The quadrupedal walking gait of the olive baboon, Papio anubis: an exploratory study integrating kinematics and EMG

The origin and evolution of primates is closely related to arboreal habitats (e.g. Cartmill, 1974; Hamrick, 2001; Toussaint et al., 2020) and many extant primates continue to live in arboreal environments, or at least invest a certain amount of time foraging in the trees on thin and flexible branches (Hunt, 2016). Such an environment obviously implies discontinuity, variability, instability and oscillations of the substrates that perturb the inherent mechanical requirements of legged locomotion, i.e. propulsion, balance and stability of the walking gait. It has been shown that primates exhibit a set of morphological adaptations and kinematical adjustments coupled to a behavioural flexibility that allow them to meet the functional requirements demanded by this complex environment (e.g. Cartmill et al., 2020; Demes et al., 1994; Druelle et al., 2020; Hunt et al., 1996; Larney and Larson, 2004; Larson et al., 2001; Schmitt, 1999; Schmitt et al., 2006; Thorpe et al., 2007). During quadrupedal walking, primates use diagonal sequence/diagonal couplet (DSDC) gaits, large limb angular excursions, large elbow yield, long stance phase, low stride frequency and a weight support shifted toward the hindlimbs (see Larson, 2018, for a review). These features can be integrated under the compliant gait model of primate walking (Schmitt, 1999). Overall, these characteristics should particularly allow for control of pitching, yawing and rolling moments, and limit the vertical oscillations of the centre of mass (CoM), and hence branch oscillation, and increase maintenance of balance (e.g. Cartmill et al., 2020, 2002; Schmitt et al., 2006).

Parallel to these functional adjustments, the muscular activity pattern also presents specificities. Electromyography (EMG) studies on non-human primates have provided significant knowledge about muscle basic activity pattern in different locomotor modes such as climbing, brachiating and bipedal walking (e.g. Hirasaki et al., 1995; Ishida et al., 1974; Jungers and Stern, 1981; Stern and Larson, 2001, 1981), but the way muscle recruitment has been altered in non-human primates in relation to the arboreal niche remains poorly understood (Boyer et al., 2007; Courtine et al., 2005; Larson and Stern, 2009; Patel et al., 2015). During primate quadrupedal walking, distal limb muscles are suggested to be under higher cortical control including different speed-related modulations of the limbs compared with other quadrupedal mammals (Courtine et al., 2005). The different inertial properties of non-human primates compared with other quadrupeds (Grand, 1977; Druelle et al., 2019) should also affect the way the work is performed, and the forces transmitted. Overall, more flexible capabilities of the neuromotor control system are suggested in primates (Courtine et al., 2005; Jungers and Anapol, 1985; Shapiro et al., 1997), with the hand and the foot having distinct functional roles (Druelle et al., 2018; Hashimoto et al., 2013; Lawler, 2006; Patel et al., 2015). Furthermore, although the phasic recruitment of muscles is generally similar to that of other quadrupedal mammals (Courtine et al., 2005; Higurashi et al., 2019; Shapiro and Jungers, 1994), specificities have been emphasized for the humeral retractors (Larson and Stern, 1987, 1989, 2007), and hip extensors might also be used for different functions (Larson and Stern, 2009; Reynolds, 1985). As a result, primate quadrupedal control and mechanics are considered atypical compared with that of other (terrestrial) quadrupedal mammals. However, basic biomechanical and physiological data regarding the typical quadrupedal walking pattern of non-human primates is, unfortunately, non-existent. Integrative studies examining the kinematics, the dynamics of the CoM and the muscular activity (EMG) all together are very rare (see Courtine et al., 2005).

To explore the quadrupedal walking gait of non-human primates and the suggested neuromotor flexibility, we conducted an integrative analysis exploring the limb kinematics, the CoM mechanics (using kinematics) and the muscle activity of six extrinsic hindlimb muscles in olive baboons, Papio anubis. Baboons can be considered as (semi-)terrestrial primates as they have developed a set of cursorial adaptations that mean they are more suited for terrestrial locomotion and for long-distance walking. For instance, they have been shown to move between 3.6 and 9.5 km per day, making them the non-human primates with the longest daily travel distance (Garland, 1983); their limbs are relatively extended with a digitigrade hand posture when walking, they display convergent natural pendular periods of their forelimbs and hindlimbs and their proximal limb mass distribution suggests good propulsive action capacities (Druelle et al., 2017a; Patel, 2009; Patel et al., 2012; Raichlen, 2004; Zeininger et al., 2017).

In this study, animals were trained to voluntarily cooperate using positive reinforcement techniques. Our refined experiment used surface probes (sEMG) that did not need surgical procedures. We first recorded the muscular activity of two female baboons walking quadrupedally using their typical DSDC gait over flat regular ground at their voluntary walking speed. This allowed us to describe their typical quadrupedal walking gait in terms of spatio-temporal parameters and muscular activity. We then recorded the gait of one female baboon when walking DSDC on a suspended horizontal and cylindrical substrate (branch) and while spontaneously using lateral sequence/lateral couplet (LSLC) gaits for a few strides on the ground. We thus provide a preliminary picture of the normal quadrupedal walking pattern in baboons and its flexibility (through kinematic modulation) in the context of different gait and substrate. First, and because the baboon is defined as an adapted quadrupedal walker, we expected it to be able to make use of an inverted pendulum mechanism when walking steadily on the ground using its preferred walking gait (DSDC; as previously observed in ring-tailed lemurs and capuchin monkeys: O'Neill and Schmitt, 2012; Demes and O'Neill, 2013). Second, we specifically evaluated whether the usual muscle activation pattern observed in DSDC walking is altered in other gait and substrate contexts. When the baboon uses the unusual LSLC gait on the ground, changes are expected in proximal muscles (i.e. in the main limb motors), whereas walking on a different substrate, such as a branch (imposing higher balance requirements and certainly higher control), should mainly affect the recruitment of distal muscles (Boyer et al., 2007; Courtine et al., 2005; Patel et al., 2012). Exploring the kinematics, the movements of the CoM and the muscle activity together should provide an integrative picture of the quadrupedal walking pattern in the baboon and how these are connected (see Courtine et al., 2005). In this context, a reduced efficiency in the mechanics of the CoM (for example, walking on a branch requires kinematical adjustments that tend to reduce the vertical oscillations of the CoM; see Gàlvez-Lòpez et al., 2011; Schmitt, 1999) should match an alteration of the muscular activity toward an increased activation.

The olive baboons, Papio anubis, were housed at the Primatology Station of the CNRS (Rousset-sur-Arc, France). The experiments presented in this study and conducted inside the primatology centre are part of a large ongoing project on bipedal walking skills in non-human primates (G.B. and F.D., unpublished data). The heterogeneous character of the quadrupedal data presented here results from opportunistic periods of data collection that were not initially intended to be considered together. All the procedures that are described in this study were approved by the ethical committee on animal experimentation no. 14 (Projet 68-19112012, CEEA-14 Marseille).

Prior to the experiments, primates were trained over a period of 18 months and focused on the habituation and desensitization of individuals to the whole experimental setup. All training was based on positive reinforcement techniques and systematic desensitization (Prescott and Buchanan-Smith, 2016; Schapiro et al., 2003) and was conducted by the same two trainers, P.M.-V. and B.R. The first training period was carried out in a large indoor cage where the two individuals were pair-housed for 18 months. During this period, individuals were mainly habituated to the close proximity of and interaction with trainers by means of training other basic behaviours such as target or parking training. After 6 months of training (46 sessions of training of approximately 30–45 min), individuals cooperated to get out of their enclosure on a leash and to walk quadrupedally beside the trainer on the experimental setup. Individuals were then moved to a bigger indoor/outdoor enclosure where they could move freely. This training allowed us to lead the baboons to the technical platform for data collection.

Training sessions started when the baboons were 1.5 and 2 years old. The collection of the data presented in this study was performed during three recording sessions for individual 1 (Id1), 4 July 2013, 31 July 2013 and 15 December 2017, and during one recording session for individual 2 (Id2), 3 December 2013. At the time of data collection, Id1 was 3 years old and weighed 8.14 kg in 2013 and 7.5 years old and weighed 13.5 kg in 2017. Id2 was 3.5 years old and weighed 8.05 kg in 2013. At 3 years of age, olive baboons have already experienced significant growth-related changes in body morphometrics and their general morphotype is very similar to that of adults (see fig. 6 in Druelle et al., 2017a). The maturation of their neuromotor control also appears to be largely complete (see Druelle et al., 2017b; Rose, 1977), as has also been observed in other cercopithecoid species at this age (e.g. Vilensky and Gankiewicz, 1990; Zeininger et al., 2017; Dunbar and Badam, 2000).

Muscle activity was recorded at 2000 Hz using a wireless Zerowire system (Aurion Srl, Milan, Italy) with MyoResearch XP Master Edition software (Noraxon^®). We focused on 6 muscles commonly studied as actuators of the three main joints of the hindlimb (hip, knee and ankle), and easily identifiable at the surface of the skin: gluteus medius, biceps femoris, rectus femoris, tibialis anterior, gastrocnemius (medialis or lateralis) and peroneus longus, of the right side (Fig. 1). To ensure accurate electrode placement, the muscular topography was previously assessed on olive baboon cadavers of a similar size and age that were available at the primatology station. Baboons were instrumented during a short period of anaesthesia (∼60 min); the anaesthesia was initiated with an intra-muscular injection of ketamine (4 mg kg⁻¹) and medetomidine (40 μg kg⁻¹) and maintained with gas isoflurane (1%). This anaesthesia was short and light enough to shorten the recovery period using atipamezole as an antagonist. After locally shaving, cleaning and degreasing the skin, electrodes were taped at the level of the muscle belly (approximately at mid-length of each muscle and electrode pairs and always parallel to fibre direction). We used fabric legwear and trunkwear (close-fitting clothes, see Fig. 6B) to ensure a good adhesion of the equipment and to protect the electrodes. In the context of the present study, we have worked on the technical platform for biomechanics (Motion Analysis of Primates, MAP) that is permanently installed in an outdoor enclosure of the primatology centre (see Berillon et al., 2011, for details). The experiments started after the baboons were totally recovered from the anaesthesia (i.e. they were fully active and able to climb and run), between 60 and 90 min after their arrival in the enclosure.

The animals walked on a straight and horizontal calibrated walkway and on a horizontal branch (diameter 0.16 m, positioned at approximately 1 m above the ground), where they could easily be video-captured using an integrated multicamera system (Norpix Inc.) with 3 synchronized HD-video cameras running at 200 frames s⁻¹ (Baumer HXC1) (see Berillon et al., 2010, for a general description of the experimental setup); synchronized recording was proceeded with Streampix 4 (Norpix Inc.). Two cameras were laterally placed perpendicular to the walkway, to accurately record in a large field of view the main sagittal spatio-temporal and joints kinematics. One additional camera was positioned obliquely to increase the accuracy for visually identifying the locomotor events, i.e. touch-down and lift-off of the limbs. EMG and video recordings were synchronized using an external digital signal; in addition, a fourth lateral video camera (running at 60 frames s⁻¹) was driven by the MyoResearch XP software and thus software-synchronized with the EMG recording.

We first qualitatively selected appropriate quadrupedal sequences during which the individual was walking steadily, along the platform or the branch. The quadrupedal strides were defined from a right hindlimb touchdown to the next right hindlimb touchdown. As we noticed some speed variation between consecutive quadrupedal strides, we calculated an acceleration/deceleration index as follows: (average speed stride_n+1−average speed stride_n)/average speed stride_n. For the present analysis, we removed the quadrupedal strides for which speed variations were above 10% between consecutive strides [note that given the small number of lateral sequences recorded (n=6), we kept one stride for which the speed variation was 26% in order to keep a sample of 4 strides and, therefore, use non-parametric statistics, see below]. Table 1 shows the dataset included in the present analysis after applying this strong selective criterion.

We calculated the following spatio-temporal parameters: cycle duration (time period between two right hindlimb touchdown events), frequency (inverse of cycle duration), duty factor (proportion of stance phase of the right hindlimb relative to cycle duration), stride length (horizontal displacement of the tip of the right foot during one stride), speed (estimated from the horizontal displacement of the CoM; see below) and limb phase (time period between the right hindlimb touchdown and the right forelimb touchdown relative to the cycle duration; <50% is a lateral sequence, >50% is a diagonal sequence). The spatio-temporal parameters were made dimensionless using geometric scaling equations (Hof, 1996) and the shank length was used as the scaling factor (Aerts et al., 2000). A total of 20 anatomical reference points were digitized by S.M. and N.S. using the DLTdv7 application developed in MATLAB by the T. Hendrick Lab (Fig. 2). Video footage was analysed for one individual and the kinematics were therefore extracted for Id1 in different contexts (see Table 1). When the tracked anatomical references were obscured by the movement of other limbs, we applied a piecewise linear interpolation to add the missing datapoints. Movement patterns were visually checked using a custom-made script in MATLAB to ensure a satisfactory digitization result, i.e. a smooth movement with no anomaly. Data were then filtered per stride using a 4th order zero phase shift low-pass Butterworth filter with a cut-off frequency of 5 Hz. We then applied a cubic spline interpolation over a time base with 100 points in order to stride-normalize the individual quadrupedal cycles. Tracking the joints and body segments during the walking strides allowed us to measure 6 internal joint angles (joint extension corresponds to their increase) (Anvari et al., 2014): the ankle [estimated between the foot taken as a whole (from heel to tip) and the shank segment], the knee, the hip, the shoulder, the elbow and the wrist; as well as 5 external joint angles (with respect to the horizontal): the shank, the thigh, the trunk, the humerus and the forearm.

We first visually checked the signals to ensure the absence of abnormalities and artefacts. We used MATLAB (R2019a) to analyse, filter and create an envelope of the EMG signals (see Supplementary Materials and Methods 1 for an evaluation of different methods, and Fig. S1). The following successive steps were applied: band-pass 4th order Butterworth filter (5 Hz to 500 Hz), centred, rectified (full-wave), low-pass 4th order Butterworth filter (10 Hz), rectified (full-wave) (see Supplementary Materials and Methods 1 for Matlab script). We then applied a cubic spline interpolation over a time base with 200 points in order to stride normalize the individual quadrupedal cycles. This provides an envelope of the muscular activity pattern on which we then applied different statistical analyses (see below). Each individual muscle amplitude was also normalized per day of experiment based on the maximum peak contraction value of the set of quadrupedal walking sequences. Because of these treatments, we were able to average the multiple strides per individual.

We proceeded with two analyses: (1) interindividual differences (Id1 versus Id2; Table 1) when walking quadrupedally on the ground using the typical DSDC gait; and (2) the effect of gait (diagonal versus lateral) and substrate (ground versus branch). For both analyses, the stride represents the independent statistical unit. Given the small sample size and the heterogeneity, we applied exact (non-parametric) tests: the Permutation tests for independent samples. (1) Duty factor, dimensionless stride length and dimensionless stride frequency (and duration) were compared between individuals using Permutation tests for independent samples and were regressed against dimensionless speed; the Pearson correlation (r) allowed us to test for significant relationships. (2) Differences in angle average values over a stride are tested between diagonal walking on the ground and diagonal walking on the branch, and diagonal walking on the ground and lateral walking on the ground using Permutation tests for independent samples, as well as the range of motion (angle amplitude over a stride), the spatio-temporal parameters (duty factor, stride length, stride duration, speed) and the CoM mechanics (%C, RA, Recovery, W_ext).

In order to evaluate potential differences between the walking contexts, we divided the EMG signals into 4 phases of equal duration, representing 25% of the cycle duration. The numerical integration of these different phases allowed us to estimate the proportion of EMG activation per phase and its distribution across phases. We conducted a Chi-square test of independence (χ²) to assess whether there is a significant relationship between the distribution of the activation pattern between individuals (1), and within individual for the different contexts studied (2), i.e. walking on the ground using a diagonal sequence, walking on a branch using a diagonal sequence and walking on the ground using a lateral sequence. Because the data collected in 2017 were obtained the same day on the same individual, we were also able to compare for this specific day and individual the maximal activation values between gaits and substrates using Permutation tests for independent samples. The statistics were conducted using the software for exact non-parametric inference StatXact 3.1 (Cytel, Inc., Cambridge, MA, USA). The significance threshold was set at P<0.05.

The data used here were from Id1, 15 December 2017 (diagonal/ground n=12; Table 1). Time plots of the hindlimb and forelimb joint and segment angles are shown in Figs 3 and 4, respectively. During baboon quadrupedal walking, the hip extends gradually during the stance phase up to ∼90 deg. It flexes rapidly during the swing phase until it reaches ∼50 deg and remains in a constant position at the end of the swing phase just before the foot touches the ground and while the knee is extending. The knee starts flexing slowly during the stance phase and reaches a plateau around mid-stance, but the flexion increases rapidly just before lift-off when the ankle is rapidly extending from 65 to 90 deg. The swing phase is characterized by a rapid flexion of the knee up to 90 deg during one-third of the swing phase and then by a rapid extension, reaching 140 deg at touch-down. The ankle shows the same pattern as the knee but is slightly delayed. The trunk angle is maintained around 3 deg on average and remains stable during the quadrupedal stride; the movement amplitude is ∼7 deg. The shoulder shows a reverse pattern compared with the hip; it flexes gradually during the stance phase of the forelimb until reaching ∼40 deg and increases during the swing phase until 80 deg. The range of motion of the humerus is 39 deg while the amplitude of the thigh (femur) is 44 deg. The elbow flexes during the beginning of the swing phase from ∼130 deg to ∼90 deg and then extends during the second part of the swing phase up to 120 deg. It increases slowly during the stance phase, but it reaches a plateau around mid-stance, as also observed in the knee. The end of the stance phase is characterized by an extension of the elbow from 130 deg to 140 deg. The wrist shows an extreme extension around 220 deg in the last third of the stance phase that precedes a significant roll off before lift-off. The wrist flexes to ∼100 deg and then extends again in preparation for touch-down at ∼180 deg. The high wrist extension corresponds to the typical digitigrade hand posture in baboons and the wrist dorsiflexes gradually during two-thirds of the stance phase until reaching an extreme extension when the ipsilateral hindlimb touches down.

The data used here were from Id1, 15 December 2017 (diagonal/ground n=12) and 4 July 2013 (diagonal/ground n=3). In our sample, the recovery rate of energy was 57% on average when the walking speed was under 0.9 m s⁻¹ and tended to decrease at higher speed (Fig. 5A). Note that although we provided a linear regression model for this relationship, a second-degree polynomial might be a better fit for the data (R²=0.34 versus R²=0.26 for the linear model). A broader speed range would enable confirmation of the linear or the upside-down U-shaped relationship between recovery rate and speed in baboons. Recovery rate was significantly negatively correlated with congruity, i.e. the more the KE and PE fluctuate out of phase, the better the recovery (Fig. 5B). The percentage recovery was also significantly correlated with limb phase. A limb phase value closer to 0.55 corresponded to a higher level of recovery (Fig. 5C). There was no relationship between recovery rate and the relative amplitude of the KE and PE (Fig. 5D).

The data used here were from Id1, July 2013 (n=3+9) and from Id2, December 2013 (n=19; see Table 1).

Table 2 shows the average values per individual of the spatio-temporal parameters for walking quadrupedally at comfort speed on the ground using DSDC gait. The two individuals studied adjusted their spatio-temporal parameters the same way according to speed (see Fig. S2). Both individuals modulated their walking pattern by altering both their frequency and stride length. Id1 tended to increase walking speed mainly by increasing stride frequency (Id1: slope=0.12; Id2: slope=0.04), whereas Id2 increased speed mainly by increasing stride length (Id1: slope=3.6; Id2: slope=5.7). Duty factor was significantly greater in Id1 than in Id2, although the absolute difference was small (67% and 64%, respectively, Permutation test: 3.538, P=0.0001). Within the limited speed range observed, duty factor was not significantly related to speed for either of the individuals studied. The dimensionless stride length was longer in Id2 than in Id1 (4.76 and 4.22, respectively, Permutation test: −3.108, P=0.0004).

The bursting EMG patterns were very consistent between the two individuals (Fig. 6). The gluteus medius (hip extensor and rotator muscle), the biceps femoris (hip extensor and knee flexor), the gastrocnemius (ankle extensor muscle) and the peroneus (foot evertor muscle) were active at the beginning of the stance phase. The maximum burst activity was approximately 25% of the stance phase, i.e. when the right hindlimb is still in a protracted posture and when the (contralateral) left hindlimb starts to swing and the left forelimb touches the ground. The rectus femoris (hip flexor and knee extensor) showed a reverse pattern, with a burst of activity when the limb starts to be retracted and the knee maintained around 120 deg. This occurred at approximately 75% of the stance phase and corresponds to the (contralateral) left hindlimb touching the ground. There was also a certain level of recorded activation of the rectus femoris in Id1 at the beginning of stance phase, when the gluteus medius, the biceps femoris and the gastrocnemius are very active. The distinct activation pattern of the rectus femoris was the only significant difference found between the two individuals (χ²=12, d.f.=3, P=0.0069). While the activity of the muscles located in the thigh showed very short bursting patterns, the burst activity of the gastrocnemius and peroneus remained during a longer period of the stance phase, i.e. until touch-down of the left hindlimb. The tibialis anterior (dorsiflexion of the foot and inversion) started its activity at the very end of stance phase, just after the (ipsilateral) right forelimb touch-down and showed a burst of activity just after the right hindlimb lift-off when the ankle is rapidly flexed. Its activity then decreased during the second period of the swing phase. There was no activity of the gluteus medius, rectus femoris, biceps femoris, peroneus and gastrocnemius during the swing phase, but the biceps femoris and gastrocnemius appeared to start their activity at the very end of swing phase, in preparation for touch-down.

The data used here were from Id1, 15 December 2017 (diagonal/ground n=12, lateral/ground n=2, diagonal/branch n=6), and to increase the sample for the lateral walking gait, also from Id1, 4 July 2013 (lateral/ground n=2; see Table 1).

Table 3 shows the average values of the spatio-temporal parameters for walking quadrupedally on the ground using diagonal sequences (DS-G), for walking quadrupedally on the branch using diagonal sequences (DS-B) and for walking quadrupedally on the ground using lateral sequences (LS-G). We observed a significant difference in stride frequency (and duration) between diagonal and lateral gaits (1.26 and 1.44 s⁻¹, respectively, Permutation test: 2.193, P=0.0209) as well as in the percentage congruity (%C) of the BCoM (38.7% and 53.6%, respectively, Permutation test: −2.272, P=0.0159). There was a significant difference in stride length and speed when walking on the ground versus the branch (stride length: 0.73 m and 0.66 m, respectively, Permutation test: 2.277, P=0.0197; speed: 0.92 m s⁻¹ and 0.79 m s⁻¹, respectively, Permutation test: 2.089, P=0.0335).

The ankle angle was more extended and the shank angle more flexed during lateral gait compared with diagonal gait (ankle: 82.1 deg and 76.4 deg, respectively, Permutation test: −2.509, P=0.0121; shank: 51.4 deg and 53.7 deg, respectively, Permutation test: 2.157, P=0.0247). The range of motion (ROM) of the knee angle was greater during diagonal walking than during lateral walking (51.5 deg and 47.3 deg, respectively, Permutation test: 1.914, P=0.0247).

On the branch, the baboon increased the ankle extension and the ROM of the ankle as well as the shank extension compared with the ground (ankle: 82.3 deg and 76.4 deg, respectively, Permutation test: −3.165, P=0.0004; ankle ROM: 44.9 deg and 38.3 deg, respectively, Permutation test: −2.08, P=0.0339; shank: 56.6 deg and 53.7 deg, respectively, Permutation test: −2.79, P=0.0023). The ROM of the trunk angle was reduced on the branch compared with the ground (4 deg and 6.3 deg, respectively, Permutation test: 2.862, P=0.0015). For the forelimb, the shoulder angle was kept more extended on the branch compared with the ground and the wrist was more dorsiflexed (shoulder: 58.4 deg and 55.3 deg, respectively, Permutation test: −2.09, P=0.0326; wrist: 187 deg and 180.8 deg, respectively, Permutation test: −2.53, P=0.0081). Fig. S3 shows the average pattern for each joint and segment angle studied in each gait and substrate context.

The bursting EMG profiles were very consistent across gaits and substrates (Fig. 7). There was no difference between the lateral and the diagonal sequences performed on the ground as far as the bursting EMG profiles were concerned. But the maximal burst of activity of the biceps femoris, gastrocnemius and tibialis anterior significantly increased in lateral sequences (Permutation test: −1.909, P=0.0467; Permutation test: −2.042, P=0.0379; and Permutation test: −2.595, P=0.0055, respectively).

With regard to the quadrupedal strides performed on the ground and on the branch, we observed a significant difference of the activation pattern at the level of the peroneus (χ²=10.62, d.f.=3, P=0.0137). While the activity of the peroneus was concentrated at the beginning of the stance phase when the baboon was walking steadily on the ground, its activity was kept during the entire stance phase when walking on the branch. Furthermore, the maximal activity of the peroneus was significantly higher on the branch than on the ground (Permutation test: −2.858, P=0.0022). The maximal burst of activity of the rectus femoris was significantly lower on the branch (Permutation test: 2.106, P=0.0231).

The present study aimed to explore the quadrupedal walking pattern of baboons using different levels of analysis, via limb and body kinematics (also used to estimate the dynamics of the BCoM) and muscular activity. Overall, our study shows that the dynamics of the BCoM in a baboon walking quadrupedally on the ground using its preferred walking gait (DSDC) is effective and can be energetically efficient. In contrast, walking on a branch appears to give priority to stability, but how it affects efficiency requires further exploration. In addition, walking using the LSLC gait appears to increase muscular effort and may reduce the potential for high recovery rates in baboons. Importantly, although our study supports specific assumptions about stability and efficiency in non-human primates, further data including a larger sample size are required.

Overall, the patterns of muscular activation during steady quadrupedal walking in baboons were consistent within and between individuals (Fig. 6B shows an overview of the muscular activity pattern during quadrupedal walking). The phasic activities of the six recorded muscles are very similar to the muscular activity profile described in macaques (Courtine et al., 2005; Higurashi et al., 2019). Furthermore, despite differences in the footfall pattern between non-primate quadrupedal mammals (lateral sequences) and non-human primates (diagonal sequences), the activity pattern of the hindlimb muscles in baboons is very similar to that observed in other primates such as chimpanzees, spider monkeys, macaques and lemurs (Ishida et al., 1974; Kimura et al., 1979; Larson and Stern, 2009), as well as in dogs (Deban et al., 2012; Goslow et al., 1981) and cats (Rasmussen et al., 1978). Therefore, we can assume that the basic functions of the muscles for retracting and protracting the limbs, as well as to support the body against gravity and inertial forces are conserved across groups.

We first hypothesized that, given the ‘cursorial’ adaptations of baboons (e.g. Druelle et al., 2017a; Patel and Polk, 2010; Raichlen, 2004), they should be able to make use of an inverted pendulum mechanism when walking steadily on the ground. Our results support this hypothesis as we observed that KE and PE mainly fluctuate out of phase (congruity=39%) during normal quadrupedal walking using DSDC gaits. The recovery rates observed in baboons in this study (22–84%) cover an important range of variation compared with other specialized quadrupeds such as dogs (50–70%; Griffin et al., 2004), but these are close to the range of values available for non-human primates, such as ring-tailed lemurs and capuchin monkeys (35–71% and 11–65%, respectively) during quadrupedal walking (O'Neill and Schmitt, 2012; Demes and O'Neill, 2013). Griffin et al. (2004) developed a model showing that the pendulum-like CoM movements in quadrupedal animals are a combination of inverted pendulums around both the pectoral and pelvic girdle. These pendulums are directly influenced by the footfall pattern and body mass distribution. For example, a BCoM located at mid-trunk coupled to a forelimb that lags behind the hindlimb by 25% would result in a flat trajectory of the BCoM (Griffin et al., 2004). Baboons have a BCoM positioned at mid-trunk compared with dogs that have a cranially positioned BCoM (Druelle et al., 2019), hence baboons may rely on limb phase only to produce ‘optimal’ fluctuations of their BCoM. In the present study, we found that the ipsilateral forelimb lagged behind the hindlimb by 61% on average (minimum 54%, maximum 69%) and there was a significant relationship between recovery rate and limb phase (i.e. the lag between forelimbs and hindlimbs; see Fig. 5C). These results support the model of Griffin et al. (2004), but do not support the ‘reduced cost zone’ of 65–72% estimated recently by Miller et al. (2019) for non-human primates. However, note that, as recognized by the latter, body mass distribution can also affect the model and specific primate conformation is likely to shift the reduced cost zone of footfall timing. In addition, given the small size of our sample, our results should be carefully considered before rejecting previous assumptions.

Our results, in addition to those provided by O'Neill and Schmitt (2012) and Demes and O'Neill (2013), do not support the hypothesis of Ogihara et al. (2012) suggesting that diagonal sequence gaits impede good recovery rates in primates. The baboon appears, however, much more efficient than the macaque (Ogihara et al., 2012) when walking on the ground. As previously suggested (Druelle et al., 2016), macaques may be less equipped for quadrupedal walking than baboons. Because baboons commonly travel at least 3.5 km per day, while macaques move less than 1.5 km (Garland, 1983), the specific ecological requirements of baboons may have shaped these species into more efficient walkers (i.e. higher potential for relying on inverted pendulum strategies) than macaques. Given the ecological niche of olive baboons and their large home range (Garland, 1983; Henzi et al., 1992; Isbell et al., 1998; Pebsworth et al., 2012), they should be under strong selection for walking efficiency (Druelle et al., 2017a). This could also be the case for ring-tailed lemurs and capuchin monkeys, which both exhibit good recovery rates during quadrupedal walking (O'Neill and Schmitt, 2012; Demes and O'Neill, 2013).

When walking on a branch, an inherent instability (rolling and pitching moments) is in action. Kinematical adjustments should reduce the vertical oscillations of the CoM (Gàlvez-Lòpez et al., 2011; Schmitt, 1999) and in this way give priority to stability over efficiency (but see Schmitt, 2011). It has been shown that animals increase their stability on arboreal substrates through employing strategies that reduce the oscillations of their BCoM (Gàlvez-Lòpez et al., 2011). However, these strategies can be of a different kind. For instance, while cats (medium sized, ∼4.4 kg) make use of slower speeds and shorter swing phase durations on a narrow runway, dogs (large sized, ∼28 kg) rely on high-speed locomotion and high frequency to gain dynamic stability (Gàlvez-Lòpez et al., 2011). The adjustments observed in the olive baboon (13.5 kg in the present study) when walking on a (suspended, non-bending) branch are a decrease in stride length and speed, a decrease in the movement amplitude of the trunk and a trend towards a longer relative contact time. There is also an increase in ankle and wrist extension that results from a slight external rotation of the distal limb segments to fit the rounded shape of the substrate. The decrease in trunk range of motion may result in part from the shorter stride length and thus from the lower speed, as well as from the shoulder, which is kept more extended on the branch. More in-phase fluctuations of the KE and PE (congruity=46%) were also observed compared with walking on the ground. These adjustments (certainly closer to the cat strategy) reduce the vertical oscillations of the BCoM, and with the increased activity of the peroneus (increasing torque generation), meet the balance control requirements. This general strategy may reduce the possibility of relying on an efficient inverted pendulum-like exchange of energy, but we did not observe a direct significant difference in recovery rate. The baboon thus appears to adjust its movement and to increase the control in the distal segment to gain stability, but more data are required to assess how much this strategy influences efficiency. Interestingly, a previous study showed that in Lemur catta, walking on a raised horizontal pole does not significantly affect energetic recovery (Schmitt, 2011). Furthermore, limb interference is important during diagonal walking on the ground. The baboons move their right hindlimb on the right (or left) side of their right forelimb, while the left hindlimb moves on the right (or left) side of the left forelimb (Fig. 6B). On the branch, there is no space for limb interference and overstriding during walking, and the hindlimbs need to be positioned behind the forelimbs at touch-down. This can also contribute to the shorter stride lengths and lower speed observed on the branch.

A study on bonobos found that when they walk on a horizontal arboreal substrate, they reduce their stance phase, take longer strides and reduce stride frequency (Schoonaert et al., 2016). This strategy appears to be very different from that used by the baboon in the present study (note that both the size of the specimens and the substrate diameter are different: 12 cm for the bonobos and 16 cm in the present study), as well as from those used by lemurs and marmosets, which all tend to decrease their speed and increase the contact time with the substrate (Franz et al., 2005; Young et al., 2016). However, we wonder whether the pole height may also be of importance. For instance, the bonobos were walking on a pole that was positioned approximately 30 cm above the ground, and hence falling was not really a risk encountered by the animals (Schoonaert et al., 2016). This emphasizes the importance of considering morphological differences and kinematic strategies coupled to risk evaluation among primates for negotiating arboreal substrates (hominoids, cercopithecoids, platyrrhines and lemuriforms all present different morphological features; Fleagle, 2013). Although balance requirements are shared between animals when walking on a horizontal branch, how these demands are met (by controlling pitching, yawing and rolling moments) can occur in very different ways, i.e. from static to dynamic stability of the BCoM. It is possible that all these strategies allow primates and other mammals to limit the vertical oscillations of the BCoM, given their respective morphology (e.g. Gàlvez-Lòpez et al., 2011; Schmitt et al., 2006; Young et al., 2016).

Muscular effort is influenced by the footfall pattern (Bishop et al., 2008; Griffin et al., 2004); hence, energetic cost is probably an important factor in footfall timings (Miller et al., 2019). We hypothesized that the usual muscle activation pattern observed in diagonal gaits should be altered towards an increased activation of the proximal muscles when the baboon uses the unusual lateral gait on the ground and of the distal muscles (for increasing control) when it walks on a substrate that imposes higher balance requirements (such as a branch). First, we observed an increased activity of the biceps femoris, the gastrocnemius and the tibialis anterior when walking using LSLC gait on the ground. Second, the peroneus activity increased substantially in intensity and duration during the stance phase when walking on the branch. Generally, the peroneus is suggested to stabilize the ankle and to be an extensor and evertor of the foot. Boyer et al. (2007) suggested that the peroneus (longus) is an important muscle for positioning the foot at touchdown and for resisting inversion throughout the stance phase in primates. The activity of the peroneus observed in our study when the baboon was walking on the branch is similar to that of a lemur species walking quadrupedally on large and small poles (Boyer et al., 2007). While it is not strongly active on the ground during diagonal and lateral gaits in baboons, it appears to be specifically recruited for resisting foot inversion during the stance phase on the branch. Our kinematics support this, as we observed an increase in ankle extension (and external rotation). The more outward and everted position of the foot should ensure a better fit between the foot and the substrate, thus increasing friction and allowing transmission of the rotating moment (Lammers and Gauntner, 2008; Preuschoft, 2004).

The biceps femoris is described as a hip extensor and a knee flexor (Swindler and Wood, 1973). It is active during the initial part of the support phase. At touch-down, the limb is in a protracted position and the knee is relatively extended (∼140 deg) (Larney and Larson, 2004); the biceps femoris is, therefore, acting as a thigh (hindlimb) retractor by flexing the knee (∼110 deg at midstance) and extending the hip (from 50 deg to 75 deg between touch-down and mid-stance). The fact that its activity increases substantially during lateral walking can be directly related to the higher stride frequency (relative stance phase does not change). The increase in activity of the gastrocnemius and the tibialis anterior may also result from the higher stride frequency. A faster retraction and protraction of the limb would require more muscular fibres to be active. Lateral walking also led to an increase in extension of the ankle that can be related to the increase in activity observed in the gastrocnemius and the tibialis anterior. But the functional reason for this higher extension remains elusive at this stage (note that such an ankle extension in this context also appears to be observed in capuchin monkey; see fig. 2 in Wallace and Demes, 2008). On the branch, stride length decreases as well as speed, and the activity of the rectus femoris (for controlling limb retraction) is reduced in this context.

The present study provides new exploratory data on the normal quadrupedal gait pattern in olive baboons. It is also an attempt to better understand the flexibility of the quadrupedal gait in different contexts. Baboons mainly walk quadrupedally on the ground using a DSDC gait and our evaluation of the dynamics of the BCoM using kinematics suggests that this is more efficient than walking using the LSLC gait. Also, from a motor perspective, a significant increase in muscular activity should result in a higher level of oxygen consumption. Hence, the most efficient baboon remains the one walking on the ground using DSDC gaits (potentially with a tight limb phase: 50%<limb phase<60% and at low speed, but data from a larger speed range are required to assess how it influences the energetic recovery rates, i.e. linearly or not). Moreover, our data suggest that the use of a different walking gait, or walking on a branch, both result in less efficient forms of locomotion. Although our analysis on this aspect is preliminary, this emphasizes a possible limitation in the flexibility of the walking pattern when efficiency is the target. In other words, walking economy could be related to a stereotyped, non-flexible pattern, while gait adjustments and flexibility ensure stability in an arboreal habitat. However, further data are required to investigate this hypothesis. As recently claimed, the effect of footfall timing on mechanical stability and mechanical energy exchange on arboreal and terrestrial substrates, as well as the way these competing demands interact (Miller et al., 2019), remains poorly known. It is also possible that these two components do not have the same ‘time frame’, as mechanical energy exchange should occur within a stride, while mechanical stability can be performed and ensured among multiple strides (Biewener and Daley, 2007; Chadwell and Young, 2015; Lammers and Gauntner, 2008). Further integrative data including larger sample size are needed to better understand the relationships between gait adjustments, BCoM mechanics and muscular activity in non-human primates.

Overall, the quadrupedal peculiarities of primates certainly represent biomechanical adjustments to arboreal life and played a crucial role in the early differentiation of primates from other mammals. However, despite these peculiarities, they also share very similar patterns of muscle activation with other quadrupedal mammals, at least when they walk quadrupedally on the ground. This reflects the shared basic functional demands (i.e. support and propulsion) on the hindlimbs for quadrupedal locomotion in tetrapods. Finally, the diagonal gait pattern appears to be strongly implemented into the baboon neuromotor system because lateral sequences are likely to result in a direct and significant increase of the energetic costs (at the level of BCoM mechanics and muscular activity). In contrast, the diagonal walking gait does not impede good recovery rates in primates.

We are very grateful to Romain Lacoste and Thomas Brochier, the director and the scientific director, respectively, of the Primatology Station of the CNRS, who provided access to the animals and facilities. We are very grateful to Josie Meaney-Ward who revised and improved the English of the manuscript. We thank the referees for their constructive and detailed comments on previous versions of the manuscript.