A coati conundrum: how variation in levels of arboreality influences gait mechanics among three musteloid species Free

Among quadrupedal mammals, the walking pattern of primates is distinguished by three major locomotor characteristics (Granatosky, 2020; Granatosky and Young, 2023; Schmitt and Lemelin, 2002). First, peak vertical substrate reaction forces are highest in the primate hindlimb (Demes et al., 1994; Kimura et al., 1979; Reynolds, 1985a), subverting the typical mammalian condition of forelimb-biased weight support during quadrupedal walking (Barclay, 1953; Granatosky et al., 2020; Gray, 1944; Lemelin and Schmitt, 2007; Schmitt and Lemelin, 2002). Second, at forelimb touchdown, the arm is characterized by a more protracted position, in contrast to a more retracted position observed in most other mammalian lineages (Granatosky, 2020; Larson, 1998; Larson et al., 2000, 2001). And third, primates employ diagonal-sequence, diagonal-couplet (DSDC) gaits during walking, whereas most mammals rely upon lateral-sequence, diagonal-couplet (LSDC) gaits (Cartmill et al., 2002, 2020; Hildebrand, 1976, 1980; Plocek and Dunham, 2023; Vilensky and Larson, 1989; Wimberly et al., 2021). These defining characteristics of primate walking are interpreted to reflect their evolutionary history, combining the more frequent use of thinner arboreal supports with an emphasis on hand use and dexterity to ‘free’ the forelimb from strictly locomotor constraints (Cartmill, 1974, 1992; Napier, 1967; Wood Jones, 1916).

The central role of this arboreal heritage in driving the locomotor adaptations of primates is supported by convergence of the walking pattern observed in other arboreal mammalian lineages. For example, several marsupial taxa, including Trichosorus vulpecula (White, 1990), Dromiciops australis (Pridmore, 1994) and Acrobates pygmaeus (Karantanis et al., 2015), employ a near-exclusive diagonal-sequence walking gait pattern. The woolly opossum (Caluromys philander), which, like many basal primates, predominantly traverses terminal arboreal substrates and makes frequent use of its hands to manipulate foods (Lemelin, 1999; Rasmussen, 1990), exhibits all three defining characteristics of primate walking (Lemelin and Schmitt, 2007; Schmitt and Lemelin, 2002). This is in stark contrast to more terrestrial marsupials (e.g. Monodelphis) that exhibit the typical gait characteristics of quadrupedal mammals (Amanat et al., 2020a; Cartmill et al., 2007, 2020; Lammers et al., 2006; Lemelin et al., 2003; Schmitt and Lemelin, 2002).

Similar gradients of arboreality can be observed across other mammalian lineages, including within Musteloidea – an ecologically diverse superfamily spanning numerous locomotor specializations from fossoriality to marine swimming (Fish, 1994; Kilbourne and Hutchinson, 2019; Kitchener et al., 2017; Shimer, 1903). Within musteloids, it is possible to isolate significant locomotor diversity within a relatively constrained phylogenetic, anatomical and body size range (Fabre et al., 2013, 2015). Indeed, the musteloids include both primarily terrestrial coatis (Nasua narica, ∼5.9 kg in body mass, and Nasua nasua, ∼3.9 kg; Gompper, 1995; Gompper and Decker, 1998) and the highly arboreal kinkajou (Potos flavus, 2.5 to 3.7 kg; Ford and Hoffmann, 1988), as well as the red panda (Ailurus fulgens, 3.7 to 6.2 kg; Roberts and Gittleman, 1984), which can be best described as a mixed terrestrial/arborealist.

Kinkajous are almost exclusively arboreal (Ford and Hoffmann, 1988; Julien-Laferrière, 1993, 1999), primarily moving and foraging on small, terminal supports less than 5 cm in diameter (Charles-Dominique et al., 1981). They are capable of object manipulation via unimanual grips like primates and arboreal marsupials (Lemelin, 1999), and use their prehensile tails to hang from branches to forage (McClearn, 1992). By contrast, coatis spend far more time on the ground and exhibit numerous anatomical features associated with terrestriality (e.g. shortened forelimbs and long manual claws for digging effectively). They do, however, retain the ability to scale smaller trees and vines and may spend much of their inactive time above the ground (Gompper, 1995; Gompper and Decker, 1998; McClearn, 1992).

Compared with coatis, red pandas appear to spend more time in the trees, but do not approach the more specialized arboreal behaviors of kinkajous in terms of time spent in the canopy and reliance on thinner branches. Their arboreal activities are primarily related to foraging for shoots and leaves (Reid et al., 1991). Namely, red pandas frequently scramble across fallen logs and tree stops, and also ascend/descend relatively large-diameter tree trunks (Roberts and Gittleman, 1984; Wei et al., 2000). Unlike kinkajous, red pandas do not have a prehensile tail; instead, their tail functions as a support and counterweight during climbing (Roberts and Gittleman, 1984). Additionally, red pandas exhibit many morphological similarities of the forelimb with other mixed terrestrial/arboreal musteloids such as raccoons and tayras (Fabre et al., 2013).

A number of studies have documented gait patterns during locomotion in red pandas, coatis and kinkajous (Hildebrand, 1967, 1976; Lemelin and Cartmill, 2010; McClearn, 1992; Plocek and Dunham, 2023). Lemelin and Cartmill (2010) has shown that kinkajous, like highly arboreal woolly opossums, exhibit primate-like DSDC gaits and increase the diagonality (i.e. limb phase) of their gaits as substrate diameter decreases. This shift may function to maximize the distance between contralateral feet during periods of diagonal bipedality, increasing the support base and thus reducing instability (Lemelin and Cartmill, 2010). In contrast, Plocek and Dunham (2023) have shown that red pandas exclusively utilize lateral-sequence gaits regardless of substrate. That said, other aspects of the gait mechanics of these three musteloids remain somewhat unknown and not well documented. To that end, we here present a comprehensive kinematic and kinetic comparison of locomotion between the highly arboreal kinkajou, mixed arboreal/terrestrial red panda and primarily terrestrial coati across a range of different experimental substrate conditions. In doing so, we test the overarching hypothesis that species exhibiting high degrees of arboreality, as observed in some musteloids, are expected to display more primate-like gait characteristics. To assess this theory, we outline the following predictions.

Highly arboreal kinkajous practice a primate-like gait pattern characterized by DSDC gaits on the ground and increased limb phase on narrower poles (Lemelin and Cartmill, 2010). Therefore, we predict that terrestrial coatis and the mixed arboreal/terrestrial red panda will not exhibit this specialized arboreal gait pattern when walking on the ground, and will instead exhibit LSDC gaits reflective of a more generalized mammalian condition (Plocek and Dunham, 2023).

Although a more protracted arm at forelimb touchdown is observed in arboreal primates and woolly opossums during walking, this forelimb segment is typically more retracted in other mammalian lineages (Larson, 1998; Larson et al., 2000, 2001; Lemelin and Schmitt, 2007; Schmitt and Lemelin, 2002). We predict higher degrees of arm (humeral) protraction at forelimb touchdown in the highly arboreal kinkajou compared with the more terrestrial coati and red panda on the ground.

Primates and woolly opossums further demonstrate a unique condition of hindlimb-biased weight support during quadrupedal walking, subverting the typical mammalian trend of forelimb-driven quadrupedalism (Demes et al., 1994; Kimura et al., 1979; Lemelin and Schmitt, 2007; Reynolds, 1985a; Schmitt and Lemelin, 2002). We predict that during quadrupedal walking, the highly arboreal kinkajou will exhibit hindlimb-biased weight support, whereas the more terrestrial coati and red panda will exhibit forelimb-biased weight support on the ground.

To address the aims of this study, we collected kinetic and kinematic data on two subjects (two females) of kinkajou [Potos flavus (Schreber 1774)], two subjects (one male, one female) of white-nosed coati [Nasua narica (Linnaeus 1766)] and two subjects (one male, one female) of red panda (Ailurus fulgens Cuvier 1825). All subjects were adults, in good health, and clear of any obvious gait pathologies. The kinkajous were acquired specifically for this study and housed at the University of Alberta Ellerslie Farm facility in Edmonton, AB, Canada. The housing enclosure was large enough to have an extensive rope system for daily locomotor and postural activity and exercise, and nesting box and synthetic fleece pouches for resting and sleeping. Subjects were transferred using a transport crate from their housing enclosure to an adjacent observation room for collection of locomotor data. A typical observation session would last approximately 30 min, with an animal handler encouraging movement of the subject on the experimental substrate using a synthetic fleece pouch as an incentive. Upon return to the housing enclosure, subjects were weighed on a regular basis using a digital scale (nearest 0.01 kg). The coatis and red pandas were part of the display animals of the Edmonton Valley Zoo in Edmonton, AB, Canada. Subjects were transferred by a zoo technician using a transport crate from their display enclosure/area to an observation room. The room was equipped with a large enclosure specifically designed and built for collection of locomotor and other behavioral data on zoo animals. The enclosure was made of metal mesh and one side had a clear Lexan wall to facilitate video recording of movement. A typical observation session would last approximately 20 to 30 min, with a zoo animal technician promoting movement of the subject on the experimental substrate using food treats or familiar pet toys. Upon return to the housing enclosure, subjects were weighed on a regular basis using a digital scale (nearest 0.01 kg). Animal housing and husbandry followed standards prescribed by the Canadian Council on Animal Care (CCAC) and the Canada's Accredited Zoos and Aquariums (CAZA). Housing, husbandry, handling and data collecting methods and procedures in the laboratory on all subjects were approved by the University of Alberta Animal Care and Use Committee (ACUC)–Biosciences (protocol no. 392).

The experimental substrate setups used for this comparative study are similar to those used by other studies (Granatosky et al., 2016; Hanna et al., 2006; Lemelin and Schmitt, 2004; Schmitt, 2003; Schmitt and Lemelin, 2002, 2004) and are described below (Fig. 1). All animal subjects were encouraged to walk across a runway with a floor painted with a mix of clear varnish and sand to increase friction. The runway was 3.05 m long, 0.56 m wide, raised 0.16 m from the concrete floor to accommodate a force platform, and had walls 0.6 m tall. The side wall closer to the camera had a clear Lexan section 1.02 m wide for a clear view of the animal subject during filming. The mid-section of the runway (0.71 m long) had a cut-off section (mask) to accommodate a rectangular block (0.4 m wide) that was instrumented to the force platform with three large screw T-bolts. The block and mask had 5-mm gaps on all sides to prevent the main runway from touching the instrumented section used to record individual limb forces. An 11-cm-long instrumented section was used to accommodate the entire hand or foot of kinkajous and a 13-cm-long section was used to accommodate the larger hand or foot of coatis and red pandas (Fig. 1). A pole setup made of three sections of painted graphite tubing (total length of 2.9 m, 0.25 cm in diameter) covered with a mix of varnish and sand to increase friction was used for kinkajous only to gather locomotor data on a simulated branch. The right and left sections were supported by metal posts with adjustable height, while the mid-section (7.5 cm long) was instrumented onto the force platform with an adjustable metal bracket with four large screw T-bolts and separated by 5 mm gaps from non-instrumented sections of the pole. The mid-section pole length was optimal to gather individual limb forces without interference between ipsilateral footfalls, a recurrent problem for mammals using diagonal-sequence gaits (Fig. 1).

Individual forelimb and hindlimb forces synchronized with digital video recordings of movement were collected on all subjects walking at constant speeds between each end of the experimental substrates. Trials were only selected when the entire hand or foot of a single limb contacted the instrumented portion of the runway or pole while moving at walking speeds, with no obvious acceleration or deceleration. Limb forces were acquired using a Bertec custom force platform (Bertec, Columbus, OH, USA) model 4060-08 (0.4×0.6 m) equipped with T-shaped grooves to ease instrumentation of the mid-section of each substrate (runway and pole). Analog force data were gathered at a 1250 Hz sampling rate for all three channels (F_x, F_y, F_z). Only data from the F_z channel (vertical substrate reaction force) are reported in this study. Analog data output from the force platform was sent to an AM6501 amplifier (Bertec) and then to a National Instrument BNS 2110 break-out box (National Instruments, Austin, TX, USA) before export to an Antec Plus 660 AMG computer (Antec, Fremont, CA, USA) equipped with a National Instruments analog data computer card. Digital videorecording of movement was achieved with a Photron Fastcam PCI R2 high-speed camera (Photron USA, San Diego, CA, USA) equipped with a 12.5–75 mm zoom lens and resting on a tripod approximately 3 m perpendicular to the center to the runway or pole apparatus. Upon manual TTL triggering, digital recording of a trial was prompted at 125 Hz sampling rate. The resulting .AVI file was reviewed, cropped and then saved to the computer using the Photron Motion Tools v. 1.3 software (Photron USA). The same software also allowed for digital images to be synchronized with the analog force data at a 1:10 ratio. Both digital video from the high-speed camera and analog data from the force platform were stopped simultaneously in accordance with specific trigger percentage settings. Both reference video frame 0 and reference force datum 0 were positioned at the trigger point of the sequence by the Photron Motion Tools data module.

As our predictions are only applicable to walking gaits (Granatosky et al., 2016; Lemelin and Schmitt, 2004; Schmitt and Lemelin, 2002, 2004; Wallace and Demes, 2008), only trials with limb duty factors greater than 50% were used for subsequent analyses. Ambling gaits (Schmitt et al., 2006) were very common for one kinkajou subject moving on the runway, and were removed from subsequent kinematic and kinetic analyses. Furthermore, although no specific analysis for steady-state locomotion was conducted, any trial in which the subject was clearly accelerating or decelerating was removed from the sample.

Following Hildebrand (1967, 1976, 1985), limb phase and duty factors were calculated (both as %) for each selected trial to differentiate walking gait patterns among the three musteloid species. From the same video recordings, the humeral angle (deg) at forelimb touchdown was calculated for each trial (Larson et al., 2000). The marker-less pose estimation program DeepLabCut (Mathis et al., 2018) was utilized to obtain positional data of the shoulder, elbow and wrist joint areas relative to the substrate from sequential lateral views of each trial. Using the program's machine learning algorithm, 150 frames (i.e. kinkajous and coatis) to 200 frames (i.e. red pandas) of these areas were labeled to train its neural network. The initial DeepLabCut output was retrained following Mathis et al. (2018) until points achieved a likelihood above 95%. These kinematic outputs were visually inspected to ensure the accuracy of the anatomical landmarks. The resulting positional data of the tracked points were used to determine forelimb touchdown (i.e. corresponding to the frame the palm first contacted the substrate) and derive the humeral angle. The line passing between the shoulder and elbow joint areas relative to a reference vertical axis passing through the shoulder joint area was used to calculate humeral angle at forelimb touchdown using a custom-written MATLAB (MathWorks, Natick, MA, USA) script. Following Larson et al. (2000), alignment of the humerus with the vertical axis passing through the shoulder joint was considered the neutral position (0 deg). Thus, positive angles indicate humeral protraction and negative angles indicate humeral retraction at forelimb touchdown.

Analog force plate output sampled at 1250 Hz was converted from raw voltage to N using ProAnalyst v. 1.6 (Xcitex, Woburn, MA, USA). Force data were filtered (high-pass Fourier 60 Hz filter) and processed for peak vertical substrate reaction force (V_pk) for each footfall using a custom-written MATLAB code. Only trials for which a clear separation between footfalls could be determined were included for subsequent analysis. Forces were normalized to the corresponding body mass recorded on a given experimental day. For all statistical analyses (see below), V_pk was analyzed as a percentage of the individual's body weight (%BW). All recordings were calibrated in affine space to calculate average speed (m s⁻¹) for each trial. As V_pk correlates positively with speed (Demes et al., 1994; Granatosky et al., 2020; Schmitt, 2003), such information was necessary for subsequent statistical analyses (see below).

A series of linear mixed-effects model analyses were performed to assess gait diagonality (i.e. limb phase), humeral angle kinematics and V_pk kinetics. All interspecific analyses were conducted as the species moved on the runway, whereas separate analyses were conducted between the pole and runway for the kinkajous (see below). To account for potential intraspecific idiosyncrasies, animal subject was treated as a random effect throughout all models. To assess diagonality differences between species, a model was created accounting for duty factor, which was included as a covariate based on recent studies showing an inverse relationship between duty factor and limb phase (Granatosky et al., 2021, 2022; Usherwood and Self Davies, 2017). A similar analysis was conducted to assess humeral protraction/retraction differences between species, without including duty factor as a covariate. Because speed correlates with V_pk (Demes et al., 1994; Granatosky et al., 2020), it was included as a fixed effect in all kinetic analyses. This series of kinetic models involved subsampling the data to first assess intraspecific differences between forelimb and hindlimb V_pk, followed by an assessment of interspecific differences in V_pk between the forelimb and hindlimb. For each of these analyses (i.e. diagonality, arm kinematics and limb kinetics), a separate sub-analysis was performed within kinkajous to assess substrate differences (runway versus pole). All initial statistical analyses were conducted in R (https://www.r-project.org/) using the packages ‘lme4’ and ‘lmerTest’ (Bates et al., 2015; Kuznetsova et al., 2017). A series of post hoc tests were subsequently conducted on model outputs between species using ‘multcomp’ (https://CRAN.R-project.org/package=multcomp).

A total of 518 trials were suitable for analysis of limb phase to assess diagonality, 230 of which were contributed by the coatis, 79 by the red pandas and 209 by the kinkajous (66 of which were collected on the runway and 143 on the pole; Table 1; Dataset 1). Our model demonstrated no significant difference in limb phase between the mixed arboreal/terrestrial red panda (21.5±2.26%, duty factor of 70.19±2.32%) and primarily terrestrial coati (21.38±3.67%, duty factor of 70.26±4.40%), but both taxa differed significantly from the highly arboreal kinkajou (54.71±2.77%, duty factor of 58.48±3.53%, both P<0.001) (Table 2, Fig. 2). An additional linear mixed-effect model accounting for the effects of duty factor and substrate within the kinkajous demonstrated non-significant differences between pole (57.11±2.89%) and runway (54.71±2.77%, P=0.973; Table 2).

A total of 563 trials were suitable for analysis of humeral position at forelimb touchdown, 272 of which were contributed by the coatis, 79 by the red pandas and 212 by the kinkajous (67 of which were on collected on the runway and 145 on the pole; Table 1; Dataset 1). Our model demonstrated non-significant differences in humeral protraction between the red panda (−0.50±20.29 deg) and coati (5.04±12.44 deg; Table 2, Fig. 3), though a bimodal distribution was observed within the red pandas as the male had protracted humeral angles at touchdown (13.14±15.12 deg), whereas the female exhibited a more retracted humerus (−17.25±11.21 deg). Wider degrees of humeral protraction at forelimb touchdown were observed in both kinkajous (29.41±7.70 deg; Table 2, Fig. 3) that were significantly greater than those of either coatis or red pandas (both P<0.002). An additional linear mixed-effect model comparing the substrates within the kinkajous demonstrated non-significant differences in humeral position between the pole (28.21±9.37 deg) and runway (29.41±7.70 deg, P=0.479; Table 2, Fig. 3).

A total of 690 trials were suitable for analysis of peak vertical substrate reaction force (V_pk) between forelimbs and hindlimbs for all three musteloid species (Table 1; Dataset 1). A total of 378 trials were contributed by coatis, with 189 forelimb and 197 hindlimb trials. A total of 98 were contributed by red pandas, with 57 forelimb and 41 hindlimb trials. A total of 214 trials were contributed by kinkajous; 67 of which were collected on the runway (with 30 forelimb and 37 hindlimb trials) and 147 of which were collected on the pole (71 forelimb and 76 hindlimb trials). Speed significantly affected all forces on the runway (all P<0.001), but not the pole substrate (P=0.066; Table 3).

Our model compared V_pk differences between forelimb and hindlimb on the ground substrate for each species while accounting for speed. In the coati, greater V_pk values were observed in the hindlimb relative to the forelimb (hindlimb=62.44±8.54%BW, forelimb=47.16±6.30%BW, P<0.001; Table 3, Fig. 4). By contrast, the red panda exhibited greater forelimb V_pk values (60.48±3.39%BW, P<0.001) relative to the hindlimb (47.76±6.37%BW, P<0.001). Finally, the kinkajou exhibited greater forelimb V_pk values compared with the hindlimb on the runway (forelimb=68.79±6.92%BW, hindlimb=66.08±5.50%BW, P<0.001), but no statistical difference was observed on the pole (forelimb=66.81±5.30%BW, hindlimb=65.71±6.16%BW, P<0.854; Table 3, Fig. 4). A significant difference was found between the coati and both other taxa in the hindlimb and forelimb (P<0.001 in all cases), but no differences were observed between kinkajous and red pandas in either limb (P=0.052 in the forelimb; P=0.153 in the hindlimb; Table 4).

In this study, we used three taxa exemplifying varying degrees of arboreality to assess whether the locomotor biomechanical features typically associated with fine-branch arboreality (i.e. the predominance of DSDC gaits, arm protraction at forelimb touchdown, and hindlimb-biased weight support) have evolved in other mammalian lineages. We confirm the findings of Lemelin and Cartmill (2010) that the highly arboreal kinkajou employs almost exclusively DSDC gaits and considerable humeral protraction at forelimb touchdown compared with other musteloids (Figs 2, 3, 5). Kinkajous do not evince hindlimb-biased weight support, although their relatively equal forelimb/hindlimb peak vertical force ratio is also highly unusual for non-primate mammals (Fig. 6). Among primates, a similarly equal distribution ratio is found in baboons, which are among the most terrestrial primates (Demes et al., 1994). The mixed terrestrial/arborealist red panda shares none of these locomotor features associated with fine-branch arboreality. Unexpectedly, the highly terrestrial coati shows some features associated with arboreality (i.e. some degrees of humeral protraction at forelimb touchdown and hindlimb-biased weight support; Fig. 5) combined with an almost exclusive use of lateral-sequence, lateral-couplet (LSLC) gaits, which typify some other highly terrestrial mammals (Cartmill et al., 2002; Hildebrand, 1985). To our knowledge, this unique combination of arboreal and terrestrial characteristics has never been observed in another mammal species, making the coati truly unusual with regards to its locomotor profile.

DSDC gaits are common in primates (Cartmill et al., 2002; Hildebrand, 1967, 1976; Vilensky and Larson, 1989), arboreal marsupials (Cartmill et al., 2020; Lemelin et al., 2003) and kinkajous (Hildebrand, 1967, 1976; Lemelin and Cartmill, 2010), but virtually absent in other mammals. Indeed, for most arboreal tetrapods, there is a general tendency to avoid DSDC gaits altogether, regardless of substrate size, likely attributed to either a neuromuscular constraint (Vilensky and Larson, 1989; Wimberly et al., 2021) and/or work minimization concerns (Miller et al., 2019; Usherwood and Self Davies, 2017; Usherwood and Smith, 2018). Some arboreal animals show the dual tendency to increase duty factor and decrease limb phase, resulting in slower-speed lateral-sequence gaits (Demes et al., 1990; Granatosky et al., 2021, 2022). For example, slow lorises and chameleons commonly rely on slow lateral-sequence gaits with higher limb duty factors and maximize tripedal limb supports as part of their cryptic prey-catching lifestyle (Demes et al., 1990; Ekhator et al., 2023; Peterson, 1984). Meanwhile, other arboreal mammals, such as tree squirrels and marmosets, adopt a greater proportion of higher-speed asymmetrical gaits, such as bounds, half-bounds or gallops, when moving on branches of all sizes, including thinner ones (Dunham et al., 2020; McElroy and Granatosky, 2022; Young, 2009; Young et al., 2016). Adoption of DSDC gaits, and by extension the amble (Schmitt et al., 2006), allows primates, arboreal marsupials and kinkajous to maintain symmetrical gait sequences across a range of speeds and substrate sizes.

Compared with the highly arboreal kinkajou, both the red panda (mixed terrestrial/arboreal) and coati (highly terrestrial) utilized a higher proportion of LSLC gaits, a finding consistent with a recent study of red panda gaits (Plocek and Dunham, 2023). This is not surprising as red pandas and coatis especially spend more time on the ground and, therefore, have walking gaits typical of other more terrestrial carnivorans (Amanat et al., 2020b,c; Dev et al., 2020; Granatosky, 2018a). Given the higher limb duty factors (and slower speeds) observed in those more terrestrial musteloids, lateral-sequence gaits maximize support polygons during the stride period (Cartmill et al., 2002; Granatosky, 2018a). LSLC gaits are more commonly observed in animals with relatively long limbs for their body size (e.g. cheetahs, camels, etc.), although a recent phylogenetic analysis has shown no significant effect of relative limb length on limb phase (Wimberly et al., 2021). Generally speaking, LSLC gaits involve almost synchronous movement of the ipsilateral forelimb and hindlimb. Although this arrangement limits the risk of interlimb interference, it also raises potential stability concerns (Cartmill et al., 2002; Granatosky, 2018a). LSLC gaits maximize unilateral (ipsilateral) bipods as limb phase approaches 25% or less (i.e. pace), increasing the likelihood of toppling over towards the unsupported side (Cartmill et al., 2002, 2007), though the slow walking speeds and increased limb contact times observed in red pandas and coatis could mitigate such risks. Moreover, the cost of maintaining potentially less stable pacing gaits at higher walking speeds may be outweighed by the benefit of minimizing limb interference, an important fact in carnivorans with long, non-retractable claws such those found in coatis. A broad phylogenetic analysis of carnivoran gaits across a range of body sizes, limb lengths, and claw shape and size could shed light on the high incidence of LSLC walking gaits in terrestrial musteloids.

Kinkajous showed greater degrees of humeral protraction at forelimb touchdown than coatis and red pandas. In this way, kinkajous very much resemble primates and woolly opossums, and deviate from most other mammals that show more retracted arm positions at forelimb touchdown (Larson et al., 2000, 2001). Greater arm protraction further increases effective limb length (Granatosky et al., 2019; Larson, 1998), which translates into a longer stance phase and higher duty factors (Schmitt, 1999; Schmitt and Hanna, 2004, but see Dickinson et al., 2022) and potentially reduces peak vertical substrate forces on the forelimb (McMahon et al., 1987; Schmitt, 1999; Schmitt and Hanna, 2004). Larger-bodied (over 1 kg) arboreal species such as kinkajous may also utilize a protracted humerus at touchdown as a strategy for speed regulation (Granatosky and McElroy, 2022). Speed regulation in animals involves two basic parameters: stride frequency and stride length (Schubert et al., 2014; Strang and Steudel, 1990). Granatosky and McElroy (2022) demonstrated that relatively larger arboreal taxa increase speed with longer strides rather than higher stride frequencies. In turn, longer strides may have the dual advantage of reducing branch oscillations (Demes et al., 1990) and minimizing overall cost of transport (Heglund and Taylor, 1988). Therefore, a protracted arm position at forelimb touchdown may serve as a strategy to increase overall stride length (Granatosky et al., 2019; Larson et al., 2000, 2001) with additional locomotor benefits. Further experimental work could partition those interrelated benefits associated with greater degrees of humeral protraction at forelimb touchdown.

It is notable that both coatis and at least one red panda (see Limitations, below) consistently showed humeral protraction at forelimb touchdown, albeit to a lesser degree than the kinkajous. As outlined above, humeral protraction at forelimb touchdown is usually associated with arboreal movement. However, Larson and colleagues’ (2000) analysis of forelimb posture across mammalian quadrupeds shows that many carnivores exhibit protracted humeral angles at forelimb touchdown. Unfortunately, the individual data points were not labeled, so it is not possible to determine the ecology of those carnivore species that show true humeral protraction at forelimb touchdown. It is also notable that, compared with ungulates and rodents, carnivores tend to show less humeral retraction at forelimb touchdown than is colloquially assumed. As such, the association between humeral protraction at forelimb touchdown may be less tied to arboreality than previously thought, and may instead be driven by phylogeny or some other aspect of anatomy. A reanalysis of limb kinematics across mammals with varying locomotor ecologies may help to better establish the role of humeral protraction at forelimb touchdown.

Perhaps no other aspect of primate locomotion has received more attention than the tendency for most taxa to exhibit a hindlimb-biased weight support pattern (e.g. Demes et al., 1994; Druelle et al., 2019; Kimura et al., 1979; Larson, 1998; Reynolds, 1985b; Schmitt and Lemelin, 2002). Such a tendency is extremely rare among tetrapods, and the observation that the highly arboreal woolly opossum displayed higher peak vertical substrate reaction forces on the hindlimbs relative to the forelimb (Schmitt and Lemelin, 2002) only added to the conclusion that a hindlimb-biased weight support is highly adaptive to a fine-branch milieu.

This hypothesis is somewhat weakened by the observation of a clear pattern of hindlimb-biased weight support in coatis. Two main functional explanations have been proposed to explain the tendency of primates to alleviate weight bearing from the forelimb. First, hindlimb bias may serve as a mechanism for arboreal animals to assess the flexibility and stability of a branch with their forelimb prior to committing the weight of a grasping hindlimb (Cartmill et al., 2002; Lemelin et al., 2003). Second, this pattern of weight support may function to reduce loads on the forelimb wherein joints are designed for mobility rather than stability (Demes et al., 1994; Reynolds, 1985a; Schmitt, 1994, 1998, 1999). Although the first mechanism is unlikely to apply to the forelimbs of coatis, it is possible that the second mechanism could apply. Like other procyonids, coatis use their forelimbs extensively for food procurement (McClearn, 1992). Although coatis do not manipulate food objects to the extent of raccoons (two-handed grips) or kinkajous (single-handed grips), they extract food by digging with the aid of their massive claws (Gompper, 1995; Gompper and Decker, 1998; McClearn, 1992). Whether such extractive foraging behavior with the forelimb has any bearing on the hindlimb-biased force pattern reported here remains unclear. Further, discussions of hindlimb-biased weight support often treat it as an active process in which the animal pitches its body weight caudally (Larson and Demes, 2011; Larson and Stern, 2009; Reynolds, 1985b; Young, 2012). However, we cannot dismiss the importance of body plan, especially regarding coatis, as a possible explanation for their unexpected hindlimb-biased weight support (Dickinson et al., 2022; Druelle et al., 2019; Granatosky et al., 2018a,b; Raichlen et al., 2009). Coatis have long tails and large haunches (McClearn, 1990; Wilson et al., 2009). Perhaps the coati demonstrates hindlimb-biased weight support simply because a larger proportion of its mass is distributed posteriorly. Unfortunately, a center of mass position is not available for the coati to test mechanisms of weight support distribution.

Finally, it is important to consider that in virtually all quadrupedal taxa, the absolute loading magnitudes experienced by the forelimb compared with the hindlimb differ by only ∼10 to 20% in peak vertical forces. Thus, at higher locomotor speeds, the ‘weaker’ forelimb may frequently experience forces of comparable magnitude to the habitual loads experienced by the ‘more stable’ hindlimb. Moreover, if a redistribution of ∼10% body weight (i.e. a few newtons of force) from the forelimb to the hindlimb is critical to mitigate the risk of breaking branches, then arboreal animals that do not shift weight backwards such as kinkajous should be at constant risk of falling, which is clearly not the case. Thus, it remains unclear to what extent either of these overarching explanations apply to arboreal quadrupedal walking on a more global scale.

As with all live animal studies, there are certain limitations that must be addressed. First, it goes without saying that the sample size in this study is low. Typically, such concerns would be mitigated by noting that these are rare animals and the feasibility of collecting such data at zoological institutions is diminishing. However, the small sample size did result in two issues that raise concern. First, on the ground, one of our individual kinkajous elected to amble (Schmitt et al., 2006) rather than walk. These ambles were not included in the sample. Consequently, for this kinkajou, the number of analyzable strides was very low, making our data from kinkajous on the ground essentially represented by one individual. Second, in terms of limb kinematics, our two red pandas showed a bimodal distribution, with one individual displaying a protracted humeral position at touchdown and the other a retracted position. Therefore, it is impossible to determine which kinematic patterns are actually representative of the species. Given these issues, we exercise caution and have attempted throughout to avoid making claims that could not be reliably ascertained from the data.

Sampling also faced certain issues. As evidenced by Figs 1 and 5, the quality of the video was not particularly high and had a narrow field of view that prevented us from seeing the whole animal. These recordings focused on the small area of the force plate, rather than the entire runway. Unfortunately, this is an unavoidable limitation, as the data presented in this study cannot be recollected because the specific individuals analyzed have since passed. The sampling period was also longer than ideal (e.g. coati data collection range: 2009–2011; red panda data collection range: 2013–2015; kinkajou data collection range: 2008). Although we did not test for the effects of sampling duration, it is possible that aging of the animals may have influenced gait, kinematic or kinetic data (Druelle et al., 2016; Young, 2012). These sampling limitations should be considered.

Most importantly, although we did collect data from the kinkajous on both the pole and a flat runway, such data were not collected for the red pandas or coatis. The extent to which animals shift gait characteristics when moving on the ground versus a pole can vary considerably (e.g. Granatosky, 2018b; Nyakatura and Heymann, 2010; Young et al., 2022). As demonstrated in this study, kinkajous modified gait, kinematic and kinetic data only minimally in response to substrate. Similarly, Plocek and Dunham (2023) have shown that red pandas do not change gait characteristics in response to substrate. With these considerations in mind, we do not believe that sampling all species on all substrates would have drastically changed the findings or interpretations of this study. Despite these assurances, throughout the paper any intraspecific differences we discuss are based on data collected on the flat runway, rather than mixing inferences drawn from data collected from kinkajous on a pole.

The data presented herein demonstrate that the trio of locomotor characteristics traditionally used to describe walking in primates (i.e. a DSDC gait, a protracted arm at forelimb touchdown, and hindlimb-biased weight support) do not represent a single suite of interdependent characteristics. Instead, they can vary independently of one another as previously observed in the slender loris (Schmitt and Lemelin, 2004). Highly arboreal kinkajous, which exhibit similarities in ecology, foraging behavior and manual dexterity to primates, show preference for DSDC gaits and greater degrees of arm protraction at forelimb touchdown, combined with an even forelimb/hindlimb weight support ratio similar to some primate taxa. Thus, kinkajous represent a second documented case of convergence upon a primate-like gait in a specialized non-primate arboreal mammal (alongside the woolly opossum; Lemelin and Cartmill, 2010; Lemelin et al., 2003; Schmitt and Lemelin, 2002).

More surprising is the presence in the coati of a primate-like hindlimb-biased weight support pattern and moderate degree of arm protraction at forelimb touchdown combined with a LSLC walking gait pattern typical of highly terrestrial mammals. Although hindlimb-biased weight support is historically considered an adaptation for arboreal life, our data on coatis suggest that hindlimb-biased weight support may not be an arboreal adaptation per se, but instead represents a mechanism to free the forelimb from a purely locomotor function and thus confers benefits that are useful to any animal that uses their forelimb in a specialized manner (Wood Jones, 1916). Coatis – with their long, non-retractable forelimb claws adapted for digging – may therefore utilize a hindlimb-biased weight support to protect these specialized limbs. It is further possible that other procyonids, such as the dexterous raccoon (Iwaniuk and Whishaw, 2000; McClearn, 1992), might exhibit similar patterns of weight support distribution to free their forelimbs for object manipulation, a hypothesis that should be addressed in future experimental studies.

We thank the three anonymous reviewers for their comments that have greatly improved the quality of the paper. We are most grateful to Dr Patrick Nation and staff of the Health Sciences Laboratory Animal Services at the University of Alberta for housing and care of the kinkajous and access to research space, the Edmonton Valley Zoo (City of Edmonton), Denise Prefontaine, and the late Dr Milton Ness for providing access to the coatis and red pandas, and newly renovated research space for the purpose of this study. Sandy Helliker and many other animal caretakers of the Edmonton Valley Zoo made time in their busy daily schedules to train and transfer animal subjects and assisted with handling during the numerous behavioral sessions. Dr Karl Berendt, Michael Nagy and Dr Lian Willetts also provided invaluable assistance with data collection in the laboratory. Sam Graziano and the University of Alberta Biomedical Workshop expertly constructed the runway and many other custom-made experimental devices used in this study.