4.1 Methodological novelty and considerations

An important aim of this paper is to introduce a novel statistical approach for analyzing covariation between serially corresponding structures. More specifically, this novel approach provides a means for assessing covariation between the elements of two series. This is achieved by applying the same landmarks within each series and performing a single Procrustes fit onto all elements of the same series, so that they are all in the same morphospace. Subsequent 2B-PLS analyses produce axes within each morphospace, which can be compared with each other. A rather simple way of doing this is to calculate the angles between the axes within each morphospace and use the angles as proxies for distance in PCoA.

The novelty of this approach lies in the fact that the various 2B-PLS analyses are directly compared, not only for magnitude of covariation, but also direction in shape-space. An important prerequisite for applying this approach is that the objects represented in block 1 of all 2B-PLS analyses are subjected to the same GPA, so that they lie in the same shape-space, and idem for all objects represented in block 2 of all 2B-PLS analyses. This approach is easily implemented using freely available software (MorphoJ and PAST). The results are easy to interpret taking advantage of the graphical output.

In the present study, because fingers and toes are very similar in morphology, the landmarks were kept homologous across the two datasets. This is, however, not a requirement for this method. This procedure would be equally applicable to two sets of serially corresponding structures that are expected to covary, but do not share a similar morphology, such as thoracic ribs with associated vertebrae. In that hypothetical case, the individual thoracic ribs would all be landmarked with the same homologous landmarks and occupy a single morphospace, whereas the thoracic vertebrae would occupy another morphospace. It would be prudent to limit this type of analysis to within species until the consequences of stacking both elements and species in the matrix have been assessed.

In this study, the GPA was performed based on relatively distinct forms (lateral toes and fingers vs. pollex and hallux). If the difference in specific landmarks is excessive between specimens, this could possibly lead to a Pinocchio effect (e.g., Klingenberg, 2021), which affects the position of other landmarks in the configuration. Furthermore, the distinct forms (ray 1) could hinder differences amongst forms that are more similar in the dataset (rays 2–5). The differences between highly similar phalanges, however, are not ultimately of interest in this study and it is not important to our study whether some differences are overshadowed by the distinctiveness of the first ray. The aim of this study is to assess the covariation between the hand and the foot by comparing the relative position of the hand morphospace and the foot morphospace using the angles between PLS axes. This is only possible with all the hand bones in one morphospace and all the foot bones in another. Moreover, Klingenberg (2021) states that there is no basis for the preference of methods that adjust for the Pinocchio effect. Furthermore, several analyses were repeated without the thumb and the big toes to show that those two bones do not have a large effect on the combined morphospace.

When working with skeletal material, the proximal phalanges of the hand are relatively easily identified (Navsa, 2010; Susman, 1979). Yet, in the present study, only the proximal thumb phalanx shows a clearly differentiated morphology from the other phalanges (Figure 3a). Phalanges of the fifth and second rays also differentiate somewhat, but there is appreciable overlap amongst all manual lateral phalanges. In the present study, the landmarks primarily describe the shape of the articular surfaces and to a limited extent the position and size of the tubercles. Additionally, size is removed by GPA. When identifying proximal phalanges in an osteological collection, however, the main aspects used are the shape of the tubercles and the overall phalanx size, causing the discrepancy.

Lateral pedal phalanges are much more difficult to attribute to the correct ray, than manual phalanges, particularly when the foot is incomplete. The general pattern found in manual phalanges (Figure 3a) is, however, also visible in the pedal proximal phalanges (Figure 3b). Although, here, there is also slight overlap between the morphology of the hallux and the fifth phalanx.

This indicates that a landmark based approach is not sufficiently detailed to differentiate between the lateral rays of both the hand and the foot. Sliding semilandmarks may be more suitable if one is interested in quantitatively discriminating between rays, for example, for the objective determination of isolated archeological finds.

4.2 Allometry

All organisms and their parts are influenced by allometry, because, as the size of a biological system changes, the relationship amongst its different components must be adjusted too to maintain proper function (Brown & West, 2000). Our results show that this also holds true for phalanges in H. sapiens. Static allometry may also include sexual dimorphism, whereas evolutionary allometry accounts for differences between populations, both of which could influence phalanx morphology. Although the fifth proximal manual phalanx is relatively small, all lateral phalanges share the same allometric scaling pattern: with increasing centroid size, the lateral phalanges are less robust (Figure 4a). This indicates that increases in size are mainly accomplished by increases in phalanx length rather than width or breadth. The thumb also displays decreasing robusticity with increasing size, but overall has a distinct shape from the lateral phalanges, indicated by a difference in regression scores, which is also confirmed by the PCA in Figure 3a.

The allometric scaling pattern shown by the lateral pedal phalanges is quite like that of the manual lateral phalanges and the thumb (Figure 4d), possibly, though speculatively, indicating shared developmental processes, or shared physics constraints related to scaling. Additionally, the cross-sections of the bone, also a product of biomechanical constraints, might influence the shape of the articular surfaces. Although the hallux also displays decreasing robusticity with increasing size (Figure 4c), its size and shape are such that the overall pedal allometric signal consists of increasing robusticity with increasing size (Figure 4b). Although it may be expected that the hallux and the thumb should display a similar allometric pattern based on comparisons with the manual and lateral pedal phalanges, which do show similar patterns, it appears that the size and shape of the hallucal first phalanx deviate from this expectation. A conceivable, but hypothetical, explanation might be that some aspect of the developmental process shared by the other proximal phalanges is not shared by the big toe or, at least, not to the same degree.

From a developmental perspective, the big toe and the thumb have one phalanx less than the other toes and fingers. There are four hypotheses of why this might be the case: the first metacarpal could be the proximal phalanx (Galen 170 in Testut & Latarjet, 1966; Valenzuela, & Y, Berríos-Loyola R, and Canals M., 2009), the first metacarpal could be the result of a fusion between the true first metacarpal and the proximal phalanx (Sappey, 1871), the middle phalanx could be missing (Paturet, 1951), or the distal phalanx could be the result of a fusion between the second and third phalanges (Pfitzner, 1900). In the latter two hypotheses the proximal phalanx of the thumb and the big toe is the true proximal phalanx and, as such, corresponding to the first phalanges of the other fingers and toes. In the first two hypotheses, the thumb is a paramorph to the other fingers (Wagner, 2014). Insights from morphometric studies support the hypothesis that the lost segment of the first ray is the true metacarpal, and the metacarpal is the true proximal phalanx (Bondioni et al., 2020; Pazzaglia et al., 2018). Nevertheless, from an evolutionary perspective, the data suggest that the first metapodial is not a true phalanx (Reno et al., 2013). Furthermore, the condition symbrachydactyly, which reduces the length of the middle phalanx, usually affects only the lateral rays, suggesting that the thumb middle phalanx is already absent (Guillem et al., 1999), which supports the hypothesis by Paturet (1951). Additionally, symphalangy, the fusion of the middle and distal phalanges, is common in the foot and present in the fifth ray in 25–33% of individuals (Freyschmidt et al., 2003) and could have occurred in the first ray as well (Guillem et al., 1999), supporting the hypothesis of Pfitzner (1900). Even though it is not clear now, which element is missing in the thumb and the big toe, this must have an influence on development. Herein, it is assumed that the proximal phalanges of the first ray are truly corresponding to those of the second to fifth rays.

4.3 Covariation and biomechanical notes

The next step is to assess and interpret covariation. Within the hand, covariation is relatively strong (Table 3), particularly between the middle three rays, whereas thumb proximal phalanx shape covaries least with the other proximal phalanges. This makes sense in terms of hand function and is expected from morphology.

The thumb has a vastly different function from the other fingers and is the only ray that is truly opposable to the others. The thenar muscle group controls the movements of the thumb (Gilroy et al., 2012; Gray, 1918; Van Sint Jan & Rooze, 1992). The musculus adductor pollicis inserts on the lateral tubercle of the pollical base. The musculus flexor pollicis brevis and the musculus abductor pollicis brevis insert on the medial tubercle. Therefore, these tubercles comprise muscle attachment sites (entheses), of which relative robusticity is experimentally shown to be affected by habitual manual activity (Karakostis et al. 2017; Karakostis & Harvati, 2021). Here we show that the robusticity of these tubercles is directly related to robusticity and proportions of proximal hand phalanges themselves. This has previously also been noted both within the hand as well as when comparing between sexes and distinct populations (Karakostis, Lorenzo, & Konstantinos, 2017). Although the morphology of muscle attachment sites on the metacarpals has been doubted to reflect muscle architecture (Williams-Hatala et al., 2016), muscle size depends on many factors besides habitual activity (muscle use), such as old age and body size. This study compares raw 3D size of entheseal surfaces with attaching muscle size, reporting no significant correlation in a sample of elderly individuals, but this could be due to old age. In contrast, Karakostis, Hotz, et al. (2017) found a significant effect of habitual manual activity on entheses, relying on the Validated Entheses-based Reconstruction of Activity method, which has been validated based both on different experimental animal studies (Karakostis, Jeffery, & Harvati, 2019; Karakostis, Wallace, et al., 2019) as well as human skeletons with lifelong occupational documentation (Karakostis, Hotz, et al., 2017). Nevertheless, a strong correlation has been shown in the combined analysis of entheses from metacarpals and proximal phalanges (Karakostis, Hotz, et al., 2017). In these studies, the entheses of the flexor pollicis brevis and abductor pollicis brevis were proven to be of special value, as they contributed to differentiating across individuals of distinct lifelong occupational tendencies, suggesting that the robusticity of these tubercles may be additionally affected by the biomechanical stress resulting from physical activity itself. A larger enthesis for the musculus flexor pollicis brevis and the musculus abductor pollicis brevis is related to thenar movement and thumb – index finger interaction and, as such, less strenuous and/or mechanized occupations (Karakostis, Hotz, et al., 2017). Consequently, the activity pattern of the individual could have influenced the size of the tubercles on the fingers. Changes in the position of the tubercles likely equals a change in the muscles' joint moment arms, which directly affect muscle force production and direction of forces (Karakostis, Haeufle, et al., 2021; Karakostis, Reyes-Centeno, et al., 2021; Maki & Trinkaus, 2011).The landmarks used herein do not provide sufficient coverage of the tubercles to determine whether changes in landmarks 15 and 16 represent relative size changes of parts of the tubercles or a true change in location of the tubercles, but it is likely that the observed variation in the medial tubercle of the thumb is associated with functional or strength changes of the musculus flexor pollicis brevis and/or the musculus abductor pollicis brevis, which are responsible for flexion and abduction of the thumb respectively. Some of the observed differences in covariation between hands and feet may be related to differential muscle use affecting the form of the tubercles and overall bone shape. The abductor hallucis inserts on the base of proximal phalanx of great toe. The function of this muscle is to abduct and flex the great toe, which contributes to the stability of the foot during walking. The adductor hallucis inserts on the lateral aspect of base of proximal phalanx of great toe. It is responsible for toe adduction, toe flexion and support of the longitudinal and transverse arches of foot. The flexor hallucis brevis inserts on the lateral and medial aspects of base of proximal phalanx of great toe. Its function is flexion of the toe and support of the longitudinal arch of foot. These muscles of the feet are essential for standard locomotion and, thus, are likely recruited far more systematically than the specialized muscles of the thumb, which are likely connected to specific grips. This difference in muscle use between the hallux and the pollex could possibly, in part, be responsible for the differences found in this study.

The little finger is controlled by the hypothenar muscles (Gilroy et al., 2012; Gray, 1918). Although the little finger cannot be opposed to the other fingers in the same way the thumb can, the presence of a saddle shaped articulation at the base of the fifth metacarpal and the combined movement of flexion, abduction and rotation, mostly due to the musculus opponens digiti minimi, at the fifth carpometacarpal joint has been regarded as a form of opposability of the little finger (Napier, 1955). These similarities with the thumb, due to both fingers being on the outside of the hand, are reflected in increased asymmetry and explain its increased morphological proximity to the thumb relative to the other lateral fingers in Figure 3a. Covariation between the little finger and the thumb is not increased relative to the covariation between the thumb and the other fingers. Although this might seem surprising at first sight, similar functional capabilities do not influence the genetic aspect of the developmental process, only the resultant morphology through loading. And as the thumb might be a paramorph to the other fingers (Wagner, 2014), similar morphologies might conceivably be achieved via different developmental pathways.

The middle three fingers are mostly governed by the lumbricals and the interossei (Gilroy et al., 2012; Gray, 1918; Parminder, 2013) and are functionally quite similar to one another. The high degree of covariation between them (Table 3) may have an (epi)genetic cause, may be a consequence of developmental plasticity, and may be due to allometric scaling. Further research into fluctuating asymmetry might be able to provide clues to the most likely explanation. As fluctuating asymmetry reflects in part perturbations to the developmental process, its study allows distinguishing between genetic and environmental processes affecting development (De Coster et al., 2013; Dongen, 2006; Klingenberg, 2003; Klingenberg, 2015; Klingenberg et al., 1998; Klingenberg & McIntyre, 1998) but confirmation bias needs to be avoided (Kozlov & Zvereva, 2015).

Sometimes, the human foot is functionally divided into a medial and a lateral column (Aiello & Dean, 1990; Kidd, 1998; Kidd, 1999; Morton, 1935; Zipfel et al., 2009). The medial column consists of rays 1, 2, and 3 and the lateral column consists of rays 4 and 5. It would, then, logically follow that the fourth and fifth proximal pedal phalanges form one functional module and the medial three phalanges form another. Covariation within these modules would be expected to be stronger than between these modules. The result presented here (Table 3) do not support this hypothesis. Rather, there seems to be a divide between the first and the lateral phalanges, which is also supported by the PCoA results (Figure 6). This is suggestive of a differential selection pressure on the medial tubercle of the hallux. The medial tubercle is the insertion area of the tendons of the medial head of the musculus flexor hallucis brevis and the musculus abductor hallucis (Kelikian, 2011). As in the thumb, it is not possible to determine whether the observed variability represents relative size differences of parts of the tubercles or a true difference in location of the tubercles. It is likely, however, that the observed variation in the medial tubercle of the hallux is associated with functional or strength changes of the musculus abductor hallucis, rather than the musculus flexor hallucis brevis, as in the latter case a similar morphological pattern for the lateral tubercle would have been expected. The musculus abductor hallucis plays an important role in gait by supporting the foot's medial arch during the toe-off phase (Wong, 2007).

Additionally, in the big toe, the dorsal margin of the articular surface retains its form, regardless of general robusticity, whereas in the lateral toes the form of this margin is much more dependent on the robusticity of the phalanx. As ground reaction forces are highest on the hallux (Trinkaus, 2005), its morphology, particularly regarding musculus abductor hallucis insertion, may be more restricted than the lateral rays to retain proper functionality of the foot, consequently limiting covariation between robusticity, tubercle size and position, and articular surface shape. The dorsal extension of the proximal articular surface of the hallux may also be limited because it experiences more dorsiflexion than the lateral rays.

4.3.1 Speculation on evolutionary implications

Figure 6 shows that the covariation between hands and feet is quite different for the first ray than for the other rays with regards to direction in morphospace. Rolian et al. (2010) used an evolutionary quantitative genetics framework to demonstrate that selection pressure was most likely on the foot, with the hand coevolving. The results presented herein (Table 1) corroborate their finding and have the appearance that an overall decrease in integration of the pedal proximal phalanges might have occurred as a response to the selection pressure on the hallux to become larger and more robust relative to the other toes. It has indeed been shown that modularity is related to increased evolutionary rates (Claverie & Patek, 2013) and that directional selection can drive the evolution of modularity (Melo & Marroig, 2015). Conversely, integration may act as a constraint on evolutionary change (Goswami & Polly, 2010; Klingenberg et al., 2011). The RV coefficients presented herein are higher in the hand than in the foot for every pairwise comparison between the rays with only one exception. The difference in RV coefficients between the hand and the foot, as presented herein, looks to be suggestive of the hand remaining more integrated than the foot because of less differential selection pressure on the thumb relative to the other fingers, than on the big toe relative to the other toes.

The patterns of covariation described herein suggest that the big toe and the thumb are less integrated with the lateral phalanges, than the lateral phalanges amongst each other. The direction of covariation between the big toe and the thumb in morphospace is also different from the other rays. Additionally, the foot appears to be less integrated than the hand. Combined, these aspects may indicate the following potential scenario. Differential selection pressure on the hallux might have driven the evolution of increased modularity in the foot between the big toe and the lateral toes, simultaneously causing a decrease of integration of overall foot covariation to release the constraint on evolution. This would allow the dramatic changes in foot morphology seen in the fossil record, such as adduction of the first pedal ray (e.g., Darwin, 1871; Latimer & Lovejoy, 1990), which is an autapomorphy of Homo. Australopithecines, however, have also been suggested to present an intermediately adducted hallux (e.g., Bennett et al., 2016). The increased modularity in the foot, as shown here, could have allowed for evolutionary changes in the hallux independent of the lateral rays and, consequently, a timely adaptation to bipedal locomotion.

Since both extremities likely share a common evolutionary origin, they are influenced by the same developmental pathways (Petit et al., 2017). In fact, other workers have also demonstrated strong selection on the lateral digits of the foot driving decreases in the length of the fingers (Rolian et al., 2010). Although the correlations described herein do not necessarily imply a causal relationship, it could speculatively be hypothesized that the evolution of the hand was influenced by the evolution of the foot. Selection pressures on the foot associated with the rise of bipedalism, such as hallucal adduction, could have resulted in changes in the hand, such as the position of the pollux, that were exaptations for increased precision grip. The results presented here (Figure 5a) suggest that an increase in robusticity of the big toe is associated with a larger articular surface and a more medial position of the medial tubercle. The same changes are visible in the thumb as well. The musculus flexor pollicis brevis and the musculus abductor pollicis brevis, which are responsible for flexion and abduction of the thumb respectively, insert on the medial tubercle. Functional opposition of the thumb requires recruitment of the abductor pollicis brevis and the superficial head of the flexor pollicis brevis. The major force of opposition comes from the abductor pollicis brevis, with the flexor pollicis brevis providing a secondary motor for opposition maneuvering (Duncan et al., 2013). A more medial position of the medial tubercle would, therefore, give both these muscles a larger moment arm, increasing efficiency. The evolution of the precision grip may not have solely been the result of the selection pressures on the foot and might also have been influenced by selection pressures on the hand directly.

Thumb morphology and its precision grasping efficiency never stopped undergoing changes. Even though the thumb's relative length was already increased in Australopithecines (Kivell, 2015; Kivell et al., 2018), Ar. ramidus likely shared primitive suspensory adaptations with modern chimpanzees and bonobos (Prang et al., 2021). In hominoids, the first metacarpal is ancestrally gracile and the first carpometacarpal joint is curved, but Paranthropus robustus or early Homo is the first to display the derived condition of a robust first metacarpal, which produces more force and tolerates higher joint stresses during tool use (Rolian et al., 2011; Susman, 1994), and less curved carpometacarpal joint, which allows for a forceful precision grip (Susman, 1988). As such, this trait may perhaps be only exclusive to Homo. Orrorin tugenensis is the first hominin to have distal phalanges with broad apical tufts, which enhances tip weight support (Almécija et al., 2010; Rolian et al., 2011). Increased mobility of the second metacarpal necessitated changes of the proximal articular facets, which are first observed in Australopithecus afarensis (Marzke, 1983; Niewoehner 1997; Tocheri et al., 2003). The thumb's overall morphology and biomechanical efficiency appear to increase substantially from ca 2 Ma (Karakostis, Haeufle, et al., 2021), likely in association with the emergence of the genus Homo. Nevertheless, thumb evolution continued within Homo, leading to high morphological variability amongst recent hominins. Examples include the mosaic morphologies of Homo naledi or Homo floresiensis. Additionally, there are noteworthy differences between the Neanderthal and the early modern human hand, including the robust proportions of the thumb's proximal phalanx, its proportional size relative to the distal phalanx, (Niewoehner, 2006; Niewoehner et al., 2003), and the smaller extent of the MC1-trapezium joint (Bardo et al., 2020; Marzke et al., 2010). Overall, several morphological aspects with direct biomechanical implications for precision grasping kept undergoing considerable selection pressure throughout human evolution. This unique combination of selection pressures and patterns of covariation between and within the hand and the foot may have allowed for the evolution of the precision grip, which we use every day.

4.4 Conclusions and future horizons

The aims and objectives of the present study were to assess morphological variation, introduce a novel statistical approach and assess covariation between the hand and foot.

With increasing centroid size, the lateral manual phalanges become less robust, indicating that increases in size are mainly accomplished by increases in phalanx length. The thumb also displays decreasing robusticity with increasing size, but overall has a distinct shape from the lateral phalanges. The allometric scaling pattern shown by the lateral pedal phalanges is quite like that of the manual lateral phalanges and the thumb, which might, speculatively, indicate shared developmental processes.

A novel landmark-based geometric morphometric approach focusing on covariation is based on a PCoA of the angles between PLS axes in morphospace. This novel approach provides a mean for assessing covariation between the elements of two series, which is achieved by applying the same landmarks within each series and performing a single Procrustes fit onto all elements of the same series. Subsequent 2B-PLS analyses produce axes within each morphospace, which can be compared with each other.

PCoAs on the angles of the PLS vectors of the five rays of the hand and the foot differentiate between the first ray and the lateral rays. In terms of shape, differences in covariation between the first ray and the more lateral rays are mostly determined by variation in the medial tubercle. The musculus flexor pollicis brevis and the musculus abductor pollicis brevis insert on the medial tubercle. A larger enthesis for these muscles is related to thenar movement and thumb – index finger interaction, and as such is important for the efficiency of the precision grip.

A potential future for this line of work would be to integrate all the elements of the hand and the foot. A good starting point would be the metapodials and the other phalanges. The podials would provide additional challenges, as their homology is not always clear. Another future perspective includes the use of semilandmarks to capture the morphology of the bones more accurately. This would allow, amongst other things, for a more detailed appraisal of variations in tubercle morphology. Moreover, the analyses should not be limited to H. sapiens, but be expanded to great apes, and the consequences of stacking both skeletal elements and species attribution in a single matrix should be assessed. Furthermore, fossils should be included in this type of morphometric analysis to support or contradict any evolutionary hypotheses derived from data from extant species.