1. Diet and cranial variation.

The selectively important foods described in H_(FMP1) that are mechanically challenging to eat may be fallback foods (resistant, less desirable foods consumed when preferred foods are not available; e.g., [34]), but they could also represent mechanically resistant preferred foods. The key requirement of H_(FMP1) is simply that the food items in question are difficult to process but nonetheless selectively important; in other words, the inability to efficiently consume the items would have a deleterious impact on fitness. In such cases, H_(FMP1) predicts that morphologies of the feeding apparatus that decrease feeding performance would be selected against, and a functionally advantageous pattern would spread in the population. The latter should result in an increase in integration intensity, because selection reduces variation. In theory, such adaptation could also be reflected in the number of elements recruited in the functional complex under selection. One could hypothesize that such differential recruitment of elements in feeding might be driven by the necessity to produce higher or more frequent bite forces, which results in the propagation of higher magnitude or more repetitive strains in more craniofacial regions.

H_(FMP1) predicts integration among traits related to two constraints – the ability to generate appropriately high or repetitive bite forces, necessitating facial morphologies associated with high muscular mechanical advantage, and the need for a craniofacial skeleton able to withstand the forces associated with feeding on resistant foods. These constraints should result in the evolution of craniofacial traits that efficiently absorb reaction forces and/or transmit loads to the food being bitten. As feeding loads either increase in magnitude or become more repetitive, there should be a greater need to evolve functionally advantageous morphological character complexes, and for decoupling feeding functions from other (e.g., sensory) functions performed by the same architectural complex. An analogy to the latter situation is suggested by Young and Hallgrímsson [35], who showed that serially homologous characters in the mammalian fore- and hind limbs are correlated, but in species like bats which possess highly derived forelimbs, such correlation is absent, presumably because selection for flight decoupled development of the forelimb from that of the hind limb.

The link between diet and craniofacial morphology is firmly established (e.g., [36]–[43]), and interspecific cranial differences indicate that the facial skeleton is subjected to strong selection (e.g., [44]). The expectation of H_(FMP1) rests on several lines of evidence. Numerous studies have found that the face, and particularly its oral subunit, contributes most extensively to cranial integration in primates (e.g., [27], [29], [30], [45]). Research carried out by Marroig and Cheverud [27] provides the strongest published evidence, to date, that diet might structure integration. In this study, each species was assigned percentages of different food types consumed year round such as fruits, insects and leaves. Then for a group of taxa, a dietary similarity matrix was built using a variable reflecting pair-wise similarity between taxon A and taxon B calculated using the formula:(1)where F stands for fruits, L stands for leaves and I stands for insects. Subsequently, the dietary similarity matrix was correlated with its corresponding morphological integration similarity matrix. However, such analyses do not specifically target the foods whose consumption might compromise the functional activity of the jaws, lead to injury and the inability to feed. Furthermore, such analyses dilute the potential effect of such resistant foods on morphological integration (e.g., by expecting that the proportion of insects consumed might induce similarity in morphological integration). Moreover, such analyses do not acknowledge Liem’s paradox and the evolutionary importance of fallback foods.

Goswami [46] reported a lack of strong and significant correlation between similarly built dietary similarity matrices and morphological integration matrices in marsupials, a result, which might be due to the lack of consumption of mechanically resistant foods by these species, or, as proposed, to the early ossification of facial structures directly involved in feeding, an adaptation stemming from the early birth of underdeveloped offspring combined with the necessity to suckle. Within placental mammals, diet quantified in a way similar to Marroig and Cheverud [27] was shown to be significantly or marginally significantly correlated with morphological integration only in Arctoidea (bears), when controlling for phylogeny whether or not the allometric component of variation is included and when phylogenetic and allometric signals are not excluded from the data, in Mustelidae (weasels) when allometric variation is preserved in the data whether or not phylogeny is accounted for, and in Musteloidea (raccoons, weasels and pandas), regardless of phylogeny and allometry [47]. Notably, these taxonomic groups differ from other Carnivora in that they contain many species that include an important fruit and/or leaf component in their diets. Importantly, at least some species of bears focus on resistant foods such as acorns, hickory nuts, beechnuts, hazelnuts and pine nuts during the autumn when tremendous weight gain is critical to ensure survival during the winter.

The factors responsible for the structuring of dietary integration as well as its underlying mechanism (i.e., heritable, selection-driven characteristic versus gene-by-environment interaction effect) remain unknown. Zelditch [25] found that the dietary and behavioral changes in placental mammals associated with the transition from suckling to mastication occur only after neonatal integration has been re-patterned into integration designed to withstand the effects of grinding. Thus, selection may act on a given ontogenetic stage and thereby lead to adult cranial integration in advance of a change in behavior. The elements of the facial skeleton in placental mammals follow a typically somatic postnatal growth pattern that extends into early adulthood [48], [11], which arguably favors the influence of gene-by-environment effects on facial anatomy and function. Skeletal growth can be influenced in many ways by many factors, and mechanical loads are among the best documented [49]. Indeed, numerous reports indicate that craniofacial morphology clearly responds to diet, in other words the physical characteristics of food can influence facial morphology during the lifespan of an individual [50]–[54]. There are no experiments suggesting that cranial integration in species adapted to a soft diet is affected by the adoption of mechanically challenging diet. Thus, it is unknown if traits that jointly experience high strains tend to covary. If this is the case, it would be unknown if a higher integration would result from the same regions experiencing higher strain or from more regions experiencing strains higher than a particular physiological threshold. An interesting study was carried out by Corruccini and Beecher [55]; [50], [53], who found that groups of placental mammals that had been fed artificially soft diets during development exhibited increased morphological variability and decreased dentognathic integration (including malocclusion) relative to groups that had been fed harder diets. This suggests that a normal loading environment might be essential for the functional integration of gnathic morphologies [56]; (see also [57] for results on limbs), and that functional integration relies at least in part on gene-by-environment interactions.

Vinyard [58] found no association between dietary type expressed as percentages of foods consumed and covariation structure of the mandible of galagos. Despite the varied diet of galagos, it was argued that the ability to obtain and digest gum might be the fundamental adaptation of this group [58]–[64]. If this is the case, the lack of association between diet and covariation structure might not be as surprising. A comparison of biomechanical performance of the skulls of tree-gouging primates indicates little evidence for increased force-production or load-resistance abilities, which led to the prediction that the skulls of tree gougers do not have to generate high biting forces [65].

The second alternative FMP hypothesis, H_(FMP2), expecting that the faces of species adapted to the consumption of mechanically-resistant foods will be less integrated, is based on a prediction deduced from research carried out by Wood and Lieberman [66]. The authors did not specifically address integration, but suggest that “skeletal structures whose development is strongly influenced by epigenetic stimuli, especially those subject to high and frequent magnitudes of masticatory strain will be more variable within species than those that are routinely subjected to lower strain magnitudes” [66], which conceivably might affect correlated variation in the facial skeleton. Extending this hypothesis from different modules to different species would imply that species whose faces typically experience low strains will have less variable morphology than species whose faces experience higher strains. The Wood/Lieberman hypothesis has engendered a lively debate (e.g., [67], [68], [56]), but its value lies in the fact that it allows one to evaluate the nature of interactions between ecological stimuli and integration from a phenotypic plasticity standpoint.

This review of the existing literature led us to think that broad-scale dietary classifications (e.g., fruits, leaves, insects, etc.) do not target a specific environmental factor that might account for changes in integration. Rather, the material properties of selectively important foods are likely to play a critical role in structuring the patterned variability of the cranium. Clearly, the potential association between food-processing behavior and facial integration requires more attention.

2. Ontogeny and morphological integration.

The second set of hypotheses tested here, the heterochrony hypotheses, link ecology to development. H_(HET1) predicts that males in sexually dimorphic species will display higher levels of facial integration than females. H_(HET2), following the Wood/Lieberman [66] prediction, has the opposite expectation. Both predictions imply that integration will be sexually dimorphic due to gene-by-environment interactions in the cranium, and follow from the observations that the face does not terminate its growth until early adulthood and that growth spurts are sex-dependent [69]. Specifically, the ossification pattern of the face is largely driven by soft tissue growth [70]–[72], including instances in which mechanical forces up regulate transcription factors in sutures to induce osteogenesis [7], [73]–[76].

The heterochrony hypotheses require that bone remodeling is responsive to biomechanical loading during development, and this premise appears to be well supported. Many studies have demonstrated increased cortical remodeling in skeletal regions experiencing elevated strains under both normal and experimental loading conditions [51], [52], [77]–[86]. Furthermore, evidence suggests that cortical bone is primarily responsive to strain prior to sexual maturity, both in terms of the rate of new bone growth and the rate of turnover [87]. Bouvier and Hylander [52] reported that the distribution of cortical bone remodeling in the macaque face coincided with peak mechanical strains during mastication in immature animals, while remodeling was observed in both high and low strain regions in the adults. A further expectation of H_(HET1) would be that integration increases during ontogeny and this argument appears to be supported [25], [30]. Thus, basic principles of bone biology and previous work on morphological integration do not contradict the heterochrony hypotheses a priori. Note that intraspecific sex-linked differences in craniofacial integration might potentially reflect the action of selection at later stages of development instead of gene-by-environment interaction. However, if such differences are consistently expressed among species characterized by differing dietary adaptations, then the likelihood of heritable sex-dependent differences decreases.

3. Cranial modularity.

The cranial modularity hypotheses will evaluate the evidence for a modularly organized cranium using both a new landmark protocol that is not reliant on suture intersections but is relevant to biomechanical performance.

The growth and development of the mammalian cranium is a complex process, involving different embryological tissue origins, modes of ossifications and hormonal influences [12]. Although defining integrated suites of characters in complex structures such as the skull is problematic because they tend to exhibit high levels of correlation not only within but also between regions [88], data collected thus far indicate that the skull has a hierarchical modular architecture, in which smaller, relatively independent subsets of highly correlated traits, or modules, are interlocked within larger, less integrated modules. For instance, the oro-facial complex is more tightly integrated than the skull as a whole (e.g., [26], [27], [45]). It has been suggested that the modular structure of the cranium roughly matches its developmental pattern. A large scale analysis of cranial modularity carried out by Goswami [45] in placental mammals identifies three highly integrated modular units, including two facial modules (anterior oro-nasal, molar) and the cranial base, and three weakly integrated modules, two facial modules (orbital, zygomatic-pterygoid) and the cranial vault. Goswami [45] further reports that theoretical correlation matrices testing specific hypotheses of modularity indicate that shared function, tissue origin, and mode of ossification are significantly correlated with observed correlation matrices for the mammalian cranium. Thus, the modular structure of the cranium roughly matches its developmental pattern. The bones of the face, the anterior cranial base, and the frontal bones of the cranial vault differentiate from neural crest-derived mesenchyme, while the posterior cranial base and posterior vault are derived from mesoderm [7], [89], [90]. The occipital region surrounding the foramen magnum differs from the rest of the skull in that it differentiates from the sclerotome of the occipital somites [4], [11], [90]–[92]. Based primarily on the type of ossification and maturation they undergo, the osseous elements forming the braincase can be grouped into cranial vault and cranial base. The basicranium and the bones surrounding the sensory capsules arise via endochondral ossification in which a cartilaginous phase exists before bone formation, while the cranial vault bones and the bones forming around the viscerocranial cartilaginous elements in the face differentiate directly into bone via intramembranous ossification, without a preexisting cartilaginous phase [7].

Despite the early differentiation of the face from the brain case and the bony orbits, growth factors and morphogens intervene in the generation of integration among those developmentally distinct units that ensures the structural integrity of the cranium [93]–[96], [11]. In fact, non-embryological hypotheses of modularity, such as models considering functional interactions coupled with an allometric component were reported to provide the best models for observed cranial phenotypic variation [25]. Furthermore, it was shown that patterns of artificial cranial deformation result in significant indirect effects on the face, the cranial base and mandible [97], [98]. In addition, several previous analyses have repeatedly suggested that the spatial modifications of the lateral basicranium have consequences for the spatial arrangements of facial structures [20]–[22]. The lateral basicranial elements, whose morphological development extends long after birth, are more strongly integrated with the mandible than are midline basicranial elements [22]. Furthermore, the depth of the middle cranial fossae appears to modulate facial length variation [22]. Specifically, an antero-lateral expansion of the middle cranial fossae rotates the face ventrally placing it under the anterior cranial fossa [23], which has been argued to lead to a shortening of the oropharynx and a decrease in projection of the upper face [4], [24].

In summary, during ontogeny integrating effects create inter-relationships among developmentally distinct cranial regions. This study benefits from a sample of closely related species characterized by differing dietary adaptations suitable for addressing questions of cranial and facial modularity in the crania of adult individuals.

1. Assessment of measurement error.

Intra-observer measurement error was assessed using the method developed by Yezerinac et al. [146] implemented in R [147]. Digitizing error was quantified by the repeated measures of a random subsample of 6 Cebus albifrons individuals, including males and females, several months after data collection. The sample of repeated measures was aligned via Procrustes superimposition and symmetrized in MorphoJ (symmetrization averages the left and right observations for bilateral characters). Measurement error was calculated as the percentage of total variance within a sample attributable to within-individual variation in landmark coordinates. An ANOVA was used to partition the variance of each landmark coordinate into within- and between-individual components, and the percentage measurement error was calculated as the ratio between the within-individual variance and the total variance multiplied by one hundred:(2)

The measurement error yielded by the comparison protocol including all cranial coordinates was 2.17%, while this associated with the protocol including only facial coordinates was 3.7%; a measurement error under 5% is widely judged as acceptable. Furthermore, the actual measurement error affecting the entire sample of each species (between 77 and 100 individuals in this study) is certainly smaller, as Yezerinac et al. [146] point out that when sample size increases, the proportion of total variance due to imprecision associated with data collection decreases.

2. PLS: the study of covariation between modules.

Partial Least-Squares analyses (PLS) were performed to test if pairs of taxa and sexes show similar patterns and degrees of oral-zygomatic, facial-basicranium and maximum modularity (not limited to spatially contiguous subsets of landmarks) correlation. Furthermore, PLS allows assessing whether correlation between facial blocks is stronger than correlation between facial and basicranial measurements, which addresses the question of cranial modularity.

PLS aims to determine the correlated pairs of linear combinations of variables within one of the a priori chosen blocks that express the greatest proportion of covariance between blocks [140], [142], [148], [1]. For each sample PLS was carried out in MorphoJ on 3D coordinates which contain an allometric component and on residuals, in which the effect of size on shape was removed, by pooled within-group regression (for subspecies and sex) of the 3D shape coordinates on centroid size (i.e., the square root of the sum of the squared distances of the set of landmarks from their centroid). In many cases, allometry is the most important source of covariation among morphological characters; therefore, simply discarding allometric shape change leads to the analysis of non-realistic integration patterns and phenotypes, that contribute to a small percentage of covariation and that are likely not the principal target of selection. Thus, PLS was used to compare allometric and non-allometric covariation axes, in terms of both integration magnitudes and integration patterns.

It has been noted that sources of variation that induce unequal levels of population heterogeneity and population sub-structuring should be held constant in integration studies [149]. To ensure a controlled comparison of integration magnitudes across samples being compared, which contain different numbers of differently related subspecies, the data were transformed prior to all analyses into pooled within-groups correlation or variance-covariance matrices. Thus, subsequent analyses focus on the covariation between the 3D coordinate residuals after removing sub-specific differences in means. In the case of the PLS analysis carried out on residuals from regression on centroid size, the matrices were also pooled within-sex.

PLS was carried out within a single configuration rather than on shapes of parts considered separately as spatially distinct blocks (which is the most frequent use of PLS), because it is more appropriate to study covariation between the purported modules of the cranium in the context of the structure as a whole [150]. Choosing PLS within configuration (e.g., [151], [152], [28]) for cranial analyses rather than the more commonly used two-block PLS is justified, because the orientation of the chosen blocks relative to each other has important architectural and arguably even functional consequences.

The permutation tests offered in the PLS procedure concern the null hypothesis of complete independence between the two blocks. The p-values of all variables used to infer integration magnitudes were obtained through 10,000 randomization rounds. The RV coefficient [153]; [150] was used as a measure of overall covariation between the two landmarks partitions. It represents a multivariate analogue of the squared correlation coefficient between two sets of variables. The RV coefficient is calculated as the sum of the squared covariances between the variables in the two blocks divided by the squared variances and covariances within the two blocks [150].

Maximum cranial modularity two-block landmark partitions, or partitions yielding the lowest between-block covariances expressed as RV coefficients were obtained through the modularity function in MorphoJ. Subsequently, these partitions were used in PLS analyses.

Shape change along axes of covariation (or pairs of singular axes for each block) allowed a visual assessment of whether or not the vectors at landmarks are indicative of similar correlation trends between the groups of interest. In all figures containing shape changes illustrated by vectors at landmarks, shape change was magnified three times, to facilitate the reading of the graphs.

In addition to PLS analyses, Principal Components Analyses (PCA) were carried out to compare patterns and magnitudes of shape change at vectors at landmarks.

3. Assessment of modularity with cluster analysis.

Five cluster analyses (one for each species) using Ward’s method of linkage were used to determine whether conserved cranial modules exist. This approach is an alternative to the maximum modularity method developed by Klingenberg [131].

Hierarchical clustering using Ward’s method of linkage was carried out on individual species distance matrices derived from corresponding correlation matrices obtained from pooled within subspecies and sex residuals from regression on centroid size. The variables subjected to hierarchical clustering were landmark coordinates. The p-values at each of the dendrogram’s nodes were estimated using a bootstrap procedure with 1000 permutation rounds using the “pvclust” package in R. Two types of p-values were computed: approximately unbiased (AU) p-values, which are obtained via multiscale resampling and Bootstrap Probability (BP) p-values, computed using normal bootstrap resampling. In this work, discussion will focus on the AU p-values. When a bootstrap value is larger than 95 the hypothesis that a particular cluster does not exist can be rejected with an alpha level of 0.05.

4. Comparison of integration indices calculated as the variance in eigenvalues of a correlation matrix. Assessment of modularity.

The variance in eigenvalues [154] was used as an alternative metric of overall facial and cranial integration magnitudes, as a quantification of integration within smaller purported modules (i.e., oral, zygomatic) and within modules obtained through a maximum modularity test in MorphoJ. Values of eigenvalue variances of purported cranial modules were compared within species, to determine whether a hierarchical organization of the cranium exists (test of the cranial modularity hypothesis), and between species and sexes, to test the FMP and the HET hypotheses. Eigenvalue variances were computed on configurations excluding redundant bilateral landmarks for the following modules: oral (K (number of landmarks) = 8), zygomatic (K = 5), rostral (K = 8), molar (K = 5), basicranial (K = 8), facial (K = 13), cranial (K = 20) and maximum cranial modularity blocks (number of landmarks differs depending on the species, see results on maximum modularity). The specific prediction of the cranial modularity hypothesis is that since in a modularly organized structure integration within modules is higher than integration between modules, the integration magnitudes of at least some of the purported facial modules (e.g., molar, oral, zygomatic, rostral), will be higher than integration magnitudes within the entire face, while the face and the basicranium themselves are expected to individually be characterized by higher magnitudes than the cranium as a whole, which includes both of them.

The analysis aiming at comparing integration magnitudes among species calculated the eigenvalue variance (EV) from a pooled-within sex correlation matrix obtained from the Procrustes coordinates residuals from a regression on centroid size while controlling for sex and subspecies. The analysis comparing differences in integration between the sexes calculated the EV from a correlation matrix from the residuals from a regression on centroid size while controlling for subspecies. In structures characterized by weak correlations between variables, the variance in the dataset will be distributed across many orthogonal axes of covariation each explaining a limited proportion of overall variance, resulting in a low EV. Conversely, in highly integrated structures numerous traits will covary strongly, thus the majority of the variance will be summarized by a small number of orthogonal axes accounting for most of the total covariation, resulting in a high EV [154], [155].

A separate Procrustes superimposition, regression on centroid size, and correlation matrix was obtained for each module, species or sex. For each EV, a 95% confidence interval (95% CI) was obtained by resampling with replacement of the dataset for 1000 iterations, and a correlation matrix was obtained for each replicate. If the correlation between the actual correlation matrix and the correlation matrix of a replicate was greater than 0.95 (and greater than 0.90 for the cranial configuration analysis), this replicate was used to compute an integration index. All computations were performed in R.

5. Comparison of integration indices at a common level of sampled population variance.

Finally, an alternative approach founded on the rationale behind the use of eigenvalue variances was implemented to quantify and compare integration magnitudes across species and sexes at a common level of sampled population variance. The method developed by Young et al. [156] was implemented. Young et al. [156] found that if two traits are highly correlated, then increasing the sampled variance will improve their estimated correlation (resulting in a positive association between measures of overall covariation among measurements and the average trait coefficient of variation (CV), whereas if the correlation is low, then increasing the sampled variance will not improve correlation estimates. The implication of Young et al.’s [156] finding for the comparative analysis of integration is that when the covariance structures of the samples being compared are broadly similar, this relationship can be taken into account in order to make appropriate comparisons among integration magnitudes at a common level of sampled population variance. On the other hand, if no relationship exists between correlation and average trait variance (measured in this study by the average trait coefficient of variation), then it is reasonable to infer that the examined variables (inter-landmark distances from the right half of the cranium in this study) are not strongly and thus consistently correlated.

To correct for variance artifacts in integration, each species was resampled with replacement 1000 times and for each replicate a pooled within-sex correlation matrix was obtained. If the correlation between the actual correlation matrix and the correlation matrix of a replicate is greater than 0.95, this replicate was used to compute an integration index and an average trait CV. The generated sample of replicates was used to plot the relationship among the values of the two computed parameters, and trace its 95% confidence ellipses. Thus, the 95% confidence intervals of the integration indices can be compared inter-specifically in view of their associated 95% intervals of the average trait CV. The coefficient of variation of the eigenvalues (ICV) [157], [158], [149] calculated as the square root of the variance in eigenvalues divided by the mean eigenvalue was the preferred integration index.

ICVs were calculated for the cranial, the facial, the basicranial, the oral, the zygomatic, the rostral and the molar blocks. For each block, data used for the ICV computation included all inter-landmark distances from the right side of the cranium obtained from the 3D coordinates transformed into residuals from pooled within-groups regression on centroid size (regression pooled within subspecies and sex to test the FMP hypotheses, and within subspecies to test of the HET hypotheses).

ICV and EV are only partially redundant measures of integration. ICV differs from EV in that 1) the use of inter-landmark distances allows comparing integration magnitudes at a particular average trait CV level, and in that 2) the use of all inter-landmark distances that can be obtained from a set of landmarks results in many highly correlated measurements sharing common endpoints and redundantly measuring identical cranial aspects, which does not allow a comparison of integration magnitudes between blocks.

It was noticed that resampling with replacement produces a distribution of integration indices whose corresponding mean is always higher than the actual ICV or EV mean associated with a particular species or sex, and whose upper and lower 95% confidence limits are often higher than the ICV and the EV means of the actual sample. This is expected because resampling with replacement creates bootstrap replicatesin which single individuals are sampled more than once which leads to higher correlations.

An adjustment of the 95% confidence intervals of the ICVs and EVs was performed to include the actual sample mean within these intervals. The 95% confidence intervals adjustment proceeded by the calculation of the difference between the bootstrap sample mean and the actual mean for the ICVs, the EVs and the average trait CVs, followed by the subtraction of this difference from the 95% confidence interval obtained from independent resampling with replacement for 1000 iterations for each species or sex. The results from 95% confidence intervals adjustment are presented in the Supplementary Information and discussed in the “Results” section of the paper.

1. Facial integration magnitudes as measured by the RV coefficient.

In all five capuchin species, the PLS analyses carried out on shape coordinates containing allometric shape change or on residuals from a regression on centroid size, be it in the framework of an oral-zygomatic modularity hypothesis or in the framework of a maximum facial modularity hypothesis, yield significant RV coefficients (used here as a multivariate generalization of the squared correlation coefficient), indicating that the permutation test against the null hypothesis of facial units independence is falsified, and that the face represents an integrated structure. Furthermore, the apelloid capuchins are characterized by higher between-block RV coefficients than C. albifrons and C. olivaceus under the oral-zygomatic PLS containing an allometric component or carried out on residuals regressed on centroid size, and under the maximum facial modularity PLS containing an allometric component.

In particular, the overall strength of association between the oral and the zygomatic allometry-containing blocks of the three apelloid capuchins is higher than this of the gracile capuchins by approximately 0.20 (Cebus apella s.s.: RV = 0.59 (p<0.0001); Cebus libidinosus: RV = 0.57 (p<0.0001); Cebus nigritus: RV = 0.53 (p<0.0001); Cebus albifrons: RV = 0.39 (p<0.0001); Cebus olivaceus: RV = 0.38 (p<0.0001)) (Table 3). When allometric variation in the landmark coordinates is preserved, in all capuchins, the landmark subdivision yielding the smallest between-module integration includes a unit associating the three molar landmarks with the TMJs, and variably including landmarks near the zygomatic roots (especially zygomaxillare superior in all species but C. apella s.s.) (Figure 2). It is interesting to note that the molar-TMJ module of C. olivaceus contains all zygomatic landmarks, while this of C. albifrons contains only zygomaxillare superior. Contrarily to the situation in the apelloids, no rostral landmarks are strongly integrated with the molar-TMJ module in both gracile species. This observed independence of 1) the molar-TMJ module from the oral-zygomatic module in C. albifrons and 2) the molar-TMJ-zygomatic module from the rostral module in C. olivaceus, contrasts with the existing inter-relationships and coordinated shape change between the molar-TMJ, the rostral and the zygomatic modules observed in the apelloids, and indicates that, in the latter group, the face as a whole is a relatively integrated structure, at least under an allometric maximum modularity scenario.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Two-block partitions yielding the minimum between-block covariance (maximum cranial modularity partition).

https://doi.org/10.1371/journal.pone.0040398.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Distribution of facial integration magnitudes under an allometric PLS, as measured by the RV coefficient.

https://doi.org/10.1371/journal.pone.0040398.t003

When allometric shape change is preserved, the overall strength of between-block association under the maximum modularity hypothesis (not limited to spatially contiguous subsets of landmarks) is higher in the tufted capuchins compared to the gracile capuchins (RV (C. apella s.s.) = 0.40 (p<0.0001); RV (C. libidinosus) = 0.33 (p<0.0001); RV (C. nigritus) = 0.31(p<0.0001); RV (C. albifrons) = 0.22 (p<0.0001); RV (C. olivaceus) = 0.23 (p<0.0001)).

In a PLS based on an oral-zygomatic block subdivision after accounting for allometric variation, the RV coefficients of all apelloid capuchins are also higher than these of the gracile capuchins, although the numerical distinction between these groups is more limited in comparison with the allometric scenario (RV (C. apella s.s.) = 0.383 (p<0.0001); RV (C. libidinosus) = 0.404 (p<0.0001); RV (C. nigritus) = 0.41 (p 0.0001); RV (C. albifrons) = 0.33 (p<0.0001); RV (C. olivaceus) = 0.32 (p<0.0001)) (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Distribution of facial integration magnitudes under a non-allometric PLS, as measured by the RV coefficient.

https://doi.org/10.1371/journal.pone.0040398.t004

When controlling for allometry, the landmark subdivision yielding the smallest between-module integration is identical in all species except for C.olivaceus, and includes a unit associating the three molar landmarks with the palatal landmark and zygomaxillare superior (Figure 3). In C. olivaceus, the three molar landmarks are associated with the palatal landmark and with the landmark at the meeting of the canine and the lateral incisor at the alveolus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Sets of landmarks producing blocks that minimize the between-block covariance (maximum facial modularity partition).

https://doi.org/10.1371/journal.pone.0040398.g003

Thus, under both the allometric and the non-allometric scenarios, the oral region does not represent an individualized module, and instead can be broken down into two distinct modules: a rostral module and a molar module.

The non-allometric maximum modularity PLS analysis differs from the three previous analyses in that it yields RV coefficients that do not allow for a distinction between tufted and gracile capuchins (C. apella s.s.: RV = 0.26 (p<0.0001), C. libidinosus: RV = 0.30 (p<0.0001), C. nigritus: RV = 0.24 (p<0.0001), C. albifrons: 0.24 (p<0.0001), C. olivaceus: RV = 0.26 (p<0.0001)) (Table 4).

In summary, non-allometric shape variation in the oral and the zygomatic blocks is coordinated to a greater extent in the apelloids, but correlation magnitudes do not support the existence of integration differences based on dietary groups under a PLS subdivision yielding minimal between-module covariation.

2. Cranial integration magnitudes as measured by the RV coefficient.

Under a face-cranial base PLS landmark partition including allometric variation, the RV coefficients of all apelloid capuchins are also higher than these of the gracile capuchins. The overall strength of association between the facial and the basicranial blocks is higher in the tufted species and, for each species, equal or slightly higher than its corresponding oral-zygomatic RV (C. apella s.s.: RV = 0.64 (p<0.0001); C. libidinosus: RV = 0.62 (p<0.0001); C. nigritus: RV = 0.52(p<0.0001); C. albifrons: RV = 0.41(p<0.0001); C. olivaceus: RV = 0.44 (p<0.0001)) (Table 3).

The maximum cranial modularity partition for data preserving allometry, in all species, involves a block including the molar, the lateral basicranial and the foramen magnum landmarks, and a second block including all but one of the zygomatic landmarks, excluding in four of the five cases zygomaxillare superior and all (or most in the case of C. nigritus) of the rostral landmarks. The details of this pattern vary (Figure 2). In all tufted capuchins, some midline basicranial landmarks (sphenobasion or the anterior tips of the petrous pyramid) are excluded from the molar-cranial base block.

Under a maximum modularity PLS partition of coordinates preserving allometry, the RV coefficients of all apelloid capuchins are also higher than these of the gracile capuchins (C. apella s.s.: RV = 0.42 (p<0.0001); C. libidinosus: RV = 0.36 (p<0.0001); C. nigritus: RV = 0.35 (p<0.0001); C. albifrons: RV = 0.25 (p<0.0001); C. olivaceus: RV = 0.27 (p<0.0001)) (Table 3). When preserving allometric variation, the overall strength of association between the maximum modularity cranial blocks in all capuchin species is slightly higher than the RV coefficients under the facial maximum modularity subdivision.

The modular architecture obtained from the analysis of non-allometric data is more variable. In all tufted species, the anterior block always includes all rostral landmarks and all zygomatic landmarks (without zygomaxillare superior in C. nigritus), while the posterior block includes most or all basicranial landmarks (along with the molar landmarks in C. apella s.s. and C. nigritus but not in C. libidinosus). In C. albifrons, the maximum modularity analysis creates a relatively good partition between the face and the basicranium, with a block including all facial landmarks without the palatal landmark, and opisthion. An almost complete face-cranial base separation exists in C. libidinosus as well. In C. olivaceus, the face is not an individualized module; instead, all dental landmarks (molar and rostral), the inferior zygomatic root and all lateral basicranial landmarks are combined in a module, while all midline basicranial landmarks group with the rest of the zygomatic landmarks and the palatal landmark.

The strict face-cranial base PLS on non-allometric data produces RV coefficients, which are higher in all apelloid capuchins and C. olivaceus compared to that of the gracile C. albifrons. The overall strength of association between the facial and the basicranial blocks in the capuchin species is as follows: C. apella s.s.: RV = 0.28 (p<0.0001); C. libidinosus: RV = 0.27 (p<0.0001); C. nigritus: RV = 0.24 (p = 0.0001); C. albifrons: RV = 0.20 (p = 0.0004); C. olivaceus: RV = 0.29 (p<0.0001) (Table 4).

In each species, the non-allometric face-cranial base RVs are numerically very similar to the maximum modularity RVs, indicating that the face-cranial base landmark partition optimizes cranial modularity relatively well. The RV coefficients from the PLS, which maximizes modularity are statistically significant and slightly lower than these of the face-cranial base subdivision, and also indicate that C. albifrons has the least integrated face (C. apella s.s.: 0.231(p<0.0001), C. libidinosus: 0.263 (p<0.0001), C. nigritus: 0.224(p<0.0001), C. albifrons: 0.196(p = 0.0003), C. olivaceus: 0.264 (p<0.0001)). The proportions of partitions yielding RV coefficients smaller than the face-cranial base partitions are very small (C. apella s.s.: 0.0003, C. libidinosus: 0.000016, C. nigritus: 0.000008, C. albifrons: 0.000008, C. olivaceus: 0.000111).

The non-allometric face-basicranium RV values are numerically very similar to both the maximum cranial modularity RV values and the RV values produced by PLS on a non-allometric facial maximum modularity partition subdividing the face into a molar-palatal block and a rostral-zygomatic block. A comparison of RV coefficients derived from facial and cranial maximum modularity PLS analyses carried out on residuals from regression on centroid size, however indicates that between-block correlation is slightly stronger between the facial modules rather than between the cranial modules in all species but C. olivaceus. This finding suggests that the face is not a very strongly individualized module within the cranium, but suggests that the face may be slightly more integrated than the cranium as a whole. Analyses presented in the following sections of the article assessing statistical significance of differences in integration magnitudes between blocks allow firmer conclusions.

In addition to the suggested covariation between the facial and cranial measurements not containing allometry, allometric variation further increases integration between the face and the basicranium through correlation between the molar landmarks, the lateral basicranial landmarks and foramen magnum landmarks.

3. Facial integration magnitudes of the allometric PLS1 patterns.

Under an allometric scenario, the first singular axes (PLS1) of the oral-zygomatic and maximum modularity PLS are significant in all species and summarize size-related shape covariation between the blocks in all species with the exception of C. olivaceus (Figure 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Distribution of the individuals’ loadings along the oral, zygomatic and maximum facial modularity allometric PLS1s.

Abbreviations : SV = singular value, %covar. = percentage of between-block total squared covariance explained by the set of PLS1 axes, r = between-block correlation coefficient.

https://doi.org/10.1371/journal.pone.0040398.g004

The strength of association between the oral and zygomatic blocks is very high and does not allow for a distinction on the basis of diet (C. apella s.s.: r = 0.894 (p = 0.0008), C. libidinosus: r = 0.91 (p = 0.0005), C. nigritus: r = 0.85 (p = 0.0022), C. albifrons: r = 0.841 (p = 0.0008), C. olivaceus: r = 0.89 (p = 0.0004)). The PLS1 of C. albifrons is characterized by a non-significant between-block correlation under a maximum modularity hypothesis, while C. olivaceus possesses the highest significant between-block PLS1 association (C. apella s.s.: r = 0.85 (p<0.0001), C. libidinosus: r = 0.70 (p = 0.0006), C. nigritus: r = 0.63 (p = 0.0016), C. albifrons: r = 0.55 (p = 0.1805), C. olivaceus: r = 0.875 (p = 0.0017)).

Despite the similarity in the magnitude of oral-zygomatic allometrically-driven correlations, PLS1 explains a smaller percentage of the total squared between-block covariance in the gracile capuchins (C. apella s.s.: 82.3%, C. libidinosus: 74%, C. nigritus: 69%, C. albifrons: 54.2%, C. olivaceus: 56%), indicating that allometric variation plays a more prominent role in the apelloids.

Under a maximum modularity hypothesis, the percentage of the total squared between-block covariance explained by PLS1 is smaller and particularly small in C. albifrons (C. apella s.s.: 69.7%, C. libidinosus: 53.05%, C. nigritus: 56.1%, C. albifrons: 31.5%, C. olivaceus: 55%). Thus, allometric shape change is less canalizing in C. albifrons’s face as it accounts for less covariation.

4. Facial integration magnitudes of the non-allometric PLS1 patterns.

All non-allometric oral-zygomatic and maximum modularity PLS1 axes are statistically significant (Figure 5). In all apelloid species and in C. olivaceus, the oral-zygomatic correlation coefficients are quite high and higher than the allometric oral-zygomatic PLS1 correlations (C. apella s.s.: r = 0.81 (p = 0.021), C. libidinosus: r = 0.91 (p<0.0001), C. nigritus: r = 0.93 (p<0.0001), C. albifrons: r = 0.65 (p = 0.23), C. olivaceus: r = 0.91 (p<0.0001)) (Figure 5). Noticeable is the non-significant oral-zygomatic correlation coefficient of C. albifrons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Distribution of the individuals’ loadings along the oral, zygomatic and maximum facial modularity non-allometric PLS1s.

Abbreviations : SV = singular value, %covar. = percentage of between-block total squared covariance explained by the set of PLS1 axes, r = between-block correlation coefficient.

https://doi.org/10.1371/journal.pone.0040398.g005

The maximum modularity correlation coefficients are statistically significant in all species, and lower than those of the allometric maximum modularity PLS1. C. nigritus, C. albifrons and C. olivaceus are characterized by the lowest correlations (C. apella s.s.: r = 0.84 (p = 0.0003), C. libidinosus: r = 0.804(p<0.0001), C. nigritus: r = 0.67 (p = 0.0098), C. albifrons: r = 0.70 (p = 0.0001), C. olivaceus: r = 0.73 (p = 0.0021)).

In all tufted species, the PLS1s under the non-allometric oral-zygomatic and maximum modularity partitions explain a percentage of the total squared between-block covariance that is approximately twice smaller than the allometric oral-zygomatic and maximum modularity PLS1s, respectively (Oral-zygomatic: C. apella s.s.: 42.3%, C. libidinosus: 33.5%, C. nigritus: 44.7%, C. albifrons: 34.7%, C. olivaceus: 42.2%; Maximum facial modularity.

C. apella s.s.: 60.7%, C. libidinosus: 44.9%, C. nigritus: 35.3%, C. albifrons: 41.04%, C. olivaceus: 43.5%). No numerical trend in the percentages of explained covariation distinguishes the tufted from the gracile capuchins.

5. Cranial integration magnitudes of the allometric PLS1 patterns.

The PLS1 axes under both the face-cranial base and the maximum modularity scenarios of all species are significant (Figure 6). The face-cranial base and the maximum modularity PLS1 correlation coefficients are statistically significant and high in all species, and slightly higher in the apelloids or in the apelloids - C.olivaceus group (Face-Cranial Base PLS1: C. albifrons: r = 0.83 (p = 0.0004), C. olivaceus: r = 0.83 (p<0.0001), C. apella: r = 0.89 (p<0.0001), C. libidinosus: r = 0.89 (p<0.0001), C. nigritus: r = 0.85 (p<0.0001), Maximum modularity PLS1: C. albifrons: r = 0.69 (p<0.0001), C. olivaceus: r = 0.73 (p = 0.0002), C. apella: r = 0.81 (p<0.0001), C. libidinosus: r = 0.76 (p<0.0001), C. nigritus: r = 0.73 (p<0.0001)).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Distribution of the individuals’ loadings along the facial, basicranial and maximum cranial modularity allometric PLS1s.

Abbreviations : SV = singular value, %covar. = percentage of between-block total squared covariance explained by the set of PLS1 axes, r = between-block correlation coefficient.

https://doi.org/10.1371/journal.pone.0040398.g006

The proportion of explained squared between-block covariance is slightly higher in the apelloids (Face-Cranial base: C. albifrons 81.2%, C. olivaceus 79.3%, C. apella 92.4%, C. libidinosus 87.9%, C. nigritus 89%, Maximum modularity: C. albifrons: 60.92%, C. olivaceus: 54.2%, C. apella: 85.5%, C. libidinosus: 77.6%, C. nigritus: 76.5%) (Figure 6).

6. Cranial integration magnitudes of the non-allometric PLS1 patterns.

The non-allometric PLS1 axes of all species under both PLS partitions are significant (Figure 7). Compared to the other capuchins, the non-allometric face-cranial base PLS1 of C. albifrons explains a smaller percentage of between-block covariance (C. albifrons: 35.7%, C. olivaceus: 46.1%, C. apella s.s.: 41.4%, C. libidinosus: 44%, C. nigritus: 45.9%) and has the lowest between-block correlation coefficient (C. albifrons: r = 0.63 (p = 0.0479), C. olivaceus: r = 0.80 (p = 0.0012), C. apella s.s.: r = 0.784 (p = 0.023), C. libidinosus: r = 0.80 (p<0.0001), C. nigritus: r = 0.69 (p = 0.0017)).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. Distribution of the individuals’ loadings along the facial, basicranial and maximum cranial modularity non-allometric PLS1s.

Abbreviations : SV = singular value, %covar. = percentage of between-block total squared covariance explained by the set of PLS1 axes, r = between-block correlation coefficient.

https://doi.org/10.1371/journal.pone.0040398.g007

Under the non-allometric maximum cranial modularity scenario, the proportion of explained between-block covariance contained in the PLS1axes does not differentiate any of the gracile species from the rest of the capuchins (C. albifrons: 34.1%, C. olivaceus: 36.7%, C. apella: 30.3%, C. libidinosus: 41.6%, C. nigritus: 39%). The non-allometric PLS1 between-block correlation coefficient is not statistically significant at the 0.05 level in C. nigritus and is marginally significant in C. albifrons (C. albifrons: 0.59 (p = 0.0596), C. olivaceus: 0.69 (p = 0.0003), C. apella: 0.66 (p = 0.003), C. libidinosus: 0.78 (p = 0.0001), C. nigritus: 0.58 (p = 0.058)). The high correlation coefficient of C. olivaceus is not surprising, because each of its blocks includes many landmarks from both the face and the cranial base. C. libidinosus and C. albifrons are characterized by very similar maximum modularity partitions in which the face and the cranial base are well distinct. Yet, consistent with the H_(FMP1) hypothesis, C. libidinosus’ correlation coefficient is higher by 0.18 units. Thus, compared to C. libidinosus, C. albifrons has a more modular cranium. Similarly, the maximum modularity partitions of C. nigritus and C. apella s.s. are very similar and both include all molar landmarks, the palatal landmarks and all basicranium landmarks with the exception of opisthion. Yet, between-block correlations indicate that C. nigritus, compared to C. apella s.s., is characterized by less coordinated variation between the molar-basicranial unit and the rostral-zygomatic unit. C. nigritus has been reported to include a high proportion (up to 73.6%) of leaves in its diet during food scarcity periods, which might imply that selective pressures related to the opening of hard-shelled nuts with the anterior dentition and associated biomechanical constrains on facial anatomy are lower in this species [113], [115].

7. The role of allometry in facial PLS1.

Shape change contained in the oral and zygomatic PLS1 correlates significantly (p<0.0001) with centroid size in all species. Furthermore, centroid size has a higher predictive power on the variation in each block in the tufted species. (Oral, % predicted: C. apella s.s. = 60.76%; C. libidinosus = 70.3%; C. nigritus = 63.5%; C. albifrons = 49.3%; C. olivaceus = 50.5%); Zygomatic, % predicted: C. apella s.s. = 66.32%; C. libidinosus = 79.5%; C. nigritus = 63.7%; C. albifrons = 61.9%; C. olivaceus = 50%).

Under a maximum modularity scenario which minimizes allometrically-driven between-block correlation, in all species except for C. olivaceus, the PLS1 axis of each block is significantly correlated with centroid size, and in all instances the molar-TMJ module contains a larger proportion, approximately two-thirds of the allometric variation, than the module containing the rest of the landmarks (% of variation predicted by centroid size: C. apella s.s.: 69%(p<0.0001), C. libidinosus:77% (p<0.0001), C. nigritus: 66% (p<0.0001), C. albifrons: 60% (p<0.0001)). In C. olivaceus, centroid size predicts only 6.6% (p = 0.023) of the total variation contained in the PLS1 scores of the molar-TMJ-zygomatic module, while 36% (p<0.0001) of the shape change contained in the PLS2 scores of the molar-TMJ-zygomatic block is allometric. The C. olivaceus’ maximum facial modularity PLS2 explains a markedly smaller proportion of the total squared covariance contained in the landmark coordinates compared to the maximum facial modularity PLS1 (17% in PLS2 versus 55% in PLS1), but its shape change is identical to the allometric change contained in the PLS1 axes of the other species (see last section).

In all species except for C. olivaceus, shape change contained in the rostral-zygomatic PLS1 is also partly allometric and explains only approximately one third of the variation (% variation prediction by centroid size: C. apella s.s.: 37% (p<0.0001), C. libidinosus: 32% (p<0.0001), C. nigritus: 33% (p<0.0001), C. albifrons: 36% (p<0.0001)). In C. olivaceus, centroid size does not predict variation in the PLS1 scores of the rostral module (% predicted: 2.3% (p = 0.1813)). The PLS2 of the rostral module of C. olivaceus does not contain allometric variation either (% predicted: 1.11%, p = 0.36).

The results presented above demonstrate (a) that allometry plays a much less important role in C. olivaceus than in the other examined capuchins and (b) that the molar -TMJ module contains two times more size-related shape change than the rostrum. In fact, the allometric molar-TMJ shape change is among the most pronounced shape changes in the face even under the maximum modularity hypothesis PLS analysis.

8. The role of allometry in cranial PLS1.

In a face-cranial base partition, in all species, approximately two thirds to three fourths of the PLS1 variation in both blocks is allometric in origin. The PLS1 axes of the faces of the tufted capuchins appear to be slightly more accurately predicted by centroid size than the PLS1 axes of the faces of the untufted species (C. albifrons: 68.3%, C.olivaceus: 60.3%, C. apella: 74.2%, C. libidinosus: 82.4%, C. nigritus, 72.1%, (p<0.0001)). The proportions of variation in the PLS1 of the cranial base block explained by centroid size do not allow for a distinction between apelloids and graciles (C. albifrons: 68.6%, C. olivaceus: 59.5%, C. apella: 71.6%, C. libidinosus: 79.6%, C. nigritus, 59.7%, (p<0.0001)).

Under a maximum modularity scenario, in all species, 66 to 85 per cent of the variation contained in the PLS1 of the molar-cranial base block is allometric in origin (% predicted by centroid size: C. albifrons: 74.44%, C. olivaceus: 66.8%, C. apella s.s.: 77.9%, C. libidinosus: 84.8%, C. nigritus: 69.2%, (p<0.0001), while only between 35 and 60 per cent of the variation contained in the rostral-zygomatic block can be accurately predicted by centroid size (% predicted by centroid size: C. albifrons: 36.1%, C. olivaceus: 59.5%, C. apella s.s.: 54%, C. libidinosus: 56.6%, C. nigritus, 38.1%, p<0.0001). In all species with the exception of C. olivaceus, the molar-cranial base block contains at least 24% more allometric variation than the rostral-zygomatic block.

The overall allometric variation in the cranial 3D landmark coordinates also indicates that allometric change is more canalizing in the tufted species (C. apella s.s.: 26.7%, C. libidinosus: 27.2%, C. nigritus: 22%, C. albifrons: 17.44%, C. olivaceus: 16.5% (p<0.0001)).

9. Comparison of facial and cranial allometric and non-allometric PLS1 and PC1.

In contrast with the high percentage of the total between-block covariation of PLS1 explained by allometry, allometry in the first facial and cranial PC1 axes accounts for a small percentage of overall variation in the dataset that is smaller in the gracile species (Table 5). Compared to the allometric facial PC1, a larger proportion of variation contained in the cranial PC1 is predicted by centroid size in all species (Table 5). Interspecific variation in the predictive power of centroid size on variation in the coordinates contained in the facial and cranial PC1s confirms the less prominent role of allometry in C. olivaceus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Variance explained and allometry contained in the first principal component axis of the facial and cranial configuration.

https://doi.org/10.1371/journal.pone.0040398.t005

The variance explained by the non-allometric facial PC1 is lower than the variance explained by the allometric facial PC1 in the apelloids but not in the gracile species. Variation contained in the non-allometric cranial PC1 is lower than variation contained by the allometric cranial PC1 in all species.

The allometric cranial PC1 accounts for slightly more variation than its corresponding facial PC1, while the non-allometric cranial PC1 is slightly lower than the non-allometric facial PC1.

10. PLS2: allometric and non-allometric facial integration magnitudes.

The PLS2 axes of all species are significant (under the allometric and the non-allometric oral-zygomatic and maximum modularity partition) (Table 6). All species possess very high and significant oral-zygomatic PLS2 correlations obtained from an analysis preserving allometry (note that it is the first axis which contains the allometric shape change) (C. albifrons: 0.93 (p<0.0001), C. olivaceus: 0.82 (p<0.0001), C. apella s.s.: 0.93 (p = 0.0052), C. libidinosus: 0.96 (p = 0.0005), C. nigritus: 0.93 (p<0.0001)). These high correlations are combined with relatively low proportions of explained squared covariance by the PLS2 axes (Table 6). Between-block correlations are lower under an allometric maximum modularity scenario and non-significant in C. olivaceus (C. albifrons: 0.80 (p<0.0001), C. olivaceus: 0.50 (p = 26), C. apella s.s.: 0.76 (p = 0.0052), C. libidinosus: 0.708 (p = 0.015), C. nigritus: 0.74 (p = 0.0003)).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Distribution of facial integration magnitudes as measured by the between-block correlation coefficient of PLS2.

https://doi.org/10.1371/journal.pone.0040398.t006

Under a non-allometric scenario, oral-zygomatic PLS2 correlations are not statistically significant at the 0.05 level in the two gracile species and in C. libidinosus (C. albifrons: 0.65 (p = 0.095), C. olivaceus: 0.67 (p = 0.19), C. apella s.s.: 0.93 (p = 0.0004), C. libidinosus: 0.74(p = 0.27), C. nigritus: 0.94 (p<0.0001)).

Under a non-allometric maximum modularity scenario, the PLS2 between-block correlations are numerically very similar in all species (C. albifrons: 0.67 (p = 0.0018), C. olivaceus: 0.67 (p = 0.007), C. apella s.s.: 0.67 (p = 0.009), C. libidinosus: 0.69(p = 0.007), C. nigritus: 0.63 (p = 0.1)).

11. PLS2: allometric and non-allometric cranial integration magnitudes.

In comparison with the facial PLS2, the allometric face-cranial base PLS2 axes of all species with the exception of C. libidinosus are not significant (Table 7). Furthermore, the allometric face-cranial base PLS2 correlations of C. apella s.s. and C. libidinosus are the only statistically significant correlations at the 0.05 level, while the correlation of C. albifrons is marginally significant and lower than these of the aforementioned apelloids (C. albifrons: 0.68 (p = 0.056), C. olivaceus: 0.71 (p = 0.13), C. apella s.s.: 0.80 (p = 0.0009), C. libidinosus: 0.80 (p = 0.0001), C. nigritus: 0.60 (p = 0.118)). Similarly to the results from the face, the allometric maximum cranial modularity PLS2s correlations are numerically very similar in all species (C. albifrons: 0.65 (p = 0.0047), C. olivaceus: 0.68 (p = 0.0017), C. apella s.s.: 0.70 (p<0.0001), C. libidinosus: 0.65(p = 0.024), C. nigritus: 0.59 (p = 0.166)).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Distribution of cranial integration magnitudes as measured by the between-block correlation coefficient of PLS2.

https://doi.org/10.1371/journal.pone.0040398.t007

The between-block correlation coefficients of the non-allometric PLS2s range between approximately 0.60 and 0.75, are significant in most species, and do not exhibit a difference in values that corresponds to dietary groupings (Table 7).

1. Unadjusted confidence intervals.

An interspecific comparison of the eigenvalue variances of the actual samples and their 95% confidence intervals obtained from a sample with 1000 bootstrap replicates indicates that the 95% CI of the facial eigenvalue variance of C. albifrons is significantly lower than these of C. apella s.s. and C. libidinosus. C. olivaceus’ facial EV 95% CI is almost identical to this of C. nigritus and overlaps with the 95% EV ranges of all species except for C. apella s.s. Both C. olivaceus and C. nigritus are characterized by significantly lower facial EVs than C. apella s.s. (Figure 10). These findings lend support to the H_(FMP1) hypothesis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 10. Variation in integration magnitude as measured by EV between species and between modules. A.

Distribution of actual EV values; B. Distribution of the 95% confidence intervals for EVs.

https://doi.org/10.1371/journal.pone.0040398.g010

The integration among the cranial landmarks is lowest in C. albifrons, followed by C. olivaceus, C. apella, C. nigritus and C. libidinosus. The cranial 95% CIs are non-overlapping between C. albifrons and the apelloids, and overlapping between C. olivaceus and all species except for C. libidinosus. The FMP1 trend is not observed in the rostral, molar and zygomatic facial subunits or within the basicranium.

A within-species comparison of eigenvalue variances from block to block allows defining the modular architecture of the cranium. The 95% CIs of the cranial EVs are non-overlapping with the facial EV CIs in all species but C. apella s.s., which is characterized by an exceptionally high facial integration. Thus, in all species but C. apella s.s., the cranium as a structure is significantly more integrated than the face. The largest difference in intensity between the facial and the cranial blocks is seen in C. libidinosus.

In C. olivaceus and C. albifrons, the zygomatic block has a very similar integration magnitude to the face (the zygomatic is slightly more integrated than or as integrated as the face under the actual eigenvalue variance comparison and the 95% CIs comparison), while in the apelloids the face is significantly (or marginally significantly in C. nigritus) more integrated than the zygomatic. The zygomatic 95% CIs are overlapping between species (except for the zygomatics of C. olivaceus and C. apella s.s.).

The 95% EV ranges of the molar, zygomatic and facial blocks overlap in all comparison pairs in the gracile species and in C. nigritus (although, as noted, the distinction between the facial and the zygomatic EVs in the latter species is marginally significant). The zygomatic and the molar 95% CIs EVs are higher than the oral EVs in all species, and significantly higher than the oral EVs in the gracile species, consistent with the finding that the oral block is composed by more than one module. The rostral-zygomatic 95% CIs for EVs are lower and non-overlapping with the molar 95% CIs in C. albifrons, C. apella s.s. and C. libidinosus. In C. olivaceus, the rostral-zygomatic EV 95% CI includes a range of values almost identical to its molar 95% CI. When the EVs from the actual species samples are compared, they indicate that the zygomatic and the molar blocks are more integrated than the rostral and the oral blocks.

The basicranial EV 95% CIs are lower and non-overlapping with those of facial 95% CIs in all species but C. libidinosus. Thus, in all species except for C. libidinosus, the face is significantly more integrated than the basicranium. The 95% CIs indicate that the basicranium is significantly less integrated than the molar block in C. albifrons, C. olivaceus and C. apella s.s. The 95% CIs for EVs of the basicranium overlap with these of the zygomatic block in the apelloids, while basicranial integration is significantly lower than zygomatic integration in the gracile species.

No specific trend is observed when interspecific variation in integration magnitudes within cranial modules maximizing modularity are examined (Figure 11), with the exception that although the cranial landmark partition is approximately identical in C. albifrons and C. libidinosus, both the facial and the basicranial modules of C. albifrons are significantly less integrated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 11. Variation in integration intensity of maximum cranial modularity blocks measured by EVs 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0040398.g011

2. Adjusted confidence intervals.

The adjusted EV values produce more important overlap between the 95% confidence intervals of different cranial blocks, but further demonstrate that the face is not more integrated than the cranium, that facial integration is particularly high and not significantly different from cranial integration in C. apella s.s., that all facial modules in C. apella s.s. are significantly less integrated than the face while in the rest of the capuchins the bootstrap distribution of the facial and molar blocks are almost identical (except for C. libidinosus in which the overlap is limited) (Figures S3).

The adjusted 95% confidence intervals for the EV values further suggest that the oral and the rostral blocks are significantly less integrated than the face and the cranium in all species (except for the rostral block of C. olivaceus), that the molar block 95% CIs are markedly higher and non-overlapping or marginally overlapping with the 95% CIs of the oral and the rostral blocks in all species except for C. olivaceus, that all purported facial sub-modules are significantly less integrated than the cranium in all species except for the zygomatic and the molar blocks in the untufted species and that the basicranium is significantly less integrated than the face and the cranium in all species except for C. libidinosus, in which the basicranial and the facial 95% confidence intervals overlap. It is further suggested that the basicranial 95% confidence interval generally extensively overlaps with the 95% confidence intervals of the oral and the rostral blocks (with the exception of the high rostral integration in C. olivaceus producing non-overlapping rostral and basicranial confidence intervals, and the high basicranial integration in C.libidinosus producing a basicranial confidence interval that does not overlap with the rostral and oral intervals). Finally, an examination of the adjusted EVs of the maximum cranial modularity blocks further confirms that the facial and the basicranial blocks of C. albifrons are characterized by significantly lower integration magnitudes than the corresponding blocks of C. libidinosus (Figure S4).

1. Allometric PLS1 shape change.

In all species, the distribution of the specimens’ scores in the PLS1(oral) - PLS1(zygomatic) shape space after controlling for the effect of subspecies, defines a male morph and a female morph. The males-to-females transformation follows an identical pattern in C. apella s.s., C. nigritus, C. libidinosus and C. albifrons.

In comparison with the average species pattern, the characteristics of the male pattern in the allometrically driven trait relationships include a narrower dental arcade at the level of the molars, more anteriorly shifted molars relative to the TMJs, relatively smaller cheek teeth and more posteriorly located incisor arc (Figure 19). Other characteristics defining this pattern include more laterally flaring zygomatic roots, a taller lower face, a slightly taller midface due to a more superior location of the infraorbital margin, and a higher position of the TMJs. The shape of the males’ zygomatic roots also places the zygomaxillare inferior in a more lateral position.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 19. Shape change contained in the allometric oral-zygomatic PLS1 and the allometric PC1.

Legend: oral-zygomatic PLS1 blocks colored in red and turquoise; PC1: landmark configuration colored in dark blue.

https://doi.org/10.1371/journal.pone.0040398.g019

In all tufted capuchin species, the males’ and females’ configurations along the allometry-containing PLS1 axes are distinct with little overlap between the sexes, unlike the situation in C. albifrons and C. olivaceus (Figure 4). This characteristic likely stems from the observed little overlap in centroid sizes between apelloid males and females, indicative of an important degree of facial dimorphism.

Each species’ extreme configurations define ranges in centroid sizes indicating that cranial size variation cannot account for the higher RV coefficients in the tufted species (centroid size ranges: C. apella s.s.: 95–127 units (range = 32), C. libidinous: 95–118 units (range = 23), C. nigritus: 95–119 (range = 24), C. olivaceus: 95–121 units (range = 26), C. albifrons: 87–114 units (range = 27). Thus, centroid size ranges are not wider in the apelloids compared to the gracile capuchins despite the more extensive overlap in cranial size between gracile males and females. In the same time, the crania of C. albifrons are on average smaller than the apelloid and the C. olivaceus crania, which indicates that the starting point for the allometric transformation in C. albifrons is different from this of the tufted species and C. olivaceus.

Despite the similarity in transformation patterns between C. albifrons and C. olivaceus and the apelloids, the morphology of the male and female configurations differs. Males belonging to the gracile species are characterized by more anteriorly projected rostrums relative to the average pattern in comparison with their apelloid counterparts (Figure 19). Consequently, rostral projection is more sexually dimorphic in the tufted capuchins.

The C. olivaceus PLS1 shape change pattern is very similar to this of the tufted species and C. albifrons. Contrarily to the other species, in C. olivaceus the PLS1 does not involve notable variation in the shape or position of the incisor arc, combined with posteriorly migrating TMJs. This observation is consistent with the high ICV and EV values of the rostral block of C. olivaceus, indicating little covariation with other units. The marked allometric covariation between the rostrum and the TMJ in the apelloids ensuring that both landmarks are concurrently displaced either anteriorly or posteriorly might be adaptive.

Compared to apelloid males and similarly to C. albifrons males, the zygomatic roots of C. olivaceus males are less laterally flaring, as indicated by the position of the zygomaxillare inferior, which creates a face, narrower than this of similarly-sized C. albifrons individuals.

Shape change contained in the allometric PLS1 under an oral-zygomatic subdivision and under a maximum modularity subdivision is very similar in each species (Figure 20). Differences between the two modularity scenarios are expressed in the magnitudes of the vectors at landmarks. In particular, shape change at the molar landmarks and the TMJs are the principal driver of covariation under the oral-zygomatic modularity hypothesis, while the maximum modularity scenario grouping the molar landmarks and the TMJs in a single block allows for the expression of rostral and zygomatic shape change.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 20. Shape change contained in the allometric maximum facial modularity PLS1.

https://doi.org/10.1371/journal.pone.0040398.g020

Slight distinctions in the orientation and the magnitude of the vectors at landmarks between the oral-zygomatic integration hypothesis and the maximum modularity scenario are noticeable despite the general similarity in pattern. In C. albifrons, C. apella, C. libidinosus and C. nigritus, the similarity between shape change under the oral-zygomatic hypothesis and the maximum modularity scenario is almost full. In C. olivaceus, the maximum modularity PLS2 contains a shape change more similar to the PLS1 of the oral-zygomatic patterning, and, as previously noted, also contains a higher proportion of allometric change (Figure 21). The C. olivaceus PLS2 pattern includes a molar segment of the larger individuals (there is a complete overlap between males and females) that is anteriorly shifted and slightly narrower; however, unlike the situation in the PLS1 of the other capuchins, the PLS2 extreme morph of the larger individuals does not have a taller lower face (molar alveoli not too inferiorly located), the infraorbital margins do not change their supero-inferior position, the superior zygomatic root and the zygomatic arch migrate inferiorly rather than superiorly, there is no variation in the position of the zygomaxillare inferior, and the TMJs are more posterior but not more medial. Non-allometric shape change associated with the maximum modularity PLS1 of C. olivaceus (under the allometric scenario) includes a posterior shift of the rostrum and more laterally flaring, slightly more anterior zygomatics, and slightly taller lower face and mid face.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 21. Shape change contained in the allometric oral-zygomatic PLS2.

https://doi.org/10.1371/journal.pone.0040398.g021

2. Non-allometric PLS1 shape change: comparison with the allometric PLS1.

There is a globally good correspondence between the allometric PLS1 and the non-allometric PLS1 of C. apella s.s. and C. albifrons, the allometric PLS1 and the non-allometric PLS2 of C. libidinosus and the allometric PLS1 and the non-allometric PLS3 of C. nigritus. In these species, the similarities between non-allometric patterns and the allometric PLS1 pattern are extensive and consist of the typical shortening of the jaws via a posterior displacement of the rostral landmarks and an antero-inferior and to an extent medial displacement of the molar landmarks, a superior displacement of the zygomatic landmarks, a posterior displacement of the infraorbital margin as measured by zygomaxillare superior (except for C. nigritus), and a superior displacement of the TMJs accompanied by a concomitant posterior shift in TMJs (except for C. libidinosus), a taller lower face (except for C. libidinosus) and a slight lateral displacement of the zygomatic landmarks (Figure 22).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 22. Shape change contained in the non-allometric oral-zygomatic PLS1 and the allometric PC1.

Legend: oral-zygomatic PLS1 blocks colored in red and turquoise; PC1: landmark configuration colored in dark blue.

https://doi.org/10.1371/journal.pone.0040398.g022

In the four capuchins (except for C. olivaceus), in comparison to the allometric PLS1 pattern, the non-allometric PLS shape change consists of less medio-lateral variation in the molar landmarks (with the exception of M1 in C. libidinosus), markedly less or no antero-posterior variation in the TMJ position, and a narrower molar arcade not associated with markedly medially displaced TMJs.

In the non-allometric oral-zygomatic PLS1 of C. nigritus, one of the extreme patterns is mostly characterized by a shorter lower face, by TMJs more superior to the occlusal plane and more posterior relative to the molars.

Non-allometric oral-zygomatic shape change contained in the PLS1 of C. olivaceus and C. libidinosus corresponds to shape change contained in the PLS2 of C. apella and C. nigritus. One of the extreme configurations combines an anterior displacement of the dental landmarks and nasospinale with a shorter lower face, a posterior and slightly medial displacement of the zygomatic landmarks and a supero-medial displacement of the TMJs (Figure 23).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 23. Shape change contained in the non-allometric oral-zygomatic PLS2.

https://doi.org/10.1371/journal.pone.0040398.g023

The maximum modularity non-allometric PLS1 contains a pattern reminiscent of both the maximum modularity allometric PLS1 and the oral-zygomatic non-allometric PLS1. In all species, the maximum modularity non-allometric pattern differs from the oral-zygomatic non-allometric pattern by a more exaggerated anterior displacement of the molars, more posterior zygomatic roots (with the exception of C. olivaceus), higher position of the zygomatic roots, and higher TMJs (Figure 24).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 24. Shape change contained in the non-allometric maximum modularity PLS1.

https://doi.org/10.1371/journal.pone.0040398.g024

One of the important conclusions to be drawn from these comparisons is that the displacements at the TMJs in the supero-inferior and antero-posterior directions, and thus gape, is governed by allometry. Furthermore, the non-allometric axes contain magnified shape change of the rostral and zygomatic landmarks, which is normally obscured by allometry at the expense of molar-TMJ variation. Overall, the absolute magnitude of shape change is higher along the allometric axis, in other terms there are individuals exhibiting more extreme degrees of the pattern.

3. Comparison between PLS 1 and PC 1.

Under an allometric scenario shape variation contained in the PC1 axes is almost indistinguishable from variation contained in the PLS1 axis (Figure 19).

Shape change contained in the non-allometric PC1 axis of C. apella, C. nigritus and C. albifrons is very similar to the pattern contained in the non-allometric PLS1 (Figure 22). The C. libidinosus PC2 shares extensive similarity with PLS1, and explains slightly lower percentage of the total variance contained in the landmark coordinates than PC1 (PC2∶ 13.13%).

In C. olivaceus, shape change contained in the non-allometric PLS1 is also fairly similar to the non-allometric PC2 shape change, which explains 12.43% of the total variance.

4. PLS2: allometric and non-allometric oral-zygomatic integration.

Under both the allometric and the non-allometric scenarios, shape change along PLS2 is of very low magnitude (Figure 21, Figure 23). In both the oral-zygomatic allometric PLS2 and the non-allometric PLS2, one of the extreme configurations of all species except for C. libidinosus includes a lengthening of the rostrum, a posterior migration of the zygomatic landmarks and the infraorbital margin and a medial migration of the TMJs. The allometric PLS2 of C. libidinosus contains similar shape change, while in its non-allometric PLS2 the displacement at the zygomatic landmarks is medial rather than posterior.

5. Shape changes within sex-specific samples.

As expected, shape change contained in the allometric PLS1 of several samples, such as C. albifrons males, C. apella males, C. apella females, C. libidinosus males and C. olivaceus males, is almost identical or fairly similar to this contained in the common allometric PLS1 (Figure 25, Figure 26). Similarity between the non-allometric PLS1 of males and the non-allometric PLS1 of females cannot be established (Figure 27, Figure 28).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 25. Shape change contained in the allometric oral-zygomatic PLS1 of males.

https://doi.org/10.1371/journal.pone.0040398.g025

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 26. Shape change contained in the allometric oral-zygomatic PLS1 of females.

https://doi.org/10.1371/journal.pone.0040398.g026

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 27. Shape change contained in the non-allometric oral-zygomatic PLS1 of males.

https://doi.org/10.1371/journal.pone.0040398.g027

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 28. Shape change contained in the non-allometric oral-zygomatic PLS1 of females.

https://doi.org/10.1371/journal.pone.0040398.g028

1. Shape change contained in the allometric PLS1 and its biomechanical consequences.

Shape change along the allometric PLS1 is similar in all species and differentiates a male cluster from a female cluster (Figure 29). The male extreme combines an anterior, inferior and medial migration of the molar teeth with anterior displacement of all zygomatic landmarks except for zygomaxillare superior which migrates posteriorly, a superior displacement of the anterior dentition and the subnasal segment, and a posterior, inferior and medial displacement of the landmarks from the middle cranial fossae and the midline basicranium, and a supero-medial displacement of the landmarks on the posterior cranial fossa. The final result is a male pattern combining a significantly narrower molar arcade and narrower cranial base, slightly more anteriorly located zygomatic roots, lower position of the molars resulting in a taller lower face, with slightly more posteriorly located infraorbital margins relative to the incisor teeth, posteriorly shorter jaws and a greater distance between the third molars and the landmarks found posteriorly to the sphenobasions, smaller petrous pyramids, sphenotemporal suture in a lower and more posterior position and posterior aspects of the pyramids and posterior end of foramen magnum in a higher position. The latter indicates that the posterior part of the cranial base faces more posteriorly rather than more inferiorly in larger crania.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 29. Shape change contained in the allometric face-cranial base PLS1 and the allometric PC1.

Legend: oral-zygomatic PLS1 blocks colored in red and turquoise; PC1: landmark configuration colored in dark blue.

https://doi.org/10.1371/journal.pone.0040398.g029

Differences in mean shape configurations among the species include more posteriorly located zygomaxillare inferiors, relatively flat in their supero-inferior dimension middle cranial fossae and a relatively higher position of the anterior midline basicranium in the apelloids species and C. olivaceus compared to C. albifrons.

Shape change between the two allometric integration partitions (face-cranial base and maximum cranial modularity is very similar in all species (Figure 29, Figure 30).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 30. Shape change contained in the allometric maximum cranial modularity PLS1.

https://doi.org/10.1371/journal.pone.0040398.g030

2. Shape change contained in the non-allometric PLS1.

Similar to the situation in the face, the non-allometric face-cranial base PLS1 pattern is partly similar to its corresponding pattern in the allometric PLS1 (Figure 31). In C. albifrons, C. apella s.s., C. nigritus and C. olivaceus, similarities include an antero-medial displacement of the molar landmarks, combined with an anterior displacement of the rostral landmarks, an antero-medial displacement of the zygomatic landmarks without zygomaxillare superior and a posterior displacement of the basicranial landmarks. In C. libidinosus, the displacement at the molars is postero-medial rather than antero-medial. Interspecific differences involve the supero-inferior orientation of the rostral landmarks and of the posterior petrous pyramid landmarks.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 31. Shape change contained in the non-allometric face-cranial base PLS1 and the allometric PC1.

Legend: oral-zygomatic PLS1 blocks colored in red and turquoise; PC1: landmark configuration colored in dark blue.

https://doi.org/10.1371/journal.pone.0040398.g031

There is a correspondence in patterns between the non-allometric face-cranial base and maximum modularity PLS1 scenarios in all species but C. apella s.s. and C. olivaceus (Figure 21, Figure 32). In C. apella s.s., the predominant shape change is concentrated at the inferior zygomatic root and the zygomaxillares, while the displacement at the molar landmarks is only lateral rather than postero-lateral. In C. olivaceus, the maximum modularity PLS1 summarizes a pattern, in which the predominant shape change is represented by an anterior displacement of the rostral landmarks and a posterior displacement of the zygomatic landmarks accompanied by a slight anterior movement of the rest of the landmarks. The C. olivaceus PLC2 contains shape change similar to this of the other species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 32. Shape change contained in the non-allometric maximum cranial modularity PLS1.

https://doi.org/10.1371/journal.pone.0040398.g032

3. Degree of similarity between the PLS1s and the PC1s.

A nearly perfect correspondence exists between the allometric face-cranial base and the maximum cranial modularity PLS1 and the allometric PC1 in all species (Figure 29, Figure 30). Shape change is exaggerated in PC1. The same rule holds for the non-allometric face-cranial base PLS1 and the non-allometric PC1s (Figure 31, Figure 32).

4. Shape pattern of the allometric and non-allometric PLS2.

In all species, the principal shape change contained in the allometric PLS2 creates an extreme morph combining an antero-medial displacement of all zygomatic landmarks (with the exception of the inferior zygomatic root in C. nigritus) (Figure 33). The displacement at the two zygomatic roots and zygomaxillare inferior is also superior, and it is combined with a smaller magnitude shape change concerning a superior displacement of the molars and an inferior displacement of the anterior basicranial landmarks including the sphenobasion, the pyramidal apex, the TMJs and the posterior point of the sphenotemporal suture (the latter is a high magnitude shape change in C. nigritus), and in some cases basion and the point at the junction between the anterior edge of the petrous pyramid and the external auditory meatus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 33. Shape change contained in the allometric face-cranial base PLS2.

https://doi.org/10.1371/journal.pone.0040398.g033

A common interspecific pattern is not obvious in the non-allometric PLS2 (Figure 34).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 34. Shape change contained in the non-allometric face-cranial base PLS2.

https://doi.org/10.1371/journal.pone.0040398.g034