The human sample used in this study consists of 12 adult femora representing two females, seven males and three individuals of unknown sex from 19th-century France (N = 6; coll. Musée de l’Homme, Paris, and University of Poitiers) and from the 1st-2nd century Imperial Roman graveyard of Velia, in southern Italy (N = 6; coll. Museo Nazionale Preistorico Etnografico “L. Pigorini”, Rome).

The chimpanzee (Pan troglodytes) sample consists of 10 adult femora from three males, four females and three individuals of unknown sex selected from the African ape skeletal collections stored at the MNHN Paris and the Senckenberg Forschungsinstitut und Naturmuseum, Frankfurt am Main. Unfortunately, information about their original living environment (captive vs. wild) is not systematically available in the files. However, while Jensvold et al. (2001) noted that captive chimpanzees living in confined and less complex environments are characterized by lower locomotor activity, Marchi, 2005 and Marchi, 2007 did not find significant differences between captive and wild apes in cross-sectional geometric properties of the tibia, fibula, hand and foot bones This conclusion was recently strengthened by Morimoto et al. (2011a) in their comparative morphometric mapping of the femoral shaft.

Because of the limited number of individuals involved in this preliminary study, data were treated regardless of sex. All material has been selected because of its excellent preservation quality and absence of any macroscopic indication of gross pathology.

Femora were scanned using medical computed tomography equipment (CT, N = 19) and the synchrotron radiation microtomographer set at the beamline ID 17 of the European Synchrotron Radiation Facility of Grenoble (SR-μCT, N = 3). CT-based acquisitions resulted in voxel size ranging from 111 to 710 μm (for the x and y axes), and from 330 to 625 μm (for the z axis), while SR-μCT acquisitions used an isotropic voxel size of 350 μm.

For each femoral specimen, in order to virtually isolate the cortical shell, to image its inner morphology, and to reconstruct the entire volume, a semi-automatic threshold-based segmentation with manual corrections has been carried out following the half-maximum height method (HMH; Spoor et al., 1993) and the region of interest thresholding protocol (ROI-Tb; Fajardo et al., 2002). On different slices of the virtual stack, repeated measurements (Coleman and Colbert, 2007) have been taken by the Avizo v.6.2. (Visualization Sciences Group Inc.) and ImageJ (Rasband, 2010) packages. The result of the segmentation is a triangulated mesh model of the endosteal and periosteal surfaces composed of 3D coordinate vertices connected by lines that generate triangular faces obtained after “unconstrained smoothing” (Yoo, 2004).

As shown by previous similar studies dealing with long bone 3D virtual modelling using different imaging systems (Bayle et al., 2011, Bondioli et al., 2010, Mazurier et al., 2010 and Puymerail, 2011), a number of intra- and interobserver tests for accuracy run by different observers revealed differences less or equal to 5%.

Two different analytical methods have been used in this analysis in order to characterize the femoral shaft structural organisation in Homo and Pan: cross-sectional geometry and morphometric mapping (for technical details, see Bondioli et al., 2010). Analyses have been run by means of custom routines developed in Scilab v.5.1.1 (The Scilab Consortium) and in R v.2.12.2 language (R Development Core Team, 2011).

In all investigated cases, cross-sectional geometric properties were assessed on virtual sections set at regular intervals of 1% along the shaft within the portion 20 to 80% of the biomechanical length. Following Ruff (2002), the femoral biomechanical length is measured from the average of the distal condyles to the proximal neck where, by convention, 80% corresponds to the relative proximal end of this shaft portion (thus, 20% is distal). At each cross-sectional level, the following parameters have been systematically measured: total area (TA, in mm²); cortical area (CA, in mm²); percent of cortical area (%CA); second moments of area about the medio-lateral (M-L) and antero-posterior (A-P) axes (Ix, Iy, in mm⁴) and the Ix/Iy ratio (see Ruff, 2008 and Stock and Shaw, 2007 for further discussion of these properties). To identify behaviourally significant differences in bone structure, it's necessary to control the effects of body size. Since individual body mass values were not available for any specimens of our sample, we choose to specifically focus on the size-free ratio %CA and Ix/Iy.

Firstly introduced by Amtmann (1971) and applied to the human fossil record by Zollikofer and Ponce de León, 2001 and Zollikofer and Ponce de León, 2005, morphometric mapping is a new way to model stress and strain distribution along the bone shaft (Bondioli et al., 2010, Morimoto et al., 2011a, Puymerail et al., 2012a and Puymerail et al., 2012b). For the specific purposes of the present study, cortical bone thickness is defined as the distance for each point from a node on the periosteal surface to its closest node on the endosteal surface (Bondioli et al., 2010). Since femoral diaphysis does not significantly deviate from a cylindrical model, the periosteal (external) surface is then mapped into a cylinder whose diameter corresponds to the maximum width of the original shaft surface. By using the (μ)CT-based 3D reconstructions, the cylinder representing each investigated femur is firstly virtually cut vertically along a predefined line along the anterior aspect, and then unrolled into a plane. As the direction of unrolling changes according to the anatomical side of the investigated specimen, it is therefore possible to invert the polarity of this variable in order to obtain fully comparable maps, regardless the original sides.

In order to statistically compare for their cortical bone topographic distribution, two samples of femoral shafts representing differently sized and shaped taxa (Homo vs. Pan), both size of each morphometric map and thickness measurements have been standardized. With special reference to overall size, the 20 to 80% investigated shaft portions have been systematically grid onto a regular mesh of 100 rows by 100 columns. The second step implied thickness standardization between 0 and 1. We then performed a generalized additive modelling (GAM). According to this method, thickness values at the gridline intersections are evaluated by means of thin plate splines regression of the morphometric maps (Wood, 2006). Since morphometric maps fully overlap following standardisation, it is possible to perform GAM in order to obtain distinct consensus maps of the human and the chimpanzee shafts, respectively, by merging all the individual maps into a single dataset (Puymerail, 2011). This protocol allows visual comparative assessments as well as statistical comparisons run by principal component analysis of thickness distribution at the gridline intersections. Two-sample Wilcoxon tests between the two locomotor groups. R v.2.12.2 has been used to run all statistical analysis.

The unstandardized values of the cross-sectional geometric properties of the human and chimpanzee femoral samples measured (in distal-proximal direction) at 20%, 35%, 50%, 65% and 80% of the biomechanical length are shown in Table 1, and the scatterplots of the percent of cortical area (%CA) and the Ix/Iy ratio (Ix/Iy) are presented on Fig. 1 and Fig. 2, respectively. In order to simplify the global reading and interpretation of the results, the statistical significance (P-values) of the differences at each cross-section for the Wilcoxon test run between the two locomotor groups is provided directly on each scatter plot.

Cortical area (CA) is a measure of strength, rigidity or resistance of the diaphyseal cross-section in pure compression or tension (Lieberman et al., 2004, Ruff, 2008 and Sládek et al., 2006). For this variable, differences between Homo and Pan are evident all along the shaft (Table 1). CA unstandardized values displayed by the human sample used in this study are higher compared to the chimpanzee ones, especially in the proximal portion of the diaphysis (50–80% of the biomechanical length), where humans express maximum variation. Compared to the human figures, variation in cortical area in our Pan sample is globally lower and more uniformly distributed along the shaft. As indicated by Raichlen et al. (2009), compared with other primates, chimpanzees support more weight on their hind limbs because they walk with a relatively protracted hind limb during both terrestrial and arboreal quadrupedalism (Raichlen et al., 2009).

Percent of cortical area (%CA), assessed with respect to the total cross-sectional area (including the medullary cavity), is a measure of diaphyseal resistance in pure compression or tension (Trinkaus and Ruff, 2012). As shown on Fig. 1, Homo and Pan are clearly distinct for their %CA values in the mid-proximal portion of the femoral diaphysis, notably from 55% to 70% of the biomechanical length. They also differ in the pattern of relative %CA distribution. More specifically, while in chimpanzee the most robust diaphyseal portion is found distally, the opposite is true in the human sample. Furthermore, while %CA values in Pan show a marked increase from 20% to 35%, then continuing into a plateau-like outline toward the proximal end, in Homo this distal-proximal increase is more accentuated and extends till 55 to 65% of the biomechanical length, where it is followed by a sharp decrease within the remaining shaft portion.

Statistical differences between the two taxa for the Wilcoxon test are significant all along the diaphysis, except for its very proximal portion (70% to 80%) and around 40% of the biomechanical length (Fig. 1). However, it should be also noted that, compared to the modern human figures (Puymerail et al., 2012b, table 3), the site-specific variation at cross-sectional level of cortical thickness assessed along the femoral diaphysis is still poorly reported in adult Pan (but see Morimoto et al., 2011a).

The Ix/Iy ratio is an index of relative bending strength of the A-P versus M-L bone distribution calculated by using the anatomically oriented second moments of area (Ruff, 2008 and Ruff and Hayes, 1983). The outlines shown on Fig. 2 indicate that, with respect to the chimpanzee condition, the human femur is fully adapted to relatively greater M-L bending loads in the proximal and distal shaft portions, and to relatively greater A-P bending loads around the midshaft region (40–60% of the biomechanical length). Conversely, as previously shown by Carlson (2005) in his study of 120 individuals, our results show that the chimpanzee femur is better adapted to relatively greater M-L bending loads all along the diaphysis. Accordingly, the main differences between the two samples for the Ix/Iy ratio are found around the midshaft region, while P-values of the Wilcoxon test are not significant for the proximal and the distal femoral portions (Fig. 2). Terrestrial and arboreal landscapes are littered with obstacles around which chimpanzees must move; but arboreal locomotion is characterized by multi-directional external forces (Carlson et al., 2006).

Concerning the chimpanzee, the strong variation observed in the biomechanical parameters of the femoral shaft could be explained by variation of sex differences characterizing the locomotor behaviour (e.g., Doran, 1993).

As exemplified by the patterns displayed by two representative specimens respectively selected from the two investigated samples, differences between Homo and Pan in cortical bone thickness topographic variation for the 20 to 80% portion of the femoral diaphysis are rendered in different anatomical views on Fig. 3. According to this visualisation perspective, marked differences in cortical bone distribution are found on each projection, but the anterior one. However, in order to summarize this spread endostructural variation within a single image of the femoral diaphysis suitable for direct functional comparison, we generated a 3D consensual standardized morphometric map distinctly for the human and the chimpanzee samples (Fig. 4).

The structural signature shown by the modern human map (Fig. 4a) features three major shaft reinforcements. The most developed one corresponds to the posterior pilaster, where the absolutely thickest bone is essentially found around the midshaft. Additional strengthening is recognizable at the level of the proximal medial and the proximal lateral aspects of the shaft, the former representing the structural continuity of the typical asymmetry in cortical bone thickness characterizing the configuration of the bipedally-adapted hominin femoral neck (Galik et al., 2004, Lovejoy et al., 2002, Matsumura et al., 2010, Ohman et al., 1997, Zebaze et al., 2005 and Zebaze et al., 2007). However, while the posterior and the proximal medial features are systematically well expressed in our modern human femora (see also Puymerail et al., 2012a), the degree of development of the lateral thickening is highly variable (cf. Bondioli et al., 2010, fig. 3). This feature consists of two components. Its most distal portion, which is rather anteriorly oriented, is mostly found between 50% and 65% of the biomechanical length, while its proximal portion, between 65% and 80%, is oriented toward the posterior aspect. Together, these structural components roughly correspond to the attachment site of the gluteus maximus muscle (Carlson, 2006, Morimoto et al., 2011b, Sigmon, 1974 and Swindler and Wood, 1973). In all cases represented in the present study, bone thickening in the modern human femur is distinctly concentrated within the proximal half of the diaphysis, while its distal portion, notably below 35% of the biomechanical length, is relatively and absolutely thin.

In contrast to this pattern, as seen on Fig. 4b, two distinct reinforcements can be identified medially and laterally on the proximal portion of the chimpanzee femoral diaphysis. The former is the most extended one, as it covers the entire medial aspect of the shaft portion between 40% and 80% of the biomechanical length. The lateral one, which is more proximally restricted (70–80%) and corresponds to the lateral spiral pilaster expressed at the external surface of the chimpanzee femur (Morimoto et al., 2011b), commonly appears as a crook-shaped imprint. Finally, a variably expressed posterior bone thickening is occasionally detected in Pan between 55% and 65% of the biomechanical length. This structural signature is consistent with the activity and distribution of the thigh muscles of the chimpanzee (Larson and Stern, 2008 and Stern and Susman, 1981). As previously noted for Homo, the thinnest cortical bone along the chimpanzee femoral shaft is again found distally, between 20% and 30% (but see Morimoto et al., 2011a for some structural differences recorded at this level between captive and wild chimpanzees). This functional pattern globally fits the condition described by Morimoto et al. (2011a) following their morphometric analysis based on 16 adult chimpanzee femora.

In order to evidence the most distinct functional areas characterizing the human and chimpanzee diaphyseal pattern, respectively, the individual morphometric maps have been used to run a principal component analysis. Results are shown on Fig. 5.

By representing 30.7% of the total variance, the first component (PC1), which mostly expresses the presence vs. absence of the medial and lateral reinforcements, does not separate humans from chimpanzees. Conversely, a distinction between the two locomotor groups is provided by the second component (PC2, 22.3% of the total variance). In this case, the most discriminant areas in terms of relative cortical bone thickness variation correspond to the posterior and to the lateral diaphyseal reinforcements, two features which typically characterize the modern human femur. Both regions are found on the proximal half of the shaft, while the distal one is significantly less informative in biomechanical terms. A wilk-lambda test run on the first six components (82.6% of the total variance) reveal highly significant differences between the two groups (P = 5.56 × 10⁻⁶, F = 17.39). This is in accordance with the proximal attachment onto the femoral shaft of various muscles that are biomechanically relevant for quadrupedalism and bipedalism (Carlson, 2006, Lovejoy et al., 2002, Morimoto et al., 2011b, Sigmon, 1974 and Swindler and Wood, 1973). As shown by Thorpe et al. (2009), with the exception of the adductors at the hip, chimpanzees are adapted for more movement at the hindlimb joint than human (Thorpe et al., 2009). At the opposite, human bipedalism requires forces to be exerted with the limbs in particular positions, which are about the same in each stride (Aiello and Dean, 1990). Arboreal locomotor mode generates strains of relatively high magnitude and variable orientation with functional consequences on features of locomotion in general (Carlson, 2005).