The origin of bipedal locomotion in lizards is unclear. Modeling studies have suggested that bipedalism may be an exaptation, a byproduct of features originally designed to increase maneuverability, which were only later exploited. Measurement of the body center of mass (BCOM) in 124 species of lizards confirms a significant rearward shift among bipedal lineages. Further racetrack trials showed a significant acceleration threshold between bipedal and quadrupedal runs. These suggest good general support for a passive bipedal model, in which the combination of these features lead to passive lifting of the front of the body. However, variation in morphology could only account for 56% of the variation in acceleration thresholds, suggesting that dynamics have a significant influence on bipedalism. Deviation from the passive bipedal model was compared with node age, supporting an increase in the influence of dynamics over time. Together, these results show that bipedalism may have first arisen as a consequence of acceleration and a rearward shift in the BCOM, but subsequent linages have exploited this consequence to become bipedal more often, suggesting that bipedalism in lizards may convey some advantage. Exploitation of bipedalism was also associated with increased rates of phenotypic diversity, suggesting exploiting bipedalism may promote adaptive radiation.

Although adaptation and natural selection are central concepts in evolutionary science, less attention is often given to the concept of exaptation (Gould and Lewontin 1979; Gould and Vrba 1982). This term refers to a feature that has a new biological function, which was not that which caused the original selection of the feature (Buss et al. 1998). Although examples of adaptations abound in the literature, fewer examples of exaptation exist. The most often cited example of exaptation is the evolution of feathers in birds. Given the evolution of feathers appears to predate the evolution of flight, it is likely that the feathers of birds first evolved for another function (e.g., thermoregulation), but were later co-opted for flight (Ostrom 1974; Bakker 1975; Ostrom 1979). Exaptations may be important because they might result in key innovations that can provide the stimulus for adaptive radiations (Simpson 1953; Losos and Mahler 2010). For example, key innovations in the toe pads of geckos (Russell 1979), phytophagy in insects (Mitter et al. 1988; Farrell 1998), and pharyngeal jaws in fish (Stiassny and Jensen 1987; Mabuchi et al. 2007) have led to adaptive radiations, as indicated via species richness. More recently, it has been hypothesized that the evolution of bipedal locomotion in lizards, may have arisen as an exaptation and like the feathers of birds, has been latter exploited (Aerts et al. 2003). By examining the evolution of bipedal locomotion in this group, we cannot only rigorously test the exaptation hypothesis, but if confirmed, compare rates of phenotypic diversity of exaptated features with phenotypic diversity in other adaptive radiations.

The hypothesis suggesting a consequential origin for bipedalism in lizards may have arisen from the lack of an obvious functional explanation for bipedalism in this group. Bipedalism has evolved independently many times among terrestrial animals, being present in primates, dinosaurs, birds, hopping marsupials, hopping placentals, insects, and lizards, yet for many of these groups structural changes associated with bipedalism are more obvious. In birds, dinosaurs, and primates, the forelimbs appear repurposed for either grasping or flight (Snyder 1962), while hopping on large hindlimbs is thought to provide an energetic advantage through the elastic storage of energy (Dawson and Taylor 1973). However, for lizards and insects, morphological changes associated with bipedalism are often more subtle (Snyder 1954). The forelimbs of lizards do not appear adapted for purposes other than quadrupedal locomotion, nor do they use a hopping gait (Snyder 1954). Further, both lizards and cockroaches are the only animals that start on all their limb pairs, then transition to the bipedal gait (Snyder 1962; Full and Tu 1991). Thus for lizards, it seems the mechanism and advantages of bipedalism are much less clear.

Instead, forward dynamic modeling in the small lacertid lizard Acanthodactylus erythrurus suggested bipedalism in lizards may be an accidental consequence of morphology and acceleration, perhaps in response to selection for increased maneuverability (Aerts et al. 2003). This model showed the pitching rotation that lifts the head and trunk is probably a mechanical consequence of the lizard's forward acceleration combined with a rearward shift in the body center of mass (BCOM). Given this origin, bipedalism need not have been selected for per se, but could have evolved as a consequence of a posterior shift in the BCOM and increased acceleration. There was some support for this hypothesis, the BCOM was shown to be correlated to the percentage of total strides a lizard runs bipedally among 18 species of agamid lizard, and acceleration showed a distinct threshold above which strides tend to be bipedal among four species of agamid lizards (Clemente et al. 2008).

However, this passive bipedal model developed by Aerts et al. (2003) suggests that acceleration is crucial for bipedal running. Yet, as noted by Aerts et al. (2003) some lizards appear able to run bipedally without acceleration. For example, the teiid lizard Aspidoscelis sexilineata was able to sustain bipedal running over a 1 m racetrack (Olberding et al. 2012), Liolaemus lutzae and Tropidurus torquatus sustained bipedal running over sand dunes (Rocha-Barbosa et al. 2008), and among four species of Australian agamid examined, two (Ctenophorus nuchalis and Ctenophorus femoralis) were capable of bipedal strides at accelerations below zero (Clemente et al. 2008). Later models exploring this, suggested that steady-state bipedalism may be possible via more active dynamic mechanisms such as changes in orientation of the body during the stride, as well as changes in the ground-reaction force profile (Van Wassenbergh and Aerts 2013).

These observations suggest that while bipedalism in lizards may have evolved as a consequence of acceleration and a rearward shift in the BCOM, lizards may be exploiting this consequence of bipedalism by further altering dynamics of the lizards stride (Aerts et al. 2003). The significance of this exploitation is extensive. If lizards are exploiting bipedalism, this might suggest that bipedalism itself may be a co-opted spandrel (Buss et al. 1998) and indeed hold some selective advantage within lizards.

To test the exaptation hypothesis, first the passive model is tested empirically. The two major predictions in the passive bipedal model suggest that bipedalism is a consequence of a rearward shift in BCOM combined with acceleration. The BCOM is examined among 124 species across 14 families to resolve if bipedal species show distally directed BCOM. The evolutionary rate of BCOM is also examined to determine if it increases in groups with a diverse number of bipedal species. Further, logistic regression models are fitted to acceleration scores to determine if a distinct threshold in acceleration exists between quadrupedal and bipedal strides.

Second, deviations from the passive bipedal model are explored. By quantifying deviations from the model the relative contribution of passive morphological versus active dynamic variation to bipedalism can be determined. This is tested by comparing the acceleration thresholds measured from lizards with those predicted from the passive model based solely on morphological changes.

Third, the exploitation hypothesis is tested. If bipedalism is being exploited through changes in dynamics, it would be expected that deviations from the model increase over evolutionary time. To examine this, deviations between measured and predicted acceleration thresholds are compared against node age in Australian agamids.

Finally if bipedalism does indeed appear to be an exaptation, we can compare rates of phenotypic diversification to examine if these parallel those seen in adaptive radiations from other taxa.



A combination of wild caught and museum specimens was used to estimate horizontal body center of mass (BCOMhori) among lizard species. In total 362 lizards from 124 species were measured. Prior to BCOMhori measurements snout to vent length (SVL) and the distance from the vent to the center of the pelvis, between the hip joints, was recorded.

To measure BCOMhori a horizontal wooden or plastic beam was suspended between two scales (for mass > 300 g, Amput APTK461 ± 0.1 g; mass < 300 g, Suofei SF-718 ± 0.01 g; Fig. 1). The length of the beam was varied to be longer than the total length of each lizard measured. The scales at each end were tared for the mass of the beam, after which each specimen was placed along the beam, in a neutral posture, with its limbs tucked under each hip/shoulder joint, and its snout tip at one edge. Because the torque around the BCOM should be equal for each scale, the forces recorded on each scale will be proportional to the radius from the BCOM. Using this, the length of the horizontal BCOM from the snout tip can be determined using equation (1);

display math(1)

where BCOMhori = length of the horizontal BCOM from the snout tip, L = length of the beam, W1 = the mass recorded at the snout tip, and W2 = the mass recorded on the distal scale. Measurements of SVL and vent to hip length were then used to determine the position of the BCOM forward of the hip (BCOMhip).

Figure 1.

The two-scale method for determining the horizontal body center of mass. The distance from the snout tip to BCOMhori, X is given by X = W2 · L/(W1 + W2), where L is the length of board, between the midpoint of the scales, onto which the lizard is placed, and W1 and W2 are the recorded mass on the scales at the snout tip and tail, respectively.

Because the BCOM is typically correlated with size, size effects were removed by calculating phylogenetically corrected residuals of the BCOMhip from SVL using the function phyl.resid.R in the phytools package (version 0.3–72) (Revell et al. 2012) in R (version 3.0.2) (R Core Team 2012). Branching patterns and branch lengths were based upon the squamate phylogeny of 4161 species reported by Pyron and Burbrink (2013) and Pyron et al. (2013), which was subsequently pruned in R, using the function from the ape package (version 3.0–11) (Paradis et al. 2004). The strength of the phylogenetic signal was determined by comparing trait change to a null hypothesis of Brownian motion using Blombergs K statistic, implemented in the function phylosig.R from the phytools package.

An extensive literature search was undertaken to classify species capable of running bipedally or quadrupedally. Because for many species it was ambiguous whether species were capable only of quadrupedal locomotion, we marked all species where bipedal strides have not been reported as “unknown,” rather than assume quadrupedal locomotion. The residual BCOMhip of “bipedal” and “unknown” species was compared using a phylogenetically informed ANOVA implemented using the phylo.ANOVA.R function in the phytools package.

The evolutionary rate change of size-corrected BCOMhip was also examined. Of interest was whether rates of evolution differ between bipedal and unknown lineages, and further whether the rate has significantly changed throughout evolutionary history, particularly toward branches with many bipedal species. This was examined using two methods.

Classification of species as bipedal or unknown was mapped onto a phylogenetic tree and the fit of seven different models of evolution were compared to best characterize the evolution of BCOMhip. Most basically a single rate Brownian motion (σ2) model was tested (BM1), along with an Ornstein–Uhlenbeck model with one optimum (θ) for all species of lizards (OU1). A second Brownian motion model was also fitted, which allowed for separate Brownian motion rates for each regime, bipedal and unknown. The OUM model fitted a two optima OU model that allowed different optima (θ) for each regime. Further three two-optima OU models were included that allowed different Brownian motion rates (OUMV), different strength of selection parameters (OUMA), and variation in both Brownian motion and strength of selection parameters (OUMVA) between selective regimes. These were applied using the OUwie.R function from the OUwie package (version 1.40) (Beaulieu et al. 2012) implemented in R. This method requires a phylogenetic tree where the internal nodes denote the ancestral selective regimes. Ancestral node states were determined using the hidden rate model implemented in the corHMM.R function in the corHMM (version 1.13) (Beaulieu et al. 2013) package in R.

This first method described above limits rate variation to be between the two regimes (bipedal and unknown) input into these models. Given that at least some uncertainty must exist in the classification into the unknown regime, we compared patterns of rate variation described above to patterns where no specific prior hypothesis for how the variation in evolutionary rate is structured throughout the history of the species is provided. This was achieved using a Bayesian Markov Chain Monte Carlo (MCMC) approach as described by Revell et al. (2012). This method identifies a possible rate change by simultaneously sampling rates and shift points in proportion to their posterior probability, and then collapsing the posterior sample into an estimate of the parameter of interest. This method was implemented using the evol.rate.mcmc.R function from the phytools package, and simulations were allowed to run for 100,000 generations, sampling every 100 generations. The first 2000 generations were excluded as burn-in, and then the functions minSplit.R and posterior.evolrate.R were used to find the split in evolutionary rate, and to determine average evolutionary rates before and after the split.


A subset of the lizards above was used to estimate the acceleration threshold in running lizards. A total of 44 adult lizards from 10 species were included. Before measurement BCOMhip, SVL and hind limb length (HLL) were determined for each individual.

Acceleration during strides was determined by filming lizards running along a racetrack. The racetrack was 3.6 m long, 0.6 m wide, and 0.6 m tall sides, composed of wood, with sandpaper adhered to the floor to provide traction. One side of the racetrack was replaced with clear Perspex sheeting, allowing strides to be recorded simultaneously from the side and the top. High-speed cameras positioned laterally to (HiSpec1, Fastec Imaging, San Diego) and directly above (Fastec IL3, Fastec Imaging or an EX-ZR200, Casio, Tokyo, Japan) the racetrack recorded views at 240 fps. Cameras were synchronized using either internal triggers, or a light pulse from an LED mounted on the racetrack. Prior to recording, eight landmarks were painted on each lizard, using Liquid Paper™, to mark the pelvis and the hindlimb joints of each lizard and facilitate digitizing the video images. A mark along the pelvis was digitized in Matlab (version R2012a, Mathworks, Inc., Natick, MA) using DLTdv3.m (Hedrick 2008), and displacement data were smoothed using the mean square error algorithm, implemented via spaps.m, because this approach was least error prone (Walker 1998). Acceleration scores were then calculated as the mean acceleration throughout the entire stride.

Each stride began at footfall of a hindlimb, and ended with the subsequent footfall (subFF) of the same limb, thus each stride consists of two hindlimb steps, and a subFF. Because lizards typically use a quadrupedal trotting gait (Farley and Ko 1997), the lateral camera was used to classify steps and subFF as being bipedal or quadrupedal, by the presence or absence of a touch of the contralateral forelimb. The stride was then classified as being quadrupedal or bipedal if both steps and subFF were classified similarly, or a transitional stride if steps and subFF changed from quadrupedal to bipedal, or bipedal to quadrupedal. We excluded the first three steps from standstill because these have previously been shown to have considerable variation in acceleration and kinematics (McElroy and McBrayer 2010).

Classifying threshold accelerations between bipedal and quadrupedal locomotion is difficult because the long tails of many species produce high moments of rotational inertia (Carrier et al. 2001). This might allow some strides to remain quadrupedal at higher than predicted accelerations and some to remain bipedal in its absence. To reduce this error, acceleration thresholds were calculated using two different methods. Most directly an average of all transitional strides from quadrupedal to bipedal or vice versa (ACCtrans) were used. However, these strides can represent a small proportion of the total number of strides. Therefore acceleration thresholds with logistic regression was implemented using the inflection point P (0.5) as an estimate of the threshold (ACClogistic). For some species accelerations of bipedal and quadrupedal strides show perfect separation, and therefore did not converge using standard logistic regression, therefore the function brglm.R, from the brglm package (version 0.5–9) (Kosmidis 2008) was applied for all species. This function uses a modified-score approach to bias reduction by iteratively fitting local generalized linear models on a pseudodata representation.

The empirically measured acceleration thresholds above were also compared to the predicted acceleratory threshold based on simple morphology for each species. This predicted acceleratory threshold (ACCpred) was based upon the model presented by Aerts et al. (2003), which used equations of motion to determine the vertical forces acting on the forelimbs of an accelerating lizard (eq. 2),

display math(2)

where Fyforelimb is the vertical force acting on the forelimbs, g is gravity at −9.81 m/sec2, Mtot is total body mass, L is trunk length, ay and ax are the vertical and horizontal acceleration of the BCOM, respectively, BCOMhip and BCOMvert are the horizontal and vertical positions of the BCOM, respectively, and xfhind is the point of application of force onto the ground by the hindlimb.

During acceleration, when the vertical force acting on the forelimb (Fyforelimb) reaches zero, the front of the body will begin to lift up passively because we assume no attachment force between the forefoot and the ground. This criterion was used to determine the acceleration at which bipedalism should occur passively by rearranging equation (2) when Fyforelimb = 0;

display math(3)

Ax was then used to represent the predicted acceleration threshold for each species (ACCpred). To calculate Ax for each species, species means for BCOMhip were used and vertical acceleration over the stride was assumed to be negligible (Ay = 0). BCOMvert was assumed to approximate hip height at midstance, and xfhind was estimated as the fore-aft position of the foot relative to the hip at midstance. Hip height and position of the foot were digitized during acceleration trials described above. Because only the influence of morphology was of interest, any potential differences in kinematics among species were removed. To do this proportional hip height and foot position from HLL were calculated for each species, to create an average for all 10 species. This average proportion was then multiplied by species mean HLL to calculate kinematically identical, but size-corrected estimates of BCOMvert and xfhind for each species.

Rates of evolution for ACClogistic, ACCtrans, and ACCpred were estimated using residual maximum-likelihood estimates using the function ace.R in the Ape package in R. Significant differences between these rates were determined by comparing a likelihood model where each acceleration threshold estimate evolved at a distinct evolutionary rate to a model where all thresholds are constrained to evolve at a common evolutionary rate, following the methods detailed in Adams (2013). To compare rates directly each acceleration threshold estimate was shifted along the number line to remove negative values, then log-transformed to reduce the effects of scale (Felsenstein 1985; O'Meara et al. 2006; Adams 2013). Finally, predicted ancestral states for acceleration thresholds and divergence times were calculated using the ace.R and the branching.times.R function, respectively, from the Ape package.



The BCOM was significantly correlated with body size (SVL) by the following function; Log10(SVL) = 0.923Log10(BCOMhip) − 0.461, where both SVL and BCOMhip are in units of mm (R2 = 0.71, P < 0.001; Fig. 2A). The change in the residual horizontal BCOMhip among squamates (here after simply referred to as BCOM) is shown in Figure 3A. Broad scale changes in the BCOM tend to be fairly small, but appear significantly associated with phylogenetic history (K* 0.29, P < 0.001). Phylogenetically informed ANOVA suggests that lizards known to run bipedally tend to show a rearward shift in the BCOM, toward the hip, when compared to lizards for which the mode of locomotion is ambiguous (F1,109 = 32.1, P = 0.007; Fig. 2B).

Figure 2.

(A) Body center of mass from the hip (BCOMhip) against snout to vent length (SVL) showing the line of best fit from which residual BCOMhip was determined (B) Differences in phylogenetically informed residual BCOMhip from SVL between lizards capable of bipedal locomotion and lizards for which the mode of locomotion is ambiguous (unknown). The bold lines indicate the 50th percentile of the data, whereas the lower and upper bounds of the box represent the first and third quartiles, respectively. Whiskers represent the 10th and 90th percentiles values, and stars represent outliers.

Figure 3.

Phylogenetic change in size-corrected horizontal body center of mass. (A) A phylogeny for the 124 species used in the current study, modified from Pyron et al. (2013). Families of lizards are indicated by black bars. Dark blue bars indicate an anterior shift in the BCOMhip, whereas red bars indicate a posterior shift. The length of the tree is given by the length of the indicator bar in the top left corner. The gray bar indicates the position of a likely shift in the rate of evolution of BCOMhip. Red boxes at the tips indicate observed bipedalism. (B) A subtree of phylogeny above showing the radiation of Agamidae. Rates and posterior densities are from an analysis of residual BCOMhip on the tree shown in (A). Posterior probabilities of the rate shift being on each edge of the tree (if >0.01) are shown by the red fraction of the pie chart. Total length of the tree is 117 Mya.

Of the seven evolutionary models used to describe variation in BCOM, the best model based upon ΔAIC and Akaike weights was the OUMA model, indicating that both the strength of selection and the optima differ between the two selective regimes, but the rate of stochastic motion tends to be similar (Table 1). Rates of evolution, strength of selection, and optima were determined for each regime using the weighted mean of the four best models, where the AIC weights of each model were used as the weights (Table 2). The optima for bipedal lizards was −0.158, suggesting selection for a rearward shift in the BCOM. The remaining species (unknown) show an optima just positive of zero (0.033), suggesting that for these species selection trends toward the mean or a weak forward shift in the size-corrected BCOM. This was reflected in a slightly higher strength of selection in bipedal species (Table 2). Rates of evolution among bipedal lizards (0.0009) were similar when compared to lizards for which the mode of locomotion was ambiguous (0.0008).

Table 1. The fit of alternative evolutionary models for the evolution of size-corrected BCOMhip among bipedal lizards and those where the ability for bipedal locomotion is ambiguous (unknown)
  1. Likelihood scores and AIC scores were determined using the function OUwie.R in the OUwie package (Beaulieu et al. 2012) in R. Delta AICc scores and AICc weights were then calculated using the qpcR function akaike.weights.R (Ritz and Spiess 2008). These results indicate the OUMA model best describes the evolution of BCOMhip, which estimates different optima (θ) and a different strength of selection parameter (α) for each regime.

Table 2. Parameter estimates and their associated 95% confidence interval (CI) for the weighted average for the four best models based on AIC weights (OUM, OUMV, OUMA, OUMVA) for the evolutionary change in residual BCOMhip data for bipedal lizards and those where the ability for bipedal locomotion is ambiguous (unknown)
 Bipedal95% CIUnknown95% CI
 estimate(lower, upper)estimate(lower, upper)
  1. 95% CIs were calculated via bootstrapping (n = 500), using simulated datasets created via OUwie.sim.R from the OUwie package (Beaulieu et al. 2012) in R. Lower and upper bounds represent the 2.5 percentile and the 97.5 percentile, respectively.

α0.030(0.018, 0.078)0.028(0.018, 0.077)
σ20.0009(0.0002, 0.0023)0.0008(0.0005, 0.0021)
θ−0.158(0.248, −0.071)0.033(−0.011, 0.069)

Supporting this, MCMC simulations indicated a shift in the evolutionary change in BCOM occurred after the node leading to agamids lizards (Fig. 3B). The evolutionary rate before the shift was estimated to be (sig21) 0.0004, but after the shift increased to (sig22) 0.0006. This rate shift coincides with a posterior shift in the BCOM.


We analyzed 534 strides from 10 species of lizards all of which were capable of bipedal locomotion. Among these species, half showed several strides of steady-state (zero acceleration) bipedal locomotion. Figure 4A shows their position across the phylogenetic tree. Species that are capable of steady-state bipedalism appear to be located on distal branches of the tree, whereas species that show only accelerating bipedal strides, appear more basal, with the exception of Ctenophorus fordii (Fig. 4A).

Figure 4.

Transitional acceleration during strides. (A) The phylogenetic relationship for the 10 species of agamid for which accelerations of strides were recorded. Filled circles represent species for which bipedal strides were observed with accelerations below zero. Open circles represent species for which bipedalism was only observed during acceleratory strides. (B) The relationship between predicted ancestral states for mean transition acceleration strides (ACCtrans) with node age. (C) The difference between predicted ancestral mean transition acceleration (ACCtrans) and the predicted acceleration based on morphological parameters (ACCpred) with node age.

For all but two species, logistic regression significantly separated bipedal and quadrupedal strides, suggesting strong support for the acceleration model in bipedal locomotion (Fig. 5, Table 3). The lack of significant separation in the two remaining species is probably a consequence of low sample size, these being represented by 12 and 14 strides, which likely limits the ability of logistic analyses to define a threshold with any certainty. Despite broad similarity in position (Fig. 5), acceleration thresholds from logistic analysis (ACClogistic) were not significantly related to acceleration thresholds calculated from the mean of transitional strides for each species (ACCtrans; R2 = 0.22, P = 0.180; Table 3).

Table 3. Analysis of the acceleration threshold between quadrupedal and bipedal locomotion
   (mm/sec2)(mm/sec2)  BCOMhipHLLACCpred
SpeciesN (ind.)N (strides)mean ± SE (Nstrides)(lower CI, upper CI)ZP(mm) ± SE.(mm) ± SE(mm/sec2)
  1. ACCtrans is the mean acceleration of all transitionary strides between quadrupedal and bipedal gaits. ACClogistic is the threshold acceleration estimated from bias reduced logistic models (brglm.R from the brglm package (Kosmidis 2008) in R) of bipedal and quadrupedal strides using the 50% probability to determine the transition acceleration. ACCpred is the predicted acceleration threshold based on morphological data input into the model of Aerts et al. (2003).

C. adelaidensis4145621 ± 1530 (2)34831.730.08311.3 ± 1.130.2 ± 0.99233
C. caudicinctus564878 ± 542 (8)351 (−172, 814)3.75<0.00114.9 ± 1.460.1 ± 4.96472
C. cristatus458−467 ± 744 (13)915 (−833, 2667)3.130.00110.6 ± 0.9101.3 ± 2.93351
C. femoralis515−2828 (1)−211.100.2707.4 ± 0.154.4 ± 0.54052
C. fordii4551141 ± 776 (8)9709 (6432, 17,371)3.120.0027.6 ± 0.546.4 ± 0.94640
C. nuchalis71034589 ± 1289 (11)6939 (4621, 11,538)2.860.00419.1 ± 4.946.7 ± 2.19927
C. ornatus3431177 ± 759 (16)1138 (−6, 12,047)2.100.03510.8 ± 0.2101.3 ± 2.94057
C. reticulatus1423101 ± 877 (10)3376 (1451, 12,047)2.150.03217.3 ± 0.042 ± 0.09877
C. scutulatus6961720 ± 854 (30)4502 (2396, 9074)3.41<0.00110.3 ± 1.380.5 ± 6.23865
I. lesureri5441726 ± 417 (15)1566 (289, 3041)2.400.01615.0 ± 1.1100.9 ± 4.84309
Figure 5.

Accelerations and predicted thresholds for (A) Ctenophorus cristatus and (B) C. scutulatus. Circles represent the acceleration of individual strides, classified as being either bipedal, quadrupedal (Quad) or a transitional stride between the two gaits (Trans). Solid vertical line represents the mean acceleration for transitional strides, with gray bars representing ± standard error. The curvilinear line represents logistic regression for quadrupedal and bipedal strides, with upper and lower confidence intervals. Vertical dashed line represents the predicted acceleration threshold based on morphological data input into the model of Aerts et al. (2003). Silhouettes to the right illustrate a typical bipedal (above) and quadrupedal (below) posture for each species.

Predicted threshold accelerations (ACCpred) using only morphological dimensions and based upon the model of Aerts et al. (2003) are shown in Table 3. For five of 10 species predicted thresholds lie above the confidence intervals (CI) for both ACClogistic and ACCtrans, suggesting lizards are running bipedally at much lower accelerations than predicted by morphology alone. For three species, Ctenophorus reticulatus, C. ornatus, and C. nuchalis, ACCpred was higher than CIs for ACCtrans but lay below the upper CI of ACClogistic, still supporting results for the five species above. For the final two species, C. scutulatus and C. fordii ACCpred were above ACCtrans but below ACClogistic. Regression between predicted and measured thresholds suggested ACCpred was not significantly related to ACClogistic (R2 = −0.03, P = 0.419), but was significantly and positively related to ACCtrans (R2 = 0.56, P = 0.008).

Differences between predicted and measured thresholds were further explored in relation to evolutionary history. Rates of evolution (σ2) for ACClogistic and ACCthresh were significantly higher than rates of evolution for predicted thresholds based on morphology (ACCpred; Table 4). When ancestral states for ACCtrans were calculated, there was a significant positive relationship between node age and ACCtrans, suggesting a trend toward lower bipedal thresholds in more derived species, and higher thresholds in basal species (R2 = 0.39, P = 0.044; Fig. 4B). Further when ancestral states for the difference between ACCpred and ACCtrans are compared to node age, there was a significant negative relationship (R2 = 0.59, P = 0.009; Fig. 4C), suggesting that basal nodes show the strongest similarity to morphologically predicted thresholds, whereas distal nodes show the greatest divergence. Statistical support for this pattern was weaker when ACClogistic was used (ACClogistic ∼ node.age; R2 = 0.09, P = 0.221; ACCpredict – ACClogistic ∼ node.age; R2 = 0.18, P = 0.139).

Table 4. Evolutionary rate comparisons for three different measures of acceleration thresholds for 10 species of Australian agamids lizard calculated using likelihood as described in Adams (2013)
σ2 observedσ2Standard errorLRTPAICobsAICcommon
  1. ACCtrans is the mean acceleration of all transitionary strides between quadrupedal and bipedal gaits. ACClogistic is the threshold acceleration estimated from bias reduced logistic models of bipedal and quadrupedal strides. ACCpred is the predicted acceleration threshold based on morphological data input into the model of Aerts et al. (2003).

σ2 common      
 0.0392 33.80<0.00164.4094.21
Pairwise analyses      
ACClogistic vs. ACCtrans  6.380.01167.0071.38
ACCtrans vs. ACCpred  15.03<0.00122.4835.51
ACClogistic vs. ACCpred  30.88<0.00139.3268.21


These results indicate strong support for the exaptation hypothesis proposed by Aerts et al. (2003), which suggests bipedalism in lizards may have evolved as a consequence of acceleration combined with a posterior shift in BCOM, but has later been exploited. Variation in bipedalism, does indeed appear to be reflected in variation in the BCOM (Figs. 2, 3) with a distinct rearward shift in the BCOM evident in bipedal lizards. Further there was strong statistical support for a distinct acceleration threshold to bipedalism for many species of agamids (Table 2). Yet as predicted by Aerts et al. (2003), some species appear to be exploiting this consequence and extending bipedal stretches, likely through changes in dynamics (Van Wassenbergh and Aerts 2013). By applying the model of Aerts et al. (2003) but excluding any dynamic variation, we can compare predicted acceleration thresholds purely from morphology, to actual thresholds seen in lizards, which combine morphological, kinematic, and kinetic variation. Differences between these models can then be thought of as the influence of dynamics on bipedalism. Comparisons between predicted morphological and actual thresholds suggest at best only 56% of the variation in bipedalism can be explained by morphological variation. Further, acceleration thresholds get lower in more distal lineages, while the difference between predicted morphological and actual thresholds increases. These patterns are consistent with a deviation from accidental, morphologically based bipedalism toward dynamically controlled bipedalism within Australian agamids.

Although these results demonstrate the process of exaptation among bipedal lizards, selection is necessary not only to explain the adaptation and by-products that were available for co-optation but also to explain the process of exaptation itself (Losos 2011). This means we must explain not only why acceleration and rearward shift in the BCOM improved fitness, but also how and why bipedalism itself has since been exploited.

Variation in BCOM may result from many sources, for example, increases in tail length, such as that associated with arboreal habitats (Jusufi et al. 2008; Thompson et al. 2009) may result in a rearward shift in the BCOM, whereas selection for larger wider heads with greater biting forces (Verwaijen et al. 2002; Lailvaux et al. 2004) could results in a forward shift in the BCOM. Although one hypothesis consistent with biomechanical models suggests that a caudal shift in the BCOM could be favored for species that are selected for higher maneuverability (Aerts et al. 2003). A caudal shift in the BCOM would benefit maneuverability because forces to change the heading and to align the body to its new heading do not conflict with one another (Jindrich and Full 1999). This combined with short bursts of acceleration, which would also benefit maneuverability, suggest a likely early evolutionary scenario for bipedalism.

However, increasing divergence from morphologically based predictions of acceleration thresholds over evolutionary time suggest that dynamically based variation becomes increasingly more important. Several kinematic parameters may act to reduce the acceleration threshold, and therefore make bipedalism achievable at lower accelerations. Changing the angle of the body or tail relative to the ground would aid in bipedalism through changes in the BCOM (Irschick and Jayne 1999). Active tail lifting during the acceleration phase will result in the trunk rotating upwards and backwards, through the increased angular momentum of the tail, making bipedalism possible at lower acceleration (Aerts et al. 2003). Further, this trunk rotation will cause the BCOM to shift posteriorly reducing the nose-down pitching effect of gravity, and extending a bipedal stretch (Van Wassenbergh and Aerts 2013). Similarly, increases in the height of the hip throughout the stance phase will increase the proportion of the stride where the BCOM lies above the line extending the hindlimbs ground-reaction force vector, which would further contribute to a nose-up pitching moment about the hip (Van Wassenbergh and Aerts 2013). Perhaps most important dynamic factors may be spatial and temporal footfall patterns. Placing the foot further forwards during the stride, or modifying the ground-reaction force profile such that a greater proportion of the vertical impulse is applied shortly after initial foot contact, would increase the duration in which the ground-reaction force vector are anterior to the BCOM, further increasing the nose-up pitching moment (Van Wassenbergh and Aerts 2013). The relative contribution of each of these variables, and variation among species remains the subject of further investigations.

Although these biomechanical models provide information on how consequential bipedalism may have been exploited to extend bipedal stretches, there is less information on why this should be the case. The sheer number of Australian agamids that have exploited bipedalism strongly suggests some advantage to bipedalism within this group, and among other families. Original studies suggested bipedal running in lizards is faster than quadrupedal locomotion, because the front limbs are no longer at risk of interfering with the hindlimbs, allowing greater stride length (Snyder 1962). However, comparisons between bipedal and quadrupedal strides showed that hindlimb kinematics do not greatly differ (Irschick and Jayne 1999). Further, speeds of bipedal and quadrupedal were not significantly different in two American species of lizard, Callisaurus draconoides and Uma scoparia (Irschick and Jayne 1998) nor did they differ significantly between quadrupedal and bipedal strides for four species of Australian agamid lizards (Clemente et al. 2008).

Alternatively bipedally running lizards were thought to be advantaged through more favorable locomotor efficiency, because no energy needs to be consumed to move the front limbs. This hypothesis had some support for hopping kangaroos, which are able to store energy elastically (Dawson and Taylor 1973), but two other studies comparing bipedal and quadrupedal running in similar sized animals, showed similar costs of transport among the gaits (Fedak and Seeherman 1979; Roberts et al. 1998). Finally when Clemente et al. (2008) compared the proportion of strides a lizards runs bipedally with endurance capacity, there was a significant negative correlation, suggesting that bipedally running lizards do not receive an energetic advantage, but instead may even incur some metabolic cost.

Instead some studies have suggested the advantage of bipedalism is the possible increase in environmental perception during locomotion by elevating the head and expanding the visual field (Kohlsdorf and Biewener 2006). Studies on the behavioral repertoire of Sceloperous woodi show some support for this hypothesis (Tucker and McBrayer 2012). Bipedal locomotion was more commonly used when approaching a smaller obstacle, over which a bipedally running lizard could see, when compared to a larger obstacle, which was above the height, and therefore obscured the vision, of even a bipedally running lizard (Tucker and McBrayer 2012).

A bipedal gait may also be advantageous during obstacle negotiation because of the increased height of the BCOM, in contrast to quadruped lizards that must elevate the body prior to an obstacle (Kohlsdorf and Biewener 2006). Despite a limited number of lizards species in which this has been tested, in a study on obstacle negotiation during steady quadrupedal running in S. malachiticus, forward speed decreased in the presence of obstacles (Kohlsdorf and Biewener 2006). However, this decrease was absent from the bipedally running teiid lizard Cnemidophorus (Aspidoscelis) sexlineata during similar obstacle negotiation trials (Olberding et al. 2012). This suggests lizards may avoid decreases in performance associated with obstacle negotiation by adopting a bipedal gait. If performance during obstacle negotiation does indeed contribute to fitness, then selection would favor bipedal locomotion, explaining its prevalence and exploitation among Australian agamid lizards.

Given that bipedalism does appear to be exploited among agamids, has this led to adaptive radiations based upon this innovation? Rates of morphological diversification in the BCOM for bipedal and nonbipedal lineages, as well as the rate for the predicted acceleratory threshold based on morphological differences, tend to be very similar (Tables 2 and 4). In comparison to this, bipedal locomotion has allowed phenotypic variation to occur at much higher rates than morphological variation alone. Rates for the evolution of transitional acceleratory thresholds were 10 times higher than for predicted morphological acceleratory thresholds (Table 4), mirroring the difference in rates changes seen in Anolis lizards inhabiting newly colonized islands (Mahler et al. 2010; Revell et al. 2012). Because transitional acceleratory thresholds are likely a result of kinematic and kinetic variation, increased rates may be the result of the inclusion of neuromuscular and behaviorly controlled phenotypes that may be more labile (Huey and Bennett 1987; Gittleman et al. 1996; Blomberg et al. 2003). The extent to which bipedalism is associated with increased phenotypic diversity in other groups of lizards remains to be explored, along with the extent to which exaptations in general can promote adaptive radiations.


I thank S. Blomberg, A. Niehaus, and R. Wilson from the University of Queensland for helpful advice on statistics and comments on earlier versions of this manuscript. I thank L. Revell and D. Adams for their statistical help and for making their R code freely available. I thank N. Wu from the University of Queensland and both A. Amey and P. Cooper from the Queensland Museum for help with measurement of museum specimens. I thank K. Harvey from the Southern Animal Hospital for help collecting acceleration data on lizards. I also thank two anonymous reviewers who contributed many helpful comments on earlier drafts. This study was funded by a DECRA fellowship awarded to CJC. Animals were collected under permits WISP11435612 (QLD) and SF009075 (WA) and experiments conducted under animal ethics approval numbers SBS/195/12/ARC (UQ) and RA/3/100/1188 (UWA). The author declares no conflict of interest.


The doi for our data is 10.5061/dryad.14mb6.