (Van Valkenburgh 1990). Although the prediction error was relatively low, it could be improved if individual body masses were used; however, this is generally not possible given the dearth of actual body masses available from museum collections.

The flat-headed cat, Prionailurus planiceps, had unusually low values for CA when compared to literature body mass estimates, but not for GM. This seems to indicate an individual (or individuals) that are heavier than the average species mass are biasing the results. However, this may also indicate the flat-headed cat has unusually thin cortical bone, although it is unclear what the functional significance of this result might be.

Interestingly, the bush dog, Speothos venaticus, was also an exception to the general trend seen in felids and canids. Relative to the rest of the Canidae, it had significantly thicker cortical bone than would be predicted by its mean species mass or humerus length. When compared to humerus length, only CA was very large in the bush dog, and this value remained high when regressed against geometric mean. Yet, this still shows that the bush dog is relatively more robust than other canids with similar bone lengths. Since CA is a poor estimator of non-axial strength, it is perplexing as to what these high CA values mean, functionally. Although the values of Ix and Iy in the bush dog did not fall outside of the confidence interval, they were relatively high, and these values would reflect functional adaptations to activities. As mentioned previously, bush dogs may frequently swim or burrow (Goldman 1920; Bates 1944), and they also restrain prey with their forelimbs (Kleiman 1972).

As a parallel to the bush dog, Biknevicius (1993) found that cortical bone is thicker in species that consistently use their limbs in strenuous activities, such as digging. A fossorial rodent, the tuco-tuco (Ctenomys), has a much higher mean moment of area (J/2) in the humerus than would be predicted for a rodent of its body mass, and considerably thicker cortical bone than similarly sized members of its family (Biknevicius 1993).

The dire wolf was not significantly different from other canids and felids. However, its presence in this study was significant. The dire wolf allowed a comparison of middle range pantherines, such as the jaguar, to canids. Without C. dirus, a false positive result may have occurred between canids and felids even when lions and tigers were removed from the data set. Even though this animal is extinct, it was not a stretch to include it in this analysis, because its ecological strategy was well known (Merriam 1912; Wang and Tedford 2008) and it was sufficiently similar to extant Canis to be placed within that genus.

Although both cats and dogs show similar scaling of cortical thickness relative to GM, cats are significantly different from isometry, whereas dogs are not for CA and Iy. Felids possibly experience more bending forces in the mediolateral plane (Iy), which may be a consequence of forces encountered when holding onto struggling large prey. This may also be due to the mediolateral distal diaphyseal accumulation of cortical bone seen in some cats as in Fig. 1. This portion of the humerus may need to be reinforced to avoid breakage during prey struggle, because of the anteroposteriorly flatter shape of the distal end of the diaphysis. This thickening may also be due, in part, to the increased use of the wrist flexor and extensor muscles that originate on the distal end of the humerus and facilitate the expansion of the humeral epicondyles observed in the felids (Meachen-Samuels and Van Valkenburgh 2009b). These flexors and extensors allow grasping of prey with the paws during the initial attack. For both families, cortical variables seem to scale with body mass; however, cats trend towards positive allometry, possibly as a result of prey-capturing techniques.

Recently, Doube et al. (2009), using CT scan data, found that larger felids showed greater cortical thicknesses in the forelimbs than the hind limbs and this difference became more apparent with increasing cat size. This suggests that mass may not be the only factor determining cortical thickness and larger cats may indeed have reinforced forelimb cortical bone to cope with the increased stresses of large prey capture. These results agree with the findings of this study that although cats and dogs are not significantly different from each other, cats show a positive significant deviation from isometry that may be functionally related to prey-killing habits. However, Doubé et al. (2009) did not examine any dog humeri, so they could not compare canid and felid scaling.

While results suggest no major significant differences between the humeral midshaft cortical thicknesses of canids and felids, there are external morphological differences in the humerus associated with different prey-capture strategies in felids. Specializations for different prey sizes are accomplished through changes in mechanical advantage (i.e., muscle size and in the size of their origins and insertions), as shown in Meachen-Samuels and Van Valkenburgh (2009b). Further differences may come to light between canids and felids if both the forelimbs (humerus) and the hind limbs (femur) are compared. Additionally, an examination of the forelimbs of these two families using CT scans may provide better resolution of the differences between them.

In conclusion, it seems as though humeral cortical thickness and distribution is a conserved character between canids and felids and that these variables are more profoundly affected by body mass constraints than by prey-killing behavior or locomotor mode. These ecological roles are instead more evident in external osteological morphology (Meachen-Samuels and Van Valkenburgh 2009b). Few significant differences were found between the humeral cortical thicknesses of felids and canids, but there was a trend towards greater positive allometry in felids that may be attributable to the additional use of their forelimbs in prey killing. A noticeable exception to this rule among living canids is the bush dog, which showed increased cortical thickness that may reflect digging and/or swimming habits, or possibly even prey-killing strategies similar to felids (i.e., using the forelimbs to hold down pacas or other larger prey) (Goldman 1920; Bates 1944; Kleiman 1972).

Future directions for this project include comparing cortical thickness of the humerus and the femur in canids and felids to see if the same patterns hold true for the hind limbs. Another current project includes examining the cortical thickness of the extinct saber-toothed felid Smilodon fatalis, to see if this species shows cortical patterns similar to living cats or if their humeri have been secondarily thickened because of differing prey-killing strategies. Preliminary results suggest that they have disproportionately thick humeral cortical bone relative to conical toothed cats.

Acknowledgments

The following curators and collection managers kindly allowed access to specimens (and digital radiographic equipment) in their care: J. Dines (Museum of Natural History of Los Angeles County), C. Shaw and S. Cox (George C. Page Museum), K. Molina (Donald R. Dickey Collection of the University of California, Los Angeles), and L. Gordon and J. Jacobs (U.S. National Museum of Natural History). Discussion with and comments by B. Van Valkenburgh, J. Samuels, W. Binder, X. Wang, D. Jacobs, R. Wayne, P. J. Brantingham, K. Koepfli, V.L. Roth, T. Roberts, P. Durst, and two anonymous reviewers greatly improved this paper. This project was partially funded by a U.S. Dept. of Education Graduate Assistance in Areas of National Need (GAANN) fellowship from UCLA and partially funded by NESCent NSF Grant # EF-0423641.

Copyright information

© Springer Science+Business Media, LLC 2010