Phylogenetic context provides a necessary source of information by which to gauge the strength of dietary inference from comparative functional studies because it may inform how constraints can limit form-function-behavior relationships (even if the precise nature of these constraints is currently undiscovered). Unfortunately, as noted by Ross and Iriarte-Diaz (2014), many biomechanical studies of primate feeding over the last few decades have neglected any consideration of phylogeny.

Genetically-determined ancestral patterns of morphology may simply be retained passively or through patterns of developmental constraint as plesiomorphies in descendant taxa, and thus what might be recognized as “robust” traits need not involve directional selection. Such “phylogenetic inertia” may be rooted in pleiotropy, genetic linkage, or developmental pathways that can impose constraints. Consideration of phylogenetic history is important in establishing the constraints under which morphology could have been adaptively modified by selection. As aptly observed by Hildebrand (1988: 728), “No amount of field observation or biomechanical analysis can fully explain form. We must know the history, the phylogeny.”

Thus, the high cusps and crests that adorn the cheek teeth of colobine monkeys are well-suited for shredding leaves, but this does not compel the conclusion that bilophodonty represents an obligate adaptation for folivory among primates. If a dedicated leaf-eating hominin had existed in the past, it would be rash to suggest that bilophodonty would be a decisive condition for inferring folivory in such a species. Because the low, bulbous and thickly enameled cusps of australopith cheek teeth are seemingly ill-equipped to process displacement-limited (i.e. tough) foods (e.g., sedges), some workers (e.g., Strait et al., 2010, Strait et al., 2009, Strait et al., 2008 and Wood and Schroer, 2012) have argued that any hypothesis that incorporates such foods can be discounted. While it should be acknowledged that Paranthropus molars are suboptimal for shredding fibrous items, these teeth can also be viewed as representing a biological entity that is contingent upon not only a range of possible functions, but also the dentition of its predecessor (Ungar and Hlusko, 2016).

The predicted outcome of a morphological trait under a model of adaptive selection may fail if that trait is genetically and/or developmentally correlated with (i.e. constrained by) others that are under conflicting pressures (Lande, 1979). At the same time, however, developmental integration could be a major factor maintaining a relatively stable covariance structure of functionally related traits during the evolutionary diversification of a lineage (Marroig and Cheverud, 2001). Constraints, which may result in modularity of traits through integration, can affect the evolution of complex morphological systems by altering levels of evolutionary flexibility (Cheverud, 1984). Comparative studies have revealed differences in magnitudes of integration and modularity in the skulls of different mammalian (including primate) lineages, where enhanced “modularization” enables a broader spectrum of evolutionary response (Porto et al., 2009 and Shirai and Marroig, 2010). Modularity that affects morphological variation can be thought of in terms of the developmental and genetic components that may impact the ability to respond to functional selective pressures. Traits that perform one or more roles interact functionally through modularity (Klingenberg, 2008).

There is strong experimental evidence that inactivation of specific genes can disrupt development of the (mouse) mandible, and that these developmental disturbances can affect different parts of the mandible either separately or in concert. Some 37 quantitative trait loci have been identified as affecting aspects of mandibular size and shape (Cheverud et al., 1997, Klingenberg and Leamy, 2001, Klingenberg et al., 2001, Leamy et al., 1997 and Leamy et al., 1999). Subsequent ontogenetic changes in form through modeling under mechanical load are similarly spatially structured (Klingenberg et al., 2001). Despite the expectation that ontogenetic growth trajectories should be under strong directional selective pressure, studies have found high-levels of standing additive genetic (co)variance for growth trajectories. Additive genetic (co)variance is higher early on and decreases with age (Irwin and Carter, 2013). On top of this, plastic responses to altered loads during growth may induce changes in gene- and protein-expression patterns, which ultimately affect anatomical and functional variation (Ravosa et al., 2007 and Ravosa et al., 2008). Interestingly, and perhaps unexpectedly, such epigenetic changes appear to influence the phenotypes of subsequent generations (Agrawal, 2001, Pigliucci et al., 2006 and West-Eberhard, 2005).

Given the confusion that may surround adaptation and constraint, Pigliucci and Kaplan (2006: 128) have proposed a minimum of five essential questions that should be addressed when dealing with a trait of interest: •

how did it arise?

Obviously, some of these questions will be resistant to exploration in the fossil record. As noted by Brakefield (2006: 367), “Integrative research on generative constraints in which genetic variation and developmental mechanisms are explored together with natural selection and the performance of phenotypes is necessary to progress beyond the level of ‘just-so stories”’. The task before the paleontologist is formidable.

Mechanical loads clearly influence the skeletal system, but it is not clear to what extent mechanical environments dictate particular morphologies. Bone material serves to equalize peak stress (or strain) levels within a narrow band of “optimal” magnitudes in long bone diaphyses (Rubin and Lanyon, 1984), but this relationship varies even among different regions of the shaft (Judex et al., 1997 and Rubin et al., 2013). The ubiquity of strain gradients in the skull (Hylander and Johnson, 1992 and Ross and Metzger, 2004) establishes that bone does not adapt to a state of maximum strength with a minimum of material. While minimum levels and frequencies of strain seem to be required for bone to produce a physiological response (Beecher et al., 1983, Bouvier and Hylander, 1984 and Lanyon and Rubin, 1985), defining these minimal values is far from straightforward (Gross and Srinivasan, 2006 and Qin et al., 1998). Moreover, peak strain values may differ significantly across the skeleton (Ravosa et al., 2010), and the presence of such gradients implies the existence of local, site-specific controls on bone formation and maintenance (Goodship et al., 1979, Hylander and Johnson, 1997 and Ravosa et al., 2010).

Adding to this complexity, patterns of bone material response to mechanical stimuli in long bone shafts do not translate easily to the skull (Daegling, 2010, Hylander and Johnson, 1992, Hylander and Johnson, 1997 and Ravosa et al., 2010). Indeed, experimental evidence strongly suggests that mechanobiological influences are not uniform across the craniofacial region (Hylander et al., 1991, Ravosa et al., 2000 and Ross and Hylander, 1996). Thus, across a range of species, significant strain disparities have been demonstrated between the lower facial and circumorbital regions of the skull (Hylander et al., 1991). The latter maintains considerably larger amounts of bone than is needed to resist masticatory loads (Hylander and Johnson, 1997 and Hylander et al., 1991). This calls into question the value of biomechanical models that assume circumorbital – and especially supraorbital – features are adapted for resisting masticatory stresses (cf., Bookstein et al., 1999 and Ravosa et al., 2000). It is unwise to assume that the entire primate facial skeleton is first and foremost an adaptation to accommodate masticatory forces (Daegling et al., 2013 and Ross and Iriarte-Diaz, 2014). The robusticity of the mandibular corpus may be related more directly to allometric considerations than to dietary factors (Ravosa, 1990), and other aspects of jaw morphology (e.g., symphyseal depth) may be artifacts of somatic scaling that secondarily affect masticatory biomechanics (Smith, 1993).

Comparative anatomy and biomechanical analyses may serve to bracket a range of behavioral capabilities, but these approaches are inherently limited in their capability to identify specific adaptation in paleobiological contexts. The application of finite element analysis (FEA) to the hominin fossil record has demonstrated that under similar loading conditions, the form of the stress field in an australopith facial skeleton may differ from that in some extant primates (Dzialo et al., 2014, Ledogar et al., 2016, Smith et al., 2015, Strait et al., 2013, Strait et al., 2010, Strait et al., 2009 and Wroe et al., 2010). Depending on one's perspective, connecting the different stress fields to different feeding adaptations is either eminently logical or a leap of faith (Daegling et al., 2013). Biomechanical studies may inform about the physical performance attributes of a morphological feature, but the biologically important question is whether these attributes have a tangible effect on fitness (which is implicit in the identification of adaptation). With regard to osseous traits involved in mastication, it is necessary to realize that both load magnitude and frequency have been implicated in bone metabolic activity that results in changes in the mass and architectural properties of skeletal elements (Judex et al., 2006, Ozcivici et al., 2009, Qin et al., 1998, Ravosa et al., 2007, Rubin et al., 1990 and Rubin et al., 1991). Thus, a given jaw structure may be responsive to high chewing loads and/or low but frequent chewing loads.

The presence of strain gradients is evidence that primate crania are not optimized to dissipate or resist the stresses engendered by feeding forces that might be associated with different diets, but rather there appears to be a relationship between skull morphology and the ability to generate feeding forces (Ross and Iriarte-Diaz, 2014). Even in this regard, however, caution should be exercised when attempting to draw conclusions about the type of diet or food mechanical properties that may relate to force generation. The crania of colobine monkeys have larger areas of attachment of the masseter muscle than do cercopithecines monkeys, and colobines are more folivorous (Ravosa, 1990). Colobine crania also exhibit greater mechanical advantage to generate feeding forces than cercopithecines. This is an effect of their having shorter faces, and Hylander (2013) has argued that canine size is a prime driver of facial length because of gape requirements. Thus, the reduced mechanical advantage of long-faced cercopithecines does not have any necessary connection to relaxed dietary requirements (however these might be defined). On the other hand, colobines that consume more seeds have better mechanical advantage of the masticatory muscles than those that do not (Koyabu and Endo, 2010), and platyrrhine monkeys that harvest sclerocarps tend to present larger chewing muscles than their close relatives that do not prey upon seeds (Kay et al., 2013).

With reference to chewing, the processing of tough, fibrous foods entails increased cycle number in association with higher occlusal forces (Hylander, 1979). As such, mandibular morphology per se cannot distinguish between potential adaptations to feeding on either tough or hard items (Daegling and Grine, 2007, Daegling and Grine, 2017 and Hylander, 1988). There is a remarkable diversity of corpus shape associated with hard object feeding in primates (Daegling, 1992, Daegling and McGraw, 2001, Daegling and McGraw, 2007 and Taylor, 2006), which means that similar feeding challenges do not necessarily have a singular morphological solution. As noted by Ross et al. (2012), because mandibular loading, stress, and strain regimes are determined by interactions among food mechanical properties, feeding behaviors, jaw kinematics and jaw morphology, it is difficult to discern whether similarities in chewing kinematics also translate to similarities in loading, stress and strain. It is not always clear what loading variables and specific morphological features of the mandible are of importance (Gröning et al., 2011 and Shi et al., 2012). Data on food material properties do not effectively map mandibular form to diet (Ross et al., 2012). The relationship between food material properties and mandibular morphology does not necessarily hold across all comparisons, which suggests that one cannot necessarily predict one from the other (Daegling and Grine, 2007 and Daegling and Grine, 2017).

This presents additional problems for dietary inference from the cranium, since the mechanical loading environment of the mandible is much better understood than is the rest of the facial skeleton (Ross et al., 2012). Distinguishing the relevant factors that influence craniofacial form in order to make informed and reliable inferences about trophic behaviors and habits is a daunting task (Shi et al., 2012 and Walmsley et al., 2013), made even more so by genetic and developmental constraints that can affect interpretation (Ungar and Hlusko, 2016). Given this present uncertainty, an agnostic stance on what facial morphology reveals about specific diets and/or dietary adaptations would seem to be warranted (Daegling et al., 2013).

Perhaps even more than for bones, the forms and structures of teeth have been tied to dietary function. This is to be expected given the incredible diversity of mammalian dentitions and diets (Ungar, 2010). No less in primates than in many other mammalian groups, there is clearly an operational relationship between tooth form and function (Kay, 1975, Lucas, 2004 and Ungar, 2010). However, as with bone, it is not clear to what extent specific mechanical environments (i.e. diets) dictate particular morphologies. Most attention to dietary relationships among primates, and especially the retrodiction of diet for extinct taxa, has been based on the form of the tooth crowns, and upon the thickness and structure of the enamel cap.

Premolar and molar crown relief – cusp height and “sharpness” – is commonly related to masticatory function, and to the types and/or physical properties of the foods that are processed (Lambert, 2010, Lucas, 2004 and Ungar, 2010). In particular, quantitative analyses of dental topography that employ measures of relief (Boyer, 2008, Ungar, 2004, Ungar and Bunn, 2008 and Ungar and Williamson, 2000), complexity (Evans et al., 2007) and Dirichlet normal energy (Bunn et al., 2011 and Godfrey et al., 2012) have differentiated among groups of primates that differ in the types of foods they consume. To date, dental topographic analysis has been applied to only a few fossil species (e.g., Godfrey et al., 2012 and Ungar, 2004). While these methods are encouraging, it is not at all clear to what extent they can differentiate between anything but the broadest dietary categories.

Because many mammalian dental features are correlated with one another and represent iterative processes that are dependent upon relatively simple alterations in signaling proteins, they are not independent of one another (Jernvall et al., 2000, Kangas et al., 2004 and Kassai et al., 2005). Because many aspects of the dentition have the developmental potential for correlated change, this may obscure functional (and certainly dietary) inference. Indeed, Gailer et al. (2016) have shown that supposedly “essential functional adaptations” in cheek tooth morphology may not be required to process food efficiently in extant bovids.

Tooth enamel thickness has long been regarded as being sensitive to diet, or at least to the mechanical properties of the foods consumed (e.g., Constantino et al., 2012, Lucas et al., 2008b, Martin et al., 2003, Schwartz, 2000, Teaford, 2007 and Vogel et al., 2008). In particular, because enamel thickness is expected to be able to respond quickly in evolutionary time to dietary/ecological change (Hlusko et al., 2004), differences in enamel thickness have been related to corresponding differences in the hardness of food items (Lucas et al., 2008b and Vogel et al., 2008). However, while there seems to be some general correspondence between these two variables, the reliability of tooth enamel thickness as a dietary indicator clearly breaks down in some cases where phylogenetically closely-related species that consume different amounts of hard items are considered. For example, the expected differences between Bornean and Sumatran orangutan molars are not apparent (Smith et al., 2012), nor is there a predictable relationship among mangabey species (McGraw et al., 2012). Of course, other factors can account for the evolution of a thicker coat of enamel. Rabenold and Pearson (2011) postulated an alternative hypothesis that relates enamel thickness to exogenous abrasives that may be found on or in dietary items. In the nine primate species sampled, they found a strong correlation between the amount of abrasive silica phytoliths in the diet and molar enamel thickness. In other words, it is not only resistance to fracture, but also resistance to prolonged wear to which enamel thickness can be related (Pampush et al., 2013).

Tooth enamel structure has been regarded as being sensitive to diet, or at least the mechanical properties of the foods consumed (e.g., Chai et al., 2009, Popowics et al., 2004, Shimizu and Macho, 2008, Teaford, 2007 and von Koenigswald and Pfretzschner, 1991). To date, however, there have been very few studies that have considered the degree of variation in structural features within teeth (e.g., prism decussation; Lynch et al., 2010). Indeed, most of the mechanical models for tooth enamel developed to date have been highly theoretical, being based on the behavior of glass, ceramics and polymer resins rather than actual dental tissues (Kim et al., 2007, Lucas et al., 2008b, Qasim et al., 2007 and Rhee et al., 2001). Yet another theoretical study by Shimizu and Macho (2008) concluded that it is not possible to predict the absolute strength of a tooth unless its chemical composition is known, although they failed to determine the critical variables at the molecular level. It would seem that predicting tooth strength is problematic, especially since teeth are complex, composite structures.

It is perhaps noteworthy that the hyper-thick enamel and massive, low-cusped molars of P. boisei, a species envisioned by many as the quintessential hard object feeder (e.g., Constantino et al., 2011, Kay, 1985, Macho, 2014 and Wood and Strait, 2004), exhibit little enamel decussation in comparison to the molars of Homo (Beynon and Wood, 1986). Nonetheless, the rounded, blunt cusps and the presence of well-developed Hunter-Schreger bands in the enamel of sea otters have led to their being proposed as a model from which to infer that species like P. boisei most likely consumed hard food objects with substantially higher biting forces than those exerted by modern humans (Constantino et al., 2010 and Constantino et al., 2011). The logic is that if sea otters, which possess thinner enamel than hominins, can process mollusk shells and the tests of sea urchins, then P. boisei must have been superbly well-adapted to crack hard objects with powerful bite forces (Constantino et al., 2010 and Constantino et al., 2011). Of course, this is not the first time that otters have been employed as analogies for Paranthropus feeding adaptations. Shabel, 2005, Shabel, 2007 and Shabel, 2008 envisioned P. robustus as a crab-eating durophage, with the Cape clawless otter (Aonyx capensis) as its trophic analogue. Here again, the critical consideration in the paleontological context is whether what could have been eaten constitutes evidence for what was, in fact, consumed. If the latter is not true, then the adaptive inference is invalid.

The postcanine dentition of Ar. ramidus, which is comparatively small and thinly-enameled, has been viewed as lacking the adaptations for heavy chewing and the consumption of abrasive foods, suggesting instead an omnivorous diet dominated by frugivory (Suwa et al., 2009). These inferences find consilience with the microwear signals and isotopic values obtained for Ar. ramidus tooth enamel (Suwa et al., 2009 and White et al., 2009). The latter are consistent with a diet dominated (85–90%) by C₃ plant sources such as the flowers, fruits, seeds and leaves borne by trees.

Australopithecus anamensis tooth morphology has been interpreted as indicating a shift in food resources to harder (Ward et al., 1999 and Ward et al., 2001), tougher (White et al., 2006), harder and tougher (Macho et al., 2005), or perhaps more abrasive (Teaford and Ungar, 2000 and White et al., 2006) items. In particular, the thick enamel caps and pattern of enamel decussation have suggested adaptation for “habitually consuming a hard-tough diet” (Macho et al., 2005: 310) or a “tougher and probably harder and abrasive diet” (Macho and Shimizu, 2010: 30).

However, thickly enameled, low-cusped molars would have difficulty processing tough foods, whereas they would easily break down hard, brittle items (Lucas, 2004 and Strait et al., 2009). As noted above, although enamel thickness and decussation have been claimed to carry an “unambiguous adaptive signal in relation to diet” (Lucas et al., 2008b: 383), neither appear to be perfectly correlated with food type. While the crack-stopping properties of decussated enamel prisms is reasonably well-established (e.g., Cuy et al., 2002, Hassan et al., 1981, He and Swain, 2007, Popowics et al., 2001, Popowics et al., 2004, Rasmussen et al., 1976 and Xu et al., 1998), quantification of the form and degree of decussation has been difficult to achieve.

The occlusal microwear fabrics documented for Au. anamensis, although determined from a paltry sample of only three individuals, does not resemble those of extant hard object feeders (Grine et al., 2006a and Ungar et al., 2010). Rather, they appear to be reminiscent of the wear pattern described by Suwa et al. (2009) for Ar. ramidus. The Au. anamensis and Ar. ramidus specimens that have been sampled exhibit δ¹³C values that differ very little (Cerling et al., 2013 and White et al., 2009). Like the values for Ar. ramidus, those for Au. anamensis have a relatively narrow δ ¹³C range that is compatible with 90%–100% of the diet being comprised by C₃ resources (Cerling et al., 2013).

White et al. (2000) argued that the postulated evolutionary trajectory from Au. anamensis to Au. afarensis involved an increase in postcanine tooth size and other adaptations to a “more heavily masticated” diet. In this context, such a diet may have involved larger forces alone (e.g., hard food items), or larger forces together with more chewing cycles (e.g., fibrous food items). Similarly, Ward et al. (2010) have conjectured that in the presumptive Au. anamensis – Au. afarensis lineage, “significant changes appear to occur particularly in the anterior dentition, but also in jaw structure and molar form, suggesting selection for altered diet and/or food processing.” Ungar (2004) observed that molar topographic relief in Au. afarensis implies that these teeth are well-suited to fracture brittle, less deformable foods but not as well-suited as the crowns of either gorillas or chimpanzees to fracture tough, more deformable foods. Because more bulbous cusps should allow transmission of higher stresses to a food item without engendering damage to themselves, Ungar (2004) suggested that the diet of Au. afarensis may have differed from that of Pan largely in hard, brittle fallback resources. In addition to the dental features enumerated above, Au. afarensis possesses very deep and robust mandibular corpora, tall mandibular rami, and robustly constructed zygomatic arches (Kimbel et al., 2004). As summarized by Kimbel and Delezene (2009: 40), Au. afarensis exhibits a number of craniodental attributes that are “conventionally associated with ‘heavy mastication’ (however imprecisely defined) compared to extant great apes”.

In light of the morphology that has been described for Au. afarensis, the occlusal microwear data recorded for this species (Grine et al., 2006b and Ungar et al., 2010) suggest that the best modern analogs for the properties of the items that were consumed by the 19 individuals sampled are those extant primates (e.g., Gorilla and/or Theropithecus) which do not consume hard objects. Although Au. afarensis may have possessed the trophic apparatus to process a fairly wide range of foods, including hard, brittle items such as “nuts, seeds and hard fruit” (Wood and Richmond, 2000: 29), its molar microwear suggests that it did not always do so (at least those individuals that have been sampled do not appear to have done so during the periods in which their microwear fabrics were being formed). In the absence of any available isotopic data for Au. afarensis, Grine et al. (2006b) interpreted the constancy of its microwear across time and space as suggesting either that this species was able to track its preferred dietary resources in the face of changing habitats and environments, or that environmentally induced shifts in diet did not involve changes in mechanical properties of the foods typically consumed.

Subsequent isotopic analysis of some 20 specimens of Au. afarensis have shown that this species displays a significant range in the consumption of C₄-based resources (Wynn et al., 2016 and Wynn et al., 2013). The δ¹³C values indicate more C₄ consumption on average than in earlier hominins such as Ar. ramidus and Au. anamensis. Wynn et al., 2016 and Wynn et al., 2013 also note a lack of any temporal pattern in the Au. afarensis isotope values, a finding consistent with the occlusal microwear data, which show little geochronological variation in this species. However, in contrast to the homogeneity exhibited by dental microwear, there is a wide range of variation in the carbon isotopic composition of Au. afarensis diet even within relatively brief temporal periods. Wynn et al., 2016 and Wynn et al., 2013 concluded that these opposing patterns suggest use of a mixture of C₃ and C₄ plant foods that may have varied little in terms of their mechanical properties.

It is unfortunate that none of the specimens of Au. afarensis from the Kada Hadar-2 submember of the Hadar Formation were found to preserve ante mortem microwear (Grine et al., 2006b), because it is in this sample that Lockwood et al. (2000) observed signs of an increase in jaw size. Fortunately, Wynn et al. (2013) were able to sample five specimens – including three associated with large jaws – from this temporal interval. With reference to the question of the relationship (or lack thereof) between isotope values and morphology, these five specimens show exactly the same spread of δ ¹³C values as the older Hadar Formation samples. In other words, there is no discernible change in diet (as deduced from stable isotopes) associated with an increase in mandibular (and presumably body) size in Au. afarensis.

The morphological attributes of Au. africanus have informed numerous speculations concerning its diet (e.g., Du Brul, 1977, Grine, 1981, Jolly, 1970, Kay, 1985, Robinson, 1954, Strait et al., 2009, Wallace, 1975 and Wolpoff, 1973). Most reconstructions have envisioned this species as having been primarily, if not exclusively, herbivorous.

Rak (1983) envisioned the “anterior pillars” of the Au. africanus maxilla as facial buttresses that linked this species with the enhanced masticatory apparatus of the South African australopith, Paranthropus robustus. Although Rak (1983) suggested that the pillars in Au. africanus were related to anterior tooth loading, this is difficult to reconcile with their presence in P. robustus, in which the incisors and canines are notably reduced. At the same time, McKee (1989) has noted that Rak's scenario is also difficult to reconcile with their absence in a form such as Au afarensis, which possessed an anterior dental battery as large as if not larger than that of Au. africanus. In fact, it has been argued that the anterior pillars of Au. africanus and P. robustus are not homologous (Villmoare and Kimbel, 2011).

Strait et al., 2009 and Strait et al., 2008 have employed FEA modeling to argue that the anterior pillars of Au. africanus were part of an adaptation to the inclusion of hard seeds and nuts in its diet, with these items being processed by the premolars. Strait et al. (2010) have elaborated this hypothesis to the inclusion of “large, hard food items” in the diet. They postulated processing by the premolars rather than the molars because they found the cranium of Au. africanus to be relatively more rigid than that of the macaque monkey during premolar but not molar biting. Differences in the magnitude and distribution of strain energy and principal strains are implicitly argued in these models to be related directly to the modulation of morphology (Strait et al., 2010, Strait et al., 2009 and Strait et al., 2008). However, the distributions of strains by themselves do not unambiguously define a state of adaptation (Daegling et al., 2013, Daegling et al., 2011 and Grine et al., 2010). ¹ 1

More specifically, they failed to specify any basic process by which mechanical signals initiate bone response, and what the morphological outcome of that response would be. If it is simply a matter of normalizing peak strain (or modeling and remodeling toward a target stain magnitude), then bone activity ought to result in a uniform or nearly uniform strain field.

Occlusal microwear evidence for the premolars and molars represents a reasonable test of the model of premolar nutcracking presented by Strait et al., 2010 and Strait et al., 2009. Because the premolars of Au. africanus display the same microwear fabrics as its molars (Grine et al., 2010), there is no evidence to support the argument for preferential premolar nutcracking in this species. In addition, a perhaps unlikely model for premolar “nutcracking” behavior exists for at least one extant primate – the sooty mangabey, Cercocebus atys (Daegling et al., 2011). This committed hard object feeder employs the ingestive strategy hypothesized by Strait et al. (2009) to explain the unique facial skeleton of Au. africanus. Strait et al. (2009: 2126) concluded that “the facial skeleton of Au. africanus is better designed to withstand premolar loads than that of Macaca fascicularis”, but the facial morphologies of M. fascicularis and C. atys are very similar to one another in essential ways (i.e. both possess relatively long faces) when juxtaposed with Au. africanus. If sooty mangabeys suffer no obvious deleterious consequences for processing large, hard seeds given the design of their facial skeletons, it would seem unnecessary for Au. africanus to require such divergent facial morphology (Daegling et al., 2011).

Au. sediba is represented by two specimens from the site of Malapa, South African (Berger et al., 2010). With regard to the craniodental remains, the more complete specimen, designated MH1, is a juvenile whose M2s had reached occlusion just prior to death. The second specimen is represented by a fragmented adult mandible. The facial and dental morphologies of Au. sediba bear similarities with Au. africanus, which has led some workers to conclude that it is exclusively linked to that species (Irish et al., 2013; but see Carter et al., 2014), or that it is simply a somewhat unusual australopith closely-related to Au. africanus (Kimbel, 2013). Others have argued that it shares similarities with Homo with respect to a number of functionally consequential craniodental features (de Ruiter et al., 2013).

With reference to paleodietary reconstruction, comparative morphological studies do not present a consistent picture for Au. sediba. Ledogar et al. (2016) have argued that the overall architecture of its skull placed this taxon at risk of distractive temporomandibular joint forces under idealized muscle recruitment patterns. They suggest, on the basis of a finite element model, that Au. sediba would have been limited in its ability to produce large bite forces, with facial strains that were absolutely low in comparison to Au. africanus. The FEA model presented by Ledogar et al. (2016) is dependent on the correctness of their estimate of adductor muscle force. This is a significant problem with FEA studies because estimation of muscle force from morphological attributes is problematic even in neontological contexts other than through direct measurement of the cross-sectional area of the muscle itself (Hannam and Wood, 1989 and Weijs and Hillen, 1985). However, muscle force has to be specified since the determination of high or low stress and strain is directly dependent on adductor force, and FEA depends on the correctness of the estimate.

On the other hand, the Au. sediba mandibles appear to retain the level of torsional strength and efficiency that sets australopiths apart from extant great apes and Homo (Daegling et al., 2016). At first glance, both studies would seem to suggest an “overdesigned” facial skeleton: i.e. needlessly strong given likely masticatory forces. As noted above, however, these forces are truly unknown in fossils.

Henry et al. (2012) evaluated isotopic, occlusal microwear and adherent phytolith data from the dental specimens of Au. sediba in order to infer the types of foods that were consumed. Collectively, these data indicate a diet that contrasts with those inferred for Au. africanus (and P. robustus). Together they suggest almost exclusive consumption of C₃ resources that probably included some hard and tough foods. The microwear signature suggests the presence of hard objects in the diet in both individuals. The most common phytoliths recovered from the dental calculus are from wood and bark, which represent relatively tough materials. Thus, to some degree the two types of food that present mechanical challenges in terms of food breakdown – but are optimally performed by different processes – were part of this species’ diet. There is no reason to expect that Au. sediba was not capable of effective breakdown of hard and/or tough items, irrespective of the constraints on maximal adductor recruitment. Indeed, Au. sediba had the capacity to produce absolutely high bite forces by virtue of body size alone.

The morphological attributes of P. robustus have led a number of workers to surmise its dietary proclivities. In particular, its characteristically diminutive incisors and canines, large and thickly enameled postcanine teeth, molarized premolars, thick palate, robust mandible, and its orthognathic face with large, anteriorly positioned malars are part of a coherent and distinctive package that has been seen as reflecting the generation and distribution of powerful chewing forces associated with a herbivorous diet (Grine, 1981, Kay, 1985, Rak, 1983 and Robinson, 1954). Most workers have interpreted the trophic morphology of P. robustus as being consistent with a diet that likely included at least some hard foods. Many of the craniodental features that characterize P. robustus are possessed also by P. boisei in even more exaggerated form. Thus, P. boisei exhibits greater molarization of the premolars, an even deeper maxilla, zygomatics that tend to be laterally bowed forming a “visor”-like configuration, and temporalis and especially masseter muscles that appear to have had more anterior attachments (Rak, 1983 and Rak, 1988). The dimensions of P. boisei premolars and molars are the largest recorded for any hominin species.

Lucas et al. (1986: 269) have suggested that the M1–M3 size ratio of P. boisei indicates the consumption of “small mouthfuls of leaves and seeds”. The comparatively large masticatory muscles would have been able to generate powerful forces, but these could have been reduced to unexceptional levels if dissipated uniformly across the postcanine occlusal table. This would suggest bulk processing of soft, pliant foods (Demes and Creel, 1988; see also Walker, 1981). While postcanine megadontia is a defining characteristic of P. boisei, tooth size per se does not explain the size of the mandible (Plavcan and Daegling, 2006). Its large, thick corpora would have been able to withstand powerful bending and twisting moments (Daegling, 1989 and Hylander, 1988). Rak and Hylander (2008) have concluded that the considerable height of the mandibular ramus in Paranthropus, the more rostrally positioned jaw muscles and resulting small gape, the topography of the glenoid fossa, the bulbous, low topographic relief of the unworn premolar and molar cusps, and the small, non-projecting canines are all linked to a “rotatory” motion of the mandible during chewing. They further postulated that these derived masticatory movements are linked to the unusually broad mandibular corpora. The question, of course, is what the mandible of P. boisei was used to chew.

The problem with interpreting mandibular form vis-à-vis diet was highlighted by Hylander's (1979) observation that the integrity of mandibular bone may be challenged both by intermittent high strains (e.g., those associated with fracturing hard objects) as well as by low, long-term cyclical strains (e.g., those producing fatigue from prolonged chewing of tough objects). The structural solution to both is the same for reducing (or maintaining) stress: increase the amount of bone. Paranthropus jaws are “robust”, being thick and exceptionally deep, but determination of diet is likely to be inaccessible from mandibular morphology (Daegling and Grine, 2017). Beyond the proposition that whatever was ingested required unusually large masticatory effort to process (numerous cycles, considerable force, or both), the diets of P. robustus and P. boisei remain uncertain on the basis of the morphology of their jaws.

Many workers have understandably viewed the craniodental features of P. boisei as specialized adaptations to a diet that likely consisted of small, hard objects. On the other hand, Cachel (1975) and Szalay (1975) independently argued that this species was adapted for meat-eating and especially bone-crushing. This view has not found much favor. Wood and Strait (2004) envisioned P. boisei as having been an ecological generalist, capable of consuming a broad array of foods with the sole exception of “tough, fibrous food items”. This is because the low, bulbous cusps of the cheek teeth would seem to be inconsistent with a diet consisting of leaves and grass blades (Lucas, 1982, Lucas et al., 2008a and Rak and Hylander, 2008). However, the combination of postcanine megadontia, mandibular corpus hypertrophy and thick enamel does not compel an interpretation that P. boisei was adapted for consumption of hard objects. It is neither necessary nor prudent to assume that the bunodonty possessed by P. boisei represents an optimal solution for the comminution of its food. Rather, phylogenetic considerations would suggest that other topographic solutions (e.g., high, pointed cusps and/or well-developed, sharp lophs) should not be expected in this species. It is equally plausible that the bulk processing of low quality fibrous foods drove selection in this lineage (Rabenold and Pearson, 2011).

Paranthropus boisei facial form has been generally interpreted as having been adapted to withstand the feeding stresses associated with high bite forces (e.g., Rak, 1978, Rak, 1983 and Rak and Howell, 1978). The notion that the diet of this “hyper-robust” species included hard objects clearly gained indirect support from the microwear of its South African cousin, P. robustus, which indicates the consumption of hard food items, at least by some individuals (Grine, 1986, Grine and Kay, 1988 and Scott et al., 2005). Most recently, FEA analyses of a P. boisei (OH 5) cranium concluded that it was structurally strong, allowing for high bite forces to be generated efficiently along the postcanine tooth row (Smith et al., 2015 and Wroe et al., 2010). This conclusion, which follows from the work of Ward and Molnar (1980) and Rak and Hylander (2008), was based on the observation that the P. boisei facial strain pattern differs substantially from that in chimpanzees (P. troglodytes). These FEA analyses substantiated what simpler models had predicted from basic mechanical principles.

The possible biomechanical significance of the large amount of overlap of the temporal and parietal along the squamosal suture in P. boisei has been a subject of interest since it was first recognized by Rak (1978). This has been most recently explored through FEA by Dzialo et al. (2014), whose models suggest that this sutural morphology was better suited to resist shear than tension. While this is yet one more line of evidence pointing to a uniquely derived morphology indicating “heavy” masticatory demands, the particulars of diet are not discernible.

An FEA analysis of a large, extinct lemur (Hadropithecus) with some superficial craniofacial resemblances to P. boisei has led to the conclusion that the hominin must have consumed hard objects because this sort of diet is inferred for Hadropithecus, a taxon that also exhibits moderately elevated δ¹³C values (Dumont et al., 2011 and Godfrey et al., 2011). However, as noted by Cerling et al. (2011b), despite the existence of certain morphological convergences between these taxa, it is far from certain that this implies similar diets. Rather, the moderately high δ¹³C values of Hadropithecus probably reflect the consumption of plants that photosynthesize utilizing the Crassulacean acid metabolism (CAM) pathway rather than the eating of C₄ plants, which have been suggested to dominate P. boisei diets (Cerling et al., 2013 and Cerling et al., 2011a). This is pertinent because the former are highly abundant in the unique spiny forests of Madagascar but comprise virtually nothing of the possible dietary staples in Africa. As concluded by Cerling et al. (2011b), inferring diet for one extinct species from the inferred albeit unknown diet of another fossil taxon, especially one that is very distantly related and inhabited a very different environment, represents an inherently uncertain line of evidence that should be eschewed.

Thus, the form of the skull in Paranthropus has been taken to indicate the reliance on hard items such as nuts and seeds (Peters, 1987), a reliance on tough, fibrous foods (Du Brul, 1977), and the processing an unexceptional diet in greater quantity (Walker, 1981). These, in turn, translate respectively to higher bite forces, an increased number of masticatory cycles per day with the likelihood of high bite forces, and more daily chewing cycles but without appreciable change in occlusal forces. Unfortunately, and despite efforts to construct FEA models to answer this question, it is currently difficult to decide among these three possibilities. The outstanding issue is that both load magnitude and frequency are implicated in bone metabolic activity that results in increased bone mass and altered osseous architecture (Judex et al., 2006, Ozcivici et al., 2009, Qin et al., 1998, Ravosa et al., 2007, Rubin et al., 1990 and Rubin et al., 1991). Bone formation, composition and architecture are influenced by dynamic variation in the magnitude and frequency of mechanical loads (Biewener and Bertram, 1993, Biewener et al., 1986, Bouvier and Hylander, 1981, Bouvier and Hylander, 1984, Bouvier and Hylander, 1996 and Lanyon and Rubin, 1985).

Finally, Scott et al. (2014) have provided at least some experimental evidence which argues that hard object feeding – whether habitual or on a “fallback” resource basis – does not explain the extreme dentognathic morphology of P. boisei. Their analysis, although based on a relatively small number of data points, suggests P. boisei morphology is better explained by reliance on tough foods that required prolonged postcanine processing and “concomitantly elevated masticatory stresses owing to higher repetitive loading and longer load durations resulting from extended bouts of milling and grinding” (Scott et al., 2014: 4). This is concordant with the argument presented by Rabenold and Pearson (2011) for extended bouts of processing of low quality, fibrous foods.

As detailed by Grine et al. (2012), molar microwear studies of P. robustus have provided some support for the notion that hard food objects comprised at least a portion of its diet, and texture analyses have revealed significant individual variation, which may reflect reliance upon less commonly eaten, but critical fracture-resistant foods by some individuals in the sample (Grine, 1986, Grine and Kay, 1988 and Scott et al., 2005). Such potential seasonality in diet is also evidenced by the carbon isotope data (Sponheimer et al., 2005 and Sponheimer et al., 2006). Tooth enamel δ ¹³C values suggest a diet comprising some 35–40% C₄ vegetation, but with considerable interindividual variability. Importantly, this variability is also manifest within individual teeth, with changes of up to 5^o/_oo in δ ¹³C values over intra-annual (seasonal) and inter-annual timescales (Sponheimer et al., 2006). In view of the fact that mineralization continues for months after initial enamel deposition, which results in isotopic overprinting that can dampen apparent dietary perturbations (Balasse, 2003 and Passey and Cerling, 2002), changes of this magnitude are quite extraordinary, and almost certainly indicate that the individuals in question changed from a C₃- to a C₄-dominated diet during the period of enamel formation.

The first examination of molar microwear in P. boisei was a qualitative SEM study by Walker (1981), who reported that the teeth exhibited fine scratches and few, small pits. As such, he suggested that the diet of P. boisei may have entailed prolonged chewing of tough or fibrous vegetable foods, perhaps with relatively little nutritional value, rather than the consumption of hard food items. This inference, which appears to have been profoundly prophetic, was largely ignored. Subsequent study has confirmed that the occlusal wear fabrics of this species are dominated by fine striae, and that they appear to show remarkable uniformity over the geochronological record of the species (Ungar et al., 2008). The carbon isotope values obtained from P. boisei specimens from a variety of sites throughout East Africa are indistinguishable from those of C₄ grass consumers from the same regions (Cerling et al., 2011a and van der Merwe et al., 2008). Like the microwear signatures, the δ¹³C values do not change over the half million years sampled, and they correspond to a diet where C₄ plants comprise, on average, some 77% of the biomass consumed, with individual values ranging between 61% and 91%. As such, the carbon isotope data for P. boisei are most consistent with its having been a reasonably devoted consumer of C₄ plants.

Paranthropus robustus provides an intriguing mixture of isotopic and microwear signatures that suggest a greater range of dietary variability than P. boisei, with the inclusion, perhaps seasonally, of hard food objects. By contrast, microwear and isotopic data for P. boisei point to a diet dominated by C₄ plants (likely grasses) that did not entail the processing of hard objects. Despite the popularity of P. boisei's nickname, these data would appear to portend the demise of “Nutcracker Man” (Lee-Thorp, 2011).

Paleoanthropology does not lack for ideas, but, given the spotty nature of the fossil record, there are relatively few “facts” with which to work. Models are essential to the enterprise of functional morphology, and they will remain vital for testing and generating new hypotheses. Box and Draper (1987: 424) famously remarked that “all models are wrong, but some are useful.” Both parts of this statement demand that models be scrutinized. Vigilance as to their underlying assumptions will lead to more useful models and perhaps fewer “wrong” ones.

This review has been critical of FEA as an approach to paleoanthropological inference – specifically the retrodiction of diet and dietary adaptation. We do not advocate the abandonment of the approach, but propose that these models should first be utilized to investigate tractable questions of importance to skeletal biology and model design, such as sutural (Wang et al., 2012), periodontal ligament (Gröning et al., 2011 and Wood et al., 2011), and muscular (Ross et al., 2005) influences on bone behavior and model accuracy. Iterative FEA models (e.g., Carter et al., 1989, Prendergast, 1997 and Tseng, 2013) show great promise for resolving how bone modeling and remodeling algorithms alter morphology and performance. Although these will not resolve every operational problem in the application of FEA to fossils, they offer an incremental approach toward understanding skeletal adaptation. This can serve to improve our discrimination, while avoiding the potential pitfalls of trying to correlate simple mechanical variables (e.g., a strain gradient) with ecological adaptations.

In effect, we recommend a less ambitious role of FEA in hominin paleobiology. Even should an FEA model be judged accurate in all details, which is unlikely (Fitton et al., 2015 and Godinho et al., 2017), it is perilous to assume that such models provide clear insight into dietary adaptation, much less paleoecology. In most applications of the method to hominin fossils, one or a handful of loadcases are sampled to infer performance. These stress fields represent a subset of the totality of loading situations that comprise the stress history of an individual skull (Curtis et al., 2008). Thus, a strain gradient from a single – even if critical – loading event provides a myopic description of performance in the context of understanding skeletal adaptation. Biting deforms the skull, but unlike a nutcracker in the kitchen drawer, the skull has to do a number of other things besides break food.

We should also be mindful that the morphology of a single individual (e.g., STS 5, OH 5) may not be an adequate guide for understanding morphological adaptation in a species. In the course of skeletal growth and modeling, not every recognizable feature will have arisen strategically to provide structural reinforcement. Features that arise as a matter of physical necessity [Gould and Lewontin's (1979) “spandrels” or Seilacher's (1973) “fabricational noise”] need not qualify as adaptations if their appearance is an inevitable product of morphogenesis. In addition, while the use of single specimens to characterize a fossil species is perfectly understandable for preservational and perhaps computational reasons, it necessarily places interpretation in an essentialist framework. A sample of one means that variance is zero. The problem this poses is thrown into sharp relief when considering the reality of phenotypic plasticity. As Ravosa et al. (2016) have noted, ontogenetic perturbations of diet that alter masticatory load magnitude and frequency can produce intraspecific variations that may be on par with the differences attributed to “species-level” adaptations. This underscores the pitfall of using a monolithic “critical” loadcase as the determinant of an adaptive response. In this context, what appears as morphological “noise” within a species may, in fact, constitutes a functional signal.