Three-dimensional shape variation of talar surface morphology in hominoid primates

Abstract

The hominoid foot is of particular interest to biological anthropologists, as changes in its anatomy through time reflect the adoption of terrestrial locomotion, particularly in species of Australopithecus and Homo. Understanding the osteological morphology associated with changes in whole foot function and the development of the plantar medial longitudinal foot arch are key to understanding the transition through habitual bipedalism in australopithecines to obligate bipedalism and long-distance running in Homo. The talus is ideal for studying relationships between morphology and function in this context, as it is a major contributor to the adduction–abduction, plantar–dorsal flexion and inversion–eversion of the foot, and transmits all forces encountered from the foot to the leg. The talar surface is predominantly covered by articular facets, which have different quantifiable morphological characters, including surface area, surface curvature and orientation. The talus also presents challenges to the investigator, as its globular shape is very difficult to quantify accurately and reproducibly. Here we apply a three-dimensional approach using type 3 landmarks (slid semilandmarks) that are geometrically homologous to determine overall talar shape variations in a range of living and fossil hominoid taxa. Additionally, we use novel approaches to quantify the relative orientations and curvatures of talar articular facets by determining the principal vectors of facet orientation and fitting spheres to articular facets. The resulting metrics are analysed using phylogenetic regressions and principal components analyses. Our results suggest that articular surface curvatures reflect locomotor specialisations with, in particular, orang-utans having more highly curved facets in all but the calcaneal facet. Similarly, our approach to quantifying articular facet orientation appears to be effective in discriminating between extant hominoid species, and may therefore provide a sound basis for the study of fossil taxa and evolution of bipedalism in Australopithecus and Homo.

Introduction

Hominoid primates display a variety of locomotor types with varying degrees of adaptation to arboreal and terrestrial locomotion. Hylobatids (gibbons and siamangs) are highly arboreal and frequently use brachiation (arm swinging) to move around forest canopies. When moving on the ground, hylobatids engage in terrestrial bipedalism, where the foot is plantigrade (flat on the ground) but the heel is occasionally kept elevated (Gebo, 1992; Crompton et al. 2008; Vereecke & Aerts, 2008). Orang-utans are also highly arboreal, engaging in quadrumanous scrambling and in less dynamic forms of brachiation than hylobatids. In orang-utans the foot is plantigrade during terrestrial locomotion, with the heel contacting the ground, and usually inverted during arboreal locomotion (Crompton et al. 2008). Chimpanzees, and particularly gorillas, are more terrestrial than hylobatids and orang-utans, and their feet are plantigrade with heel contact during terrestrial locomotion (Gebo, 1992; Crompton et al. 2008; Dunn et al. 2014). Modern humans are obligate terrestrial bipeds, walking with a heel contacting plantigrade foot (Gebo, 1992; Crompton et al. 2012). Relative to other hominoids, the foot and whole body soft and hard tissue anatomy of modern humans is highly derived and adapted to terrestrial walking and running (Lewis, 1989; Bramble & Lieberman, 2004).

The talus is particularly relevant for studying relationships between morphology, mechanics and locomotor mode for several reasons.

1. Externally, it is composed almost entirely of articular surfaces (Fig. 1). Small changes in the morphology of these articular surfaces may have direct consequences for the possible range of motion at the joint and the range of pressures the articulations can withstand (Ruff & Runestad, 1992; Rafferty & Ruff, 1994; Hamrick, 1996; Lieberman et al. 2001; Polk, 2002).

2. The talus contributes to all three tarsal joints: the talocrural joint, the subtalar joint, and the transverse tarsal joint, which are strongly involved in dorsi/plantar flexion, inversion/eversion and adduction/abduction movements of the foot (Lewis, 1980a,b, 1989; Gebo, 1993; Novacheck, 1998; Leardini et al. 1999). Subtle combinations of these movements allow for efficient bipedal walking and running gaits in modern humans (Czerniecki, 1988; Novacheck, 1998).

3. Of all the tarsal bones, the talus is the best represented in the hominoid fossil record, including several key hominin specimens (Leakey, 1960; Day & Napier, 1964; Wood, 1974a,b; Johanson et al. 1982; Latimer et al. 1982; Gebo & Schwartz, 2006; Pearson et al. 2008; Jungers et al. 2009b; Lovejoy et al. 2009; Zipfel et al. 2011). These fossil tali may hold key information on the timing and sequence of functional change in the foot during the evolution of human bipedalism.

Fig. 1.

Male Gorilla gorilla talus anatomy. (a) Dorsal aspect, (b) plantar aspect. The subtalar joint consists of calcaneal and sustentaculum facets; the talocrural joint consists of trochlea, medial and lateral facets; the head facet is part of the transverse tarsal joint.

There has been a long history of research on the evolution of the hominoid talus, reaching back to the 19th century. Past enquiries have tended to focus on specific aspects of talar morphology that are distinct in different species and/or are expected to affect specific aspects of function in relation to bipedal vs. quadrupedal gaits, notably the orientation and/or shape of the talocrural joint, the angle of inclination and horizontal angle of the neck, head torsion and estimates of the upper and/or lower ankle joint axes (Aeby, 1878; Virchow, 1903; Volkov, 1903; Sewell, 1904, 1906; Morton, 1922, 1924, 1927; Elftman & Manter, 1935; Barnett, 1955; Day & Napier, 1964; Lisowski, 1967; Day & Wood, 1968; Decker & Szalay, 1974; Lewis, 1980b, 1989; Latimer et al. 1987; Gebo, 1992, 1993; DeSilva, 2009a,b; Jungers et al. 2009a,b; Lovejoy et al. 2009; Zipfel et al. 2011; Parr et al. 2012). The interpretation of isolated aspects of talar morphology in a comparative evolutionary context can depend on the choice of reference. For example, modern humans are distinct from chimpanzees in having a larger angle between the articular surface of the talar head and the trochlea, in anterior view. This can be interpreted as an increased rotation of the talar head relative to the axis of the upper ankle joint in modern humans, or as a decreased medio-lateral inclination of the trochlea relative to the axis of the lower ankle joint (Elftman & Manter, 1935). Analysis of complete lower leg skeletons or cadavers can address this issue by quantifying talar morphology in relation to whole foot posture, but for interpreting fossil specimens, which are most commonly found as isolated elements, methods of analysis that can focus on single bones are preferable (Parr et al. 2012). Multivariate analyses of talar metrics may hold particular promise for identifying key morphological adaptations, and for distinguishing between taxa and habitual modes of posture and locomotion due to the increased number of shape variables that can be included. Previous analyses have successfully distinguished between taxa and locomotor modes on that basis, and some have successfully used resulting models for interpreting the affinities of fossil specimens (Day & Wood, 1968; Lisowski et al. 1974; Kidd et al. 1996; Kidd & Oxnard, 2005; Gebo & Schwartz, 2006; Jungers et al. 2009a,b; Kanamoto et al. 2011; Zipfel et al. 2011).

Three factors are likely to influence overall bone morphology in inter-specific comparison: phylogenetic constraint (the expectation that more closely related species will share more similar genetic and developmental blueprints, thereby constraining adult shape variation); size (including allometric shape variation); and function (variation in locomotor and postural behaviour). In addition, at specimen level, behavioural patterns may affect shape through bone remodelling during an individual's lifetime (Pearson & Lieberman, 2004).

Constraints on articular facet morphology

A number of previous studies have been conducted on facet surface areas of long bones, but the methodology for data acquisition was time consuming (Swartz, 1989; Godfrey et al. 1991; Lieberman et al. 2001). Moreover, to date, it has been difficult to accurately quantify the three-dimensional (3D) orientation and curvature of articular facets (see Latimer et al. 1987; Latimer & Lovejoy, 1989; Hamrick, 1996; Tocheri et al. 2003; Polly, 2008; DeSilva, 2009a,b; Matsuura et al. 2010; Kanamoto et al. 2011; Ward et al. 2011, for successful attempts). Yet these very characteristics are often cited in qualitative studies as being important, and even definitive attributes of particular taxa, or demonstrative of a particular function of the tarsals (e.g. Morton, 1927; Gebo & Schwartz, 2006). With the advent of 3D data capture by laser surface and computer tomographic scanning, it is now possible to image the surface morphology of bones very accurately. This allows additional new measurements to be taken that would have been very time consuming or impossible to achieve using traditional methods.

The morphology of articular facets can be described in terms of their surface areas, including their outline morphology, surface curvatures and orientations. Functionally, facet surface area is controlled by two main factors: the requirement to withstand forces transmitted through the joint; and a requirement to provide sufficient mobility, or range of motion, at the joint (Godfrey et al. 1991; Hamrick, 1996; Polk, 2002). However, these two selective pressures are not independent of each other (Rafferty & Ruff, 1994). For example, if a species becomes larger we would expect the surface area of facets to increase due to increased force transmission, resulting from increased body mass. In fact, as species get larger it has been shown that they adopt increasingly straight-legged gaits (Biewener, 1983, 1989a,b; Polk, 2002). Consequently, the range of motion at their joints might be expected to decrease, in turn leading to a decrease in relative joint surface area in larger species. To complicate matters further, as an animal straightens its leg during a gait the relative forces transmitted through the joints by the muscles to maintain joint posture decrease. This may lead to a decrease in the relative joint surface area due to decreased muscular force being transmitted across the joint.

Alexander (1980, 1981) predicted that maximum articular joint facet stresses would be of the same order of magnitude across mammals, irrespective of body mass. This presented a biomechanical foundation for the null hypothesis of isometric scaling between articular surface area and body mass (Ruff, 1988; Jungers, 1991). Godfrey et al. (1991) further noted that the concave and convex joint surfaces of an articulation operate with slightly different constraints and predicted that, of the articular pair, the concave facet surface area should be more highly influenced by body mass, whereas the convex surface should reflect differences in mobility at the joint and, hence, inter-specific differences in locomotor type. These predictions were mostly not upheld in an allometric study of talus articular surface areas in hominoids (Parr et al. 2011b). Based on Alexander (1980) and previous empirical studies (Swartz, 1989; Godfrey et al. 1991; Hamrick, 1996; Ruff, 2003), Parr et al. (2011b) had predicted that, intra-specifically, concave and convex articular surface areas would scale with similar scaling exponents, with the exception of convex articular facet areas in species where a substantial degree of sexual body mass dimorphism may be reflected in differences in locomotor strategies. Instead, while the null hypothesis of isometric scaling could not be rejected for any of the convex or concave facets in the non-dimorphic gibbons and siamangs, isometric scaling was rejected for some facets in the somewhat dimorphic chimpanzees and humans, and in the highly dimorphic orang-utans and gorillas, but this was the case for both convex and concave facets (Parr et al. 2011b). Similarly, inter-specifically, Parr et al. (2011b) found that all convex articular facets scaled isometrically, but the concave calcaneal articular facet (part of the subtalar joint) scaled with positive allometry, contrasting with the predictions that concave articular facets should scale isometrically with body mass across species, while convex facets might scale with positive or negative allometry, as a result of inter-specific differences in locomotor strategies. In conclusion, it appeared that, even though surface areas of articular facets are significantly correlated with body mass, their scaling patterns do not follow simple theoretical predictions (Parr et al. 2011b).

The mobility of a joint in a single plane can be increased in two ways. First, increasing the length of the facet curve will increase the distance one bone can translate over another (Latimer & Lovejoy, 1989; Godfrey et al. 1991; Hamrick, 1996). This will increase mobility at the joint, but may lead to increased surface area of the joint due to the lengthening of the facet curve. Second, increasing the curvature of the facet surface will increase the degree of rotation occurring around the joint. The confounding nature of different selective pressures on facet surface area and curvature makes it difficult to determine exactly which influence, if any, is the primary determinant of individual characteristics in individual specimens, and of differences between different species (Godfrey et al. 1991).

Articular facets are expected to be oriented so as to ensure that the resultant vectors for the majority of forces transmitted across the joint are as perpendicular to the joint surface as possible (Ruff & Runestad, 1992). This avoids the generation of potentially damaging shear forces in the articular cartilage (Lieberman et al. 2001; Dunn et al. 2014). Additionally, the articular facets will be oriented to maximise the range of movements in the directions required to enable efficient mobility for the locomotor type of the species (Ruff & Runestad, 1992; Tocheri et al. 2003). Thus, the multiplicity of confounding influences that control facet curvatures and facet surface areas are not expected to complicate interpretations of relative facet orientations.

Earlier studies using methods such as comparing ratios of linear measurements taken from anatomical regions deemed important have often sought to identify how these measurements are related to size or function (Day & Napier, 1964; Day & Wood, 1968; Lisowski et al. 1974; Gebo & Schwartz, 2006), contrasting in part with the approach employed by studies using more qualitative, or descriptive, methods (Morton, 1927; Le Gros Clarke, 1947; Lewis, 1980a,b,c, 1989; Gebo, 1992, 1996). Principal components analysis (PCA) is commonly performed on multivariate morphological datasets, and has in the past been used to interpret data on talar morphology consisting of either 2D linear measurements (Day & Napier, 1964; Day & Wood, 1968; Lisowski et al. 1974; Gebo & Schwartz, 2006) or, more recently, landmark or semi-landmark coordinate points. PCAs identify ‘types’ of shape variation in a dataset. Recent studies to have used 3D landmark data to quantitatively compare talar shape include an inter-specific comparison of extant and fossil hominoids (Harcourt-Smith, 2002), a study of developmental plasticity in modern humans (Hellier & Jeffery, 2006), a comparative study of adaptive morphology in the ankle of pinniped carnivorans (Polly, 2008), a study of articular facet orientation in humans and African apes (Kanamoto et al. 2011), and a comparative analysis of nine extant catarrhine taxa (Turley & Frost, 2013). However, of those, only Polly (2008) captured and analysed whole talar bone shape in high resolution, as Harcourt-Smith (2002) and Turley & Frost (2013) used a traditional landmark approach, and Hellier & Jeffery (2006) used a greatly reduced subset of landmarks from their original scans.

Aims

While previous approaches, both univariate and multivariate, have provided many important insights into hominoid talar morphology, function and evolution, most were necessarily constrained by difficulties in quantifying potentially important morphological characters of the tarsal bones, such as articular facet curvature and orientation. Additionally, the globular nature of the talus has meant that previous studies based on 2D and 3D landmarks have struggled to identify anatomically homologous landmarks that are present across entire comparative samples. Finally, high intra-specific variability has been highlighted as a key complication in earlier attempts to reliably establish form–function relationships and discrimination between functions and taxa from talar morphology (Lovejoy, 1978).

New data acquisition methods that allow for digital analysis of whole bone virtual models now make it possible both for talar morphology to be analysed in a comparative context without a priori selection of morphological characteristics, and for the quantification of previously unavailable metrics. Here we provide the first comprehensive 3D characterisation of variation in the morphology of the hominoid talus, including whole talus shape, articular facet curvatures and orientations. We discuss the results in a comparative evolutionary context, and in the context of their utility for interpreting the fossil record.

Materials and methods

Sample and data capture

The modern comparative sample consisted of tali of Hylobates moloch, Symphalangus syndactylus, Pongo pygmaeus, Gorilla gorilla, Pan troglodytes and Homo sapiens (see Table 1 for origin of specimens and sample sizes). The fossil sample consisted of casts housed in the Earth Sciences (previously Palaeontology) Department of the Natural History Museum, London, UK, and included EM3519 and SPB4 (both assigned to H. neanderthalensis), the Clark Howell Omo specimen (a previously unreported specimen labelled Clark Howell Omo in the Museum collection and found in association with early anatomically modern human material), A.L.288-1as (Australopithecus afarensis), and OH8, KNM-ER 1464 and KNM-ER 1476, which have been variably assigned to Australopithecus or early Homo (Table 2; see Parr et al. 2011b for further details). A cast of the LB1 (Homo floresiensis) talus was also included in the fossil sample.

Table 1.

Extant specimens and sample sizes

Species/populations	Sample size	Collections
Hylobates moloch	9	ZSM
Symphalangus syndactylus	10	ZSM
Pongo pygmaeus	14	ZSM, NC, UMZC, OUMNH
Gorilla gorilla	31	PCM, NHMZ, NC
Pan troglodytes	30	PCM, NHMZ, NC
Homo sapiens	59
Andomanese islander	13	NHMP
Australian aboriginal	5	NHMP
Sri Lankan (Osman Hill collection)	8	NHMP
African bushwomen	7	NHMP
Industrial Briton (Spitalfields collection)	12	NHMP
Romano Briton (Poundbury collection)	13	NHMP

NHMP, Department of Earth Sciences (previously Palaeontology), The Natural History Museum, London; NHMZ, Zoology Department, The Natural History Museum, London; OUMNH, Oxford University Museum of Natural History; PCM, Powell Cotton Museum, Kent; NC, Napier Collection, Department of Anthropology, University College London; UMZC, University Museum of Zoology, Cambridge; ZSM, Zoologische Staatssammlung, Munich.

Table 2.

Casts of fossil specimens

Fossil taxa and specimen numbers	Collection
Australopithecus afarensis: A.L. 288-1as	NHMP
Unknown hominin: OH-8, KNM-ER 1464, KNM-ER 1476	NHMP
Homo floresiensis: LB1	NHMP
Homo neanderthalensis: SP4B (Spy 4), EM 3519 (Tabun C1)	NHMP
Homo sapiens: Clarke Howell OMO*	NHMP

NHMP, Department of Earth Sciences (previously Palaeontology), The Natural History Museum, London.

^*The Clarke Howell OMO specimen stems from the Omo deposits in Ethiopia, and specifically from the ‘pelvic corner’ region of the deposits. These same deposits yielded other fossil specimens that were assigned to early anatomically modern Homo sapiens (Stringer, 2003; McDougall et al. 2005; Pearson et al. 2008). These other fossil specimens have been dated to between 104 ± 7 kyr and 196 ± 2 kyr (McDougall et al. 2005). The Clark Howell material from Omo has not yet been dated, and so the exact age of this specimen is unknown. The specimen was classified as anatomically modern Homo sapiens.

A Konica Minolta Vivid 910 surface laser scanner was used to capture the entire surface morphology of each talus. The scanner is accurate to X: ± 0.22 mm; Y: ± 0.16 mm; Z: ± 0.10 mm. Scanning and reconstruction protocols followed Parr et al. (2011b). Each virtual talus was centred with its centroid at the origin of the 3D Cartesian coordinate system. The tali were then registered with one another using an iterative closest point (ICP) algorithm (Besl & Mckay, 1992), which minimises differences in tali orientation using surface morphology. Due to the considerable size differences between, for example, a H. moloch and a male G. gorilla specimen, all models were scaled to the same surface area before ICP registration was performed. This ensured accurate registration between all models despite the large natural size differences in this inter-specific sample. After ICP registration, all models were returned to their original sizes whilst maintaining their registered orientations, which is important for both the canonical sampling and for calculating the relative orientation of the articular facets (see below).

Quantification of articular facet morphology

Articular facets were isolated following protocols provided by Parr et al. (2011b), so that their curvatures and orientations could be calculated. Briefly, the surface of the virtual bone models consisted of triangulated point clouds. Articular curvature was estimated by fitting a sphere to each of the articular facets. The fitting process was achieved using least squares minimisation to ensure a good fit of the sphere to the facet curvature (Fig. 2). Our method is similar to that of Polly (2008), except that we fit spheres to the facet surface, whereas Polly fitted quadratic curves. A sensitivity analysis showed that the results were very similar in terms of the residuals of the two different types of curves fitted to the surfaces. We use sphere fitting as this method also allows the centre of rotation around the joint surface to be calculated (see Parr et al. 2012). The trochlea was split into three facet components: the lateral component (which articulates with the distal fibula facet); the medial component (which articulates with the medial malleolus of the tibia); and the central region. The three components were separated along the medial and lateral rims at the point of maximum medio-lateral curvature. A sphere was then fitted to the whole central region. In this way, trochlea grooving is essentially ignored as the radius of the sphere is controlled by the anterior–posterior curvature of the trochlea. Once the sphere was fitted, the radius of the sphere was recorded. The curvature data (radius) could then be analysed to determine how curvature scales with respect to size by regression analysis of Ln-transformed facet sphere radii against Ln-transformed size. Because there is potential error in measuring dependent (radius) and independent (size) variables, reduced major axis (RMA, Model II) regression was used.

Fig. 2.

Calculating articular facet curvature. Sphere fitting and curvature estimation for the calcaneal articular facet of a Hylobates moloch talus. (a) Plantar surface of the talus with the calcaneal facet highlighted in blue. (b) Sphere fitted to the facet curvature using a least squares procedure. (c) Posterior view of the talus, the fitted sphere and the centre point of the sphere. (d) Enlargement of the sphere fitted to the 3D coordinate points of the calcaneal facet, with the black line illustrating the radius of the sphere – this is the measurement used to quantify facet curvature, with an increased sphere radius signifying a decreased facet curvature. (e) Dorsal view of the 3D coordinates of the facet with fitted sphere and centre point. (d and e) Illustrate the ‘goodness of fit’ of the sphere to the facet curvature, and show that the method is appropriate for determining facet curvature.

To allow for intra-specific analyses and analyses of the whole dataset using individual specimen measurements rather than species means, it was necessary to use individual size estimates for each specimen. Body mass data were not available for the present sample, and cubed centroid size of canonically sampled tali was used as body mass proxy instead. Canonical sampling (Ruiz et al. 2002; Buxton et al. 2003; Ruto et al. 2006; Ruto, 2009) is described in Parr et al. (2011a) for the present sample. In short, each 3D bone model was divided into 50 slices along its length. Each of these 50 slices was then sampled 50 times along its circumference forming the canonical sample comprised of 2500 points for each bone. These canonical samples were sub-sampled to avoid over-sampling the proximal and distal ends of the bones, resulting in the final canonical samples consisting of 1617 points for each bone (see supplementary fig. S1 in Parr et al. 2011b, and the Results section for the description and validation of the use of cubed centroid size as a body mass proxy).

For inter-specific scaling of articular surface curvature, phylogenetic RMA models with likelihood fitted lambda were derived with the phytools package (Revell, 2012) in r vs. 2.15.1 (R Core Team, 2012). Confidence intervals for all model parameters were derived from intra-specific variability through a resampling routine where, for each analysis, model parameters were derived 10 000 times, each time based on one randomly selected individual from each species. Species mean residual values were compared through Kruskall–Wallis tests.

Due to the methodological difficulties involved in measuring articular facet orientations, many previous studies on tali have relied on qualitative comparisons. A major problem with qualitative comparisons is that it is impossible to characterise variation for any particular character in an entire sample in a probabilistic framework. Without quantitative measurements it can be difficult to define the range of a morphological character within a typical sample of specimens for a given species. In the present study we introduce a facet orientation measurement technique that quantifies the orientation of the articular facets across the entire sample. Three orthogonal vectors were fitted to each of the articular facets isolated from the ICP registered tali. These vectors were fitted using a singular value decomposition (SVD) technique. The use of SVD in this context constitutes a total least squares minimisation solution in that the resulting 3D orthogonal vectors are least squares fitted to the length, breadth and normal of the facet. A PCA was then performed on the length and normal vectors. The width vector was excluded, as all the orientation information in this vector is already present in the length and normal vectors.

Quantification and analysis of whole talus morphology

The canonical sampling process (Douros et al. 2002; Buxton et al. 2003; Ruto et al. 2006; Ruto, 2009) gives an accurate representation of the whole talar surface morphology, and the resulting surface models were used in a PCA of whole talar shapes (following Parr et al. 2011a). The canonical sampling procedure used here is similar to the eigensurface method described by Polly & Macleod (2008) where the surface points per slice (50 slices) were obtained by projecting radii from the major axis at equally spaced angles (360^o/50). For the PCA of whole talar shapes, canonical models were scaled to the same centroid size, to minimise differences in geometric size of the models, and Procrustes superimposed (translated and rotated), to minimise orientation differences between models (Rohlf & Slice, 1990).

Results

Talus centroid size as proxy for body mass

Scaling of species mean talus centroid size with species mean body mass resulted in a tight fit model (R²= 0.96), with a slope of 1.096, which did not depart significantly from isometry (P = 0.403). This confirms the suitability of talus centroid size as a proxy for body mass in inter-specific analyses. Species residual values were generally low, except for humans (H. moloch: −0.010; S. syndactylus: −0.108; P. pygmaeus: −0.045; G. gorilla: −0.084; P. troglodytes: 0.098; H. sapiens: 0.311). This is in line with previous results (Parr et al. 2011b) and suggests that centroid size is influenced by overall mechanical loading, resulting in terrestrial bipedal humans having substantially larger tali than other extant hominoids. It also suggests that for studies that aim to characterise talus morphology in a biomechanical context, as is the case here, talus centroid size might be a more suitable size proxy than body mass.

Inter-specific scaling of articular surface curvatures

Raw articular facet curvature data for extant and fossil specimens are listed in Tables 3 and 4. Inter-specific regression of articular facet curvatures against talus centroid size gave mean R² values ranging from 0.65 for the sustentaculum facet to 0.92 for the trochlea and the head facets (Table 5). The null hypothesis of isometric scaling could not be rejected for the trochlea, sustentaculum and calcaneal facets, but was marginally rejected for the head facet in favour of positive allometry, meaning that larger species have relatively less curved head facets than smaller species (Table 5). Species residuals derived from these models are summarised in Table 6 and Fig. 3. Kruskal–Wallis tests showed that residual values differed between species for all facets (trochlea: chi-squared = 56.341, P < 0.0001; head facet: chi-squared = 77.731, P < 0.0001; sustentaculum: chi-squared = 79.882, P < 0.0001; calcaneal facet: chi-squared = 15.604, P = 0.0081). Orang-utans stand out in having more curved trochlea, head and sustentaculum facets than other species, while humans and H. moloch have the least curved head and sustentaculum facets relative to their size (Table 5; Fig. 3).

Table 3.

Summary statistics of raw facet curvature data for extant specimens, measured as radius length (mm); higher values indicate lower curvatures

	Homo sapiens	Pan troglodytes	Gorilla gorilla	Pongo pygmaeus	Hylobates moloch	Symphalangus syndactylus
Trochlea
Mean	18.48	17.00	20.21	13.17	6.72	7.95
Standard deviation	2.49	1.52	2.94	2.72	0.50	0.83
Maximum	24.39	21.27	25.99	18.41	7.19	9.12
Minimum	13.88	13.88	15.33	9.79	5.53	6.87
Head
Mean	15.62	11.96	15.82	9.89	4.90	5.45
Standard deviation	2.41	1.19	2.30	1.06	0.49	0.39
Maximum	21.68	15.00	22.63	11.39	5.64	6.241
Minimum	11.05	8.86	12.10	8.37	3.96	4.94
Sustentaculum
Mean	22.20	14.42	18.46	11.27	6.52	6.45
Standard deviation	5.17	2.12	3.57	4.52	0.81	1.13
Maximum	36.36	21.70	29.51	18.75	8.51	8.66
Minimum	14.93	10.00	14.20	4.98	5.83	4.79
Calcaneal
Mean	17.42	14.06	18.63	14.59	5.74	7.78
Standard deviation	2.38	1.91	3.45	2.84	0.70	1.28
Maximum	24.89	21.00	26.63	18.59	7.04	9.80
Minimum	13.49	10.18	13.02	10.92	4.70	6.20

Table 4.

Raw curvature data for fossil specimens, measured as radius length (mm); higher values indicate lower curvatures

						H. neanderthalensis
	KNM-ER 1464	KNM-ER 1476	OH8	A. afarensis A.L. 288-1as	H. floresiensis LB1	EM3519	SP4b	H. sapiens Clarke Howell Omo
Trochlea	17.01	15.85s	14.28	12.17	13.49	17.97	21.30	17.04
Head	15.23	14.43	14.21	10.52	15.61	14.88	18.34	15.52
Sustentaculum	19.42	21.00	16.73	15.99	15.70	17.40	33.59	15.535
Calcaneal	18.48	14.46	12.88	13.04	11.54	16.91	23.75	20.24

Table 5.

RMA models of articular surface curvature radii vs. cubed talus centroid size

	Slope	Intercept	Lambda	R2
Trochlea	0.359 (0.315; 0.410)	−4.285 (−5.254; −3.471)	0.945 (0.4400; 0.9999)	0.916 (0.786; 0.991)
Head	0.395 (0.341; 0.458)	−5.264 (−6.439; −4.264)	0.921 (0.4001; 0.9999)	0.917 (0.808; 0.983)
Sustentaculum	0.463 (0.326; 0.661)	−6.369 (−10.143; −3.766)	0.891 (0.3309; 0.9999)	0.652 (0.290; 0.940)
Calcaneal	0.399 (0.291; 0.499)	−5.095 (−6.959; −3.045)	0.814 (0.2359; 0.9999)	0.893 (0.731; 0.987)

Summary (mean and 95% confidence limits) of RMA model parameters derived from 10 000 resampling bouts; 95% confidence limits (in brackets) are derived from the 250th and 9750th values of the resampling distributions.

Table 6.

Summary (mean and 95% confidence limits) of individual specimen residual values by species, derived from mean phylogenetic RMA model of articular surface curvature radii vs. talus centroid size (see Table 5)

	H. moloch	S. syndactylus	P. pygmaeus	G. gorilla	P. troglodytes	H. sapiens
Trochlea	0.044 (−0.030; 0.118)	0.022 (−0.056; 0.099)	−0.131 (−0.315; 0.054)	0.021 (−0.099; 0.141)	0.102 (0.000; 0.203)	0.025 (−0.108; 0.158)
Head	0.083 (0.001; 0.165)	−0.016 (−0.124; 0.093)	−0.131 (−0.272; 0.011)	0.021 (−0.119; 0.161)	0.018 (−0.092; 0.129)	0.108 (−0.026; 0.241)
Sustentaculum	0.303 (0.103; 0.502)	0.041 (−0.295; 0.376)	−0.297 (−1.023; 0.429)	−0.110 (−0.364; 0.145)	−0.031 (−0.232; 0.170)	0.186 (−0.133; 0.505)
Calcaneal	0.006 (−0.266; 0.277)	0.095 (−0.125; 0.315)	0.002 (−0.295; 0.299)	−0.068 (−0.276; 0.139)	−0.067 (−0.239; 0.106)	−0.026 (−0.275; 0.224)

95% confidence limits are derived from the standard deviations of species-specific residual distributions.

Fig. 3.

Box plots of species residuals derived from regressions of facet curvatures against talar centroid size. (a) Trochlea facet, (b) head facet, (c) sustentaculum facet, (d) calcaneal facet, residuals.

Fossil specimen residuals derived from the regression models are summarised in Table 7. KNM-ER 1476, OH8 and H. floresiensis stand out among our sample in that their head facet curvature radii exceed the 95% confidence limits for the mean value in any of our modern species samples, including modern humans, and in the case of the latter two are the largest of our entire sample (Tables 4 and 5; Fig. 3). OH8 and H. floresiensis also have sustentaculum facets with radii whose values exceed those in gorillas and chimpanzees, but are within the upper ranges of the other species, while the sustentaculum value for A.L. 288-1as (A. afarensis) exceeds those for the African apes and siamangs.

Table 7.

Individual fossil specimen residual values derived from mean phylogenetic RMA model of articular surface curvature radii vs. talus centroid size (see Table 5)

	KNM-ER 1464	KNM-ER 1476	OH8	A. afarensis (A.L. 288-1as)	H. floresiensis	H. neanderthalensis (EM 3519)	H. neanderthalensis (Spy 4)	H. sapiens (Clarke Howell Omo)
Trochlea	0.027	0.081	0.079	0.009	0.047	0.056	0.010	0.009
Head	0.185	0.267	0.365	0.163	0.486	0.132	0.104	0.181
Sustentaculum	0.190	NA*	0.333	0.403	0.301	0.046	NA*	−0.060
Calcaneal	0.130	0.022	NA*	0.133	−0.061	0.012	0.112	0.199

^*NA, facet is damaged.

Intra-specific scaling of articular surface curvatures

There were considerable differences between intra-specific scaling exponents, although 95% confidence intervals (CIs) are generally wide, with R²-values for the models ranging from 0.904 to as low as 0.125 (Table 8). All species, except H. moloch, showed positive allometric scaling (larger specimens having less curved facets) in articular facet curvatures for two or more facets, but there were no consistent differences between the scaling patterns of convex and concave facets.

Table 8.

Intra-specific scaling properties of articular facet curvature radii relative to cubed centroid size derived from RMA regression models

Facet	Facet type	Scaling exponent (95% CI)	R2	Reject isometric scaling with centroid size?
Hylobates moloch
Trochlea	Convex	0.338 (0.205, 0.470)	0.808	No
Head	Convex	0.437 (0.281, 0.593)	0.841	No
Sustentaculum	Convex	0.489 (0.140, 0.838)	0.361	No
Calcaneal	Concave	0.511 (0.064, 0.958)	0.043	No
Symphalangus syndactylus
Trochlea	Convex	0.482 (0.360, 0.603)	0.904	Yes (positive allometry)
Head	Convex	0.328 (0.160, 0.496)	0.605	No
Sustentaculum	Convex	0.803 (0.190, 1.416)	0.125	No
Calcaneal	Concave	0.748 (0.363, 1.133)	0.601	Yes (positive allometry)
Pongo pygmaeus
Trochlea	Convex	0.546 (0.408, 0.684)	0.855	Yes (positive allometry)
Head	Convex	0.294 (0.202, 0.387)	0.775	No
Sustentaculum	Convex	0.766 (0.418, 1.114)	0.531	Yes (positive allometry)
Calcaneal	Concave	0.545 (0.275, 0.815)	0.442	No
Gorilla gorilla
Trochlea	Convex	0.430 (0.364, 0.495)	0.839	Yes (positive allometry)
Head	Convex	0.412 (0.335, 0.489)	0.759	Yes (positive allometry)
Sustentaculum	Convex	0.527 (0.388, 0.665)	0.523	Yes (positive allometry)
Calcaneal	Concave	0.540 (0.426, 0.654)	0.691	Yes (positive allometry)
Pan troglodytes
Trochlea	Convex	0.474 (0.373, 0.575)	0.683	Yes (positive allometry)
Head	Convex	0.533 (0.423, 0.642)	0.706	Yes (positive allometry)
Sustentaculum	Convex	1.156 (0.499, 1.813)	0.267	Yes (positive allometry)
Calcaneal	Concave	0.684 (0.517, 0.852)	0.587	Yes (positive allometry)
Homo sapiens
Trochlea	Convex	0.382 (0.331, 0.432)	0.756	No
Head	Convex	0.429 (0.378, 0.480)	0.803	Yes (positive allometry)
Sustentaculum	Convex	0.622 (0.500, 0.744)	0.467	Yes (positive allometry)
Calcaneal	Concave	0.376 (0.293, 0.458)	0.332	No

CI, confidence interval.

PCA of articular facet orientation 3D vectors

PC 1 explained 34.9% of the variance in inter-specific talar articular facet orientation, with PC2 explaining 14.4%, PC3 8.8% and PC4 7.4%. The first four PCs explained 65.4% of total shape variation, and none was highly correlated with centroid size (r-values for PC1–4: −0.34, 0.22, 0.10, −0.44, respectively). PC1 separates P. pygmaeus (with high positive PC1 scores) from African apes and hylobatids (with PC1 scores closer to and around 0) and humans (with all negative PC1 scores), as shown in Fig. 4. The greatest differences in articular facet orientation identified by PC1 for length vectors were in the head, sustentaculum, calcaneal and trochlea facets (blue arrows in Fig. 5). For the normal vectors, PC1 identified the greatest orientation differences in the trochlea and medial facets (green arrows in Fig. 5). On PC2, the greatest differences in facet orientation for length vectors were in the sustentaculum, trochlea and medial facets. For the normal vectors the greatest differences were in the medial and sustentaculum facets. Due to the variable shape of the lateral facet in some species, this facet does not lend itself to the reliable calculation of a length vector. This vector is therefore omitted from the vector orientation analyses.

Fig. 4.

Scatter plot of PC1 against PC2 scores for the PCA of articular facet orientations (length and normal vectors).

Fig. 5.

PC1 positive (solid) and negative (dashed) extreme facet length (blue) and normal (green) vector orientations shown on a human talar model. Articular facets are coloured accordingly: orange calcaneal facet; blue medial facet; green trochlea facet; purple lateral facet; burgundy head facet; pink sustentaculum facet. Non-articular regions of the talar surface shown in opaque yellow. (a) Plantar aspect, (b) dorsal aspect, (c) anterior aspect, (d) medial aspect, (e) lateral aspect.

The PCA failed to clearly distinguish all groups, with large overlaps in the range of PC1 and 2 scores between P. troglodytes, G. gorilla and H. moloch (Fig. 4). PC1 does not reflect body mass but does, to some extent, rank species along a functional gradient from tali that are the least to those that are the most engaged in arboreal locomotion, although no distinction can be made between the African apes and Hylobates (Fig. 4). All of the fossil specimens fell within the modern human range of facet orientations, with the exception of the A. afarensis (A.L. 288-1as) specimen, which fell outside of the modern human range for both PC1 and PC2 scores. A.L. 288-1as lay outside the ranges of all extant species, with its PC1 score being closest to the P. troglodytes and H. moloch mean scores, and its PC2 score being closest to the S. syndactylus mean score.

PCA of canonically sampled whole talar surface morphology

PC 1 explained 28.9% of the total shape variance in scaled canonically sampled whole talar surfaces, with PC2 explaining 16.8%, PC3 8.8% and PC4 6.4% of shape variance. The first four PCs explained 60.8% of total shape variation, and none was highly correlated with centroid size (r-values for PC1–4: −0.49, 0.25, −0.53, 0.32 respectively). As with the analysis of articular facet orientation, PC1 broadly separates more arboreal non-human hominoids (P. pygmaeus and S. syndactylus high positive PC1 scores) from more terrestrial hominoids including humans (negative PC1 scores), as shown in Fig. 6.

Fig. 6.

PCA of scaled canonically sampled talar surfaces for extant hominoids. Positive extreme talar shapes (dorsal view) for PC 1 and PC 2 are shown in yellow, negative extreme shapes in blue. Species means are shown (see key). Coloured polygons show the range of PC scores for each species.

PC1 identified shape differences in the relative size of the posterior calcaneal facet, talar neck thickness, head facet medio-lateral and plantar–dorsal width, the lateral extension of the lateral facet, and the medial width of the medial side of the talus. The negative PC1 extreme shape has a stockier, more robust overall appearance. The PC2 high-scoring extreme shape model had increased dorsal height of the medial rim of the trochlea facet. The negative PC2 extreme shape showed increased lateral projection of the lateral facet and expansion in the posterior medial region of the trochlea body, as well as a shorter and less angled talar neck. Our analysis did not clearly distinguish groups due to substantial overlaps in the range of PC scores between adjacent groups. Figure 6 shows that the species are not clearly ordered according to phylogenetic relationship or body mass, but that there is an inter-specific functional signal present, at least in PC1, which is similar to the one reflected by PC1 in the PCA of articular facet orientations. All of the fossil specimens except for the H. neanderthalensis Spy 4 specimen fell within the range of modern human tali shapes and outside the ranges of the other hominoids. The Spy 4 specimen scored the lowest (most negative) score of any specimen on PC1. The OH8 and KMN-ER1476 specimens fell on the outer extreme margin of the modern human talar shape range.

Discussion

Articular facet curvatures

Within species (intra-specifically), scaling of articular facet curvatures with size was considerably stronger in some taxa than others (with R² ranging from 0.043 for the calcaneal facet in Hylobates to 0.904 for the trochlea in Symphalangus; Table 8). This range in curvature variation with respect to size is particularly apparent when compared with the relationship between articular surface area and size, which shows a high correlation at an intra-specific level (R² range 0.538–0.973; Parr et al. 2011b). Positive allometry in facet curvature radii means that larger individuals within the species have relatively larger curvature radii, or relatively less curved facets than smaller individuals; and all species, except H. moloch, showed positive allometric scaling in articular facet curvatures for two or more facets. It is particularly notable that all facets are characterised by positive allometric scaling in the two species of African ape. This may be the result of a combination of sexual dimorphism (mild in P. troglodytes, pronounced in G. gorilla) with a semi-terrestrial lifestyle. Specifically, in both chimpanzees and gorilla, females spend substantially more time in the trees than males (Doran, 1993). Positive allometric scaling across all articular facet curvatures in those two species may reflect the difference between locomotion and substrate use between smaller females and larger males.

Across species (inter-specifically), the null hypothesis of isometric scaling could not be rejected statistically, except marginally for the head facet, meaning that, as for intra-specific scaling, the expectation of a consistent difference between convex and concave articular facets was not met. Overall, and as for articular facet areas (Parr et al. 2011b), scaling of articular facet curvature appears therefore to be context-specific, and the functional difference between convex and concave articular facets not as simple as predicted. In this light, analysis of residual values (Table 6; Fig. 3) suggests an influence of locomotion and habitual substrate use on articular facet curvature. The particularly low curvature values for the human head and sustentaculum facets are very likely a reflection of the adaptation of the human foot for terrestrial bipedalism. The low curvature of the head facet is most probably linked to the stabilisation of the transverse tarsal joint (Gebo, 1992), while the lower curvature in the sustentaculum facet may reflect an increased emphasis on translation as opposed to eversion–inversion in the distal part of the subtalar joint. These differences reduce mobility of the transverse and subtalar joints, and create a more stable lever suitable for habitual bipedal locomotion in humans (Gebo, 1992; DeSilva, 2009a,b). The low head and, particularly, sustentaculum facet values in H. moloch are more difficult to explain, especially in the absence of similarly low values in Symphalangus. Hylobatids are unusual among apes in employing a semi-plantigrade posture during bipedal locomotion (Gebo, 1992), which may have consequences for talar morphology. Recent work has shown that the mid-tarsal break typical of non-human primates occurs to various degrees in the tarso-metatarsal joints as well as at the level of the transverse tarsal joint in chimpanzees and gorillas, as well as in mandrills and macaques (DeSilva, 2009a), and it may be that if the mid-tarsal break occurs primarily at the tarso-metatarsal joint in gibbons, they would benefit from a stiffening of the transverse tarsal joint during bipedal leaping. It should be noted that this result is consistent with our other analyses, discussed below, where H. moloch are more similar to humans than S. syndactylus in both articular facet orientation and overall talar morphology. Little has been reported on siamang locomotion in the wild, but available sources suggest similarities in locomotor strategies with orang-utans (Fleagle, 1976; Collis, 2001). This aligns well with our analyses, where siamangs tend to align more closely with orang-utans than Hylobates in head facet curvature, general facet orientation and overall talar morphology. In contrast to humans and H. moloch, P. pygmaeus are characterised by the highest curvature values for all facets except the calcaneal facet (Table 6; Fig. 3). This makes sense as, of all the types of locomotion seen in extant hominoids, the clambering arboreal locomotion of orang-utans clearly benefits the most from a high range of motion in the tarsal joints. Thus, although at the inter-specific level facet curvature is strongly related to body mass, species-specific variations related to locomotor mode are evident in the residual values derived from the underlying scaling relationship.

The residuals of the head facet curvatures in KNM-ER 1476, OH8 and H. floresiensis lie above the 95% confidence limits of any of our modern species samples, including modern humans, and are indicative of particularly low curvature values in those fossils, even compared with modern humans (Tables 4 and 5; Fig. 3). The trochlea, sustentaculum and calcaneal residuals lie within the 95% confidence limits for modern humans. This suggests a mosaic functional morphology for these specimens. The less curved head facets in KNM-ER 1476, OH8 and particularly H. floresiensis indicate decreased mobility at the medial transverse tarsal joint, even when compared with modern humans. Reduced transverse tarsal joint mobility leading to a more rigid foot and the absence of a mid-tarsal break is widely considered to be an adaptation to bipedal terrestrial locomotion (Elftman & Manter, 1935; Gebo, 1992; DeSilva, 2009a). Assuming that they are not the result of taphonomic modification, the implied extreme adaptation for terrestrial bipedalism in OH8 and H. floresiensis may, in the case of H. floresiensis, be related to its unusual foot proportions and provide a contrast to the more non-human-like aspects of its foot morphology (Jungers et al. 2009a,b). Similarly, the low curvature seen in OH8 may relativise the conclusions of a number of previous studies that suggested the talus to be the least human-like of all the elements of the OH8 foot bone assemblage (Harcourt-Smith & Aiello, 2004). More detailed quantitative analyses of the rest of the OH8 and H. floresiensis foot skeletons will be needed in order to fully understand the significance of the apparently derived head facet morphologies. In any case, our results suggest that the species represented by KNM-ER 1476 and OH8, as well as H. floresiensis (LB1), all had relatively stiff transverse tarsal joints, implying a primary role in propulsion rather than flexibility.

Previous research reported that joint curvatures were not correlated with either joint size or body size in the wrist joints of strepsirrhine primates, and concluded that, while joint size was correlated to body size, joint curvature was related to locomotor type alone (Hamrick, 1996). Our results agree in part. We confirm a relationship between facet curvatures and locomotor type and substrate use for the hominoid talus. In terms of scaling, our earlier work has also found that joint surface areas in the hominoid talus were correlated with size (Parr et al. 2011b). However, the present results contrast with results from the strepsirrhine wrist in showing that there can also be correlation between joint curvature and size at both inter- and intra-specific levels. In particular at the inter-specific level, articular facet curvatures in the hominoid talus are highly correlated with size.

Articular facet orientations

Results from the PCA of facet orientation (Fig. 4) suggest that a functional signal is present in talar facet orientations. High PC1 scores separated P. pygmaeus, with highly mobile and habitually inverted feet, from the hylobatids, P. troglodytes and G. gorilla. H. sapiens showed the lowest PC1 scores. The groupings in Fig. 4 do not follow phylogenetic or size order. The functional signal in facet orientation is not unexpected, as previous qualitative studies have concluded that the joints align to facilitate the range and extent of movement encountered during locomotor modes habitually used by a species (Morton, 1927; Tocheri et al. 2003). The most noticeable differences in facet orientations between the PC1 extremes, with H. sapiens at one end and P. pygmaeus at the other, are a dorso-lateral (anticlockwise when viewing a right talus from the anterior aspect) rotation of the trochlea, medial, head and sustentaculum facets relative to the lateral and calcaneal facets when moving from low (H. sapiens) to high (P. pygmaeus) PC1 values (Fig. 5c). The rotation is most pronounced in the long axis of the head facet, in line with previous studies that described human tali as characterised by greater torsion of the head (Volkov, 1903; Sewell, 1906; Day & Napier, 1964; Day & Wood, 1968; Kidd et al. 1996; Jungers et al. 2009b; Kanamoto et al. 2011; Zipfel et al. 2011). Talar head torsion has been reported as showing pronounced intra-specific variability in humans (Lovejoy, 1978), but in combination with other elements of articular orientation it clearly contributes to distinguishing modern human morphology from that of other apes. The sustentaculum facet largely follows the torsion of the head facet. The second most pronounced element of this rotation concerns the medial facet, indicating a more inclined medial facet in the flexible and habitually supinated foot of highly arboreal orang-utans, compared with the steeper facet of other apes and humans. A related element of torsion along the longitudinal axis of the talus has been reported in the literature as variation in the angle formed between the tangent to the superior surface of the talus, at the anteroposterior midpoint of its trochlea, and the line connecting the inferior-most points of the tibial (medial) and fibular (lateral) facets (Latimer et al. 1987; DeSilva, 2009b), termed the talar axis angle by Lovejoy et al. (2009). The closest approximation of variation in the talar axis angle in our data is the rotation of the trochlea normal vector relative to the fixed normal vector of the calcaneal facet around the longitudinal axis of the talus (Fig. 5c).

A second pronounced shift can be seen in lateral and medial views (Fig. 5d,e), where the trochlea normal axis (green) is oriented more posteriorly and the trochlea long axis (blue) more plantarly at extreme negative PC1 values, and more anteriorly and dorsally, respectively, at extreme positive PC1 values, relative to the normal axes of the head, sustentaculum and calcaneal facets. This translates into a more plantar orientation of the talar head relative to the trochlea in modern humans, and is related to the previously reported increased inclination of the talar neck in humans compared with other apes (Day & Wood, 1968; Kidd et al. 1996; Kanamoto et al. 2011; Zipfel et al. 2011). It is worth noting, however, that inclination of the head facet does not increase relative to the normal vectors of the calcaneal or sustentaculum facets, but instead decreases slightly. When the right talus is viewed from its medial aspect, both the head facet and the trochlea are rotated anticlockwise from low to high PC1 values. The increased angle of inclination of the talar neck in humans is due to the more pronounced clockwise (posterior) rotation of the trochlea (Fig. 5d). On the strength of this evidence, the angle of inclination of the talar neck might, hence, more accurately be described as the angle of reclination of the trochlea. The more posterior orientation of the trochlea in humans mirrors the differences observed between humans and the African apes in the antero-posterior tilt of the distal articular facet of the tibia or tibial arch angle (Davis, 1964; Stern & Susman, 1983; DeSilva & Throckmorton, 2010), and may be a useful indicator of the level of rear-foot longitudinal arching in individuals, as a relationship, albeit weak, has been demonstrated between orientation of the tibial arch angle and degree of longitudinal arching in humans (DeSilva & Throckmorton, 2010).

Finally, in plantar view (Fig. 5a), the negative PC1 values associated with humans are indicative of a more transverse orientation of the calcaneal and sustentaculum long axes (blue arrows), while the positive extreme values associated with orang-utans are indicative of a more oblique orientation of those axes relative to the long axis of the talus and to the normal axis of the head facet, in line with previous reports (Keith, 1929; Lewis, 1980b, 1981, 1989). In summary, therefore, our quantification and representation of articular facet orientations successfully visualises previously identified elements of variation in the morphology of the hominoid talus that relate to facet orientation, but has the advantage of visualising all elements of variation simultaneously, making it possible to assess the relative contributions of individual elements to overall variation.

All fossil tali in our sample grouped with the human sample in the PCA of facet orientations, with the exception of A.L. 288-1as (A. afarensis), suggesting that in terms of overall relative facet orientations, only A. afarensis deviates from the modern human morphology (Fig. 4). Within our sample, A. afarensis occupies a unique position in the morphological space defined by the first two PCs, with intermediate scores on PC1, and among the lowest scores on PC2 (Fig. 4). As suggested by extant species variation, facet orientation likely informs on functional aspects of the talus, suggesting that the ankle (talocrural), subtalar and medial transverse tarsal joints of A. afarensis may have functioned differently to those of extant hominoids, including modern humans. It is possible, but we think unlikely, that the reason the AL-288-1as specimen lies out of the human range is due to a low angle of declination (potentially a flat-footed signal). We think this is unlikely to be the reason for the unique placement of AL-288-1as in Fig. 4, as the modern human sample contained unshod flat-footed individuals with declination angles similar to the AL288-1as specimen. Further, the metrics for the facet orientation vectors do not indicate that it is the angle of declination alone that makes the A.L. 288-1as specimen unique. The analyses show that it is the combination of facet orientations that is unique. Future detailed interpretations of articular facet orientation and comparisons with other fossil specimens and taxa are, hence, likely to produce important new insights on the evolution of bipedalism.

Canonical samples of whole talar surface shape

A recent study of nine catarrhine taxa, including modern humans, chimpanzees, gorillas, orang-utans and non-specified hylobatids, identified significant differences between hominoids and cercopithecoids, and roles for size and substrate preference in determining talus shape (characterised by a set of 30 Procrustes transformed landmarks; Turley & Frost, 2013). In addition, landmark subsets representing the proximal and the distal talus were found to be influenced primarily by substrate preference and superfamily affiliation, respectively (Turley & Frost, 2013). As with the PCA of articular facet orientation, our PCA of scaled talar canonical representations suggested a predominantly functional element to shape variation within hominoids. The specimens ranked along PC1 as in the analysis of facet orientation, with modern humans at one extreme and orang-utans at the other. This is perhaps unsurprising as, as shown in Figs 1 and 5, the majority of the talar surface is covered by articular facets. The results are also similar to those of Turley & Frost (2013), although we achieve a better separation of orang-utans from the other apes, likely thanks to the larger number of landmarks and, hence, more detailed representation of surface morphology, in the canonical representation of tali.

In our analysis of whole talar shapes, all fossils fell within the range of modern humans, with the exception of the Spy 4 Neanderthal, which had the lowest PC1 score of any specimen; although the OH8 specimen lies on the very edge of modern human variation with respect to PC2 (Fig. 6). The negative PC1 extreme shape has a stockier, more robust overall appearance. The Spy 4 Neanderthal is considered to be a male, and the stocky appearance of the talus is likely a reflection of the individual being a particularly large, strong and powerful representative of an obligatorily bipedal species (Trinkaus, 1983; Ruff et al. 1993; Parr et al. 2011b), suggesting an element of allometric scaling in overall talus shape variation. Unlike in the analysis of facet orientations, the A. afarensis specimen (A.L. 288-1as) groups well with modern human variation, implying an overall similarity in shape between the two species, perhaps due to their close phylogenetic relationship, but with potentially important functional differences as indicated by articular facet orientations.

Concluding remarks on the approach and its applicability to the fossil record

The method introduced here directly resamples the surface morphology of the initial talar scan to accurately capture the morphology of the whole talar surface. PCA then determines how the whole talar surface varies. Using more ‘traditional’ landmarks, previous analyses have been able to determine how the relative position of landmarks changed between species (Harcourt-Smith, 2002), but have not captured the whole talar geometry in high resolution. As demonstrated above, the canonical approach adopted in the present study allows further exploration of the bone's 3D shape as a means to gain insight into potentially useful morphological traits and a basis for further separate analyses on these traits.

In combination with previous research, our results show that inter- and intra-specific variation in surface area, curvature and orientation of articular facets in the hominoid talus are affected to different, but at least to some extent predictable, degrees by both size and function. Three-dimensional quantification of facet curvature and relative facet orientation in fossil specimens can contribute to inferring locomotor function in extinct species, although for many of the characters, high levels of intra-specific variability renders inferences based on single elements and/or measurements tentative. Intra-specifically, consistent allometric scaling of facet curvatures likely reflects intra-specific differences in locomotion and substrate use, as seen in extant chimpanzees and gorillas, and could, with sufficiently large samples, be detected in fossil taxa. The KNM-ER 1476, OH8 and H. floresiensis specimens possessed reduced head facet curvatures even compared with modern humans, suggesting reduced mobility at the transverse tarsal joint. This is consistent with foot adaptation to habitual bipedal gait, and agrees with earlier interpretations of the OH8 and KNM-ER 1476 tali (Day & Wood, 1968; Gebo & Schwartz, 2006). Apart from the reduced curvature of the head facet, H. floresiensis was not separated from modern human talar anatomy by any of the other analyses, suggesting a predominantly modern talar anatomy and ankle function in this specimen, as found by Raichlen et al. (2010). In contrast, the differences in facet orientations between A.L. 288-1as (A. afarensis) and extant hominoids suggest that the biomechanics of talar elements of the bipedal gait in this specimen differed from those in any extant species, including modern humans.

Author contributions

WCHP conceived of the study, undertook all data collection and most analyses, and prepared the manuscript. TR prepared the canonical representations of tali. JS and CS contributed some further analyses. LC contributed to fossil specimen preparation. HJC, CS and SW participated in data interpretation and manuscript preparation, and oversaw supervision of the research. All authors have approved the final article. The authors declare no conflicts of interest.