Cranial biomechanics of Diplodocus (Dinosauria, Sauropoda): testing hypotheses of feeding behaviour in an extinct megaherbivore
Sauropod dinosaurs were the largest terrestrial herbivores and pushed at the limits of vertebrate biomechanics and physiology. Sauropods exhibit high craniodental diversity in ecosystems where numerous species co-existed, leading to the hypothesis that this biodiversity is linked to niche subdivision driven by ecological specialisation. Here, we quantitatively investigate feeding behaviour hypotheses for the iconic sauropod Diplodocus. Biomechanical modelling, using finite element analysis, was used to examine the performance of the Diplodocus skull. Three feeding behaviours were modelled: muscle-driven static biting, branch stripping and bark stripping. The skull was found to be ‘over engineered’ for static biting, overall experiencing low stress with only the dentition enduring high stress. When branch stripping, the skull, similarly, is under low stress, with little appreciable difference between those models. When simulated for bark stripping, the skull experiences far greater stresses, especially in the teeth and at the jaw joint. Therefore, we refute the bark-stripping hypothesis, while the hypotheses of branch stripping and/or precision biting are both consistent with our findings, showing that branch stripping is a biomechanically plausible feeding behaviour for diplodocids. Interestingly, in all simulations, peak stress is observed in the premaxillary–maxillary ‘lateral plates’, supporting the hypothesis that these structures evolved to dissipate stress induced while feeding. These results lead us to conclude that the aberrant craniodental form of Diplodocus was adapted for food procurement rather than resisting high bite forces.
KeywordsFinite element analysisPalaeobiologyHerbivorySauropod dinosaur
Sauropod dinosaurs were the largest terrestrial tetrapods in Earth’s history. Throughout the Jurassic and Cretaceous, their size continued to increase, reaching maximum estimated body masses of 70 tonnes or more (Upchurch et al. 2004). Remarkably, in the Late Jurassic numerous species of these multi-tonne herbivores exhibited high degrees of sympatry, as shown by the sauropod faunas preserved in the Morrison Formation of western North America, the Tendaguru Beds of Tanzania and the Upper Shaximiao Formation of China (Upchurch and Barrett 2000; Mannion et al. 2011). Recent studies have suggested that niche partitioning, maintained via morphological differentiation, enabled this high biodiversity (Upchurch and Barrett 2000; Whitlock 2011). In the Morrison Formation and elsewhere, this differentiation relates to craniodental morphology and function, as well as to differences in postcranial anatomy, such as fore to hindlimb length ratios and neck function. The Morrison sauropod fauna includes at least eight genera potentially containing around 18 species: Amphicoelias, Apatosaurus, Barosaurus, Brachiosaurus, Camarasaurus, Diplodocus, Haplocanthosaurus and Suuwaasea (Mannion et al. 2011). Among these, diplodocoids, best characterised by Diplodocus itself, have extremely unusual craniodental morphologies. Among other features, the rostrum is elongate with the external nares strongly retracted posterodorsally to lie above the orbits; the tooth row is restricted to the anterior-most margin of the upper and lower jaws; the tooth crowns are apicobasally elongate with slight-to-moderate labiolingual compression and oblique wear facets on the labial surface (see the electronic supplementary material, Fig. S1), and the articular fossa was rostrocaudally elongate and shallow (a morphology often associated with translational mandibular movement or propaliny) (Holland 1906; Barrett and Upchurch 1994; Calvo 1994; Wilson and Sereno 1998; Upchurch and Barrett 2000). The overall craniodental morphology and tooth macro- and microwear have generated numerous conflicting hypotheses of feeding behaviour for Diplodocus. For example, it has been suggested that the teeth of Diplodocus were simply used during standard vertical occlusion for the slicing of vegetation (Calvo 1994). Other authors have argued that the procumbent teeth were incapable of occlusion (Barrett and Upchurch 1994). This, coupled with unusual patterns of tooth wear, in which both upper and lower tooth wear facets face labially (electronic supplementary material, Fig. S1B), gave rise to a unique ‘branch stripping’ model of feeding, with Diplodocus raking or combing its teeth through plant matter to pull leaves and shoots from branches using either the mandible or upper jaw independently (unilateral branch stripping; Barrett and Upchurch 1994) or using upper and lower jaws concurrently (bilateral branch stripping; Barrett and Upchurch 1994; Coombs 1975). Other suggestions have included precision plucking (associated with molluscivory; e.g. Sternfeld in Holland 1910), raking seaweed from rocks (Holland 1906), piscivory (Tornier 1911) and bark stripping (Holland 1924; Bakker 1986). More recently, Whitlock (2011) examined dental microwear patterns across numerous diplodocoid species and concluded that branch stripping was less plausible than other hypothesised feeding behaviours. Understanding the function and ecology of extinct organisms poses particular challenges, especially in the case of sauropods where no direct extant analogue exists.
One way to quantify, compare and better understand the function and mechanical behaviour of unusual morphologies is through biomechanical modelling. One such technique is finite element analysis (FEA), a computational method that reconstructs stress and strain within a structure after the application of performance-related loads. Finite element analysis has been used with increasing frequency in palaeontology and zoology to assess biomechanical performance (e.g. Dumont et al. 2005; Richmond et al. 2005; Rayfield 2007). Results of an FEA reveal ‘hot spots’ of functional stress, strain or deformation, the intensity of which can then be related to morphological features and the loading environment. This offers a means to test the biomechanical consequences of unusual or extreme morphologies. In the simplest sense, making bones thinner or thicker can increase or decrease stresses accordingly, and changing the load magnitudes and loading profile will modify the mechanical behaviour of the structure. It follows, therefore, that the different feeding regimes postulated for Diplodocus will generate different signature mechanical behaviours in the skull. ‘Finite element structure synthesis’ has previously been applied to principles of sauropod skull architecture (Witzel et al. 2011), but finite element methods have not been used to provide rigorous tests of sauropod feeding hypotheses thus far.
We used FEA to subject the skull of Diplodocus to simulated feeding loads for three hypothesised feeding behaviours: muscle-driven static biting (occlusion), unilateral branch stripping and bark stripping. The aim of our analysis is to test how occlusion and branch stripping comparatively influence skull biomechanics. We then test how behaviours, such as bark stripping, further compromise skull behaviour. Although bark stripping has received little support in the literature, testing of this hypothesis was deemed appropriate in order to elucidate how the Diplodocus skull would have responded to a range of feeding-induced mechanical forces. Finally, we comment on the relative likelihood of each postulated feeding scenario.
Materials and methods
A three-dimensional model of a complete Diplodocus longus skull (Carnegie Museum (CM) of Natural History, Pittsburgh, PA, CM 11161; electronic supplementary material, Fig. S1) was created using computed tomographic scanning. The skull was scanned at the O’Bleness Memorial Hospital (Athens, OH, USA), producing 290 coronal slices, with 512 transaxial slices in both perpendicular planes, separated by 0.2-mm intervals. The fossil bone was digitally extracted from the surrounding rock matrix using the three-dimensional imaging software amira (v. 4.1.2, Mercury Computing Systems, USA; electronic supplementary material, Fig. S2A). It was necessary to digitally correct some bones in amira due to breakage, distortion and other imperfections of the fossil (Electronic supplementary material for further details). The 3D model was compared to cranial material held in the Carnegie Museum and the National Museum of Natural History (Washington, D.C.) to aid reconstruction and to improve accuracy. Segmented slice data from the amira 3D model was imported into scanip v. 2.1 Build 149 (Simpleware Ltd, UK) to produce a smoothed skull surface model. In scanfe v. 2.0 (Simpleware Ltd, UK) the three-dimensional surface model was meshed, creating a solid geometry consisting of approximately 91 % tetrahedral and 9 % hexahedral elements. The final mesh had 906,257 elements (electronic supplementary material, Fig. S2B).
Material properties and boundary conditions were assigned to the mesh using the finite element (FE) software abaqus (v. 6.7, Simulia, USA). The exact material properties, constraints and loading regime in extinct taxa cannot be modelled with complete accuracy (Richmond et al. 2005), and FE studies on extinct taxa are also hindered by the inability to directly measure material properties from fossil bone (Rayfield 2007). Nevertheless, as adult Late Jurassic neosauropod skeletons (such as that of Diplodocus) are characterised by Haversian bone (Curry 1999), material properties of extant histological analogues were used as a proxy. Three tissue types were modelled: cranial bone, dentine and enamel, all of which were treated as homogenous and isotropic materials to avoid introducing additional assumptions (see the Electronic supplementary material for further details). Both the constraints and loading regimes must be as biologically realistic as possible to ensure feeding biomechanical performance is adequately tested. To model standard occlusion, the apical surfaces of the premaxillary tooth crowns were constrained from moving dorsoventrally using a distributing coupling constraint (DCC) in abaqus v. 6.7. The DCC evenly distributes the boundary constraint across the apical surfaces of the teeth, but rather than directly fixing nodes; it uses a series of rigid links to connect the teeth to a constraint control point located ‘in space’, ventral to the apex surfaces (see the electronic supplementary material, Fig. S3). This type of constraint minimises artificially high stresses generated at the teeth when this region is directly constrained from movement, thus producing a more realistic distribution of cranial stress. We present here data from models constrained at four premaxillary teeth (two left and two right), thereby simulating biting a piece of vegetation roughly equivalent in size to a small tree branch. Sensitivity analyses constraining either 8 or 14 teeth that are intended to model bites on larger objects are explained in the Supplementary information. The mandibular condyles of both quadrates were constrained using a DCC, preventing transverse and translational motion (see the electronic supplementary material for condylar reaction forces and Fig. S3A). The loading regime in the FE model simulation replicated the contraction of six jaw-closing muscles (musculus (M.) adductor mandibulae externus superficialis, M. adductor mandibulae externus profundus, M. adductor mandibulae posterior, M. pterygoideus dorsalis, M. pterygoideus ventralis and M. pseudotemporalis superficialis) (see the electronic supplementary material, Fig. S6).
The method used to create the jaw-closing muscle reconstructions has previously been outlined (Holliday 2009) and is explained further in the Electronic supplementary material. In order to visually compare the FE models, von Mises stresses were plotted, as they indicate regional deformation as a function of the three principal stresses, and are good predictors of failure under ductile fracture (Dumont et al. 2005), which elastic materials, such as bone, undergo (Nalla et al. 2003). In order to model the branch- and bark-stripping feeding regimes, an additional force was applied directly to the apices of the premaxillary teeth. For the branch-stripping model, the additional force was estimated from the shear strength of plant parenchyma, 1 MN m−2 (Niklas 1992; K. J. Niklas, pers. comm.). The area of tooth-food contact was calculated for a single tooth by measuring the surface area of the apical surface of the tooth. This area was then multiplied by the number of teeth considered to contact the plant matter (4, 8 and 14 teeth, see above). Forces required to shear parenchyma (Niklas 1992) were calculated as: area of tooth apical surface (s) in contact with plant matter (m2) × shear stress. For the ‘four-tooth’ contact model, this equated to a force of 100 N (for other tooth contact models, see calculations in the electronic supplementary material, Table S2). This branch-stripping force was directed anteriorly (i.e. parallel to the long axis of the skull; see Electronic supplementary information). This was to simulate resistance of the parenchyma as the animal retracted its head to detach plant matter. For the bark-stripping model, the additional force was estimated from the shear strength of linden wood (a wood of intermediate strength (see Niklas 1992; 730 N for four-tooth contact). This force was directed along the apicobasal axis of the loaded teeth (see Electronic supplementary information).
Due to uncertainties in the exact mechanical properties of the Diplodocus skull and associated soft tissues, we do not rely on the absolute values of the von Mises stresses generated by the analyses, as these are likely to differ somewhat from measurements that could have only been obtained in vivo. Instead, we use relative differences in the extent and magnitude of the stresses identified when comparing the results from our models—such differences still allow major functional trends to be identified with confidence.
The regions with the thickest bone, the dentigerous portion of the maxilla and the bones forming the braincase and skull roof experience very low levels of stress, although the dentigerous portion of the premaxilla exhibits higher stress. Stresses at the synovial basipterygopterygoid and quadratosquamosal joints are low (Fig. 1a; electronic supplementary material, Fig. S7). The low stress at these joints and the hypothesised reduction of the M. protractor pterygoideus are congruent with the interpretation that the Diplodocus skull was akinetic (Upchurch and Barrett 2000; Holliday and Witmer 2008). The teeth experience high stresses, yet these may be artificially inflated in this region by excluding the periodontal ligament from our model (Toms and Eberhardt 2003; Cattaneo et al. 2005; Gröning et al. 2011; Panagiotopoulou et al. 2011) and the proximity to the constraint fixing the teeth. Further regions of moderate stress, which may also be artificially inflated (because of their proximity to the fixed jaw joint), are the proximoventral surface of the basipterygoid processes and along the dorsal edge of the left quadratojugal-quadrate suture (Fig. 1a; electronic supplementary material, Fig. S7). The overall patterns of von Mises stress in the static biting and branch-stripping models are very similar (Fig. 1a, b; electronic supplementary material, Figs. S7 and S8). Minor differences in the magnitudes of the stresses in these models result from the additional load that was applied to the branch-stripping model in order to represent the force needed to detach leaves/stems from the parent plant.
By contrast, the bark-stripping model (Fig. 1c, electronic supplementary material, Fig. S9) has the same regions of peak high stress as the other feeding regimes, and yet, stresses are of a far higher magnitude. Generalised low stress is more extensive across the skull in the bark-stripping model, and large regions of moderate stress are present in the thinnest bones of the skull. Moderate levels of stress occur at: the palatal surface of the maxilla, especially around the maxillary-ectopterygoid suture; the lateral surface of the maxilla (especially along the ventral margin) and the lateral surface of the quadratojugal. There are localised peaks of high stress present along both the dorsal and ventral margins of the quadratojugal (Fig. 1c; electronic supplementary material, Fig. S9). A notable increase in stress magnitude in the bark-stripping model occurs at the dentigerous region of the premaxilla and palatal surface of the premaxilla and maxilla adjacent to the teeth. A region of low von Mises stress along the lateral margin of the maxillae is also more extensive in the bark-stripping model than in the other two models (Fig. S10). Comparisons between all three feeding models for bites, involving 4, 8 or 14 teeth, are provided in Fig. S10.
The skull of Diplodocus was generally subjected to low stress when muscle-only contractile forces are modelled. Although we cannot state that our modelled stresses are generated by the exact loads that the skull experienced during life, the adductor muscle reconstructions are as accurate as currently possible, and the resulting loads are low enough in magnitude that the skull appears not to be compromised by muscle-driven biting, consistent with the ‘horizontal slicing’ hypothesis (Calvo 1994). Indeed, it would be unexpected and indicative of error in our models if the skull could not withstand the load generated by its own adductor muscles. These results are even more interesting when we consider that 100 % muscle contraction and the highest specific tension of adductor muscles are assumed. These assumptions likely overestimate the muscle contractile forces. Our results are also fully consistent with the hypothesis that Diplodocus could branch strip (Coombs 1975; Barrett and Upchurch 1994), given the low levels of stress that are observed in these feeding models. On the basis of these results, it is plausible that Diplodocus was capable of using both standard vertical biting and branch stripping to harvest foliage (Christiansen 2000). However, the absence of a precise occlusion (Barrett and Upchurch 1994; contra Calvo 1994) suggests that under either of these scenarios, the teeth were primarily used to grip (rather than shear through) vegetation, which would then be detached from the parent plant by retraction or rotation of the head relative to the parent plant. In both models, the skull is not ‘overloaded’, as even with very high muscle tension of 392 kPa stresses in the skull do not, by some margin, exceed what is physiologically unsafe for bone.
The bark-stripping model produced a very different magnitude of stress. In the thinnest bones of the skull, moderate to very high peaks of stress were observed. In conjunction with the remarkably high stresses that the teeth endured, it is unlikely that the dentition or teeth-bearing bones of the skull could have withstood the forces involved in stripping bark from trees. Consequently, we reject the hypothesis that Diplodocus could have fed by bark stripping. Nevertheless, the bark-stripping FE models indicate the upper limits of Diplodocus skull performance, which is of value given the much lower stresses encountered during biting and branch stripping.
In all simulations, localised peaks of high stress are observed in the lateral plates immediately adjacent to the premaxillary teeth (electronic supplementary material, Fig. S1B). The lateral plate is a thin lamina of bone that extends from the main bodies of the dentigerous portions of the tooth-bearing bones to partially cover the bases of functional tooth crowns (Upchurch 1995). Our results support the hypothesis that these structures assist in dissipating feeding-induced stresses that are acting on the bases of the teeth (Upchurch and Barrett 2000). In addition, localised peaks of high stress are absent from the synovial joints, supporting the hypothesis that Diplodocus had a functionally akinetic skull (Upchurch and Barrett 2000; Holliday and Witmer 2008).
High stresses occur at the tooth bases, suggesting that these areas may have been vulnerable to mechanical failure, irrespective of the feeding behaviour adopted. Interestingly, diplodocoid sauropods had the highest tooth replacement rates of any vertebrates; in Nigersaurus, new teeth erupted every 30 days, whereas in sauropods with broad crowned teeth (e.g. Camarasaurus), the replacement rate has been estimated as 62 days (Sereno et al. 2007). It has been suggested that the narrow crowned tooth morphology of diplodocoids, which allows close packing of the teeth within the jaws, evolved in concert with this increase in tooth replacement rates, and that these features were correlated with high rates of tooth wear generated during browsing close to ground level (Chure et al. 2010). However, it is equally plausible that high tooth replacement rates were correlated with a need to regularly replace teeth that were subjected to high stresses, either as a result of static biting or branch stripping, and that might have suffered consequent high rates of tooth loss. Neither of these hypotheses is mutually exclusive.
Diplodocus is not the first extinct taxon with a skull seemingly over engineered for muscle-driven biting. The FE analysis of the contemporaneous theropod dinosaur Allosaurus obtained a similar result (Rayfield et al. 2001). While the skull must be able to accommodate a variety of different functions (making it unlikely to be optimised exclusively for feeding), it is unusual that the skull is so resistant to feeding stresses. However, it should be noted that the analyses modelled herein are static, and the cervicocranial musculature would have either retracted or stabilised the head during feeding. The results of our analyses lead us to postulate that the unusual craniofacial form of Diplodocus was plausibly an adaptation for certain behavioural strategies associated with food procurement (e.g. branch stripping) and not simply a response to resisting the bite forces produced during jaw closure.
We thank Clint Davies-Taylor and Neil Gostling for their computing assistance, Karl Niklas and Steven Vogel for their discussion on plant biomechanics, Mike Brett-Surman and Amy Henrici for specimen access, Heather Rockhold for CT scanning and John Whitlock and an anonymous reviewer for their comments on a previous version of this article. We gratefully acknowledge the financial support of the Natural Environment Research Council (NER/S/A/2006/14058) and the Natural History Museum London (awarded to EJR and PMB) and the National Science Foundation (IBN-0407735 to LMW and CMH, and IBN-0343744 and IOB-0517256 to LMW).