Introduction

The human chin is a unique morphological feature of our species and is often used as a (if not the) diagnostic criterion for identifying Homo sapiens in the hominin fossil record [1–5]. In simplest terms, a chin is an anterior bony projection of the lower jaw (mandible) extending beyond the lower incisors [1]. However, there is surprising variability in how chin morphology is delineated [6], which has contributed to the plethora of hypotheses put forward to explain the evolution of this distinctive human morphological trait [2]. If we simply characterize the possession of a chin by the degree of protrusion of the mandibular midline (symphysis) beyond the dentition (e.g., [7]), this can result in some modern humans being deemed “chinless” if their symphyseal protrusion is relatively slight and/or their lower dentition has a relatively strong forward projection [8]. For this reason, more specific anatomical definitions of the human chin have been proposed. For example, Schwartz and Tattersall [4] argue that a true chin is an inverted T-shaped bony structure composed of a triangular protrusion at the anterior base of the mandible (trigonum mentale), a vertical midline keel, and two depressions (mental fossae) on either side of the keel. This combination of anatomical features is found only in modern humans (i.e., it is a human autapomorphy) and not in other recent hominin taxa, such as Neanderthals [4]. Others emphasize the ontogenetic formation of the human chin as being the byproduct of interaction between the forward growth of the lower (basal) part of the mandibular symphysis, and the constrained growth of the upper (alveolar) part of the anterior mandible containing the dentition [7,9,10]. Given the need to maintain occlusion between the upper and lower teeth, the alveolar region develops relatively autonomously with respect to other regions of the mandible (sensu [11]), and it is the higher rates of bone deposition in the basal region combined with higher rates of bone resorption in the alveolar region that result in the possession of a bony projecting chin.

Debates over how best to define chins are not trivial, as characterization of chin morphology has a direct impact on how we understand its evolutionary origins [12]. In reality, chins are complex phenotypes with an underlying polygenic and pleiotropic genetic architecture that is not well understood [13]. Thus, in order to subject chin morphology to comparative evolutionary analyses, the continuous variability of symphyseal morphology across humans and other primates must be characterized using several quantitatively measured traits. As noted by Pampush [14], if the human chin is treated as an anatomical singularity or morphological autapomorphy (sensu Schwartz & Tattersall [4]), disconnected from mandibular variation in other primate taxa, its evolutionary origins cannot be ascertained [9].

Several evolutionary hypotheses for the origins of the human chin have been proposed. It has been suggested to be adaptive for biomechanical and functional reasons [1,15,16], which implicates the action of direct selection. Specifically, the masticatory stress hypothesis states that mandibular form is designed to resist biomechanical loads during chewing [1], and studies have found that the presence of a chin results in lower symphyseal strains (e.g., [15]), although it does not counter the effects of wishboning (lateral-transverse bending) [16], which is a source of strain on the symphysis during mastication in cercopithecoids [17–19]. The more vertically oriented symphysis and reduced symphyseal tori (the “shelves” of bone on the posterior part of the symphysis) of humans imply that wishboning became a less dominant source of masticatory stress throughout hominin evolution, presumably due to changes in incision that resulted in smaller anterior dentition in hominins compared with other hominoids [12]). Other adaptive hypotheses have been proposed centered on the notion that the demands of human speech caused changes in the stresses applied to the anterior symphyseal region due to the oblique contraction of the posteriorly anchored tongue musculature, leading to the creation of the human chin through remodeling [20] or that chins are a sexually selected signal of human mate quality [21], but none of these hypotheses have received strong support for a variety of theoretical or empirical reasons [2].

Simultaneously, it has long been suggested that the human chin is a “spandrel” (sensu [9]) in the sense that it is the by-product of selection on other aspects of craniomandibular and/or dental morphology, such as the reduction in the size of human teeth [5], or the differential growth of the alveolar and basal portions of the mandible [22]. Alternatively, it has been suggested that human mid-facial size decreased due to a reduction in androgen levels during ontogeny [23], which may have had the co-incidental effect of exposing the chin as an anterior projection. Another proposal is that the human chin is a by-product of selection on oral and pharyngeal soft tissues in response to increased basicranial flexion due to bipedalism and a more retracted face, which would have led to a constriction in the airway had the symphysis remained posteriorly inclined [24]. Under this scenario, chin morphology results from strains placed on the midline and basal anterior symphysis due to the need to pull the posteriorly attached tongue and other soft tissues forward and away from the airways, which would have been a necessary morphological change to prevent asphyxiation, although this hypothesis is almost impossible to test with fossil taxa.

Despite the wide range of hypotheses that have been proposed to explain the origins of the human chin, they all typically assume that selection (natural or sexual) is the primary force driving evolutionary changes in the mandible, and do not test a basic null hypothesis of neutral or stochastic evolution via genetic drift. This is despite the large body of literature illustrating that at least some aspects of hominin cranial shape evolved under neutral conditions [25–29]. A recent study [12] found evidence of a rate-shift towards a faster rate of evolution in the hominin lineage relative to other apes using two quantitative measures of chin-ness (chin-angle and chin-index), suggesting that the unique symphyseal morphology seen in modern humans is the product of selection on mandibular form in hominins, but it is unclear whether this selection was direct or indirect via responses to selection on other craniomandibular regions. Moreover, while a rate shift is indicative of the action of natural selection on these two quantitative measures of symphyseal angle, the role of genetic drift in shaping variation in overall mandibular shape is not currently well understood.

Here we apply a quantitative genetic approach to hominoid craniomandibular traits to test the null hypothesis that the unique symphyseal morphology of the human mandible is a neutrally-evolving product of genetic drift (H₀), with two alternative hypotheses: (H_A1) human symphyseal morphology is formed under direct selection, or (H_A2) human symphyseal morphology is a by-product of selection on other mandibular and/or cranial regions. It should be noted that these analyses operate within a quantitative genetics framework that assumes that all traits under consideration are subject to a polygenic and pleiotropic model of inheritance that is complex and unknown to us [30]. As such, the term “selection” is being used here to describe the non-stochastic processes of either stabilizing or directional selection acting on the statistical properties of multivariate morphological covariance matrices resulting in a degree of morphological difference between taxonomic groups that is larger or smaller than expected under a stochastic model of neutral divergence. This was achieved by calculating a phenotypic proxy for Lande’s [31] generalized genetic distance (GGD) between all branch segments of the phylogeny shown in Fig 1 following a reconstruction of mean ancestral craniomandibular trait values at each internal node [26,32–34]. Under a null hypothesis of neutrality (H₀), Lande’s GGD follows a chi-square distribution, such that rejection at p < 0.05 signifies that morphology is evolving either slower (lower tail) or faster (upper tail) than expected, indicative of stabilizing or directional selection, respectively. If directional selection were to be detected along the branch between the last common ancestor (LCA) of humans and chimpanzees and modern humans, selection gradients and trait responses were calculated to ascertain whether traits related to symphyseal (and therefore, chin) morphology were evolving under direct (selection gradient and trait response are in the same direction; H_A1) or indirect (selection gradient and trait response are in opposite direction; H_A2) selection [27,28,34].

Download:

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

Fig 1. Phylogenetic tree of the fifteen hominoid taxa employed here.

Maximum likelihood branch length estimates (in italics) based on Perelman et al. [35] and von Cramon-Taubadel and Smith [36]. Internal nodes (labelled 1-14) represent hypothetical ancestors.

https://doi.org/10.1371/journal.pone.0340278.g001

Under H_A1 (direct selection) it was predicted that nine mandibular traits, directly related to symphyseal morphology, would exhibit significant selection gradients and trait responses in the directions shown in Fig 2. It should be noted that these traits encapsulate more than the specific morphology of the human chin autapomorphy and also include mandibular traits homologous across hominoids, where changes would be expected to be concentrated if the human-specific chin had evolved under direct selection. Therefore, while the human chin would be better characterized by a more localized set of traits (including for example, the human-specific anatomical landmark – pogonion), our goals are to identify whether the symphyseal morphology related to the emergence of the human chin evolved stochastically, or under direct or indirect selection, within the wider context of hominoid craniomandibular evolution. This necessitated the use of a more generalized set of homologous mandibular traits, including a sub-set that specifically relate to symphyseal morphology. Specifically, the following traits would be expected to be under direct selection to decrease: the length of the alveolar portion of the mandible (INFR-ALV), the height of the midline symphysis (INFR-GNA), the distance from the front of the anterior dentition to the mental foramen (INFR-MEN), the distance from the front of the anterior dentition to the back of the inferior transverse torus (INFR-LIN), the distance from the mental foramen to the back of the inferior transverse torus (MEN-LIN), the distance from the back of the anterior dentition to the back of the inferior transverse torus (MO-LIN), and the distance from the back of the inferior transverse torus to the bottom of the symphysis (LIN-GNA). Simultaneously, the following chin traits would be expected to be under direct selection to increase: the distance between the mental foramen and the bottom of the midline symphysis (MEN-GNA), and the internal length from the back of the inferior transverse torus to the most posterior point on the basal portion of the mandible (LIN-GON). A hypothesis of indirect selection H_A2 would be supported if chin traits were either under no selection or under significant selection but with trait responses in opposite directions.

Download:

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

Fig 2. Craniomandibular landmarks with all 22 cranial and 24 mandibular interlandmark distances considered here.

Landmark definitions can be found in S1 Table. Solid colored lines show the expected selection gradients to increase (red) and to decrease (blue) for the nine mandibular symphyseal traits related to the evolution of a chin under a hypothetical model of direct selection along the Homo sapiens branch since the last common ancestor with chimpanzees.

https://doi.org/10.1371/journal.pone.0340278.g002

Materials and methods

The dataset comprises 532 matched adult crania and mandibles of 15 hominoid taxa representing every hominoid genus besides Nomascus (Fig 1; S2 Table) [34,36]. No permits were required for the described study, which complied with all relevant regulations. A total of 32 landmarks (15 cranial; 17 mandibular, S2 Table) were captured directly from museum specimens using a hand-held Microscribe™ digitizer by a single observer (NvCT) and 46 (22 cranial; 24 mandibular) interlandmark distances were extracted from these landmarks (Fig 2; S1 Table). Intra-observer error was found to be less than 1 mm for all landmarks [36], which is deemed acceptable for biological landmark configurations of this size [37]. All data were logged prior to analysis to account for difference in size among hominoid species and among traits, while still preserving information about the relative form (shape and size) of the specimens [34,38]. Craniomandibular data can be found in S3 Table and hosted by Dryad.

The same fully-resolved genetic phylogeny for the 15 hominoid taxa as was used in Schroeder & von Cramon-Taubadel [34] was employed here (Fig 1). The tree is based primarily on maximum likelihood (ML) branch lengths calculated by Perelman et al. [35], supplemented with ML branch length estimates obtained from other published sources [39–42]. A complete description of how the tree was constructed can be found in von Cramon-Taubadel & Smith [36]. The pairwise ML genetic distance matrix used to construct the tree can be found in S4 Table.

A Brownian-motion maximum likelihood approach [43] was used to estimate 95% mean ancestral values for each of the 46 craniomandibular traits at each of the 14 internal nodes shown in Fig 1. This method assumes that each trait evolved under a constant stochastic Brownian motion evolutionary model, such that the sum of the squared changes along each internal branch is minimized, and evolutionary change is proportional to branch length. Schroeder & von Cramon-Taubadel [34] also tested a multiple variance Brownian motion approach to ancestral trait estimation (e.g., [44]), which allows for different rates of change across branches, and found that ancestral trait estimates did not change substantially, nor did it impact the assessment of which branches or which traits were under directional selection. Therefore, while we acknowledge that different methods of ancestral trait estimation can produce biased results depending on the assumptions of the underlying model (e.g., squared-change parsimony, Ornstein-Uhlenbeck (OU) or Brownian motion), the structure of the tree, and changes in rates of evolution in different parts of the tree [44–47], for the purposes of this study we chose to use a standard Brownian motion approach for ancestral trait estimation such that the results obtained here would be more directly comparable to those of Schroeder & von Cramon-Taubadel [34], which used a similar cranial trait dataset but did not include mandibular data. Ancestral form estimation was carried out using the ace function in the ape package in R 4.1.2 [48]. Estimated mean trait values for each ancestral node can be found in S5 Table.

The phenotypic proxy for Lande’s “generalized genetic distance” (GGD) [31] was calculated for each section of the phylogenetic tree shown in Fig 1, in order to estimate rates of morphological change among internal nodes and between shared ancestral nodes and tip taxa. Lande’s GGD is effectively a squared Mahalanobis’ distance scaled by the effective population size divided by the number of generations since divergence and can be written as:

where N_e is effective population size, t is the number of generations since divergence, [z_i – z_j] is the difference between trait means of ancestor/descendant pair i and j, and W^-1 is the inverse of the pooled within-group phenotypic variance-covariance (V/CV) matrix (P) substituting for the additive genetic covariance matrix G [49]. The practice of substituting the P-matrix for the G-matrix in cases where estimates of genetic covariance are unattainable, known as Cheverud’s conjecture, has generally been found to be valid for morphological traits as long as the average heritability is reasonably high and the sample size used to construct the P-matrix is ≥ 40 [50]. Pooled within-group phenotypic V/CV matrices (W) were calculated using all extant taxa involved in each ancestral trait estimation. This approach is justified based on previous studies that have found highly conserved patterns of cranial trait covariance across hominoids [51,52] and indeed mammals on the whole [53]. For example, W for the GGD calculation between node 9 (Pongo ancestor) in Fig 1 and tip taxon Pongo pygmaeus was based on the pooled within-group V/CV using all Pongo specimens. The minimum sample size used to construct any phenotypic V/CV matrix was 61 (combination of two Pongo species), which is above the sample size threshold generally needed to generate stable P-matrix estimates [54]. The potentially confounding effects of sex and taxonomic structure on W were corrected for using the residuals from a MANOVA with sex and taxon as independent variables. In addition, W was multiplied by 0.4 to take account of the incomplete heritability of primate cranial traits [34,55]. The value of 0.4 was taken as an average measure of primate cranial trait heritability based on the limited number of relevant studies conducted [49,55–59]. Assuming incomplete heritability is conservative in this analysis, as assuming full heritability would risk overestimating the rate of evolution (i.e., generate larger GGD estimates). GGD follows a chi-square distribution under a model of stochastic genetic drift, with degrees of freedom equivalent to the number of traits [31]. Small values that reject a null hypothesis of drift indicate the effects of stabilizing selection as the amount of morphological divergence is less than expected under drift, while large significant values indicate the action of directional selection (i.e., more morphological change than expected). Here rates were designated as “very slow”, “very fast”, “slow” and “fast” in a two-tailed test at 99.9% and 95% confidence intervals, respectively.

Estimates of effective population size (N_e) and number of generations since divergence (t) are the same as those used by Schroeder & von Cramon-Taubadel [34] for each ancestral node in order to provide comparable results with the latter study. t is calculated as time since divergence corrected for average generation length. It should be noted that a recent study based on whole genome data [60] provides updated estimates of divergence times for some of the same nodes considered here. The means and ranges from this recent study are included in S6 Table, which provides all the parameter values used and a description of where the data were collated from. It is worth noting that except for the divergence of the two orangutan (Pongo) species, all of our assumed divergence times fall within the confidence intervals estimated by Shao et al. [60], providing further support that these divergence estimates are relatively robust. Even in the case of Pongo, our divergence time estimate (1.3 mya) falls just outside the lower bound of the range (1.52–3.88 mya) provided by Shao et al. [60], which is considerably older than other published divergence times (e.g., 0.97 mya) for orangutan species [61]. As in Schroeder & von Cramon-Taubadel [34], we calculated 95% confidence intervals for each estimate of N_e, and GGD distances were also estimated for these confidence intervals to ensure that GGD values were not biased by any errors in the calculation of effective population size.

In instances where GGD values indicated fast rates of morphological evolution, and neutral divergence could be rejected (i.e., “fast” or “very fast” GGD values), the magnitude and pattern of selection required to produce the differences between estimated ancestral and descendant taxon means was reconstructed. The differential selection gradient [31,62] was estimated using the following equation:

Where β is the multivariate selection gradient, is the inverse pooled within-group phenotypic V/CV matrix, and is the vector of mean differences between ancestral and descendent taxa i and j [28,34,51]. 95% confidence intervals were calculated using a resampling approach to calculate 1000 vectors of trait means that were then used to estimate a distribution of 1000 selection gradients (β) with the differential selection gradient equation above. Selection gradients were considered statistically significant if their 95% confidence intervals did not overlap with zero. Overlapping with zero would indicate that the estimated selection gradient was negligible and therefore non-significant. These selection gradients were used to differentiate between direct and indirect selection, as the selection gradient is a vector of partial regression coefficients of relative fitness onto the traits [63]. Positive selection gradients signify a positive relationship with fitness if the trait response is also positive (i.e., selection is favoring the trait to increase in size and the trait is responding appropriately). Negative selection gradients signify a negative proportional relationship with fitness if the trait is also negative (i.e., selection is favoring the trait to decrease in size and the trait is responding appropriately). Both of these indicate direct selection with a matched trait response. However, if the trait response is in the opposite direction of the gradient, then it indicates that the trait has no relationship with fitness and is not under direct selection. Instead, it is likely that the change in the trait mean is uncorrelated with the selection gradient and instead is responding to selection on other traits with which is covaries.

Results

Rates of morphological evolution across apes

Rates of evolution indicate that, for the majority of branches across the hominoid phylogeny, craniomandibular morphological divergence was slow and departures from the neutral model occurred mainly in the form of stabilizing selection indicated by green GGD values in Table 1. This is comparable to what was found by Schroeder & von Cramon-Taubadel [34] for hominoid cranial form. There were a few instances where genetic drift could not be rejected (marked in yellow in Table 1), most notably between node 3 (Pan ancestor) and Pan paniscus, and between node 8 (Pongo ancestor) and Pongo abelii. Fast rates of morphological change, indicative of directional selection (red GGD values in Table 1), were detected in the branch between node 4 (LCA of Homo and Pan) and Homo sapiens. While not as strong, there was also some evidence of directional selection in the divergence of the great apes from their last common ancestor with the lesser apes (between nodes 13 and 14), and the divergence of mountain gorillas (Gorilla b. beringei) from the other eastern gorilla subspecies. However, unlike the results for humans, these latter two cases of directional selection were only found for GGD values based on mean and upper bound estimates of effective population size scaled by number of generations since divergence (N_e/t), suggestive of the influence of changing effective population sizes in the evolution of these lineages [34].

Download:

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

Table 1. Generalized genetic distances (GGD) along each branch of the phylogenetic tree shown in Fig 1.

https://doi.org/10.1371/journal.pone.0340278.t001

Assessing direct and indirect selection in the human cranium

Given that the null hypothesis (H₀) of neutral morphological divergence under genetic drift could be rejected for the human cranium and mandible (and by extension the chin), the two alternative hypotheses (H_A1 and H_A2) of direct and indirect selection were tested. In the cranium (Table 2, Fig 3), significant direct selection for trait size increase was found in traits related to basicranial flexion (e.g., lengthening of SP-BA, SP-ZI) and expansion of the neurocranium (SP-SP), while significant direct selection for trait size decrease was primarily related to facial retraction (IS-C/P3, ZI-C/P3, IS-PNS), and the anterior migration of the foramen magnum and increased basicranial flexion (SB-BA, HOR-SB, PNS-HOR) associated with the shift towards a bipedal mode of locomotion. Only one trait (ZS-AM1) was under indirect selection whereby the height of the midface was under selection to increase but the trait actually decreases. This makes sense as the trait also reflects the angle of the face (the position of the first molar relative to the base of the orbit), which is more orthognathic in humans compared with other great apes, yet the absolute size of the face also decreases in the human lineage resulting in an absolutely smaller trait value even if the trait is under selection to increase.

Download:

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

Table 2. Selection gradients (β) and responses (% change in trait means) for all traits from the LCA of Pan-Homo to humans arranged in descending order of β.

https://doi.org/10.1371/journal.pone.0340278.t002

Download:

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

Fig 3. Craniomandibular traits found to be under significant direct selection (solid lines) to increase (red) and decrease (blue).

Traits marked in either red or blue dashed lines were found to be under indirect selection. Traits in light grey did not return significant selection coefficients (Table 2).

https://doi.org/10.1371/journal.pone.0340278.g003

Discussion

Conclusions

Our results suggest that human symphyseal morphology that enables the possession of a chin is an evolutionary side-product or “spandrel” (sensu Gould & Lewontin [9]) resulting from direct selection on other craniomandibular regions, including increased basicranial flexion, facial reduction, neurocranial expansion, and a more gracile mandible with a more upright symphysis. These are familiar craniomandibular changes associated mostly with the early hominin fossil record, which documents the evolution of a lineage of bipedal apes with decreased anterior dentition relative to other great apes, and a concomitant reduction in the size of the maxilla and alveolar region of the mandible (e.g., [92]). However, the chin is not a singularity but rather the result of multiple changes in the relative position of different aspects of mandibular morphology. So, while a more vertical symphyseal angle associated with smaller anterior dentition was an important evolutionary change within hominins that allowed for the human chin to eventually evolve, selection on other craniomandibular regions also contributed to the emergence of this unique human trait. This underscores the importance of assessing the evolution of morphology with trait integration in mind.

Supporting information

Acknowledgments

We thank the editor and anonymous reviewers for their constructive comments that helped improve our manuscript. We are grateful to the following institutions for allowing access to materials in their care: Natural History Museum (London), Natural History Museum (Vienna), Duckworth Collection (Cambridge), Musée de l’Homme (Paris), American Museum of Natural History (New York), Powell-Cotton Collection (Kent), Museum of Comparative Zoology (Harvard), Royal Museum of Central Africa (Tervuren), Smithsonian National Museum of Natural History (Washington D.C), Bavarian State Collection for Anthropology, and Paleoanatomy (Munich).