We collected 70 upper second molars from 32 species of extant anthropoid primates from the following institutions: collections of the iPHEP, Université de Poitiers, France; Muséum National d'Histoire Naturelle, Paris, France; Royal Museum of Central Africa, Tervuren, Belgium; Senckenberg Museum of Frankfurt, Germany. Of course, dental wear would decrease enamel thickness at the tip of the cusps and on dental wear facets. Hence, only juvenile specimens and subadults were selected, so that the enamel was characterized by a minimal level of dental wear.

We included 25 specimens of apes (Hominoidea), 32 specimens of Old World monkeys (Cercopithecoidea), and 13 specimens of New World monkeys (Platyrrhini). Species were selected in order to encompass as wide a range of diets as possible. When possible, they were also classified as stress-limited or soft food eaters sensu Lucas et al. (2000). To do so, we followed the methodology presented in Thiery et al. (2017) and combined reports of dietary composition, including seasonal variation in food item consumption, with studies on the physical properties of primates food. Whenever a species was reported to consume stress-limited food on a regular basis, we classified it as a hard food eater. If stress-limited food consumption was only reported as marginal despite several reports on diet composition for a given species, we classified it as a soft food eater. Species for which data were too scarce or contradictory were classified as undefined. This does not necessarily mean no data on their diet could be found. For instance, the diet of chimpanzees (Pan troglodytes) is detailed in numerous studies, but it was classified as undefined because there are some discrepancies between reports involving forest chimpanzees (Vogel et al., 2008) and savannah chimpanzees (Suzuki, 1969; Peters, 1993).

No living animal was involved in this work, and no animal was killed specifically for this study. Every crania used in this study belong to historical osteological collection gathered, for the most recent specimens, in the beginning of the 20th century. Each specimen has been collected more than one hundred years ago, hence no approval from an ethics committee was required.

Computing 3D enamel thickness requires access to the inner part of the tooth and the EDJ. Since the teeth used in this study come from valuable museum specimens of juvenile primates, a non-invasive method was mandatory. The teeth were scanned using x-ray high-resolution micro-computed tomography (HR-μCT) at the Centre de Microtomographie of Poitiers, France. Scans were acquired using an EasyTom HR-microtomograph. Isovoxel resolution spans from 10 to 30 μm depending on tooth size.

The resulting array of 2D slices was stacked to build a 3D reconstruction of the teeth. Both OES and EDJ surfaces were then extracted as polygonal 3D meshes using Avizo. Using Geomagic Studio, these polygonal meshes were re-tesselized into meshes composed of 55,000 triangles of normalized area, which removed scaling effects on triangle geometry. While resulting in a large decrease in the number of polygons used to describe the surface, this level of tessellation has been shown to describe dental surfaces as accurately as surfaces composed of a larger number of triangles (Lazzari and Guy, 2014).

Still using Geomagic Studio, OES, and EDJ surfaces were paired together and their orientation was standardized. The axis formed by the paracone-protocone dentine horn tips was aligned with the x-axis of the 3D space, and the surfaces were translated so that the lowest point of the molar cervix was set to z = 0. Following Guy et al. (2015), OES and EDJ occlusal surfaces were subsampled as the regions above a plane parallel to the (xy) reference plane and passing respectively by the lowermost point of (i) the occlusal enamel basin for the OES, and (ii) the enamel-dentine junction basin for the EDJ (Figure 2C).

Enamel thickness has been measured using both volumetric and geometric approaches. Volumetric average enamel thickness (AET) was computed as the quotient of 3D volume of the enamel cap over 3D area of the EDJ (Figure 2A):

For every triangle of the OES, geometric enamel thickness was computed as the minimal euclidean distance between each node of the OES to the closest triangle of the EDJ following normal direction (Figure 2B). Afterward, geometric AET was computed as the mean distance for the whole surface:

Be it volumetric or geometric, AET is a scale-dependant variable. Because our sample includes a wide size range, from the tiny common marmoset Callithrix jacchus to the largest living ape Gorilla gorilla, comparisons require a standardized estimation of enamel thickness. Following Martin (1985) and Kono (2004), we calculated the relative enamel thickness (RET) as the AET divided by the cubic root of the volume of dentine within the enamel capsule:

While some authors combined volumetric AET or RET with topographic maps of enamel-dentine distance, a comparison of volumetric and geometric thickness is yet to be done. We thus estimate the correlation between geometric and volumetric approaches for both AET and RET, using both Spearman's and Pearson's coefficients.

Along with traditional topographic maps, thick enamel distribution has been graphically characterized using pachymetric profiles. A pachymetric profile corresponds to a bivariate plot of enamel-dentine distances for each triangle of the mesh (y-axis) vs. their cumulated frequency (x-axis). To compare teeth regardless of differences in thickness range and/or in the number of triangles, each value is expressed as a percentage of the maximal value for both variables. This approach was inspired by hypsometric curves that are used to characterize the distribution of elevation in traditional topography (Schumm, 1956).

Afterwards, we used pachymetric profiles to characterize the evenness of enamel thickness distribution, hereby defined as the proportion of similarly-thick enamel over a tooth surface. While the very notion of evenness is qualitative, an evenly thick enamel is expected to have a large proportion of similar enamel-dentine distances. This is precisely this large proportion that would make the enamel look evenly thick on topographic maps. In terms of enamel distribution, this would result in a short thickness variation over a large number of triangles. Hence the following proposition: enamel distribution evenness is proportional to the slope of pachymetric profiles, which corresponds to the thickness vs. number of triangle variation. That is, the more evenly thick is enamel, the lower the slope is expected to be.

We computed the slope of the profile at average enamel thickness as the thickness variation between the points located 10 points away at both sides of the geometric AET:

Volumetric 3DAET ranges from 0.0838 to 0.9826 mm. On the other hand, geometric 3DAET ranges from 0.1169 to 1.1960 mm (Table 1). Volumetric 3DAET and geometric 3DAET show a significant linear correlation (ρ = 0.92; r² = 0.82; df = 68; p-value < 0.001) which is attested graphically by the distribution of the points in the geometric vs. volumetric 3DAET bivariate plot (Figure 3A).

Volumetric 3DRET ranges from 0.0890 to 0.2609, while geometric 3DRET ranges from 0.1407 to 0.3672 (Table 2). The correlation between volumetric 3DRET and geometric 3DRET is lower though significant (ρ = 0.84; r² = 0.67; df = 68; p-value < 0.001), which is reflected by the dispersion of the points in the bivariate plot of geometric vs. volumetric 3DRET (Figure 3B).

In the whole sample, qualitative assessment of thick enamel distribution evenness is strongly consistent with the slope of the pachymetric profile at the mean enamel thickness. When intermediate thickness values (usually green or yellow) are spread on topographic maps, the slope of the profile at mean enamel thickness is typically around 0.2; on the other hand, when the range of colors is wide and when extreme thickness values (dark red) are widespread, the slope of the profile at mean enamel thickness is higher, ranging between 0.6 and up to 1.5 (Figure 4).

This result is independent of enamel thickness itself, be it AET or RET, as thick enameled specimens may have a high slope at mean enamel thickness e.g., Cercocebus torquatus (Figure 4A) but also a low slope at mean enamel thickness e.g., Pongo pygmaeus (Figure 4B). Conversely, thin-enameled specimens may have a low slope at mean enamel thickness e.g., Colobus guereza (Figure 4A) but also a high slope at mean enamel thickness e.g., Ateles sp. (Figure 4C).

Thick enamel distribution does not seem to be more uneven in stress-limited food specialists, except in Old World monkeys (Figure 5). Durophagous Old World monkeys have a significantly higher slope (Kruskal-Wallis Analysis of Variance, H = 11.48; df = 2; p-value < 0.005). Note that the highest slope values in the “undefined food hardness” category are assigned to Cercopithecus diana, while the lowest values in the “undefined food hardness” category are assigned to C. cephus, C. pogonias and both Papio anubis and P. cynocephalus. In apes, thick enamel distribution is significantly more even in durophagous species (H = 12.66; df = 2; p-value < 0.05). New World monkeys show no significant difference between durophagous and non-durophagous species.

In addition, pachymetric profile slope was compared with the dispersion of enamel thickness computed as the coefficient of variation (CV) of geometric enamel thickness (Figure 5). Based on enamel thickness dispersion alone, stress-limited food consumers could not be separated from soft food consumers in any taxonomic group. Note however that in apes the “undefined” category had a significantly higher CV (H = 7.76; df = 2; p-value < 0.05).

While volumetric and geometric 3DAET are strongly correlated (r² = 0.82), correlation between volumetric and geometric 3DRET is lower (r² = 0.67). This lower correlation contrasts with the fact that both variables are expected to measure the same anatomical feature, that is, relative enamel thickness.

Several explanations can account for this difference. Firstly, the amount of enamel involved in the computation of 3DAET differs between the two methods. The volumetric approach divides enamel cap volume by the 3D surface area of the EDJ, which means that it is an estimate of the average volume of enamel per element of EDJ. Martin (1983) postulated that it is a good estimate of average enamel volume synthesized by a single ameloblast, since he expected the size of ameloblasts to be similar between small and large primates. In contrast, the geometric approach consists in measuring a Euclidean distance for only the ~20,000 points that compose the mesh of the occlusal portion. Thus, a portion of the enamel volume is not involved in the computation of geometric 3DAET, which might slightly affect the final result.

Secondly, the two methods do not measure thickness in the same direction. While the geometric approach measures thickness from the OES toward the EDJ, the volumetric approach measures thickness from the EDJ toward the OES. Because OES and EDJ are not perfectly concurrent, this might result in a slight variation of angle for every distance estimation, which in turn would affect average thickness.

Finally, standardization by the cubic root of dentin volume reduces the effects of allometry, which is indeed strong in our sample. This might in turn boost the existing thickness variability between the two methods, which might explain why the difference is more visible for 3DRET.

Nonetheless, since volumetric and geometric 3DRET are not perfectly correlated, the methodology selected to estimate enamel thickness is expected to influence the results. This is corroborated by the fact that the species with the greatest volumetric 3DRET (Lophocebus aterrimus) and the one with the greatest geometric 3DRET (S. apella) do not match (Table 2). The difference is especially visible for the latter (volumetric 3DRET = 0.1956; geometric 3DRET = 0.3672). Hence, the necessity to carefully select the method that is best adapted to one's investigation:

At least in the present study and assuming that pachymetric profile slopes are a good estimate of enamel thickness evenness, the hypothesis of an unevenly thick enamel in durophagous primates can be rejected except for Old World monkeys. Durophagous cercopithecids such as Cercocebus or Lophocebus are all characterized by high slopes and by an unevenly thick enamel (Figure 5). Furthermore, C. diana, the species from the “undefined” category with the highest pachymetric profile slopes, has been reported to consume a large proportion of seeds in both Bia (Curtin, 2004) and Tai Forest localities (Kane, 2012). In contrast, non-durophagous Old World monkeys such as Colobus are characterized by low slopes and by evenly thick enamel (Figure 5). This is particularly visible on the topographic map of C. guereza (Figure 4C). Cercopithecus cephus, the species from the “undefined” category with the lowest average pachymetric profile slope, has been reported to consume softer foods in both the localities of Makokou (Gautier-Hion et al., 1980) and Lopé (Tutin et al., 1997; Tutin, 1999). Note however that C. nictitans, which is characterized by a high profile slope, and C. pogonias, which is characterized by low profile slopes (Figure 5), have both been reported to consume a large proportion of seeds at the Makandé location (Brugière et al., 2002). On the other hand, P. anubis and Papio cynocephalus are also characterized by low profile slopes even though they might regularly consume challenging underground storage organs (Dominy et al., 2008).

Concerning durophagous apes, the orangutan (P. pygmaeus) is reported to consume very challenging, stress-limited food such as the seeds of Mezzetia parviflora (Vogel et al., 2008; Lucas et al., 2012). Taking the hypothesis of Lucas et al. (2008) into account, its profile slope is therefore expected to be high, which is not the case (Figure 5). The enamel of orangutan appears to be evenly thick. As such, topographic maps of enamel thickness show a large proportion of intermediately-thick enamel (yellow polygons) but a small proportion of very thick enamel (red polygons; Figure 4B). In other words, the pachymetric profile drifts toward thicker values, which results in a flatter curve and values closer to the thick 3DAET characteristic of this species (Vogel et al., 2008; this study).

A similar trend is observed in the saki (Pithecia pithecia), a notorious seed-eating New World monkey (Norconk and Veres, 2011). Pachymetric profiles of P. pithecia are thus characterized by a low slope at mean enamel thickness (Figure 5). Topographic maps of enamel thickness for this taxon present, in relative terms, more intermediately-thick enamel (yellow polygons) and less thick enamel (red polygons) than the maps of other New World monkeys (Figure 4C). Given its low AET values, the enamel of P. pithecia can be thus described as evenly thin.

In orangutan and sakis, the even enamel distribution might result from the presence of crenulations on the occlusal surface of molars (Figures 4B,C). Several functional interpretations have been proposed for these crenulations, including a better grip for the manipulation of slippery hard food e.g., seeds of juicy fruit (Lucas and Teaford, 1994) or multiplication of contact points which would improve the ability to fracture fibrous seeds (Lucas and Luke, 1984; Vogel et al., 2008). Our observation is consistent with both interpretations, since a multiplication of contact points with stress-limited food could result in a multiplication of locally thick-enameled structures, which would ultimately make the whole enamel look evenly thick. Still, this assumption requires further investigation, as both P. pygmaeus and P. pithecia are only represented by a couple of specimens in the sample.

Similarly, S. apella and to a lesser extent Cebus capucinus are also characterized by relatively low profile slopes despite being known to consume stress-limited food while having no crenulations (Freese and Oppenheimer, 1981; Terborgh, 1983; Galetti and Pedroni, 1994; but see Mosdossy, 2013). In both species though, topographic maps of the enamel thickness present a very different aspect, with only a small proportion of thick enamel on the hypocone, their enamel being evenly thick over the rest of the tooth crown (G. Thiery, pers. obs.). Enamel is evenly to unevenly thin in the soft food eater Ateles sp. (Figure 5), although uneven distribution probably comes from a thick lateral enamel on the functional cusps (Figure 4C). Overall, several modalities of evenly thick or thin enamel distribution are present in New World monkeys.

Our results imply that the distribution of thick enamel follows different patterns, possibly from one family to another. This might indicate that primates have developed different durophagous strategies to answer the selective pressures exerted by stress-limited food. It also suggests that characterizing enamel thickness distribution requires a phylogenetic context, especially when making dietary inferences for extinct species, since such inferences can not be confronted to behavioral data.

Nonetheless, a feature that was not taken into account is the feeding action performed to access or process stress-limited food, which needs to be considered when evaluating the form-function relationship between diet and dental morphology (Thiery et al., 2017). For instance, P. pithecia does not crack open the most challenging food it consumes with its molars, but with its strong and proclive incisors and canines (Kinzey and Norconk, 1993; Norconk and Veres, 2011). The seeds it crushes with its molars might therefore be tough, but they are significantly softer (Kinzey and Norconk, 1993). While this might have affected the results for New World monkeys, we assume this is not the case for apes since P. pygmaeus is known to use its molars to crush the shells of stress-limited foods (Lucas et al., 2012).

Enamel decussation was not taken into account either. Indeed, the model proposed by Lucas et al. (2008) mentioned that species consuming large food objects of high modulus which required intermediate or high forces to fracture (i.e., stress-limited foods) were expected to show some decussated enamel. This would require further investigation, as the proportion of decussated enamel might compensate for evenly thick enamel in some durophagous primates. For instance, P. pithecia presents an evenly thin enamel (Figure 4C), but its enamel is also characterized by narrow, well-defined Hunter-Schreger Bands extending throughout its thickness (Martin et al., 2003). This might increase enamel resistance to the fibrous, possibly stress-limited seeds it masticates on a daily basis and compensate for an evenly thin enamel.

To conclude, this study shows that enamel thickness can be estimated using either a volumetric approach or a geometric approach. The former should be used to assess rate and speed of enamel secretion and more generally the amount of enamel topping the EDJ. The latter should be used to measure enamel thickness as the depth of enamel under OES and is hypothesized to better suit biomechanical models.

Furthermore, topographic maps of geometric enamel thickness and pachymetric profiles combine well for the interpretation of enamel distribution in both qualitative and quantitative terms. Slope of the pachymetric profile appears to be an especially fair estimate of enamel distribution evenness. In contrast, descriptive statistics such as the CV of enamel-dentine distance failed to detect differences in distribution evenness (Figure 5).

Overall, the methods introduced in this work make a powerful tool for testing form-function hypotheses related to enamel thickness. They can also be adapted to a wide range of studies focusing on the variation of tissue thickness across a whole surface, be it enamel or not. When investigating enamel however, the phylogenetic context should be taken into account, as enamel distribution patterns seem to depend on the family which is considered.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.