On the Relationships of Postcanine Tooth Size with Dietary Quality and Brain Volume in Primates: Implications for Hominin Evolution

Brain volume and cheek-tooth size have traditionally been considered as two traits that show opposite evolutionary trends during the evolution of Homo. As a result, differences in encephalization and molarization among hominins tend to be interpreted in paleobiological grounds, because both traits were presumably linked to the dietary quality of extinct species. Here we show that there is an essential difference between the genus Homo and the living primate species, because postcanine tooth size and brain volume are related to negative allometry in primates and show an inverse relationship in Homo. However, when size effects are removed, the negative relationship between encephalization and molarization holds only for platyrrhines and the genus Homo. In addition, there is no general trend for the relationship between postcanine tooth size and dietary quality among the living primates. If size and phylogeny effects are both removed, this relationship vanishes in many taxonomic groups. As a result, the suggestion that the presence of well-developed postcanine teeth in extinct hominins should be indicative of a poor-quality diet cannot be generalized to all extant and extinct primates.

The sample analyzed of H. sapiens includes only Late Pleistocene specimens from Europe and the Near East (see Table S2).

Postcanine tooth size
Following previous studies, PCTA was estimated by summing the occlusal areas of three maxillary teeth, the fourth premolar (P 4 ) and the first and second molars (M 1 and M 2 ). The area of each tooth was obtained multiplying its mesiodistal length (MD) by its buccolingual breadth (BL). Most measurements were collected from the seminal work of Swindler [3] on the dentition of primates. The database was completed with metric data of Lemur catta [4] and Callitrichinae [5]. The measurements for fossil hominin crania were taken from several bibliographic sources (Table S2). The only exception was holotype LB-1 of Homo floresiensis, whose MD values for P 4 , M 1 and M 2 were measured on a photograph [ref. 46: Fig. 1]. It is worth making clear that after measuring the buccolingual dimensions of selected teeth for fossil Homo on published photographs [64,65], the differences with the measurements obtained from the bibliography were of <0.1 mm in all cases.
It could be argued, however, that the use of only three teeth of the postcanine dentition could introduce a source of error and bias in our analyses, except for those constructed within a narrow taxonomy. For example, if M 1 :M 3 size ratio is compared between platyrrhines and cercopithecoids, it is apparent that New World monkeys devote relatively little area to the M 3 in comparison to Old World monkeys. This is particularly evident in callitrichids, which mostly lack an M 3 , accounting the M 1 and M 2 for 100% of the molar area. In contrast, the M 3 represents a substantial proportion of total molar area in the catarrhines, reaching >33% in some species. This is a potential source of fluctuation for the slopes adjusted when different primate clades are involved (i.e., the slopes would not only reflect differences in tooth size relative to body size, but also tooth sizes relative to teeth not included in the calculation of postcanine tooth area). However, the reason for analyzing exclusively the area of P 4 , M 1 and M 2 is that McHenry [66] used these teeth for calculating the molarization coefficient and measurements for a high number of primate species are available. In contrast, the use of other postcanine teeth would have resulted in a substantial decrease of sample size. In addition, the first molar is usually placed at 7:10 of the jaw lever arm, the point where the maximum bite force is exerted during chewing, while the fourth premolar and the second molar are situated immediately before and after it, respectively, which means that these teeth are also crucial for mastication [66]. Finally, we are aware that teeth are complex structures and the use of postcanine tooth area oversimplifies a number of histological, morphological and topological aspects that play a key role in food processing. However, there is a vast number of meta-analyses in which this type of variables (i.e., body size, brain size and tooth size) have been used (e.g., for the specific case of postcanine tooth area [67].

Dietary Quality
Dietary quality (DQ) was defined as follows: where α is the percentage of leaves and structural parts of plants in the diet, β is the percentage of fruit and reproductive parts of plants (including nuts and seeds), and γ is the percentage of animal items. In this way, DQ reflects the dietary contribution of those foodstuffs considered in more traditional dietary categories (i.e., folivores, frugivores and faunivores). DQ was estimated for primate species using data on diet composition published in a number of studies (see Table S1). This variable measures some physical properties of foodstuffs, for example those derived from fiber contents, but others (e.g., toughness and hardness) are not clearly reflected in DQ values. However, DQ has been widely used in this type of approaches, including some recent studies [e.g., 69,70].
In addition, we must recognize that our analyses may oversimplify the relationship between tooth size and food type by focusing exclusively on dietary quality. The reason is that enamel thickness and tooth morphology and topology may reflect a balance between the need to fracture foods without fracturing the tooth itself: for example, the large teeth of Paranthropus could have been used for processing foodstuffs with quite divergent mechanical properties, which opens the possibility that tooth morphology reflects "fallback" food items rarely consumed instead of the most commonly masticated items [71][72][73][74].

Phylogenetic Control
The procedure used of Phylogenetic Generalized Least-Squares (PGLS) works as follows [75]: imagine eight species showing the phylogenetic relationship depicted in Figure S2. Two phenotypic traits (X, Y; for example, body mass and brain volume) has been measured in each of these species for analyzing the relationship between the traits. The problem is that it is not possible to assume that the measurements taken in any species will be independent from those obtained in others, as they may be subject to phylogenetic inertia.
The method of phylogenetic contrasts consists of transforming the original metric variables into new variables (CX, CY), which preserve the same covariance as the original ones. According to the tree topology of the hypothetical phylogeny analyzed (Fig. S1), the differences X1-X2, X3-X4, X5-X6, and X7-X8 will be independent of each other. For example, the difference X1-X2 will depend exclusively on the evolutionary events that took place in branches 1 and 2, while the difference X3-X4 will be the result of those that account for branches 3 and 4. Given that both sets of events are independent, this situation guarantees data independence for statistical analyses.
According to this reasoning and deepening into the tree nodes, we can affirm that the difference [(X1+X2)/2] -[(X3+X4)/2] is independent from the difference [(X5+X6)/2] -[(X7+X8)/2] and thus, the original variable X (with N species) is finally transformed in CX (with N-1 cases, dots in Fig. 1). The same procedure is applied to the variable Y for transforming it in CY. Given that the method ensures that cov(X,Y) = cov(CX,CY), this implies that if the original variables were correlated the new contrasts will show also a statistical correlation [for technical details, see ref. 76], but now it is guaranteed that the cases studied are independent (i.e., free of phylogenetic inertia). COMPARE calculates the relationship between traits, while also taking phylogeny and within-taxon variation into account. As with estimation of ancestral states, weights are a function of the within-species variation, the phylogeny, and the model of character evolution.
Among other things, the PGLS method implemented in COMPARE assumes that: 1. -Character evolution can be described using a model of phenotypic evolution, which leads to either a linear or an exponential increase in between-taxon divergence with phylogenetic distance. Exponential models are appropriate for evolution under some constraints (e.g., stabilizing selection about a fixed optimum). A linear model (e.g., Brownian motion) is approximated by setting the constraint to zero. This is more commonly used in population genetics for describing the evolution of characters undergoing random genetic drift, or directional selection when the direction of selection is shifting back and forth at random.
2. -Expected similarities and differences between all taxa (including hypothetical ancestors) are known. These similarities or differences will be obtained directly from the phylogeny as part of the calculations.
3. -Standard errors of the measurements for each taxon are available, and they adequately represent within-species variation in the trait.
4. -Relationships between traits are well described by a generalized least-squares model regression or correlation.