Functional approaches classically used in palaeontology may be limited for investigating the form-function relationships in complex osteomuscular systems. Experimental biomechanics that involve human experimenters have recently been used to investigate the upper-limb kinematics and grip strength requirements associated with stone tool production and use (Key and Dunmore, 2015, Rolian et al., 2011, Williams et al., 2010, Williams et al., 2012 and Williams et al., 2014). However, such approaches, although of interest, are limited in proposing indirect inferences on fossil morphology, whereas the functional impact of morphologies with no extant analogues can be directly and quantitatively estimated with musculoskeletal modelling simulations. This methodological tool is used in the clinical field to test “what-if” scenarios for tendon transfer surgeries (Holzbaur et al., 2005), as well as in ergonomics to study the influence of an object shape on the risk of muscle fatigue or repetitive strain injuries (Vigouroux et al., 2011), to name but a few examples.

In the paleontological field, musculoskeletal simulation has allowed proposing functional inferences in terms of locomotion and bipedalism, even if very partial skeletons are available (Hutchinson, 2004, Nagano et al., 2005, Nicolas et al., 2009 and Wang et al., 2004). By contrast, no investigation on hominin hand functions has been carried out to date. In this context, the main advantage of 3D musculoskeletal simulation consists in its capacity of testing hypotheses on organization and shape of unpreserved soft-tissues (mainly muscles and ligaments) from fossil taxa in order to quantify the muscle forces needed for exerting an external force and ensuring joint stability during specific actions.

However, even in well-studied extant taxa such as H. sapiens, the obtaining of a realistic musculoskeletal model remains a scientific challenge (Hicks et al., 2015). This is due to the high level of complexity associated with the multi-scale approach necessary to account for a tremendous number of parameters, such as: external interaction with the environment, joint kinematics in relation with bone geometry (Valero-Cuevas et al., 2003), ligament constitutive equations, muscle moment potential on each degree of freedom (DoF) (Lee et al., 2015), muscle contraction dynamics (Romero and Alonso, 2016), etc. As a result, numerous assumptions are generally made and the small mobility of intercarpal and carpometacarpal joints (excepted for the thumb CMC-I) is always neglected in current musculoskeletal models of the hand (Goislard de Monsabert et al., 2014, Lee et al., 2015, Rossi et al., 2015 and Sancho-Bru et al., 2003). This limitation needs to be overcome in order to address the role of the CMC-V joint in the production of a forceful finger and palm grips, such as cradle grip (Marzke and Shackley, 1986).

Modelling of extinct species also faces up to the absence of soft tissues quantitative data, such as their moment potential calculated as the product of the physiological cross-sectional area (PCSA, combination of muscle fibre length, pennation angle, etc.) and muscle moment arm. Therefore, before being able to draw trustful conclusions on the capabilities of A. afarensis hand, the sensitivity of simulation outcomes to model parameters (bone dimensions, muscle paths and PCSA), as well as input data (external force and joints posture), need to be evaluated.

This contribution has two aims: •

to develop, present and evaluate a simulation framework that includes a musculoskeletal model of A. afarensis hand. This model is established and evaluated in comparison to a H. sapiens and a P. troglodytes model;

Opensim 3.3 software (Delp et al., 2007) was used to develop hand models versions for the three taxa: H. sapiens, P. troglodytes and A. afarensis. Opensim is an open-source software platform especially suited for developing models of musculoskeletal structures and creating dynamic simulations of movement. All models were adapted from an original model available on the Opensim repository (https://simtk.org/projects/hand_muscle) based on Lee et al. (2015). The initial model gathers works from numerous studies from the anatomic, biomechanical and physiological fields including those from An et al. (1979). Several levels of modelling are considered. The reference to the kinematic model specifically refers to the mechanical interpretation of the DoF of a joint (e.g., a cardan joint with two intersecting and orthogonal axes of rotation). The constitutive parameters of the anatomical model (sometimes named “anthropometric model” by other authors) include joints parameters (number and range of motion of the DoFs, location and orientation of the rotation axes), bones geometry (length, breadth, height), and musculotendinous attachment sites. The addition of muscles parameters (in particular PCSA) leads to the constitution of the so-called musculoskeletal model. Only modified parameters, as well as details of the 5th ray deemed critical for the simulation output are reported here.

Bones specific to the 5th ray are: hamate (Ham), metacarpal 5 (MC-V), proximal phalanx 5 (PP-V), medial phalanx 5 (IP-V), distal phalanx 5 (DP-V). Bone geometry (length, breadth, and height), mass and inertia of H. sapiens are taken from the original model and correspond to an average size defined as the 50th percentile of each measurement. Bone geometry of P. troglodytes is derived from published dimensions (Drapeau and Ward, 2007 and McFadden and Bracht, 2005). Mass and inertia are assumed similar to H. sapiens, given that their influence is negligible with regard to the small values compared to the external forces and the static nature of the simulation (Isler, 2006).

The A. afarensis bone geometry is based on fossils that were discovered on the A.L. 333 locality of the Hadar formation, in Ethiopia. All specimens date to c. 3.2 Ma and were scattered on a restricted area of few square meters (Johanson et al., 1982). They share an exceptionally good state of preservation. For the present study, we selected bones based on Alba et al. (2003)’s hypothesis of ray bones re-association of the left A. afarensis hand. In order to document musculotendinous attachments sites of the 5th digit muscles, the Australopithecus anatomical model was implemented with bone geometries of the right hamate (A.L. 333-50) and the left 5th metacarpal V (A.L. 333w-89) from the same site. Bone geometry was derived from published dimensions (Alba et al., 2003, Bush, 1982, Marzke, 1983 and Stern and Susman, 1983), enhanced with author personal observations and measurements of the original fossils and 3D reconstructions (laser scans, NextEngine). Given that all bones cannot be attributed to the same individual, bone dimensions were scaled and symmetrised (for the right hamate) in order to obtain consistent bone dimensions for the 5th ray.

No change was made to the interphalangeal and metacarpophalangeal joints kinematic models and they have been kept identical to the original human hand model (Lee et al., 2015) for the three taxa. Interphalangeal joints (distal DIP-V - and proximal PIP-V) are modelled as hinge joints (one DoF for flexion-extension), and the metacarpophalangeal joint (MCP-V) as a cardan joint (two DoF: flexion-extension and adduction-abduction). Location and orientation of the axes were defined following An et al. (1979), with flexion-extension rotation axes oriented perpendicularly to the parasagittal plan of the bones. Similarly, ab-adduction rotation axis of the MCP-V is perpendicular to the frontal plan. The wrist joint is modelled as a cardan: one DoF in flexion-extension and a second DoF in radio-ulnar deviation.

Intercarpal joints were originally modelled as welded joints (0 DoF; in Lee et al., 2015). In this study, H. sapiens CMC-V kinematic model is transformed in order to better represent the natural mobility previously reported in anatomical studies (Batmanabane and Malathi, 1985, Dubosset, 1981 and El-Shennawy, 2001). Given the limited literature and absence of existing model, two versions of CMC-V are tested in order to add mobility (Fig. 1): 3-DoF, a ball-and-socket/gimbal joint (three independent axes intersecting and orthogonal to each other), and 2-DoF, a saddle joint whose axes are not orthogonal and not intersecting, as it has been suggested for the CMC-I (trapeziometacarpal) joint (Hollister et al., 1992 and Li and Tang, 2007).

3-DoF joint centre is located at the most proximal point of MC-V and rotation axes are perpendicular to the anatomical planes of MC-V. 2-DoF associates an ab-adduction rotation axis to another tilted rotation axis that couples flexion-extension and pronation-supination. More precisely, and based on the qualitative description of El-Shennawy (2001), the ab-adduction rotation axis passes through the distal portion of the hamate and is oriented perpendicular to the frontal plan of the palm (approximated as the frontal plan of the third metacarpal). The second rotation axis passes through the proximal end of MC-V, so that the two axes are not intersecting to each other (resulting in an inter-axis distance of 2 mm, and then the absence of unique joint centre). The orientation of the second rotation axis was set to 5° and -12° with respect to the X and Z axes of the palm of the hand, respectively. The two models (2-DoF and 3-DoF) add potential mobility to the CMC-V joint around its rotation axes and allow the 5th ray to better orient towards the thumb for a pulp-to-pulp pinch (opposition movement), resulting in a more realistic behaviour.

There is no quantitative data published on the mobility of CMC-V for P. troglodytes. Based on personal observations and general acceptation in the literature of a restricted mobility (Marzke, 1983 and Tuttle, 1969), CMC-V range of motion was considered null in terms of flexion-extension, pronation-supination and abduction-adduction. Similarly, based on the suggestions of Marzke, 1983 and Marzke, 2013, A. afarensis CMC-V range of motion was assumed to be identical to P. troglodytes.

Twelve muscles whose parameters are primarily based on the original model of Lee et al. (2015) are included in the model: flexor carpi ulnaris (FCU), extensor carpi ulnaris (ECU), flexor digitalis superficialis (FDSL), flexor digitalis profondus (FDPL), flexor digiti minimi brevis (FDMB), extensor digiti minimi (EDM), extensor digitorum communis (EDCL), opponens digiti minimi (ODM), abductor digiti minimi (ADM), 3rd palmares interossei (IDP), 4th dorsales interossei (ID4), 4th lumbricales (L4).

For the H. sapiens model, numerical data of muscles attachments sites are based on cadaveric records, experimental studies and imaging techniques (An et al., 1979 and An et al., 1983). Intermediate via points are included to represent bony contours, as well as muscles volume and large insertion sites (Fig. 2). Muscles insertions and paths remained unchanged from the original H. sapiens model (Lee et al., 2015).

P. troglodytes musculoskeletal geometry was adapted from the H. sapiens model based on qualitative observations (Diogo, 2013; Diogo et al., 2012). For example, differences include the flexor digiti minimi brevis origin on the radial side of hamulus (vs. on its convex surface) and the extensor carpi ulnaris insertion at the base of PP-V (vs. on MC-V).

Given the absence of soft-tissue data, the A. afarensis model was tested with respect to both the H. sapiens and P. troglodytes musculoskeletal geometry. In the simulations, the only distinctive features accounted for were the dimensions of the bones. Muscle attachment sites are scaled to bone dimensions and plausible associations can be made with specific muscle attachment sites. For example, the long and robust hamulus of A. afarensis (Bush, 1982) was interpreted as an origin of the opponens digiti minimi located slightly more palmarly compared to H. sapiens.

Muscle properties (muscle mass, fibre length, pennation angle, etc.) and resultant force-length-velocity relationships implemented in the Opensim (Hill-type) muscle model originate from several imaging and experimental studies. Given the limited literature, and even if the influence of muscle properties on model output has been deemed modest (Bolsterlee et al., 2015; Valero Cuevas et al., 2008), three databases of PCSA were investigated (Table 1): PCSA-Homo1, the database used in the original model (Lee et al., 2015) that gathers data from different collections (Jacobson et al., 1992 and Lieber et al., 1992); PCSA-Homo2, from data published in Chao (1989); and PCSA-Pan, based on P. troglodytes data from Ogihara et al. (2005).

Muscles forces necessary to exert a 1-N external force are calculated for the three CMC-V kinematic models (0-DoF vs. 2-DoF vs. 3-DoF). Overall, mean muscle forces are 1.8 ± 2.4 N, 2.8 ± 3.8 N, and 4.4 ± 5.3 N for 0-DoF, 2-DoF, and 3-DoF, respectively. There is a significant effect of the kinematic model on muscle forces (F[2,11] = 5.27, P = 0.013). More precisely, the 3-DoF leads to significantly higher muscle forces than 0-DoF, while no significant difference was found between 2-DoF and 0-DoF. As an example, Fig. 3 compares muscle forces (as a percentage of their maximal isometric force) for the three kinematic models in two external force directions (palmar-radial direction not represented for clarity). Again, it illustrates that 3-DoF (green bars in Fig. 3) leads to higher muscle forces. It also highlights specific discrepancies, such as for the 4th dorsales interossei muscle (ID4) force, which is especially high (∼32%) for the 3-DoF + palmar force direction condition.

ID4 is the primary limiting factor in the estimation of maximal external force. The maximal external force the model is able to exert (Table 2) is smaller under the 3-DoF model. The difference is especially high when the force has to be oriented palmarly; in such a case, the external force is 4.2 N, which is 50% less than calculated under the two other conditions.

Before running the simulations and calculating muscle forces, muscles PCSA of the three databases (Table 1) were compared. On average, PCSA-Pan are 6% higher than PCSA-Homo2 and 51% than PCSA-Homo1, with notable differences for some muscles such as the FDSL (2.1, 0.4 and 1.54 cm² for PCSA-Homo1, PCSA-Homo2 and PCSA-Pan, respectively) and the ID4 (1.1, 0.91 and 2.3 cm² for PCSA-Homo1, PCSA-Homo2 and PCSA-Pan, respectively). The root mean square deviation (RMSD) between PCSA-Homo1 and PCSA-Homo2 database is higher (2.07) than between PCSA-Homo1 and PCSA-Pan (0.99) and between PCSA-Homo2 and PCSA-Pan (1.98). As a matter of course, these discrepancies lead to notable differences in the estimated muscle forces.

Fig. 4 illustrates muscle forces (as a percentage of their maximal isometric force) to exert a 1-N external force for 2 PCSA databases: PCSA-Homo1 (left) and PCSA-Homo2 (right) and 2 external force directions (palmar and radial). A large difference can be observed for the FDSL muscle force (there is also a large difference in the PCSA of FDSL between the two models: 2.1 cm² for PCSA-Homo1 vs. 0.4 cm² for PCSA-Homo2). In general terms, and beyond specific differences, results gathered from the numerous simulations exhibited a similar muscle forces distribution pattern whatsoever the PCSA database used. The influence was mainly on the amplitude of the maximal external force the model is able to exert: PCSA-Homo1 consistently leads to a smaller maximal force.

The %RMSD between muscle forces estimated from simulations based on the anatomical model of H. sapiens and simulations based on an anatomical model of P. troglodytes is 13% (average across force direction; CMC-V posture and PCSA database conditions). Overall, simulations run by using an anatomical model based on P. troglodytes demonstrated a higher capacity of external force, but muscle force patterns were rather similar (the greatest differences were observed for ECU and ADM).

As illustrated in Fig. 5, there is a notable influence of both CMC-V joint posture and external force direction on muscle forces to exert a 1-N external force. Overall, muscle forces are decreased when CMC-V can flex and supinate (green and yellow bars on the graph), compared to the condition where the joint is in a 0° of flexion, supination and abduction (blue and red bars). This result is exacerbated when the external force is oriented radially (blue vs. red bars and green vs. yellow bars). For example, mean muscle forces (average across all simulations tested) to exert a 1-N force are: 4.1 ± 5.5 N, for CMC-5 0° flexion and palmar force; 2.7 ± 3.8 N, for CMC-5 10° flexion and palmar force; 3.7 ± 3.5 N, for CMC-5 0° flexion and radial force; and 2.5 ± 2.6 N, for CMC-5 10° flexion and radial force. Furthermore, the maximal external force the model is capable to exert tends to be 50% higher (15.5 ± 3.9 N) in conditions associated with a palmar-radial force orientation, than with a radial force orientation (7.7 ± 2.5 N), or a strict palmar force direction (9.8 ± 2.1 N).

Several limitations other than those already exposed must be acknowledged before going into the interpretation of the results. The external force was assumed here as a single force vector located on DP-V (the distal phalanx), while a more accurate simulation would include multiple interactions that arise at the interface between the digit and a large-sized object (Goislard de Monsabert et al., 2012 and Rossi et al., 2015). However, such data are still unavailable for the 5th digit in the frame of Lomekwian tool making replication.

Also, no passive structure was included in the original model. The absence of passive structure contribution in the stability of unwelded DoF is frequent in musculoskeletal modelling, and assumed to be reasonably realistic when joints are positioned at mid-range of joint excursion, which seems reasonable here in the context of a forceful hand grip posture of a tool. Even if difficult to identify, an effort should be made to model passive constraints, especially at the CMC-V joint, whether using an analytical (Majors and Wayne, 2011) or a more global approach (Domalain et al., 2010 and Gabra and Li, 2016). Lastly, the constitution of the musculoskeletal model was based on the last published data, such as those from Lee et al. (2015), that beneficiate from both anatomical records of tendon locations and experimentally calculated moment arms. However, all the complexity of the hand muscular system cannot be perfectly reproduced and the model includes several simplifications whose consequences need to be assessed. For example, the intrinsic muscle connection to the dorsal aponeurosis and FDP tendon could not be modelled. Instead, muscle attachment points were set so that moment arms matched experimentally measured moment arms (Lee et al., 2015). The simplification of the extensor mechanism has also been found to have little influence on flexor forces estimations when the location of external force application is distal to the PIP joint (Li et al., 2001).

The choice of a kinematic model of CMC-V joint was a major modelling issue to obtain simulations that include a more realistic H. sapiens 5th ray. The few hand musculoskeletal models that exist consider CMC-V as a fixed/welded joint, so that joint stability is brought by passive structures only. The estimation of muscle forces is thus facilitated, but inexact. Besides, this assumption prevents the 5th ray to adequately orient towards the thumb.

Our results demonstrate that adding mobility using a 3-DoF (gimbal) kinematic model may lead to a 50% decreasing in maximal external/grip force in H. sapiens (Table 2). On the contrary, the specific 2-DoF (saddle joint) model we developed does not decrease maximal force capacity and is thus likely to be more realistic, both in terms of mobility and stability. The 2-DoF kinematic model was therefore adopted for the rest of the study for the H. sapiens CMC-V joint. This is in line with previous studies on CMC-I (trapeziometacarpal joint) that similarly recommended against assuming a full 3-DoF model (Domalain et al., 2011, Hollister et al., 1992 and Valero-Cuevas et al., 2003).

P. troglodytes and A. afarensis CMC-V were modelled as a 0-DoF joint. While it may seem arbitrary and a little mobility may exist, the results show a limited difference of external force potential between the 0-DoF and the 2-DoF models. This means that it has little importance whether the A. afarensis CMC-V joint is assumed to be equilibrated by passive structures only (0-DoF), or by a combination of passive structures and muscle forces (2-DoF). Ultimately, it is the ability to adequately orient the 5th ray towards the hand-held object that influences the most the potential of maximal external force.

Facing the absence of soft tissue data for extinct species, databases from an extant analogue has to be used (Hutchinson, 2004 and Nicolas et al., 2009). In order to ensure that the influence of this parameter on simulation outcomes was limited, we tested three PCSA databases related to H. sapiens (PCSA-Homo1 and PCSA-Homo2) and P. troglodytes (PCSA-Pan). Surprisingly, intrataxic variations (between the two H. sapiens PCSA databases) are larger than intertaxic variations (between the two H. sapiens PCSAs and PCSA-Pan). Using PCSA-Pan leads to the highest potential of external grip force but, overall, muscle force patterns remain similar for the three databases.

The difference between the two H. sapiens PCSAs highlights the difficulty to accurately estimate this parameter and the need for a unified database, rather than data gathered from different works and specimens. Nevertheless, the RMSD between PCSA conditions remains smaller than between CMC-V kinematic model condition and between anatomical model conditions. Altogether, these results would suggest a rather limited sensitivity of simulation outputs to muscle parameters (fibre length, pennation angles, etc.) and tend to argue in favour of using simulation approaches to study extinct species even with the unavoidable absence of soft-tissue data.

Two versions of the A. afarensis anatomical model based either on the H. sapiens and the P. troglodytes anatomical models were compared. For both, a few specific characteristics resulting from the scaling to fossil bone dimensions were introduced, or were postulated from their direct observation (e.g., origin of opponens digiti minimi based on specific shape of A. afarensis hamulus).

Overall, simulations that were run with an anatomical model based on P. troglodytes demonstrated a higher external force capacity than those adopting a H. sapiens model. The results also showed that the anatomical model has a greater influence than PCSA on outcomes data. The cross-analysis of all the simulations further suggests that the anatomical parameter leading to the greatest difference in outcomes data is muscle attachment sites. Muscle attachment sites influence muscle moment arm and thus its moment potential (capacity to exert a “rotation force” at a given DoF). Attachment sites are expressed in a bone reference system (see Fig. 2), thus, the consistency between muscle attachment sites and the kinematic model of the joints (rotation axes location/orientation) is as critical as the precision of the joint model itself.

The use of an accurate and consistent database, even if obtained on a single individual (see Mirakhorlo et al., 2016), would likely contribute to more accurate simulations for H. sapiens. For extinct taxa, such as A. afarensis, a more accurate morphofunctional analysis of the bones (Chaudhari et al., 2014, Daver et al., 2012 and Orr et al., 2010) associated with a validation procedure on extant taxa, should also help refining putative muscle attachment sites.

The putative low mobility of A. afarensis CMC-V joint leads to low flexion, abduction and supination angles that restrict the ability to optimally orient the pulp of the 5th ray towards the hand-held object. Indeed, in the frame of a forceful and cup hold grip, there exists a logical relationship between joint posture and external force orientation, such as the less abducted and pronated the 5th ray, the more radial the direction of the grip force is.

Simulation results (Table 2) show that grip force potential is maximal when oriented in a palmar-radial direction, and decreases by ∼50% when directed in a true radial direction. This highlights the importance of being able to adequately orient the pulp of the 5th ray so that a palmar-radial force is applied towards the object. Moreover, whatsoever the direction of the external grip force, results tend to show that, when CMC-V can be flexed (10° in our examples), muscle forces are lower for a given amount of external force (Fig. 5).

Those two phenomena are linked to the complex interaction of muscle moment arms that vary as a function of joint angles and direction of the external force that influences joint moments. Altogether, this suggests that A. afarensis inability to ideally orient the 5th ray could have constituted a limiting factor for exerting a grip force strong enough for large-sized stone tool making, which is in line with previous speculations (Marzke, 1983 and Marzke, 2013).