All three Neandertal specimens (Spy II, Neandertal 1 and Kebara 2) exhibit slightly different morphologies. The Spy lower skeleton (femur and tibia) and the Kebara 2 pelvis have no common skeletal elements that would enable straightforward scaling. Scaling methods found in literature process objects (i.e., bones) of the same nature: for example, a femur is scaled to another femur (Van Sint Jan et al., 2002). To the authors’ knowledge, no previous method in the anthropological sciences has reported a scaling method to process objects of a different nature (i.e., for example, scaling a pelvis to the dimensions of a particular femur). The fusion method presented here aimed to achieve this. The Neandertal 1 femur was used as an intermediate scaling stage between the Kebara and the Spy II dimensions (Fig. 1, Scaling A). The method is based on the assumption that relative pelvis size compared to femur and tibia size is the same in different individuals.

Model scaling was performed in lhpFusionBox via two stages using spatial data registration from the previously located ALs. The first scaling stage (Fig. 1, Scaling A) scaled the Kebara 2 iliac bone to the Neandertal 1 iliac bone (Table 2, steps 1 to 6). The second stage (Fig. 1, Scaling B; Table 3, steps 7 to 14,) scaled the result of the first stage to the Spy II femoral dimensions via the Neandertal 1 femoral dimensions. The scaling procedure was based on standard singular value decomposition, or SVD, algorithms (Challis, 1995 and Horn, 1987).

Stage 1 in Table 2 (steps 1 to 6) is similar to standard spatial registration performed in the biomechanics field, i.e., registering bone models of the same kinds together (e.g., iliac with iliac, femur with femur). Steps 1 to 4 dealt with virtual palpation and mirroring of the various models. Step 5 scaled the original Kebara 2 right iliac bone (RK2) to the dimensions of the mirrored Neandertal 1 iliac bone (RN1 ^mirror ). The same procedure was adopted for the heterolateral side (step 6). At the end of this stage, the Kebara 2 iliac bones showed Neandertal 1 proportions. The fitting of the joint is important for motion representation and to enable further motion analysis, it is therefore important that the joint registration is accurately scaled.

The above-described method enabled us to obtain a full lower limb model of a Neandertal (see results). In order to allow further musculoskeletal analysis, the entire model has been further registered to joint kinematics and motion data of an anatomically modern human (further details available from Sholukha et al. (2006)). The registration of the Neandertal model to the animated human data gives a detailed description, including a six degrees-of-freedom mechanism at knee and ankle joint levels, of motion pattern (Chapman et al., in press and Sholukha et al., 2006). This information also includes the hip, knee and ankle instantaneous helical axis (Van Sint Jan et al., 2002). The motion selected for this study was a squatting motion as this covered a large range in all joints. A similar model of a modern human was available for comparison with the Neandertal model. To analyse whether the robusticity of the Neandertal pelvis and femur produced a significantly greater mechanical advantage than the human specimen (Ruff, 1995), the size of the Neandertal model and Modern Human model were scaled to the same size.

On the two models, muscle attachments were manually located on the bony features available from the bone surfaces. Origins and insertions of the hamstring muscles (i.e., semi-membranosus m., semi-tendinosus m., long and short head of biceps femoris m) were located on the 3D models and processed to estimate the muscles’ line of actions. Hamstrings knee moment arms were then processed from the distance between the hamstring muscle line of actions and the knee instantaneous helical axis (Fig. 4) (Brand and Hollister, 1992). Instantaneous helical axis, muscle line of action and muscle moment arms were directly processed within the LhpFusionBox interface.

This section presents the results related to the first and second aims of this paper, which is the validated reconstruction of a model of the Neandertal lower limb.

Information on the dispersal of ALs between bones was given by the root mean square (RMS) error of the various ALs processed by a singular value decomposition (SVD) algorithm (Challis, 1995). The root mean square error is the square root of the average of the squared differences between landmarks. The dispersion of ALs between the Kebara 2 right iliac bone (RK2) and mirrored Neandertal 1 iliac bone (RN1 ^mirror ) iliac bones led to a RMS error of 10.8 mm (Fig. 2).

Stage 2 in Table 2 (steps 7 to 14) further scaled the Kebara 2 iliac bone, scaled at this stage to Neandertal 1, to the Spy II morphology. The common reference point between Neandertal 1 and Spy II was the presence of an almost complete single femur in both skeletons. Scaling of the Kebara 2 iliac bone (scaled to Neandertal 1) to Spy II dimensions was performed using the femoral bones as intermediate scaling.

After virtual palpation and mirroring (steps 7 to 10), the transformation matrix A L N 1 → L S 2 m i r r o r $ {A}_{LN1\to LS{2}^{mirror}}$ allowed scaling from the Neandertal 1 femur (LN1) to the mirrored Spy II femur (LS2 ^mirror ) (step 11). Results of this transformation led to a RMS error of 6.0 mm (Fig. 3). The same procedure was applied for the right. All bones were mirrored in the study and therefore RMS error was equal for both the left and right side.

The final step scaled Kebara 2 iliac bone model to Spy II dimensions using the available transformation matrices. The results of scaling stage 2 were the Kebara 2 iliac bones scaled to the Spy II morphology.

At this stage, the Kebara 2 pelvis was scaled to the dimensions of the Spy II remains and a pair of lower limbs, including the pelvis, femora and tibiae was available.

The Neandertal model integrated with the motion data of a human enabled visualisation, creation of anatomical axes, motion representation and the processing of specific biomechanical parameters with graphic output, e.g., related to muscle moment arms (Fig. 4).

Comparison of the muscle moment arms obtained from the Neandertal model and modern human model shows that the hamstring muscles seem to have a higher mechanical advantage in the Neandertal model (Fig. 5, Table 3). For example, the biceps femoris–short head muscle moment arm was on average 24% larger for Neandertals. The same trends were also found for the other hamstring muscle: semitendinosus = +21%, long biceps = +12%, and semimembranosus = +5%. These results seem to demonstrate that the available Neandertal specimen shows hamstring mechanical advantages compared to a modern human of the same size.

The scaling of Neandertal fossil material from different specimens is the first step to creating a scaled, full lower limb skeleton to enable accurate motion analysis. Standard spatial registration registers bone models of the same kinds together (e.g., pelvis with pelvis, clavicle with clavicle, cranium with cranium) (Benazzi et al., 2009, Gunz et al., 2004, Gunz et al., 2009, Kalvin et al., 1995, Neubauer et al., 2004, Ponce de León and Zollikofer, 1999, Van Sint Jan et al., 2002 and Weaver and Hublin, 2009). The new development in LhpFusionBox presented in this article enables the scaling of a particular bone using two iterations of different bones via spatial registration. The accuracy of the various geometrical transformations in which the models are processed were tracked via RMS errors.

Spatial data registration from the Neandertal 1 femur to the Spy II femur produced an RMS error of 6.0 mm. Kepple et al. (1998) found a mean RMS error of 6.6 mm from the comparison and registration of femora from 52 male and female anatomically modern human specimens. The results of Kepple et al. (1998) cannot be directly comparable to the RMS error demonstrated in the reconstructed Neandertal model as bony landmark palpation differed, although they do give an indication of the scale of RMS errors within a single species. The comparison of this mean value of RMS errors from the registration of femora between humans in the small study from Kepple et al. (1998), with the value of the registration between femora in Neandertals indicates that the morphology between the Neandertal 1 femur and the Spy II femur demonstrated an RMS error within the limits of registration of femora between humans. The comparatively small RMS error of 6.0 mm found between the Neandertal femora therefore gave us the justification of using the Neandertal 1 skeletal remains as a scaling reference for registration to Spy II.

Kepple et al. (1998) stated that there is a wide range of pelvic differences within a single species, not only between males and females but also between females of different geographic origin. Pelvis RMS errors were therefore divided into separate models (male model (RMS errors of 8.6 mm), and different female geographic origins models (7.0 mm) and (7.3 mm)) as RMS errors were too high to report between all specimens (Kepple et al., 1998). Spatial data registration from the Kebara 2 pelvis to the Neandertal 1 pelvis gave an RMS error of 10.8 mm from one iliac bone. There are several factors, which may account for this slightly higher error. The Kebara 2 pelvis was found in Israel and the Neandertal 1 pelvis was found in Germany (Schmitz et al., 2002). There is a time difference of approximately 10,000–15,000 years between the two pelves, which could account for changes (Bar-Yosef et al., 1992 and Schmitz et al., 2002). The Kebara 2 pelvis has been widely documented as being a male (Bar-Yosef et al., 1992) although no information is available for the sex of the Neandertal 1 skeleton, which could potentially be male or female. While every effort is made to minimize errors in the data processing pipeline there may be errors due to differences in palpation (Van Sint Jan and Della Croce, 2005). Another source of error could be linked to the specimen preservation: for example, when bone areas are missing. However, the method presented here focuses on anatomical landmarks that are the same in all specimens, therefore reducing the probability of error due to missing areas.

The final aim of this study was to apply a particular motion pattern, collected from a modern day human, to the Neandertal skeleton to determine if the final results support the initial hypothesis that bone morphology and joint surfaces of Neandertals demonstrate similar motion patterns to anatomically modern humans. Standard registration techniques of bone morphology to motion data analysis often lead to unsatisfactory motion simulation because of discrepancies during the location of anatomical landmarks in the datasets. The adoption of a method that allows integration of six DOFs in the joint (Sholukha et al., 2006) enables a more realistic analysis of gliding of joint surfaces during motion and detection, e.g., unrealistic bone collisions or joint dislocations due to incompatible joint surfaces within a given motion pattern. Observation of the reconstructed simulation demonstrated that the joint surfaces of the Neandertal model show a good compatibility with general motion patterns performed by a modern human. For example, the Neandertal joint surface remains congruent during the full range of motion. This is obviously an assumption since real motion measurements cannot be collected on the extinct Neandertal individuals. However, the above joint congruence is an important step towards the obtained fusion results for further biomechanical muscle study as performed in this paper.

The robust skeletal morphology of the Neandertal has been interpreted as an adaptation for cold (Weaver, 2003), for frequently elevated levels of biomechanical stress (Trinkaus, 1983) or consistent with higher mechanical loads (Ruff et al., 1993, Trinkaus and Ruff, 1989 and Trinkaus et al., 1994). Stock (2006) in a recent study on hunter-gatherer groups found that robusticity of segments correlate negatively with climate and positively with patterns of terrestrial and marine mobility. This is also a conclusion drawn by Trinkaus (1983) who stated that the robustness demonstrated across all Neandertal specimens meant that is was most likely an adaptation to behaviour.

Hamstring moment arms found in this paper for modern humans are similar to previously reported results (e.g. Bonnefoy et al., 2007, who utilize several different musculoskeletal models) indicating the method is sound. In Neandertals, robustness seems to be proportional to mechanical advantage, at least for the hamstring muscles, and mechanical advantage was found to be much greater in Neandertals than in modern humans. This result is consistent with earlier findings that if we assume a modern human-like locomotion in Neandertals then they may have had a higher mechanical advantage, greater bending moments and larger moment arms as opposed to modern humans (e.g. Ruff, 1995, Ruff, 2009, Trinkaus, 1983 and Trinkaus and Ruff, 1989).

This preliminary study was limited to the analysis of the hamstring muscles. Further studies will concentrate on developing a full muscular skeletal analysis of the lower limb of the Neandertal. In order to be more representative we also aim to include more than one Neandertal specimen, as we did in this study. The method presented and software tools allow for further analytical studies to be performed to enable the study of the relationships between hominid morphology and mechanical efficiency of the musculoskeletal system. This software will enable further comparative analysis between all hominid species.