Michel Laurin, Comparative Osteohistology, UMR CNRS 7179, Université Pierre et Marie Curie (Paris 6), Paris, France. Tel.: (33) 1 44 27 36 92; e-mail: email@example.com
The prevailing hypothesis about grasping in primates stipulates an evolution from power towards precision grips in hominids. The evolution of grasping is far more complex, as shown by analysis of new morphometric and behavioural data. The latter concern the modes of food grasping in 11 species (one platyrrhine, nine catarrhines and humans). We show that precision grip and thumb-lateral behaviours are linked to carpus and thumb length, whereas power grasping is linked to second and third digit length. No phylogenetic signal was found in the behavioural characters when using squared-change parsimony and phylogenetic eigenvector regression, but such a signal was found in morphometric characters. Our findings shed new light on previously proposed models of the evolution of grasping. Inference models suggest that Australopithecus, Oreopithecus and Proconsul used a precision grip.
Grasping behaviour is a key activity in primates to obtain food. The hand is used in numerous activities of manipulation and locomotion and is linked to several functional adaptations (Godinot & Beard, 1993; Begun et al., 1997; Godinot et al., 1997). In particular, the hand is involved in prehension, such as gripping of static foods (fruits, leaves) and dynamic foods such as insects or other prey (frogs, rodents, small antelopes). Some primates such as chimpanzees (Pan troglodytes) and capuchins (Cebus apella) use their hands to manipulate tools, to crack nuts, for example (Boesch & Boesch, 1990; Fragaszy et al., 2004), whereas gorillas (Gorilla gorilla) use their hands to extract food from holes (Pouydebat et al., 2005). The evolution of primates (humans included) is linked to the development of those behaviours allowing organisms to exploit the resources in their environment. A general model of grasping in primates proposes an evolution from a ‘power grip’ towards a ‘precision grip’, supposed to have taken place in hominids; the precision grip has been suggested to appear with Australopithecus afarensis (Marzke, 1997) or with Homo (Napier, 1956, 1960). The power grip is defined as a grasp with the palm, and is probably a very old behaviour, as it occurs in anurans, crocodilians, squamates and several therian mammals (Gray, 1997; Iwaniuk & Whishaw, 2000). On the contrary, the precision grip, in which an object is held between the distal surfaces of the thumb and the index finger, is usually viewed as a derived function, linked to tool use and human morphological autapomorphies (Napier, 1956; Tuttle, 1965; Schultz, 1969; Susman, 1979, 1989; Marzke et al., 1992; Clark, 1993). The precision grip has been considered the most important hand function of all prehensile movements (Napier, 1980).
Our aim is to reconsider this simple model of grasping evolution in the light of morphometric data from numerous species of primates and behavioural considerations such as areas of contact between the fingers and the food grasped by extant primates. Therefore, the possible presence of a phylogenetic signal in the behavioural and relevant morphometric characters is investigated and the correlation between morphometric and behavioural characters is also determined. We also present models that enable inference of behaviours from morphological characters, which we use to infer behaviours in three extinct primates: Proconsul africanus, Oreopithecus bambolii and Australopithecus afarensis, three species considered to have divergent grasping abilities. Proconsul africanus and Australopithecus afarensis are from Africa, which is probably the cradle of hominoid diversification (Arnason et al., 2000; Folinsbee & Brooks, 2007). Oreopithecus bambolii was found in Tuscany, Italy. It is included here because its prehensile behaviour has been inferred in the literature (Moyá-Soláet al., 1999; Susman, 2004).
Material and methods
Quantification of areas of contact
The data represented in this study are based on a wide variety of primates observed in captivity (Appendix S1): nine capuchins (C. apella), nine macaques (Macaca fuscata), nine baboons (Papio papio), three gibbons (Hylobates lar), seven orang-utans (Pongo pygmaeus), three gorillas (G. gorilla) and 14 chimpanzees (P. troglodytes). We also have observations from nine children, 2–5 years of age (Homo sapiens), and nine adults (H. sapiens). Data for three other species were collected from the literature (Christel, 1993; Christel et al., 1998): black mangabey (Cercocebus aterrimus), geladas (Theropithecus gelada) and bonobos (Pan paniscus). These species represent a wide array of body size, hand morphological traits and anthropoid taxa. Indeed, capuchins do not possess an opposable thumb and none of the studied primate species except humans has morphological traits usually associated with precision grip.
Protocol of observations
The observations of grasping of small and large objects have been made in various groups of animals belonging to zoological gardens in France. All individuals observed in any given species belong to a single group and the hierarchical position of each specimen was established. The animals were observed without modification of their social (within their group) or environmental (e.g. logs, rocks, ropes) context to maintain: (i) all behavioural interactions between the members of the group; (ii) all constraints in relation to the environment; and (iii) all possibility of opportunistic manipulation (Parker & Gibson, 1977).
All observations of the animals were made for 7 months (Pouydebat, 2004). The duration of observation for each specimen was standardized following the usual methods suggested in comparative ethology (Lehner, 1996). A preliminary analysis was conducted by ‘ad libitum sampling’ (Altmann, 1974) that permits the individual recognition of all subjects for each species and the identification of a wide variety of areas of fingers in contact with the presented objects. During the study, each individual was observed according to the method of ‘focal animal sampling’ (Altmann, 1974). We filmed the animals during two sessions of 2 h each for chimpanzees, baboons, capuchins and macaques and six sessions of 2 h each for orang-utans, gorillas and gibbons. Every 15 min, sequences of grasping which lasted 5 min were analysed to determine the area of the finger in contact with the object by using frame-by-frame analysis in the laboratory. We obtained a minimum of 90 min of observation of grasping behaviour for each chimpanzee, baboon, macaque and capuchin, and 180 min for each orang-utan, gorilla and gibbon.
Frame-by-frame analysis was performed with a Basler camera (Basler, Ahrensburg, Germany), recording 250 images per second. Each prehension technique was characterized by contacts between one or several lateral or ventral areas of a minimum of two digits or the complete palm. From this analysis, we determined five categories of object prehension.
Size and nature of the objects
For all primates except humans, the objects were small and scattered on the ground; the objects involved spherical cereals and fruits. In humans, the objects were spherical pearls. It was necessary to standardize the diameter and the volume in order to calibrate these parameters according to the length of the hand of the species studied. In this paper, we always presented spherical objects to the animals and determined their diameter. The diameter of the objects was calibrated according to the length of the hand of the species. As we knew the length of the hand of the smallest studied species (76.2 ± 5.3 mm for capuchin) and the diameter of the smallest object (3.0 ± 0.1 mm) grasped by this species, we deduced the diameter of objects for other species as follows (D = diameter, L = length, all units in mm):
For example, to calculate the diameter of objects to be grasped by chimpanzees, we used the length of the chimpanzee’s hand (235.0 mm) and that of the smallest hand (the capuchins’ hand: 76.2 mm) and the diameter of the smallest object (3.0 mm). In this example, Dxc corresponds to the determined diameter of the small object for chimpanzees (c). We calculated the following value for objects in chimpanzees: Dxc is equal to 9.0 mm (235.0 mm × 3.0/76.2). We followed the same method to calculate the diameter of objects for each species (Appendix S2).
Number of grasps
A total of 5549 grasps were recorded for the eight studied species (Table 1). The percentage of each prehension category was calculated on the basis of the total number of grasping observed in each species.
Table 1. Use of the grasps from all categories in anthropoid species.
Category 1 Precision
Category 2 Thumb-distals
Category 3 Thumb-lateral
Category 4 Without thumb
Category 5 Power
N, Number of grasps of small objects.
The mean number of observed grasps per individual for each species is also given. All other columns represent percentages.
Morphometric data were obtained from hand skeletons belonging to the collection of the Muséum National d’Histoire Naturelle (Paris). Our sample consisted of 17 measurements of the hand of 26 taxa (Appendix S3). A mean of 10 specimens per taxon (males and females) was measured.
Categories of contacts
Each of the six species of primates uses from 5 to 21 of 26 different modes of contacts between areas of digits and the objects. This large number of contacts can be classified into five main categories of grasping behaviour (Table 1). In order to facilitate our comparison with the previous literature, each category is named (Fig. 1) as suggested by Napier (1956) and Jones-Engel & Bard (1996):
Category 1: contact between the distal phalanx of the thumb, the distal part of the index finger and the object (precision).
Category 2: contact between the distal phalanx of at least three fingers and the object (thumb-distals).
Category 3: contact between the distal part of the thumb, the lateral side of the middle and proximal phalanges of the index finger and the object (thumb-lateral).
Category 4: contact between one or several fingers, except the thumb, and the object (without thumb).
Category 5: contact between the palm, one or several fingers and the object (power).
Data distribution and transformation
The length of the manus (LM) is used in the analyses below to remove some of the body size effect on the remaining characters. For all analyses, this character (LM) was log-transformed because body size usually follows a log-normal, rather than normal, distribution. All other morphometric characters were divided by LM before all the analyses below (Appendix S3).
Detection of phylogenetic signal
To determine whether or not the phylogeny needed to be incorporated into the analyses, and whether or not squared-change parsimony could be used to study character evolution, we performed two types of tests of phylogenetic signal. The first one consists of comparing the squared length of a character over the reference tree to the squared length of multiple (in this case 10 000) random trees. For quantitative characters optimized through squared-change parsimony, branch length data are critical. Thus, the most appropriate way to create random trees is to reshuffle the terminal taxa on a tree of fixed topology and branch lengths (Laurin, 2004; Laurin et al., 2004). However, squared-change parsimony optimization requires the same assumptions as independent contrast analysis; thus, we checked if these assumptions were met using the PDAP (Phenotypic Diversity Analysis Programs) module for Mesquite (Midford et al., 2003). This module performs four relevant tests. The first three regress the absolute value of standardized contrast against: (i) their expected standard deviation (the square root of the sum of corrected branch lengths); (ii) the estimated value of the base node; and (iii) the corrected height of the base node. The fourth and last test is a regression of the estimated value of the base node against the corrected height of the base node. We performed all these tests (four) for all characters (21) for all trees (five). No corrections for multiple tests were made, which makes our procedure more stringent by rejecting trees which yield artefactual relationships which are significant when taken in isolation, but which might no longer be significant if such corrections were made. Furthermore, it is not clear if such corrections should be made because these four tests evaluate different statistical artefacts, and results for one character have no bearing on other characters. Deviations from the assumptions were detected in several cases; this prevented analyses of the relevant characters on a given tree. To maximize the number of tests that could be performed and to test the presence of a phylogenetic signal, we produced four alternative trees (trees 2–5 in Appendix S4) that derive from our main tree. Tree 2 differs from tree 1 in having older divergence dates for some nodes, especially among hominoids (Fig. 2, grey). Tree 3 was produced by a natural logarithmic transformation of the branch lengths of tree 1. Tree 4 was produced by setting all branch lengths to 1. Tree 5 is ultrametricized from tree 4. For the behavioural data, the same trees were used, but taxa with missing data were pruned from the trees. These trees deviate increasingly from the starting tree that, we believe, includes plausible divergence times; thus, for all analyses, we used the tree with the lowest designator possible (tree 1, and if not possible, tree 2, and so on). All these trees can be used to the extent that ‘the statistical adequacy of any proposed branch lengths should be viewed as an empirical issue’ and ‘In general, any transformation of possible use for tip data…might also be tried for branch lengths’ (Garland et al., 1992: pp. 23–24).
A second set of tests of phylogenetic signal was performed using phylogenetic eigenvector regression (PVR) analysis (Diniz-Filho et al., 1998). This method relies on a principal coordinate analysis of the phylogenetic distance matrix to extract eigenvectors that are used in a standard linear regression against the character of interest. The eigenvectors represent the position of the taxa on the various principal coordinate axes. However, for n taxa, n – 1 principal coordinate axes are produced, and they cannot all be used in a linear regression analysis or there would be no degrees of freedom left. The axes we used were selected using a broken-stick model (Diniz-Filho et al., 1998), because it would have taken too long to test for a significant relationship between all axes and all characters separately (21 characters and up to 26 taxa yield 546 tests). Furthermore, none of the behavioural characters exhibits a phylogenetic signal according to the squared-change parsimony analysis (see below), and that method is usually more powerful than PVR (Cubo et al., 2005). Thus, the broken-stick model was the only method applicable to all of our data. In all our analyses of phylogenetic signal, only the first two axes were used. They represent 60.7% of the phylogenetic variance.
The phylogenetic distance matrix was obtained from tree 1 using the Stratigraphic Tools module (Josse et al., 2006). The principal coordinate analysis was performed in Progiciel R (Casgrain et al., 2004). Linear regressions were tested for statistical significance using 9999 permutations of the dependent variables (here, the morphometric or behavioural data) in Permute! (Casgrain, 2005). A regression is significant (at a 0.05 threshold value) if fewer than 5% of the data sets have an R2 value at least as large as the original data set (the original, unpermuted set is included). The advantage of using permutations to test the significance of the relationship is that this method requires far fewer assumptions about the distribution of the data. Thus, contrary to the other method, this test could be applied to all characters.
Phylogenetically independent contrasts
We assessed correlations between the behavioural (dependent) and morphometric (independent) characters using phylogenetically independent contrasts (Felsenstein, 1985), whenever the assumptions of that method were met on at least one of our trees. These tests were performed using the PDAP module for Mesquite (Midford et al., 2003). As for the test a phylogenetic signal using squared-change parsimony, we performed this test on the tree with the lowest designator (tree 1 if possible; if not, tree 2, etc.) that gave adequate contrast standardization for both characters analysed.
Variance partitioning with phylogenetic eigenvector regression
Correlations between the behavioural (dependent) and morphometric (independent) characters were also tested using variance partitioning with a PVR analysis (Diniz-Filho et al., 1998; Desdevises et al., 2003). This method incorporates the phylogeny into the analysis in the form of principal coordinate axes, as explained in the section on phylogenetic signal detection (above). We performed some exploratory analyses (results not shown) in Permute (Casgrain, 2005) to choose the characters to analyse through PVR. Correlation between these characters was then tested through variance partitioning with PVR, using principal coordinate axes that were significantly correlated with the dependent (behavioural) characters. These axes were selected by performing a simple regression of all axes against the relevant behavioural characters, because in this case only three characters are involved.
Linear regression models
Simple or multiple linear regressions were used to produce inference models of the behavioural characters. These models are based on the morphometric characters that are significantly correlated with the behavioural characters according to the variance partitioning analysis with PVR (Appendix S5) or according to linear regressions, but they were constructed without incorporation of the phylogeny. This is unavoidable because principal coordinate axes have no absolute meaning (they differ when a taxon is added or removed, or if the topology or branch length is changed), so incorporating them into predictive models would preclude their use in palaeobiological inference, which is self-defeating. These models could not be used on some of the extinct taxa included in our study because the relevant morphometric characters are not known. To maximize the number of inferences which could be drawn about these taxa, we built additional inference models for several combinations of these taxa and available osteological characters (Appendix S6). For this purpose, we used a forward selection procedure in Permute to select the characters, among those that were significantly correlated with each other according to the variance partitioning or the independent contrast analyses. In two cases, to produce the models, we had to extend character selection to other characters (because extinct taxa are incompletely known); in such cases, we used a forward selection procedure in Permute (with p to enter of 0.1) to build the model.
Taxon-specific distribution of categories
The platyrrhine and all catarrhines were able to modulate their grasping behaviour for small food items. However, some clusters of species emerge for all categories of contact. These groups can be compared on the basis of the mean percentage (Table 1).
From these data, five main conclusions can be drawn. First, great apes (except humans) and capuchins use grasping category 4 (without thumb), whereas humans and other primates never do. All primates except humans use the lateral part of their index (category 3, thumb-lateral), whereas capuchins hardly ever do and humans never do. Secondly, adult humans and capuchins are similar as regards a majority of grasping categories. Thirdly, human children do not show clear similarities with other primate species for any of the grasping categories. Fourthly, cercopithecids (i.e. macaques and baboons), capuchins and humans use the precision grip most often (Table 1). Fifthly, humans present some unique characteristics in their selection of grasping categories. Human adults are the only primates to use exclusively the tips of the fingers (precision and thumb distals). Human children and capuchins are more similar behaviourally to human adults than to other primates in our sample, because they mainly use the tips of their fingers (93%). One key result from our data is that all species were able to grasp small objects with the precision grip corresponding to the contact between the tips of thumb and index finger. This applies even to capuchins, although their thumb is only pseudo-opposable.
Detection of phylogenetic signal
Globally, about 40% of the characters display a phylogenetic signal (Appendix S4). This proportion holds when using both tests, but is affected by taxonomic sampling; when only the 11 taxa for which behavioural data are available are tested, this proportion decreases to 20%. This may explain why no phylogenetic signal was found in the behavioural data using random taxon reshuffling and squared-change parsimony or PVR analysis using the first two principal coordinate axes selected by the broken-stick model (Appendix S4). However, when each axis was regressed separately against the behavioural characters, axis 1 had a significant effect for precision, without thumb and power grasp frequency.
Character correlation assessed using phylogenetic independent contrasts
Only three of the behavioural characters were analysed using independent contrasts because for the others, the assumptions of that method were not met (Appendix S7). Some of the morphometric characters are clearly correlated with behavioural data. The behavioural character ‘precision’, involving the contact of the distal phalanges of the thumb and the index, and the character ‘thumb-lateral’, involving contact of the distal phalanx of the thumb and the lateral side of the index, are correlated with the length of the carpus and the first ray (the thumb and its metacarpal). Power grasping is linked with second and third digit length.
Character correlation assessed using variance partitioning with PVR
The forward selection test in Permute (with a p to enter of 0.1) resulted in only one or two morphometric characters being selected for each behavioural character (results not shown), which implies that grasping behaviour can be inferred fairly precisely using few morphometric data. No principal coordinate axis was selected. An additional (non-phylogenetic) test using simple linear regressions confirms that all the selected characters are correlated with behavioural characters and explain more than 50% of the variance in the behavioural characters (results not shown). In addition, variance partitioning analyses of the characters which were correlated with at least one of the phylogenetic principal coordinate axes indicate that most of the explained variance is genuinely explained by morphometry rather than by covariation with the phylogeny (Table 2).
Table 2. Variance partitioning with PVR showing the correlation between behavioural (dependent) and the morphometric (independent) characters that were selected by the forward selection procedure using linear regressions with permutations (Appendix 5) and the phylogenetic principal coordinate axes (we included those that were correlated with behavioural characters based on several simple regressions with permutations).
Portion of variance explained (probability)
Variance explained by covariation of morphometry and phylogeny
Portion of variance explained (probability)
Linear regression models of inference
Inference models obtained through linear regressions include one or two morphometric characters (Appendix S5). Inferences were obtained from these models for extant species in which morphometric data (but no behavioural data) are available (Appendix S8), and for three extinct species (Table 3).
Table 3. Inferred behaviour frequency (%) of extinct primate taxa based on linear regression inference models (Appendix 5).
As frequency of a behaviour cannot exceed the interval 0–100%, values outside this interval represent modelling errors and should be interpreted as implying values close to the nearest bound (given in parentheses). Note that as these are inferences, the sums of percentages on a given line do not necessarily add up to 100% (the difference between the total and 100% represents modelling errors).
Compilation of a time-calibrated tree
For many of the analyses performed below, a phylogeny incorporating an estimate of branch lengths is needed. Thus, we compiled a time-calibrated tree (branch lengths reflect estimated evolutionary time). The topology follows Goodman et al. (2005). Divergence time estimates are much more contentious because the affinities of several primate species based on fragmentary material are poorly constrained (Ross et al., 1998), and because divergence time estimates based on molecular data are often considerably older than the minimal times of divergence based on fossils (see Marjanović & Laurin, 2007; for a review). Estimates of divergence times based on molecular data also often differ substantially between studies; for instance, Arnason et al. (2000) estimated that the divergence between strepsirhines and haplorhines at about 90 Ma (in the Turonian, in the Early Upper Cretaceous) and that the anthropoid radiation started at 70 Ma, whereas Yoder & Yang (2004) estimated these events to have occurred at about 80 and 50 Ma, respectively. Yoder & Yang (2004) furthermore estimate the divergence between Lorisiformes and Lemuriformes at 70–75 Ma, but Roos et al. (2004) estimate it at about 60 Ma, although the oldest fossil in that clade dates from the Priabonian (Stucky & McKenna, 1993: p. 757), no more than 37.2 Ma, according to the geological time scale of Gradstein et al. (2004). Given the poor fossil record of (crown group) strepsirhines (Stucky & McKenna, 1993: p. 757), the paleontological date here is likely to seriously underestimate the actual divergence dates within Strepsirhini. Considerable differences between molecular dates are not unexpected because several methods can be used to obtain molecular dates. The choice of the calibration date also influences the results (Brochu, 2004a, b; Poux & Douzery, 2004; Marjanović & Laurin, 2007), and there are many other pitfalls in such analyses (Shaul & Graur, 2002; Graur & Martin, 2004; Marjanović & Laurin, 2007). Furthermore, evolutionary rates are quite variable in primates (Grossman et al., 2004; Steiper et al., 2004), which makes molecular dating more difficult.
For all clades, we have adopted a compromise that uses minimal divergence ages from the fossil record in the taxa where this record is reasonably abundant, and molecular ages for taxa such as Strepsirhini, for which the fossil record is poor. When using molecular data, we have tried to use studies that obtained ages compatible with the fossil record. We assembled the tree in Mesquite (Maddison & Maddison, 2006) using the Stratigraphic Tools module (Josse et al., 2006). To facilitate comparisons with palaeontological literature, whenever molecular dates fall close to geological stage boundaries, we used the age of the boundary itself (Fig. 2); this also facilitates tree manipulation in Stratigraphic Tools (Josse et al., 2006). More information about individual divergence ages can be found in Appendix S9.
Ontogenetic and taxonomic distribution of grasping behaviours
The data presented above show that precision grasping can be used by all arboreal and terrestrial primate species in our study. Indeed, the precision grip was recently reported in capuchins (Spinozzi et al., 2004). However, according to that report, this grasping technique was less frequently used by immature individuals. In our study, all individuals were observed in their social groups, and we did not observe any obvious effect of ontogenetic age for this behaviour (data not shown).
Comparisons between great apes and human children refute the idea that they have similar sensorimotor organization (Parker & Gibson, 1977). In our study, this similarity is not great. Categories of grasping used for small objects differ strongly between human children and great apes. Great apes use precision grips and the distal phalanges of their digits less often than children. In addition, they use the lateral side of their index and the grip without thumb, contrary to children. These differences between great apes and children may be explained by the neural and morphological variability existing between humans and the other species, regardless of their age. Finally, the comparison between great apes and human adults does not show strong similarities. Human adults use precision grips with small objects and the distal phalanges of their digits to grasp large objects much more often.
Relationship between morphometry and grasping behaviour
The results of this study refute some well-established ideas about the relationship between morphometry and prehensile behaviour. Our data show that great apes use precision grips less often than cercopithecoids when handling small objects and that the contrary pattern is observed when handling large objects (Pouydebat et al., 2006a). Furthermore, Pongo and Pan use the grip without the thumb, contrary to cercopithecoids. These differences are partly explained by morphometrical data such as the shorter thumb of great apes (Marzke et al., 1992), as shown by our analyses (Appendix S7, precision and characters 3 and 5; Table 2, without thumb and character 5). Therefore, the length of the thumb, including the first metacarpal, is an important morphometrical character in the behaviour of precision grasp, as previously suggested (Napier, 1956; Schultz, 1969; Susman, 1989). These morphometrical parameters are also negatively correlated with the behavioural character ‘thumb-lateral’ (Appendix S7), involving the contact of the distal phalanx of the thumb and the lateral side of the index. Therefore, the length of the first ray does not reflect precision grasping only.
Other morphometrical data are specifically correlated with a single behavioural character. For instance, the power grasping behaviour, involving the palm of the hand during the grasp, is strongly correlated with the lengths of the digits one to three (Appendix S7, power vs. characters 7–9, 12, 15). The length of digits 2 and 3 is correlated only with power-grasping (among the behaviours studied here); short index and third digits seem to favour power-grasping.
A few species show prehensile patterns which could not have been inferred from their hand morphology. For instance, gorillas show a high percentage of use of the precision grip in spite of their short thumb (Pouydebat et al., 2006a). This reflects the fact that the length of the thumb does not explain all the variance in use of the precision grip. A quantitative, statistical approach is required because of the complexity of the relationship between morphometry and prehensile behaviour.
Finally, we wonder why carpus length is correlated with precision and thumb-lateral grips. It would be interesting to test if this can be explained by soft anatomy, such as muscle or tendon morphology, or by locomotory behaviour (such as arboreality).
Unexpected similarities between capuchin and human prehensile behaviours
Our study reveals similarities in prehensile behaviours between capuchins and humans. Capuchins use precision and thumb-distal grips as often as human adults. Similar to humans, capuchin monkeys almost exclusively used the distal phalanges of their digits to grasp small objects. These results can be compared with those of Spinozzi et al. (2004), who observed a wide variety of grasping patterns in capuchins. These include various forms of precision and power grips. Contrary to our results, Spinozzi et al. (2004) reported that capuchins use precision and power grips with the same frequency to grasp small food items. We found that they opted more often for precision grips (almost 80%). This difference between our results could be due to the population and individual variability or the protocol of observation which was not the same. Capuchins display a wide variety of prehensile abilities that confirm their capacity, apparently atypical among New World monkeys, to use their hands dexterously during extractive foraging and object manipulation (Fragaszy et al., 1991;Fragaszy & Boinski, 1995; Christel & Fragaszy, 2000; Pouydebat et al., 2006b), although they do not possess the true opposable thumb typical of catarrhine primates.
Inferences about the grasping behaviour of extinct primates
The inference of grasping behaviour from morphological analyses of the hands of fossils is a complex problem. Some authors opposed the hand of extant apes to the hand of humans and argued that extant apes are unable to grasp objects with a precision grip or pad-to-pad gripping. However, many extant ape species use precision gripping (Christel, 1993; Pouydebat, 2004) without meeting all the morphological criteria usually considered to be linked with precision gripping. We have shown that precision gripping can be performed by hands showing a greater morphological diversity than previously thought.
Our sampling of behavioural characters in anthropoids largely restricts our discussion of their possible evolution within this clade, as suggested by the extant phylogenetic bracket principle (Witmer, 1995), which was extended into the context of continuous characters by Laurin et al. (2004). The absence of a phylogenetic signal in the behavioural characters (Appendix S4, characters 17–21) precludes tracing their history over the tree using optimization procedures because the results would not be reliable (Laurin, 2004). This limitation may reflect the relatively low number of taxa for which behavioural data are available as analyses of our data with the same taxonomic sampling show a similar absence of a phylogenetic signal in most morphometric characters (Appendix S4, central columns). Thus, it might be possible to reconstruct the evolution of these behavioural characters by sampling the same clade more densely.
Another possibility is to use the linear regression inference models that we have produced (Appendix S5) to infer the behaviour of extinct primate species (Table 3): A. afarensis, O. bambolii and P. africanus (Fig. 3). Thus, we can infer that the Plio-Pleistocene hominin A. afarensis (3–4 Ma old) from Hadar (Ethiopia) exhibited frequent precision behaviour (Table 3). The linear regression inference for thumb-distals behaviour in A. afarensis suggests the infrequent occurrence of this behaviour. Results about precision behaviour are close to those obtained by Marzke (1997) and Panger et al. (2002); on the contrary, several authors suggested that A. afarensis use power grasping most often (Bush et al., 1982; Stern & Susman, 1983; Susman, 1991, 1994), which we infer to have occurred infrequently. Australopithecus may share with humans the absence of without-thumb behaviour, which may be a synapomorphy of Australopithecus and Homo among hominoids (although cercopithecoids show a convergent similarity).
The Miocene ape Oreopithecus may have exhibited slightly less precision behaviour than Australopithecus, but like the latter, it resorted to that behaviour much more often than the power grip, which was rarely displayed. This result is congruent with the assessment of Moyá-Soláet al. (1999). On the other hand, Susman (2004) suggested that O. bambolii emphasized the power grip over the precision grip. Similar to most extant hominoids and Proconsul, but unlike Homo and Australopithecus, Oreopithecus may have displayed the without-thumb behaviour (Table 3).
The linear regression inference models suggest that the Early Miocene (Burdigalian) stem-hominoid P. africanus (16–18 Ma) from Kenya used the precision grip frequently (Table 3) and thumb-distals grip more rarely. P. africanus resorted to thumb-lateral grip relatively infrequently and used the without-thumb grip even less frequently. These results about P. africanus are very different from those published in the literature, which suggest that P. africanus did not use a precision grip (Napier & Davis, 1959;McHenry, 1983;Walker & Pickford, 1983; Begun et al., 1994). The widespread distribution of the precision grip in the primate species sampled in this study support our palaeobiological inferences.
The evolution of grasping behaviour
The classical model proposes a late appearance of precision grasping, often considered unique to hominids. Napier (1956, 1960) suggested that grasping objects with precision requires opposability of the thumb and favourable relative lengths of digits I and II. Furthermore, some authors suggest that precision grasping is linked with brain organization and the development of cognitive processes (Napier, 1960; Jones-Engels & Bard, 1996). Under such hypotheses, hominoid fossils presenting morphological characters associated with precision grasping have been argued to be able to use tools (Susman, 1989; Marzke, 1997). The data presented in this paper refute this evolutionary scenario because species with highly different hand morphologies and brain structure use a precision grip. The evolution of grasping abilities in platyrrhines and catarrhines is much more complex than a simple trend from power to precision grasping. Indeed, several species (i.e. P. troglodytes, P. pygmaeus, M. fuscata, P. papio and C. apella) use the tips of their thumb and index finger for grasping small objects, demonstrating that morphological criteria previously used for deducing grasping ability are not reliable (Susman, 1998). Furthermore, in our quantitative analysis based on the percentages of use of five simplified categories of grasping, capuchins are similar to humans (mainly adult humans) even though they possess a pseudo-opposable thumb (rather than a truly opposable thumb, capable of adduction and rotation of its carpo-metacarpus joint). This functional similarity of two species which diverged about 34 Ma (Fig. 2) is surprising.
Our findings are in concordant with the evolutionary model according to which a primitive power grasp was subsequently modified into a derived precision grasp (Susman, 1979), although this transition occurred before the appearance of hominoids. Our data suggest that both behaviours were already present in the first anthropoids. It would be interesting to obtain some data on Tarsius and strepsirhines to determine when precision grasping appeared.
All our results suggest that grasping has evolved in a more complex manner than previously realized. Our observations show that precision grip is far more widespread than previously thought. This is coherent with the findings that skilled forelimb movements are also present in other mammals (Ivanco et al., 1996; Whishaw et al., 1998) and even in amphibians (Gray, 1997). These movements may not be homologous and represent convergent evolution of motor patterns that superficially resemble reaching (Bracha et al., 1990). However, the similarities in reaching among different mammalian taxa suggest that the movements are homologous within mammals. It would be interesting to apply comparative methods to a far greater range of taxa to assess broader-scale evolutionary patterns of various grip patterns.
We would like to thank the Foundation Singer-Polignac for their financial support, P. Piras for helping us with some statistical analyses, D. Marjanović for numerous stylistic corrections, and all the staff of the Zoo of Beauval and of the Monkey Valley (France).