Gait-specific metabolic costs and preferred speeds in ring-tailed lemurs (Lemur catta), with implications for the scaling of locomotor costs

Animals

Three adult ring-tailed lemurs, Lemur catta, (two males, one female; mass: 2.31 ± 0.19 kg) were trained to walk, canter and gallop on a motorized treadmill. The animals were selected from a breeding colony maintained by Duke University at the Duke Lemur Center (DLC), Durham, NC, and housed in a system of large outdoor enclosures. The respirometry experiments were conducted in a climate-controlled research room, onsite at the DLC. The Duke Institutional Animal Care and Use Committee approved all procedures.

Respirometry

Metabolic measurements were made while each lemur exercised on a motorized variable-speed treadmill (Jog-A-Dog, Model DC5, Ottawa Lake, MI; belt area: 1.52 m × 0.36 m). A digital tachometer fastened to the axel of the rear drum tracked belt speed. Each animal was trained to exercise at seven speeds ranging from 0.25 to 2.25 m s⁻¹. The training regime consisted of mixed periods of exercise and rest over approximately 1 hr d⁻¹, 3–5 d wk⁻¹ for at least 6 weeks. Training familiarized the animals with the treadmill and improved their fitness levels so that at the end of the training period all lemurs could sustain 6–10 min of steady-state locomotion at each speed.

Measurements of rate of oxygen consumption were collected using an open-circuit indirect calorimetry system (Fedak et al.,1981) as each lemur exercised in a custom-built metabolic chamber (Fig. 1). The metabolic chamber (0.87 m × 0.23 m × 0.51 m) was constructed of clear Plexiglas and rested on brackets 1 cm above the moving treadmill belt. A small ribbon of weather stripping was attached to the bottom of the chamber so that the chamber effectively slid across the belt surface and was only open to the room through a 0.15 m diameter intake cut into its back wall. The main pump pulled air through the chamber at a rate of 75–100 L min⁻¹. These rates were high enough to ensure that all of the expired gas from the lemur was collected. A sub-sample of the mixed expired gas was pulled from the main flow using a small pump and regulator. The gas was then passed through columns of Drierite and Ascarite to remove H₂O and CO₂, respectively. Oxygen content was analyzed using a Sable Systems FC-1B Oxygen Analyzer (Sable Systems International, Las Vegas). The measurement system was calibrated using a nitrogen dilution technique (Fedak et al.,1981), which enabled calculation of oxygen consumption rate.

All locomotor trials lasted 6–10 min and oxygen consumption was measured throughout the trial. Rate of oxygen consumption (ml O₂) was determined for the last 3–5 min of the trial after oxygen consumption had reached a steady-state plateau. During data collection, the order of the speeds was randomized and no more than five trials were collected from any one lemur on a single day, ensuring that the lemurs were sufficiently well rested for each locomotor trial. For each individual, 2–5 trials were measured at each of the seven speeds. Gross metabolic rate was calculated as 20.1 J ml O₂⁻¹ (Blaxter,1989).

The resting rate of oxygen consumption was measured using the same settings as during locomotion, with the principal modification being that air was drawn through the chamber at rates of 75–80 L min⁻¹. During resting data collection, the room was darkened and remained quiet; typically, each lemur lay quietly on the tread surface throughout the experiment, although occasional grooming was observed. All resting measurements were recorded over a 10 min period in the morning before feeding to help ensure that each lemur was in a post-absorptive state. For each individual, 2–5 nonconsecutive resting trials were recorded.

Net locomotor costs

The net locomotor costs—calculated either per time or per distance—are partitioned from the gross metabolic rate by subtracting out some quantity representing the metabolic rate of tissue function not associated with the movement task (i.e., non-locomotor metabolism). Most previous studies of non-human primates have used the y-intercept of the gross metabolic rate versus speed linear regression as an estimate of this quantity (e.g., Taylor et al.,1982). However, a measure of resting metabolism, taken during quiet (motionless) standing, sitting or lying, has been used in more recent studies of humans and chimpanzees (e.g., Willis et al.,2005; Sockol et al.,2007; Weyand et al.,2010). Here, we used both approaches to calculate the mass-specific net COL and net cost of transport (COT), and consider their effects on the speed- and gait-specific estimates of locomotor costs.

For the COL (W kg⁻¹), estimates for each individual were obtained by subtracting the y-intercept of a linear fit to the mean gross metabolic rate per speed (over the full speed range) (intercept baseline) and mean resting metabolic rate (RMR baseline) from the gross metabolic rate. The COT (J kg⁻¹ m⁻¹) was obtained by dividing the COL by tread speed. These values were compared with the COT estimated from the “slope approach” of earlier primate studies (Taylor and Rowntree,1973; Parsons and Taylor,1977; Mahoney,1980; Taylor et al.,1982), in which the mean gross metabolic rate versus speed for all three lemurs is fit using a single linear function and the slope is taken as the COT. This slope-based estimate of the COT is sometimes referred to as the “minimum” or “incremental” COT (Full,1991) and is directly comparable with the size-scaling equations of Taylor et al. (1982). In contrast, a COT estimated from the RMR baseline is comparable with Rubenson et al. (2007).

Overground speed measurements

All overground trials were conducted on a 9.0 m × 2.0 m × 2.3 m runway enclosed by fine-screen mesh strung from wooden posts that were fastened to the floor. The mesh closed the runway at both ends and along the top and sides to completely contain the animal. A canvas blind was set in the middle of the runway to divide it into two 1 m wide corridors, effectively creating an 18 m roundabout. The floor of the runway was smooth particleboard covered with a coating of polyurethane to seal it and prevent slipping by the animal. The lemurs were placed in the runway area, and then video recorded as they moved through the enclosure. Walking speeds were self-selected without prompting; cantering and galloping speeds were usually elicited by a DLC animal technician positioned at the end of the runway, taking a step toward the animal.

Trials were filmed in lateral view with a digital high-speed video camera (Redlake Motionscope PCI-500, San Diego, CA) recording at 125 Hz. The camera was positioned approximately 3 m from the runway to reduce parallax effects. The eye and tail base were digitized in MATLAB (R2008b, The MathWorks, Natick, MA) using DLT Dataviewer (Hedrick,2008). The coordinate data were filtered using a quintic smoothing spline (Woltring,1985,1986; Walker,1998) with a constant error tolerance (“spaps,” tolerance = 0.005; MATLAB, R2008b). The average horizontal velocities of the eye and tail base were used to calculate speed as the animal walked, cantered, or galloped through the camera field of view.

Model fitting and statistics

Linear mixed model (LMM) regressions were used to describe the changes in metabolic costs against speed, with all regression models fitted through the mean values per lemur per speed. For one lemur, it was not possible to collect metabolic costs during 0.25 m s⁻¹ walks, and as a result the final dataset was unbalanced. In all models, speed was set as a fixed effect (N = 7) and individual lemur (N = 3) was set as a random effect.

To identify the best-fitting statistical model for changes in metabolic cost against speed within gaits, the Akaike Information Criterion (AIC) was used to compare the performance of linear, quadratic, and cubic regression fits. The lowest AIC indicates the best-fitting model (Akaike,1974; Sokal and Rohlf,2012). The goodness of fit of the LMM regressions was quantified using a modified coefficient of determination for mixed models (R₁²; Vonesh and Chinchilli,1997), which has been shown to perform well in simulations where LMMs were used for parameter estimation (Orelien and Edwards,2008).

Log-likelihoods (ln L) were calculated for the fit of single linear functions and for the fit of gait-specific functions to the gross metabolic rates, COL and COT against speed. Likelihood-ratio tests (LR tests, Λ; Sokal and Rohlf,2012) were then used to compare the single linear fit to the gait-specific fit, where the ln L from the single linear model fit and the sum of the ln L from the walking and cantering/galloping (gait-specific) model fits were used to compute the change in log-likelihood from one model to the other. The likelihood-ratio test probability distribution was approximated as a chi-square distribution (χ²) with degrees of freedom (df) equal to df₂ – df₁, where df₂ is the number of free parameters in the gait-specific model and df₁ is the number of free parameters in the single linear model (Sokal and Rohlf,2012). Model fitting and associated statistical testing were performed in R (version 2.9 for Mac OS) using the base, “nlme” and “lmmfit” libraries. Descriptive statistics given in the text are means ± s.d. In these cases, data were first averaged within each animal, and then averaged across animals for the species. Statistical significance was α = 0.05.

Gross metabolic rates

Gross metabolic rate increased with speed and across gaits for all individuals, consistent with most previous studies of terrestrial locomotion (Taylor et al.,1970,1982) (Fig. 2). Each lemur walked at speeds of 0.25 m s⁻¹ to 1.0 m s⁻¹, then cantered and galloped at speeds of 1.5 m s⁻¹ and above. It is important to note that during treadmill locomotion at 1.0 m s⁻¹, lemurs would switch from a walk to a canter then back to a walk for some strides. This is consistent with overground observations of ring-tailed lemur speeds and gaits, including a predicted gait transition at ∼1 m s⁻¹ (Froude number; Fr = 0.5) (O'Neill and Schmitt,2012).

A single linear regression captured much of the variation in gross metabolic rates over speed (LMM, R₁² = 0.915). The slope-based estimate of the COT is 5.26 ± 0.35 J kg⁻¹ m⁻¹ for all gaits and speeds, with an intercept of 4.05 ± 0.56 W kg⁻¹. Nevertheless, gait-specific trends were evident in the gross metabolic rates of ring-tailed lemurs. From the slowest walking speed at 0.25 m s⁻¹ to the fastest walking speed at 1.0 m s⁻¹, a quadratic model best fit the data (Table 1). In contrast, from the slowest canter at 1.5 m s⁻¹ to the fastest gallop at 2.25 m s⁻¹, a linear model best fit the data (Table 1). Comparing the single linear model with the gait-specific (piecewise) model, the gait-specific approach provided a significantly better fit to the data (Λ = 20.948; χ²_0.05 = 5.991) (Fig. 2). This suggests that the gait-specific model fits provide a more robust characterization of metabolic costs across speeds and gaits than a single linear regression.

Baseline metabolism

The intercept metabolic rate was 4.05 ± 0.71 W kg⁻¹, while the RMR was 2.77 ± 0.21 W kg⁻¹. Both of these baseline values represent substantial percentages of the gross metabolic rate, especially at slow walking speeds (Fig. 3). At 0.25 m s⁻¹, the intercept and the RMR were 81% and 55% of the gross metabolic rate, respectively. These values decreased with increasing speed. Across all four walking speeds, the percentages ranged from 81% to 39% when using the intercept baseline and from 55% to 26% when using the RMR baseline. On average, these two baselines accounted for 61% and 41% of the walking gross metabolic rate, respectively. Across the measured cantering and galloping speeds, the percentages ranged from 36% to 27% using the intercept and from 23% to 18% using RMR for the three speeds. These values were much less speed-dependent, and averaged 31% and 20% for the intercept and RMR subtractions, respectively. These data demonstrate that baseline quantities have a much larger effect on estimates of net walking costs than on net cantering and galloping costs in ring-tailed lemurs.

Net locomotor costs

Mass-specific net locomotor costs (COL, COT) were calculated using both baseline quantities (Fig. 4A,B). As expected, the different values altered the magnitude, but not the slope of the COL (Fig. 4A). The linear fits across all gaits and speeds provide slope-based estimates of the COT of 5.24 J kg⁻¹ m⁻¹ using the intercept subtraction (LMM, R₁² = 0.900) and 5.29 J kg⁻¹ m⁻¹ using the RMR subtraction (LMM, R₁² = 0.899), almost identical to the values from gross metabolic rates. However, as before, gait-specific fits were evident in the dataset, and significantly improved on the single linear model fits (COL: Λ = 19.744; χ²_0.05 = 5.991; COT: Λ = 53.584; χ²_0.05 = 9.488).

The COT calculated from the slope of gross metabolic rates assumes no change in cost across gaits and speeds, while the COT from an intercept or RMR baseline subtraction found distinct, gait-specific patterns (Fig. 4B). In particular, the slope-based estimate of COT missed the curvilinear functions across walking speeds apparent from the intercept and RMR subtraction estimates. Not surprisingly, the slope approach provided a good approximation of cantering and galloping costs estimated from an intercept baseline, but tended to overestimate the COT at 0.25–0.5 m s⁻¹ and underestimate the COT at 1.0 m s⁻¹. In contrast, the slope approach underestimated the COT at all speeds based on an RMR baseline. Using either the intercept or RMR baseline quantity, a curvilinear function best fit the walking data, while a linear function best fit the cantering/galloping data (Table 1). In both cases, the minimum COT measurement was 6.69 J kg⁻¹ m⁻¹ during walking at 0.5 m s⁻¹. It is notable that the highest walking costs were at the slowest walking speeds and near the ∼1 m s⁻¹ transition from a walk to a canter. In contrast, the slopes of the LMM fits to the cantering and galloping data were not different from zero, indicating that there is no effect of speed on cost per distance traveled, at least within the measured 1.5 m s⁻¹ to 2.25 m s⁻¹ range.

Overground speeds

A total of 140 overground trials were recorded in which the ring-tailed lemurs walked, cantered, or galloped along an indoor runway (Fig. 4C). The walking speeds of ring-tailed lemurs were tightly clustered between 0.5 and 0.6 m s⁻¹, with a mean speed of 0.59 ± 0.09 m s⁻¹ (median = 0.57 m s⁻¹; mode = 0.49 m s⁻¹). The ring-tailed lemurs almost never traveled at speeds between 0.75 and 1.5 m s⁻¹. Above 1.5 m s⁻¹, ring-tailed lemurs used a much wider range of speeds, but some clustering was evident around 2.0 m s⁻¹. For cantering and galloping, the mean speed was 2.06 ± 0.28 m s⁻¹ (median = 2.01 m s⁻¹; mode = 1.92 m s⁻¹).

Interspecific comparisons

Compared with the predictions from scaling studies of terrestrial birds and mammals, ring-tailed lemurs have low COTs for their body mass (Fig. 5), suggesting that their terrestrial locomotion is relatively economical. The general vertebrate scaling equation of Taylor et al. [1982; Eq. (8)] predicted 8.22 J kg⁻¹ m⁻¹ at all speeds and gaits, while their “primate equation” predicted 8.19 J kg⁻¹ m⁻¹. Both predictions were 36% higher than the measured COT of 5.26 J kg⁻¹ m⁻¹ for ring-tailed lemurs. The gait-specific equations of Rubenson et al. [2007; Eqs. (2)–(4)], which are based on an RMR subtraction, similarly overestimated ring-tailed lemur locomotor costs, with predictions of 9.66 J kg⁻¹ m⁻¹ for canters/gallops, and 12.0 or 11.62 J kg⁻¹ m⁻¹ for walks (using either the minimal cost or preferred-speed criteria, respectively). These predictions overestimated the COT of preferred speed canters/gallops (2.0 m s⁻¹) by 42% (5.61 J kg⁻¹ m⁻¹), and the minimum/preferred speed walking (0.5 m s⁻¹) by 42 and 44% (6.69 J kg⁻¹ m⁻¹), respectively.

Baseline subtractions

Of particular importance in the estimation of locomotor costs is how to partition the net from the gross metabolic cost, which (when determined from O₂) is necessarily an aggregate measurement. Direct determinations of the metabolic rates of non-locomotor tissues are quite difficult (Stainsby et al.,1980; Marsh and Ellerby,2006), and different baseline quantities make different assumptions regarding non-locomotor metabolism (Full,1991; Rubenson et al.,2007).

For ring-tailed lemurs, the intercept baseline is 1.46 times higher than the RMR baseline. This is consistent with other studies of terrestrial animals, which report 1.3–1.7 times higher values (Taylor et al.,1970; Schmidt-Nielsen,1972; Taylor,1977). This higher value likely reflects the statistical dependence of the intercept and is not necessarily linked to non-locomotor metabolic processes at 0 m s⁻¹. This is most apparent in studies where the intercept is negative (e.g., Chassin et al.,1976; Bransford and Howley,1977; Saunders et al.,2004). In contrast, RMR is independent of the locomotor sample.

Of course, RMR has been measured under several different conditions, including quiet (motionless) standing, sitting, and lying. Most recent locomotor studies have used the cost of standing (e.g., Griffin et al.,2003; Rubenson et al.,2007; Sockol et al.,2007; Maloiy et al.,2009); however, standing costs can be 10–25% higher than the cost of sitting or lying (e.g., Parsons and Taylor,1977; Parker et al.,1984; Levine et al.,2000). It has been suggested that this higher metabolism reflects the cost of balance, posture, and bodyweight support, which should be included in the net locomotor costs (Weyand etal.,2009,2010). For ring-tailed lemurs, the 1.46 times difference between intercept versus RMR baseline quantities implies that earlier estimates of the COL and COT that used an intercept baseline (Taylor et al.,1982) would have underestimated the cost of skeletal muscle function during locomotion, especially at walking speeds. This is of particular relevance for studies on the determinants of locomotor costs, which seek to link skeletal muscle function to metabolism.

Net locomotor costs, preferred speeds, and lemur biology

Ring-tailed lemurs have relatively low costs for walking, cantering, and galloping. Specifically, the 5.26 Jkg⁻¹ m⁻¹ COT is 36% below the predictions of Taylor et al. (1982), and the 6.69 J kg⁻¹ m⁻¹ (walk) and 5.61 J kg⁻¹ m⁻¹ (canter/gallop) are 42–44% below the predictions of Rubenson et al. [2007; Eqs. (2)–(4)]. These percentages are the lowest deviations among non-human primates (Fig. 5A–C).

These results are relevant to the view that both gait choice and gait-specific preferred speeds correspond to locomotor cost minima. First, these data suggest that the minimum COT for walking is higher than for cantering or galloping, using an RMR-based subtraction. This is consistent with the view that terrestrial animals below ∼20 kg have walking COTs that are greater than their COTs for faster-speed gaits (Rubenson et al.,2007). As a means of reducing the net locomotor contribution to daily energy expenditure then, it would be cheaper to canter/gallop than to walk a unit distance. Future studies might examine whether ring-tailed lemurs preferentially use cantering/galloping gaits in the wild. Unfortunately, current behavioral data on ring-tailed lemur travel and foraging do not differentiate among quadrupedal gaits (Sussman,1974; Simmen et al.,2010). Second, preferred speeds correspond well to the locomotor cost minima within the walking gait, indicating that ring-tailed lemurs prefer to walk at their most economical speeds. For cantering and galloping, it was not possible to sample locomotor costs over the full speed range, and therefore the linear function must be viewed as provisional. If a more complete sampling of the cantering/galloping speed range revealed a nonlinear function, as in the galloping gait of horses (Hoyt and Taylor,1981; Minetti et al.,1999), then the mean speed of 2.0 m s⁻¹ may indeed correspond to locomotor cost minima. At present though, it can only be stated that the preferred speed is similar to the speed of lowest measured cantering/galloping COT. Finally, it is notable that ring-tailed lemurs avoid their slowest speeds as well as the speed at which they transition from a walking to a cantering gait, where their measured COTs are highest.

Implications for COT scaling

The gait-specific approach indicates that ring-tailed lemurs have different COTs for walking and cantering/galloping gaits. This contrasts with earlier studies (Taylor and Rowntree, 1972; Parsons and Taylor,1977; Mahoney,1980; Taylor et al.,1982), but corroborates the more recent finding that small animals have higher COTs for walking than for other, higher-speed gaits (Rubenson et al.,2007). This is opposite the pattern in humans, for which minimum walking costs are nearly half those of running (e.g., Saibene and Minetti,2003; Rubenson et al.,2007).

The overall pattern of changing COT with gait from a small primate (e.g., ring-tailed lemur) to a larger one (e.g., human) seems to reinforce several limitations of earlier scaling studies of the COT. First, as noted by Rubenson et al. (2007), the M_b^−0.316 COT scaling slope reported in Taylor et al. (1982) is more representative of the cost of intermediate and high-speed gaits, such as running, trotting, and galloping, than the cost of walking. Second, studies that use a mass-scaling exponent alone to predict locomotor cost (reviewed in Pontzer et al.,2010) likely underestimate the cost attributable to active skeletal muscle function. This bias may be corrected, in part, by estimating the total COT from the sum of the intercept and slope exponents, and then performing a post-hoc RMR subtraction, in a manner similar to Marsh et al. (2006). However, only a few studies of non-human primates report both resting and locomotor costs for their animals. These two issues represent important limitations to the general use of the “primate equation” of Taylor et al. (1982), especially with regards to studies of walking. The criticisms of Steudel-Numbers (2003), who noted the use of tree shrews, some juvenile subjects and an atypical human running cost value as potential limitations should also be considered.