Origins of interspecific variation in lizard sprint capacity

Sprint speed measurements

Several methods have been used to measure maximal sprint speed in lizards. Most studies use a racetrack equipped with photocells positioned at set intervals (see Huey et al. 1981; Miles & Smith 1987 for descriptions). Racetracks differ among studies in length, substrate, inclination and distance between the photocells. The length of the tracks varies between 1 and 6 m, but most studies use tracks between 2 and 3 m long. It is unclear exactly how long a track should be to obtain reliable maximal speeds. Sprinting in lizards is usually explosive, and animals will reach their top velocity within milliseconds of their departure (Huey & Hertz 1982; Irschick & Jayne 1998). The substrate used also varies, but most studies employ materials that are thought to provide good traction (e.g. rubber, cork, foam board, sandpaper, window screening, linen cloth, rough-cut hardwood). Other studies, especially of desert lizards, prefer sand because it would better resemble the natural substrate of the animals. The effects of substrate on running speed have seldom been tested explicitly. Running speeds of Uma scoparia on sandy and rubberized substrates proved highly similar (Carothers 1986). Some lizards, for unknown reasons, seem to run more readily on substrates that are (slightly) inclined. Therefore, a number of authors tilt their racetrack to some degree (van Berkum 1986; Losos et al. 1989, 1991; Losos 1990; Irschick & Losos 1998). The distance between the photocells is usually 0·25 or 0·5 m.

A second category of studies uses tracks similar to the racetracks mentioned above, but uses different ways to measure the speed of the lizards. Some have used stopwatches (e.g. Snell et al. 1988; Losos et al. 1993; Zani 1996; Klukowski, Jenkinson & Nelson 1998), others have filmed or videotaped the lizards (e.g. Daniels 1983; Avery et al. 1987; Farley 1997; Márquez & Cejudo 1999). In the latter case, it is often unclear over what distance the speed was calculated.

Finally, several studies use high-speed treadmills (John-Alder, Garland & Bennett 1986; Beck et al. 1995; Dohm et al. 1998; Bonine & Garland 1999; Irschick & Jayne 1999). The speed of the belt is varied until it matches the apparent maximal running speed of the lizard. Alas, studies with high-speed treadmills often yield higher estimates of maximal sprint speed than do studies with photocell-timed racetracks (see Table 6 in Bonine & Garland 1999). Therefore, we choose not to use treadmill estimates of speed in our analysis.

Body temperature has a profound effect on sprint speed (e.g. Bennett 1980; Crowley 1985; Marsh & Bennett 1986; van Berkum 1986, 1988; Van Damme et al. 1989, 1990; Bauwens et al. 1995), and maximal running speed will be attained only at near-optimal body temperatures. Most authors acknowledge this fact and state that lizards were tested at optimal temperatures, or at temperatures close to that of animals in the field. In the latter case, it is assumed that animals in the field are active at near-optimal body temperatures. This may not always be true, but (at least in diurnal lizards, see Huey et al. 1989), field body temperatures are probably a good proxy for optimal body temperatures. We disregard data from one older study (Urban 1965) because the author admits that the temperatures of the animals in his photographic cage were not controlled.

Sprint speed may also vary with age (e.g. Garland 1985; van Berkum et al. 1989; Carrier 1996; Elphick & Shine 1998), sex (e.g. Huey et al. 1990; Dohm et al. 1998), reproductive condition (e.g. Van Damme et al. 1989; Cooper et al. 1990), hormone levels (Klukowski et al. 1998), feeding status (Huey et al. 1984) and tail loss (e.g. Ballinger, Nietfeldt & Krupa 1979; Pond 1981; Punzo 1982; Arnold 1984; Formanowicz, Brodie & Bradley 1990; but see Daniels 1983, 1985; Huey et al. 1990). Many studies do not provide information on some of these factors. We will assume that their effects are small in comparison to the interspecific variation studied here. When sprint speeds of males and females of a species are given separately, we calculate the weighted average. Data from juveniles, gravid females, males with experimentally elevated testosterone concentrations and lizards without tails are not used in the analysis.

Body mass estimates

Some studies report the snout–vent length (SVL), rather than the mass of the animals used. In these cases, the mass is calculated from the following equation:

This empirical allometric equation is based on 123 species or populations in our database, for which we had both SVL and mass. The coefficient of determination of this regression is 0·92.

Temperature data

The mean body temperature of animals active in the field was used to characterize the thermal biology of the species. In a few cases (see Table 1), where these data were not available, selected body temperatures were used.

Ecological data

The ecological data (climate, activity, microhabitat use, foraging mode) were obtained from various sources. Apart from the papers on sprint speed themselves, these include Arnold, Burton & Ovenden (1978); Cogger (1992); Cooper (1994); Vitt et al. (1995, 1998); Vitt, Zani & Caldwell (1996); Leal et al. (1998); Vitt & Zani (1998). One species in the data set, Amblyrhynchus cristatus, is a herbivorous lizard. Therefore, it was not included when testing for differences in sprint speed according to foraging mode.

Phylogenetic analyses

In recent years it has repeatedly been stressed that comparative data need to be analysed in an explicit phylogenetic context (Felsenstein 1985, 1988; Harvey & Pagel 1991; Garland et al. 1993). Because species share parts of their evolutionary history, they cannot be considered independent data points in statistical analyses and thus traditional (i.e. non-phylogenetic) tests are invalid. In this study, two different approaches were used to circumvent the problem of non-independence.

To evaluate the importance of body mass and temperature in explaining interspecific variation in sprint speed, the phylogenetic independent contrasts of these three variables were calculated (pdtree computer program, Garland et al. 1999). A multiple regression was then performed with sprint speed contrasts entered as the dependent variable and body mass and temperature contrasts as independent variables (SPSSwin 10·0; SPSS Inc., Chicago, IL, USA). The regression was forced through the origin (see Garland, Harvey & Ives 1992).

Phylogenetic simulations (Garland et al. 1993) were used to test whether sprint speed differs among sets of species with different climate (tropical, Mediterranean, xeric or cool), microhabitat use (terrestrial, arboreal or saxicolous), activity patterns (diurnal or nocturnal) and foraging mode (sit-and-wait or active foraging). In phylogenetic simulations, F statistics are compared with empirical F distributions, rather than to standard tabular values. The empirical null distributions are obtained by performing analyses of variance on the results of computer simulation models of continuous traits evolving along a known phylogenetic tree. The pdsimul computer programs by Garland et al. (1999) were used to simulate evolution of speed, mass and temperature, assuming Brownian motion as the model of evolutionary change. The means and variances were set to the means and variances of the original data. The procedure was repeated 1000 times. No limits to the simulated values of the variables were imposed. The pdanova program was used to perform traditional one way analyses of variance (anova) on the simulated data sets. The F-statistics of these 1000 anovas were used to set up the null distribution. The differences among sets of species were considered significant if the F-value exceeded the upper 95th percentile of the simulated F-distribution. The F-value at the lower end of this 95th percentile will be called ‘the critical F-value’ in the results. This procedure was repeated for each ecological variable (i.e. climate, activity, microhabitat use, and foraging mode). The pdsimul program was also used to check the results obtained with regression of independent contrasts, following procedures outlined by Garland et al. (1999).

Both methods require input on the topology and branch lengths of the phylogenetic tree. A ‘currently best’ tree was compiled from literature (Fig. 1; Arnold 1983, 1989; Garland, Huey & Bennett 1991; Joger 1991; Dial & Grismer 1992; Garland 1994; Kluge & Nussbaum 1995; Reeder & Wiens 1996; Zani 1996; Irschick et al. 1997; Wiens & Reeder 1997; Cullum 1998; Harris et al. 1998; Bonine & Garland 1999). Some unresolved polytomies remain, however. This was taken into account by subtracting one degree of freedom for each unresolved node (Purvis & Garland 1993; Garland 1994). As data on the divergence times are scattered, all the branch lengths were set to unity. It has been shown that the actual length of the branches does not usually affect the outcome of the statistical analyses to a great extent (Martins & Garland 1991; Walton 1993; Irschick et al. 1996; Díaz-Uriarte & Garland 1998). Moreover, checks of branch lengths with the pdtree program did not show any significant correlation between the absolute values of the standardized contrasts and their standard deviations (Garland et al. 1992). Because it is most likely that divergence times among the families in our data set differ strongly from those between genera and species, we also performed the phylogenetic analyses on trees of which branch lengths were proportional to the taxonomic level of the groups they connect. Divergence times were set to 5, 10, 20, 50 and 100 units for families, and to 1, 2, 3, 4 and 5 for genera (divergence times between species were always kept to 1 unit). These branch length manipulations did not alter the outcome of the tests qualitatively, and therefore only results for the tree with all branch lengths set to unity are reported.

SELECTION OF DATA

Over 50 papers reporting sprint speeds of lizards were found. Data obtained with treadmills, and from racetracks if the distance over which speed was calculated over more than 50 cm were disregarded. In addition, some material could not be used because data on mass and SVL or body temperature were missing, or because the phylogenetic position of the species concerned was unclear. Species used in this study are given in Table 1.

Effects of body mass and body temperature

Non-phylogenetic analyses

Log₁₀ maximal sprint speed correlates with log₁₀ body mass (Fig. 2, r = 0·45). The slope of the ordinary least-squares regression line has a value of 0·177 (±0·031 SE). Reduced major axis regression yields a slope of 0·39 (95% confidence interval: 0·33–0·46). Inspection of the residuals of the regression reveals two outliers: the two chameleon species are obviously slow for their body size. Removing these data points improves the fit of the regression line considerably (now r = 0·58). Ordinary least-squares regression now produces a slope of 0·202 (±0·025), reduced major axis regression a slope of 0·35 (95% confidence intervals: 0·30–0·40).

Log₁₀ maximal sprint speed also correlates with body temperature (Fig. 3, r = 0·52). The estimated slopes are 0·020 (ordinary least-squares regression) or 0·037 (reduced major axis, 95% confidence interval: 0·038–0·054). Again, the chameleons stand out for having strikingly low sprint speeds for their activity temperatures. Removing them from the analysis improves the correlation (r = 0·52) and returns slope values of 0·019 (ordinary least-squares regression) and 0·037 (reduced major axis).

Multiple regression on all species with known speed, body mass and temperature yielded a model with a significant contribution of body temperature (partial correlation = 0·41, P < 0·00001), but not of body mass (partial correlation = 0·15, P = 0·14). However, when the two chameleon species are omitted from the analysis, both log₁₀ body mass (partial correlation = 0·28, P = 0·001) and body temperature (partial correlation = 0·45, P < 0·00001) contribute significantly to the variation in log₁₀ sprint speed (see Fig. 4a). Together, they explain 34% of the interspecific variation in sprint speed. The partial regression coefficient for log₁₀ body mass estimates the allometric scaling exponent: 0·092 (± 0·027 SE).

Effects of climate, activity period, foraging mode and microhabitat use

Bigger is better

Our results seem to refute Hill’s (1950) prediction that speed would be independent of body size. The maximal sprinting speed (v) of lizards increases with body size, at least up to a certain point. There are several other predictions on the allometry of speed (elastic similarity model: v ∝ Mass^0·25; static stress similarity: v ∝ Mass^0·40 (McMahon 1974, 1975; Huey & Hertz 1982); dynamic similarity: v ∝ Mass^0·17 (Gunther 1975; Garland et al. 1987)). However, because of the large amount of scatter present in the data, it cannot be decided which of these other scaling models is more fitting. The exponent obtained by ordinary least-squares regression (0·18, or 0·20 if the chameleons are omitted) is temptingly close to the value predicted by dynamic similarity theory (0·17). Garland (1983), also using ordinary least regression, obtained a highly similar value (0·165) for 106 mammal species with body masses ranging from 0·016 to 6000 kg. However, several authors have argued that in allometric studies, reduced major axis regression may be a more suitable technique than ordinary least-squares regression (e.g. Rayner 1985; McArdle 1988; Christian & Garland 1996). The exponent obtained through reduced major axis regression on the data presented here (0·39, or 0·35 without the chameleons) is closer to that predicted by the static stress similarity model (0·40).

In mammals, none of the theoretical scaling models describe the actual relationship between speed and body mass very well (Garland 1983). Log(speed) does not increase monotonically with log(body mass), as suggested by the biomechanical models, but takes a curvilinear path, reaching an ‘optimum’ at a body mass of about 119 kg (Fig. 5). Following this line, a polynomial regression equation was fitted through the lizard data. It took the following form (with speed in m s⁻¹ and body mass in g; see also Fig. 5, chameleons omitted):

The fit was slightly higher for this curve (r² = 0·28) than for the linear regression (r² = 0·20). The equation suggests an ‘optimal’ body size for lizards (with regards to running ability) of 48 g. Observe, however, that our data set contains very few speeds of large species. Only seven species have body masses above the ‘optimum’ of 48 g. In addition, observations of lizards running in the field suggest that race track measurements may underestimate maximal performance, especially in larger lizards (Jayne & Ellis 1998). Therefore the curvilinear nature of the log(speed)–log(body mass) relationship remains uncertain; data on the maximal velocities of truly large lizards (e.g. large varanids) are badly needed to solve the case.

Measurement error is undoubtedly one of the reasons for the large amount of scatter around the body mass–speed relationship. In spite of our attempt to restrict our data set to studies using similar techniques for measuring sprint speed, it is clear that variation among experimental set-ups and protocols (e.g. the fanaticism with which lizards are chased through the tracks) is bound to introduce some error. Interspecific variation in ‘design’ (morphology, physiology, biochemistry) most probably also contributes to the scatter. For instance, biomechanical models predict a positive relationship between (relative) limb length and sprint speed (see Garland & Losos 1994 and references therein), and several empirical studies have corroborated this prediction (Snell et al. 1988; Losos 1990; Sinervo, Hedges & Adolph 1991; Sinervo & Losos 1991; Bauwens et al. 1995). However, limb lengths and other design characteristics are not routinely reported in the literature, so this line of investigation could not be pursued here.

How do sprint speeds of lizards compare with those of mammals? Of course, the difficulties encountered when comparing lizard data from different studies multiply when comparing lizards with mammals. Probably even more than our lizard data set, the mammal data in Fig. 5 (taken from Garland 1983) are a varied assortment, collected using widely different techniques and degrees of accuracy. In addition, the body size ranges for which speed data are available differ between the two animal groups (Fig. 5), further jeopardizing a statistical comparison. Therefore, the conclusions below must remain speculative. Moreover, the outcome of the comparison depends largely on the regression techniques used to summarize the data. When ordinary least-squares regression is used, the effect of body mass on maximal sprinting speed seems similar in lizards and mammals (Fig. 5a). As noted above, the exponents of the relationships are highly similar for the two groups. However, Fig. 5(a) also suggests that for a given body mass, lizards tend to be slower than mammals. This would corroborate the idea that, in terms of maximal attainable speed, the locomotor apparatus of lizards (sprawling gait, anaerobic fuelling, etc.) is inferior to that of mammals (erect gait, aerobic fuelling, etc.). When polynomial regression is used, a different pattern emerges (Fig. 5b). Now, sprint speeds of lizards tend to be similar to the speeds predicted for mammals of a similar body mass. This suggests that, for small body sizes, a lizard-like type of locomotion may allow speeds comparable to those of mammals (see also Biewener 1989, 1990; Blob 2000). Speed data for small mammals and for large lizards are needed to test this unexpected finding.

HOTTER IS BETTER

Our results confirm the hypothesis that ‘hotter is better’ (Huey & Kingsolver 1989), at least within the temperature range considered here. Species that are active at high body temperatures run faster than species with low mean field body temperatures. Most species in this study are said to be tested near optimal body temperatures, so it seems unlikely that the correlation between speed and field body temperature is an artefact of slow lizards being tested at suboptimal temperatures. Rather, we think that our results corroborate the idea that adaptation of the thermal physiology to lower body temperatures is at the expense of performance at the optimal body temperature. The thermodynamical properties of the constituents of the cell (particularly those of water, Calloway 1976) are usually invoked to explain this phenomenon. However, this hypothesis should be tested more carefully, comparing field body temperatures, selected body temperatures and optimal temperatures with maximal performances.

ECOLOGICAL CORRELATES OF SPRINT SPEED

Traditional statistical analyses suggest that three of the four ecological variables considered (climate, time of activity, microhabitat use) explain a significant part of the variation in sprint speed among lizard species. Some of the expectations formulated in the introduction are met. Lizards from Mediterranean and xeric climatic regions sprint faster than lizards from cool or tropical climates; diurnal lizards are faster than nocturnal lizards. Non-phylogenetic analyses also indicate differences in speed among lizards from different microhabitats, but here the prediction that climbing species should have lower (horizontal) running capacities than cursorial species proved incorrect. Instead, rock-climbing lizards sprint faster than both arboreal and ground-dwelling species. Foraging strategy (sit-and-wait vs actively foraging) did not influence maximal sprint speed. Traditional analyses of covariance also suggest that the differences in sprint speed between climates, activity periods and microhabitats could be explained through differences in body mass and body temperatures.

While it is tempting to explain the variation in maximal sprint speed in terms of differences in morphology, thermal physiology and general ecology, the results of the phylogenetic analyses strongly warn against such adaptive story-telling. When the genealogical relationships among the species considered are introduced into the analyses, the effects of the ecological factors are no longer statistically significant. This result once more stresses the importance of phylogenetic information in comparative analyses. Inspection of Fig. 1 shows that ecology and phylogeny are highly confounded within lizards, that is, phylogenetic related species tend to live in similar ecological conditions. This strongly suggests that the ecological characteristics considered are evolutionary stable. The ‘clustering’ of species with the same ecology reduces the statistical power of the tests to a great extent and differences among the ecological groups need to be very large to be significant (Garland et al. 1993; Vanhooydonck & Van Damme 1999). We conclude that the current data set and level of investigation does not allow formulating ultimate explanations of the variation in sprint speed in lizards. This will require finding a set of species for which ecology and phylogeny are not confounded. This may not be possible for the broad ecological classes used in this paper. Analyses at a more fine-grained level could be more fruitful. For instance, rather than dividing animals into such broad categories as ‘sit-and-wait’ and ‘actively foraging’, the foraging behaviour of particular species could be expressed in percentage of time spent moving, or home range size. This would allow testing the effect of the ecological parameter within a closely related group of lizards, and would circumvent the confounding effect of phylogeny.

Received 15 May 2000; revised 23 August 2000; accepted 1 September 2000