Shape and mechanics in thalattosuchian (Crocodylomorpha) skulls: implications for feeding behaviour and niche partitioning

Authors


Stephanie E. Pierce, University Museum of Zoology, Downing Street, Cambridge, CB2 3EJ, UK. T: +44 (0)1223336649; F: +44 (0)1223336679; E: sep55@cam.ac.uk

Abstract

Variation in modern crocodilian and extinct thalattosuchian crocodylomorph skull morphology is only weakly correlated with phylogeny, implying that factors other than evolutionary proximity play important roles in determining crocodile skull shape. To further explore factors potentially influencing morphological differentiation within the Thalattosuchia, we examine teleosaurid and metriorhynchid skull shape variation within a mechanical and dietary context using a combination of finite element modelling and multivariate statistics. Patterns of stress distribution through the skull were found to be very similar in teleosaurid and metriorhynchid species, with stress peaking at the posterior constriction of the snout and around the enlarged supratemporal fenestrae. However, the magnitudes of stresses differ, with metriorhynchids having generally stronger skulls. As with modern crocodilians, a strong linear relationship between skull length and skull strength exists, with short-snouted morphotypes experiencing less stress through the skull than long-snouted morphotypes under equivalent loads. Selection on snout shape related to dietary preference was found to work in orthogonal directions in the two families: diet is associated with snout length in teleosaurids and with snout width in metriorhynchids, suggesting that teleosaurid skulls were adapted for speed of attack and metriorhynchid skulls for force production. Evidence also indicates that morphological and functional differentiation of the skull occurred as a result of dietary preference, allowing closely related sympatric species to exploit a limited environment. Comparisons of the mechanical performance of the thalattosuchian skull with extant crocodilians show that teleosaurids and long-snouted metriorhynchids exhibit stress magnitudes similar to or greater than those of long-snouted modern forms, whereas short-snouted metriorhynchids display stress magnitudes converging on those found in short-snouted modern species. As a result, teleosaurids and long-snouted metriorhynchids were probably restricted to lateral attacks of the head and neck, but short-snouted metriorhynchids may have been able to employ the grasp and shake and/or ‘death roll’ feeding and foraging behaviours.

Introduction

One of the most intriguing groups of crocodylomorphs to evolve and diversify through the Mesozoic was the Thalattosuchia – a specialized clade of longirostrine, marine-adapted crocodiles. Although the morphological adaptations of teleosaurid thalattosuchians to marine life were limited, those of metriorhynchid thalattosuchians were much more extensive. Teleosaurids retained limbs that could function in walking and heavy dermal armour, whereas metriorhynchids evolved hydrofoil-like forelimbs, a hypocercal tail (formed by the downwards deflection of the posterior-most caudal vertebrae), reduced girdle size, and a complete loss of dermal armour (Buffetaut, 1982). These anatomical differences indicate divergent locomotory habits, implying that teleosaurids were restricted to nearshore environments, whereas metriorhynchids could exploit the open seas (Hua, 1994).

In addition to anatomical characteristics, histological and physiological data support the separation of habitat space between these two thalattosuchian families (Hua & Buffrenil, 1996; Fernández & Gasparini, 2000, 2008; Gandola et al. 2006). Teleosaurids display a bone histology that is very similar to that of modern crocodiles, which possess no special adaptations for sustained pelagic life (Hua & Buffrenil, 1996). Conversely, metriorhynchids have a modified bone histology that includes extensive osteoporotic lightening of the skull, ribs, and limbs providing increased buoyancy within the water column (Hua & Buffrenil, 1996). In addition, metriorhynchids likely possessed a hypertrophied nasal salt excretion gland to maintain constant plasma osmolality even when seawater or osmoconforming prey were ingested (Fernández & Gasparini, 2000, 2008; Gandola et al. 2006).

The combination of anatomical, histological, and physiological data indicates that teleosaurids and metriorhynchids most likely had different preferred feeding and foraging zones and thereby were able to avoid direct competition. Teleosaurids were semi-aquatic crocodiles that lived in shallow coastal waters, occasionally entering the open seas, and most likely foraged near the substrate. Metriorhynchids were open pelagic swimmers that would have hunted within the water column (Hua, 1994; Hua & Buffrenil, 1996). Within each family, however, competition for resources would have been likely.

Remains of thalattosuchians are abundant in the fossil record and in many instances multiple species have been shown to exist sympatrically. An excellent example of this can be seen in the Middle Jurassic (Callovian) Oxford Clay Formation of Peterborough, UK (Hudson & Martill, 1994), which preserves a diverse assemblage of similarly sized thalattosuchian species belonging to the genera Steneosaurus and Metriorhynchus (Andrews, 1909, 1913; Adams-Tresman, 1987a,b; Pierce et al. 2009). The stability of such crocodile-rich ecosystems would have almost certainly depended on an intricate partitioning of resources – potentially through the process of character displacement (i.e. competitive interactions leading to morphological differentiation; Adams & Rohlf, 2000). Character displacement resulting from niche partitioning also helps to explain why some closely related thalattosuchian species have more disparate snout morphologies than would be expected based on their phylogenetic proximity (Pierce et al. 2009).

Differences in diet and feeding behaviour have often been considered to be of prime importance in morphological differentiation of the vertebrate skull (see Schwenk, 2000; Aerts et al. 2002). Variation in resource utilization promotes niche partitioning, allowing closely related species to coexist; it also stimulates the acquisition of morphological specializations in order to withstand the functional demands imposed on the skull by different food types, thus moulding the form–function complex. Ecomorphological studies have supported this idea by documenting statistical relationships between the morphology of a species, its functional attributes, and the ecological niche it occupies (e.g. Dodson, 1975; Herrel et al. 1999, 2001, 2002, 2005; Monteiro & Abe, 1999; Adams & Rohlf, 2000; Cakenberghe et al. 2002; Hulsey & Wainwright, 2002; Verwaijen et al. 2002; Claude et al. 2004; Metzger & Herrel, 2005; Stayton, 2006).

The extant crocodilian skull has often been used as a model system to examine correlations between structural design, functional parameters (e.g. mechanical performance, bite force), and ecology (e.g. diet, microhabitat) (e.g. Langston, 1973; Dodson, 1975; Webb et al. 1983; Magnusson et al. 1987; Taylor, 1987; Busbey, 1995; Daniel & McHenry, 2000; Erickson et al. 2003, 2004; Ross & Metzger, 2004; Metzger et al. 2005; McHenry et al. 2006). Until recently, however, little was known about the interplay between broad-scale patterns of skull shape variation in crocodilians and factors underlying those patterns. Pierce et al. (2008) analysed modern crocodilian skulls within a quantitative framework using geometric morphometrics and finite element modelling to evaluate the relationship between skull shape, phylogeny, biogeography, and mechanical performance. The study demonstrated that skull shape variation was more complex than often portrayed and that this variation is not a by-product of phylogenetic structure or biogeographic range, but rather stems from a form–function relationship. Thus, ecological specializations in terms of feeding and foraging strategies were suggested to be major controls on the morphological diversity of modern crocodilian skulls.

Thalattosuchians are distantly related to modern crocodiles (Clark, 1994) but display a similar pattern of shape variation and morphological integration of the skull as observed in extant forms (Pierce et al. 2009). A recent geometric morphometric study by Pierce et al. (2009) found that the main sources of variation in thalattosuchian skulls are the relative length and width of the snout, and that changes in these dimensions were correlated with size of the supratemporal fenestrae and length of the nasal bones. The same trends were observed in modern crocodiles (Pierce et al. 2008). In addition, evolutionary proximity was either not correlated with skull shape (explaining < 0.002% of the observed shape variation in teleosaurids) or only moderately so (explaining 20% of the observed shape variation in metriorhynchids). This implies that factors other than phylogenetic relationships play important roles in determining skull shape in the clade. Given the similarities in skull morphology between thalattosuchians and modern forms, an obvious question is whether a similar form–function relationship might exist in thalattosuchians. If such a relationship does exist, then the morphological diversity of thalattosuchian skull shape may be controlled by a similar combination of ecomorphological and functional pressures, suggesting these factors may have a strong, consistent influence on a large part of crocodylomorph evolution.

This study examines thalattosuchian skull shape and mechanical performance within a dietary context to facilitate our understanding of the link between morphological differentiation and ecological role in these unique marine crocodylomorphs. Our focus on diet stems from the fact that feeding is a major driver of skull shape in extant crocodilians and the similar patterns of skull shape variation in both modern forms and thalattosuchians suggest that feeding was of comparable importance in the latter taxon. We draw on thalattosuchian skull shape data from Pierce et al. (2009), and explore it using a unique combination of finite element modelling and multivariate statistics. Our main objectives are to quantify the biomechanical consequences of skull shape variation, and to assess the interplay between diet, mechanical performance, and skull shape. In addition, we investigate how thalattosuchians partitioned resources within their marine environment, and compare patterns of mechanical performance with modern crocodilians (Pierce et al. 2008) to evaluate potential thalattosuchian feeding behaviours.

Materials and methods

Defining morphospace

Pierce et al. (2009) used landmark-based geometric morphometrics to determine and describe the distribution of thalattosuchian skulls within principal components-based morphospaces, and also revised the taxonomy of the group. Shape coordinates on statistically significant axes of variation (only PC1 and PC2) from Pierce et al. (2009) were used in this study to calculate the mean skull shape of each valid thalattosuchian species. This new dataset was re-plotted along principal components 1 and 2 within teleosaurid (Fig. 1A) and metriorhynchid (Fig. 1B) morphospaces. These adjusted principal component scores are used as a measure of global skull shape when assessing the relationship between shape, mechanics and diet (see below).

Figure 1.

 (A) Teleosaurid morphospace and (B) metriorhynchid morphospace, showing the mean shape of each species plotted along principal components 1 and 2. Deformation grids indicate extreme shapes along PC axes. [See Pierce et al. (2009) for a complete analysis and description of morphospace.]

In teleosaurid morphospace, PC1 describes 66.5% of the total variance in the dataset and separates species based on length of the snout (especially the length of the maxilla), relative size and shape of the supratemporal fenestra and relative size and shape of the frontal bone; PC2 describes 14.5% of the total variance and separates species based on width of the snout, relative size and shape of the nasal bone, relative size and shape of the premaxilla lateral to the narial opening, and relative size and shape of the orbit [see Pierce et al. (2009) for a full analysis and description of morphospace]. The remaining PC axes did not describe significant amounts of shape variation among teleosaurids, and are not considered further here.

In metriorhynchid morphospace, PC1 describes 65.2% of the total variance in the dataset and separates species based on length of the snout (especially length of the maxilla), relative size and shape of the supratemporal fenestra and relative size and shape of the frontal bone; PC2 describes 8.95% of the total variance and separates species based on width of the snout (especially across the nasal bones), relative size and shape of the premaxilla, and position of the narial opening [see Pierce et al. (2009) for a full analysis and description of morphospace]. The remaining PC axes did not describe significant amounts of shape variation among metriorhynchids, and are not considered further here.

Because PC1 primarily describes shape differences associated with snout length and PC2 primarily describes shape differences associated with snout width in both groups, we refer to these components as ‘relative snout length’ and ‘relative snout width’, respectively, for ease of discussion. As noted above, however, these PC axes did capture sources of skull shape variation besides snout length and width. Furthermore, we use the four quadrants formed by PCs 1 and 2 to define four qualitative snout morphotypes (short/broad; short/narrow; long/broad; and long/narrow; Fig. 1) to assess whether such generalizations accurately describe the relationship between skull form and function in thalattosuchians.

Establishing diet and phylogeny

Preferred diet

The dietary preferences of thalattosuchian species were modified from Massare (1987). In her analysis of tooth form and stomach contents in Mesozoic marine reptiles, Massare defined seven working predator types or ‘feeding guilds’, and used them to discuss resource partitioning. Marine reptiles were found to fill a very specific adaptive zone – that of large, mobile, pelagic predators of macroscopic prey – and partitioning available food was essential for species to co-exist. Thalattosuchian species fall within six of the seven identified feeding guilds (Fig. 2):

Figure 2.

 Phylogenetic relationships of (A) Teleosauridae and (B) Metriorhynchidae. General skull morphology and feeding guild are plotted for the terminal taxa. Cladograms represent the consensus trees recovered from a branch-and-bound search using a modified version of the character matrices of Mueller-Töwe (2006) and Young (2006). Feeding guilds are as described by Massare (1987) for marine reptiles based on tooth form and stomach contents. Asterisks indicate a departure from Massare’s original guild allocation, whereas question marks indicate species where tooth morphology is poorly known.

  • 1Crunch guild, robust blunt teeth for feeding on armoured fish, crustaceans, and ammonites;
  • 2Cut guild, robust pointed teeth with cutting edges for capturing large fish and reptiles;
  • 3Smash guild, teeth bearing rounded points for grasping belemnoids and soft cephalopods;
  • 4Pierce I guild, long, delicate and sharply pointed teeth for piercing small fish and soft cephalopods;
  • 5Pierce II guild, smooth pointed teeth for piercing fish;
  • 6General guild, moderately pointed teeth with longitudinal ridges for piercing fish and smashing belemnoids.

As skull morphology was not used by Massare to construct dietary guilds, overestimation of correlations between diet, skull shape and mechanical performance should not be a fundamental problem in this study.

Phylogenetic relationships

Species-level taxonomy of the thalattosuchian families Teleosauridae and Metriorhynchidae were taken from Pierce et al. (2009) and the phylogenetic relationships were determined by modifying the taxon and character sets of Mueller-Töwe (2006) for the Teleosauridae and Young (2006) for the Metriorhynchidae. The interrelationships for each family are slightly different to those recovered by Pierce et al. (2009), as several species (Geosaurus araucanensis, Geosaurus vignaudi, Machimosaurus mosae, Metriorhynchus acutus and Metriorhynchus hastifer), which were excluded from Young’s and Muller-Töwe’s original datasets, were included here to ensure phylogeny could be controlled for in our analyses (see below). Figure 2 shows the phylogenetic relationships recovered for each family and the general skull morphology of each species. In addition, the feeding guild of each species is plotted on the cladograms to visualize the relationship between phylogeny, skull morphology, and preferred diet.

Analysing mechanics

Model construction and properties

Two-dimensional (2D) finite element models (FEM) of teleosaurid and metriorhynchid skulls representing the extreme shapes along PC1 and PC2 were created as a heuristic tool to assess the relationship between mechanical performance and skull shape variation. In addition, 2D models of all 23 species (9 teleosaurids and 14 metriorhynchids) were created to test the viability of using PC end-points as a proxy for interpreting the relationship between the pattern of species distribution in morphospace and skull strength, as well as to gain insight into the mechanical properties of real morphologies. Skull strength is here considered to be inversely proportional to stress (σ) when all other factors are held constant; thus skulls that experience very little stress under certain loading conditions are regarded as ‘stronger’ than those that experience higher stresses.

Both the ‘extreme morphology’ and species models were created using the methodology outlined in Pierce et al. (2008), and were loaded into the Geostar geometry creator component of the Cosmosm FEA package (v. 2.8 for Unix; SRAC Corp. CA, USA and CenitDesktop Ltd, UK). A generalized skull shape, without any sutural contacts, was created by linking landmarks around the external perimeter of the skulls and the boundaries of internal cavities. The inner region was ‘meshed’ to produce an interconnected grid of three-noded triangular FEs representing projections of the dorsal surface of the cranium into a 2D plane. The models were fixed along the posterior edge of the left and right quadrate during loading; as such, stress patterns at these points will be magnified and not accurate. Elastic isotropic properties were assumed to be in the range of modern crocodilian skull bones: E (Young’s modulus) = 6.65 GPa and v (Poisson’s ratio) = 0.35 (Reilly & Burstein, 1974; Currey, 1987). These values are identical to those used in Pierce et al. (2008) and permit direct comparisons between extant crocodilians and thalattosuchians.

Loading conditions and stress

To measure stress distribution and magnitude, a standard bite force of 5000 Newtons (N) was used, which is the expected bite force of a 2.5 m-long individual of Alligator mississippiensis (based on Erickson et al. 2003). This is considered a conservative estimate, as thalattosuchians range in body size from 1 to 7 m (Mueller-Töwe, 2006). Furthermore, the scale of force is of little consequence, as we are not trying to determine the absolute value of stress in the models, only the relative differences in stress between various skull shapes when size is held constant. Bite force was applied at the premaxilla–maxilla contact, instead of a specific tooth or set of teeth, as this is an easily identifiable and homologous point on each model. Muscular forces were excluded.

FEAs were carried out on three separate load cases (for each of the four theoretical models and all species) to quantify stress response in relation to behavioural loading conditions:

  • 1 Load Case 1 (LC 1) – a bilateral bite at the left and right premaxilla–maxilla suture with 2500 N applied in the z-direction on each side to bend the skull dorsally;
  • 2 Load Case 2 (LC 2) – a unilateral bite at the left premaxilla–maxilla suture with 5000 N applied in the z-direction to induce bending and superimposed torsional loading;
  • 3 Load Case 3 (LC 3) – lateral loading at the left premaxilla–maxilla suture with 5000 N applied in the y-direction to generate a within-plane lateral bend to the snout.

These three load cases are designed to reflect dorsoventral, torsional, and mediolateral loads, respectively. Again, these loading conditions are the same as those used by Pierce et al. (2008) and will permit direct comparisons between modern crocodilians and thalattosuchians.

Stress distribution and magnitude were recorded for each model and load case, with the Von Mises stress value reported here. Von Mises stress is a good predictor of failure for material such as bone (see Dumont et al. 2005). It is a scalar function of principal stresses σ1, σ2, and σ3, is directly proportional to the strain energy of distortion, and is also mathematically related to maximum shear stress. Hence it was chosen as an encompassing metric of skull strength in the FE-models, with lower stress corresponding to greater skull strength. To avoid ‘artificial noise’ created by fixing the quadrates and to quantify snout strength, stress values through the snout were also calculated at specific nodes on the surface of each model that corresponded to the mid-point of snout, mid-point of snout from bite, and anterior limit of the orbit; these measurements were taken along the mid-line of the snout for bilateral bites and along the lateral border of the snout (on the same side as applied load) for lateral loading.

Correlation statistics

Non-phylogenetic

The degree of concordance between skull shape, skull strength, and dietary preference in the Teleosauridae and Metriorhynchidae was assessed by first investigating the raw data, without taking the possible effects of phylogeny into account. This was done by conducting linear regressions of PC1 scores (i.e. relative skull length) and PC2 scores (i.e. relative skull width) on skull strength to ascertain whether functional patterns and shape were related. Skull strength was determined by calculating the mean Von Mises stress along the snout (based on nodal values from the mid-point of snout, mid-point of snout from bite, and anterior limit of orbit) during both a bilateral bite and lateral loading. Subsequent univariate anovas and post hoc Tukey HSD tests were carried out using the factors ‘feeding guild’ and ‘morphotype’ to investigate whether different diets and broad shape categories are significantly different with respect to skull shape and/or skull strength. anovas on feeding guild (Fig. 2) were conducted all on variables (i.e. PC1, PC2, bilateral stress and lateral stress), whereas anovas on the qualitative grouping morphotype (Fig. 1) were only conducted on bilateral and lateral stress to avoid circularity in the data. Due to small sample size and the potential for Type II error, significant differences were assessed at an α level of 0.05 and a relaxed α level of 0.1.

Phylogenetic

In addition to the tests run on the raw data, a subsequent set of tests was run using data that were adjusted to account for the possible effects of phylogeny to ensure that any apparent patterns were not artefacts caused by patterns of relationship among the species. As species are related to one another via descent from a common ancestor, and share inherited attributes, they cannot be considered to be statistically independent observations (e.g. Felsenstein, 1985; Harvey & Pagel, 1991). Therefore, when examining correlations between phenotypic traits across species, or between traits and functional or ecological factors, it is necessary to take phylogenetic relatedness into account. Many approaches have been developed to control for phylogeny in comparative analyses (e.g. Grafen, 1989; Lynch, 1991; Diniz-Filho et al. 1998); here we used linear regressions of phylogenetically independent contrasts (PICs) (Felsenstein, 1985) and phylogenetic anovas (Garland et al. 1993) to perform a series of tests analogous to those carried out on the raw data.

All analyses were carried out in the context of the tree topologies for teleosaurids and metriorhynchids shown in Fig. 2, with two sets of branch lengths. In the first set, all branches were considered to have a length of one. In the second set, branch lengths were set to be proportional to the number of character state changes implied by the tree topology/phylogenetic data matrix in question. These branch lengths were calculated using parsimony and ACCTRAN optimization in paup* 4.10b (Swofford, 2001) with characters being optimized directly on the trees. For the phylogenetic anovas, unresolved branches were given extremely small, non-zero branch lengths, which is functionally equivalent to making the polytomies ‘hard’. Polytomies were retained in the PIC analyses, but the degrees of freedom used to calculate significance were reduced following Garland & Diaz-Uriarte (1999). A significant relationship between the standardized independent contrasts and their standard deviations was not apparent for any of the branch length/topology combinations, indicating that the different topology/branch length combinations and a Brownian motion model of evolution fit the tip data adequately for the use of independent contrasts (e.g. Diaz-Uriarte & Garland, 1996; Diaz-Uriarte, 1998).

As with the raw data, contrasts for PC1 scores and PC2 scores were regressed against contrasts for mean Von Mises stress along the snout measured during a bilateral bite and lateral loading. Regression lines were constrained to pass through the origin in all cases, and all analyses involving PICs were carried out in the PDAP:PDTREE module (Midford et al. 2008) of mesquite (Maddison & Maddison, 2007). Phylogenetic anovas were carried out using the factors ‘feeding guild’ and ‘morphotype’ to investigate whether different diets and broad shape categories are significantly different with respect to skull shape and/or skull strength. Phylogenetic anovas on feeding guild (Fig. 2) were conducted on all variables (i.e. PC1, PC2, bilateral stress and lateral stress), whereas phylogenetic anovas on the qualitative grouping morphotype (Fig. 1) were only conducted on bilateral and lateral stress to avoid circularity in the data. Null distributions for the test statistic were calculated under a Brownian motion model of evolution using 25 000 simulated datasets in the Geiger package (Harmon et al. 2008) of r 2.8.1. Due to small sample size and the potential for Type II error, significant differences were assessed at an α level of 0.05 and a relaxed α level of 0.1.

Results

Teleosaurids

Stress patterns

Colour-coded stress distribution plots for the extreme PC end-point shapes (PC1 = −0.20 and 0.16; PC2 = −0.12 and 0.08) in morphospace illustrate the pattern of peak stress in teleosaurid skulls during bilateral biting (Fig. 3A), unilateral biting (Fig. 3B), and lateral loading of the snout (Fig. 3C). There is no qualitative difference between stress distribution and magnitude in the bilateral and unilateral models. As such, the unilateral bite does not induce a marked torsional response in the 2D models, although there is a slight increase in stress and asymmetry in the patterning that indicates the model is registering the unilateral load. The pattern of stress distribution is very similar between the four extreme PC end-point models (within each loading regime), although the magnitude of stress changes.

Figure 3.

 Stress magnitude and distribution in teleosaurid skulls representing the extreme shapes of PC1 (−0.20 and 0.16) and PC2 (−0.12 and 0.08). (A) LC 1: bilateral bite; (B) LC 2: unilateral bite; (C) LC 3: lateral load applied to the snout. Units are Pa. See text for details.

Stress patterns during a bilateral and unilateral bite (Fig. 3A, B) show that stress increases through the skull in an anteroposterior direction during biting, peaking around the supratemporal fenestrae. Examination of stress magnitude (along the midline) through the snout during a bilateral bite (Fig. 4A) shows PC1neg (extreme short snout) to be the strongest snout shape, with an average snout stress of 69 900 Pa, followed by PC2pos (extreme broad snout) at 112 000 Pa, PC2neg (extreme narrow snout) at 115 000 Pa, and finally PC1pos (extreme long snout) at 178 000 Pa. In terms of stress distribution through the snout (Fig. 4A), PC1neg and PC2pos show an increase in stress through the snout during biting, as opposed to PC2neg and PC1pos, which show an initial increase through the snout and then a decrease in stress towards the anterior limit of the orbits.

Figure 4.

 Stress magnitude and distribution profiles through the teleosaurid snout for the extreme shapes of PC1 (−0.20 and 0.16) and PC2 (−0.12 and 0.08), and the average shape in morphospace. (A) LC 1: bilateral bite. (B) LC 3: lateral load applied to the snout. Stars on teleosaurid skull images indicate where the measurements were taken (i.e. along the midline for LC 1 or along the lateral margin for LC 3).

Conversely, during lateral loading of the snout (Fig. 3C) the pattern of stress distribution increases anteroposteriorly along the lateral margin of the snout, peaking at the posterior constriction of the snout (i.e. the area where the snout starts to broaden) and along the posterior end of lateral bar of the supratemporal fenestrae. Examination of stress magnitude (along the lateral margin) through the snout during lateral loading (Fig. 4B) shows that PC1neg (extreme short snout) is the strongest snout shape with an average stress of 5260 Pa, followed by PC2pos (extreme broad snout) at 9980 Pa, PC2neg (extreme narrow snout) at 11 500 Pa and PC1pos (extreme long snout) at 23 000 Pa. In terms of stress distribution through the snout (Fig. 4B), all models show an initial increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbits, except PC2pos, which shows a steady decrease in stress.

Calculating stress through the snouts of all nine teleosaurid species during a bilateral bite and the application of a lateral load (Fig. 5) shows that Machimosaurus mosae has the strongest snout shape and Steneosaurus gracilirostris has the weakest. During a bilateral bite (Fig. 5A), all species increase in stress posteriorly through the snout, except for Pelagosaurus typus and Steneosaurus bollensis, which show an initial increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbit. During lateral loading (Fig. 5B), all species show an initial increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbit, except for M. mosae, which shows a steady decrease in stress through the snout. Although the majority of species maintain their ‘ranking’ on the stress spectrum for both load types, a few species increase or decrease in stress more noticeably. For example, Platysuchus multicrobiculatus experiences a more marked increase in stress posteriorly through the snout as compared to other species during both a bilateral bite and lateral loading. Alternatively, S. bollensis and Steneosaurus heberti experience a more marked decrease in stress posteriorly through the snout as compared to other species during a bilateral bite and lateral loading respectively.

Figure 5.

 Stress magnitude and distribution profiles through the snout for the mean shapes of all nine teleosaurid species delineated by Pierce et al. (2009). (A) LC 1: bilateral bite. (B) LC 3: lateral load applied to the snout. Measurements were taken in the same places as in Fig. 4.

Taking into consideration the FE results recovered for the extreme PC end-point shapes, we can predict that teleosaur species in the short/broad quadrant of morphospace would be the strongest, and species in the long/narrow quadrant of morphospace the weakest. However, ranking each species by mean stress through the snout (by averaging nodal values from the mid-point of snout, mid-point from bite and anterior limit of orbit) demonstrates that there is not a close correspondence between snout strength and the distribution of species within morphospace. For instance, P. multicrobiculatus and Steneosaurus leedsi have stronger snout shapes than expected based on the extreme PC end-point shape results and their positions within morphospace: P. multicrobiculatus experiences less stress in the anterior portion of the snout than S. heberti, a species that has a shorter snout morphology, and the narrow snouted S. leedsi experiences less stress through the snout than the broad snouted species P. typus, even though they both display a comparable snout length (Fig. 1). Results of the univariate anovas with the factor ‘morphotype’ support this discrepancy (Table 1) by indicating a non-significant relationship between morphotype and stress during a bilateral bite (P = 0.115) and lateral loading (P = 0.131), even with a relaxed significance level of 0.1. However, when the data are adjusted for phylogenetic relationships, teleosaurids do show a significant relationship between morphotype and stress ( 0.06).

Table 1.   Results of univariate anovas on the uncorrected raw scores and phylogenetic anovas of the mean Von Mises stress through the snout in teleosaurids, using teleosaurid morphotype as a categorical variable.
 R2PMean
Uncorrected raw dataPhylogenetic branch lengthEqual branch length Short/ broad Short/ narrow Long/ broad Long/ narrow
  1. Significant differences are indicated in bold (< 0.05) or with an asterisk (< 0.1). A relaxed level of significance was used to compensate for small sample size and low power of the test.

Bilateral stress (Pa)0.6660.1150.0490.0307.6E + 41.1E + 51.4E + 51.5E + 4
Lateral stress (Pa)0.6460.1310.0450.060*5.3E + 31.1E + 41.5E + 41.6E + 4

Diet, mechanics, and shape

Regressions of the raw data PC scores representing teleosaurid morphospace (Fig. 1A) against mean stress through the snout (Table 2) show a strong and significant linear relationship between relative snout length (PC1) and bilateral (= 0.001) and lateral (= 0.003) stress. However, relative snout width (PC2) does not show a significant relationship with stress ( 0.97). A similar result is obtained when the raw data are corrected for phylogeny using independent contrasts (Table 2): snout length (PC1) shows a strong and significant relationship with stress ( 0.018), whereas snout width (PC2) shows a non-significant relationship ( 0.325). Univariate anovas (Table 3a) on the uncorrected raw scores with the factor ‘feeding guild’ indicate a significant relationship between feeding guild and PC1 (= 0.004), as well as bilateral stress using a relaxed significance level of 0.1 (= 0.082). This relationship is maintained when phylogeny is taken into account, although lateral stress also becomes significant ( 0.045).

Table 2.   Results of linear regressions of the uncorrected raw scores and phylogenetically independent contrasts between mean Von Mises stress along the snout in teleosaurids and PC1 and PC2.
 PC1PC2
R2FPR2FP
  1. Significant correlations (< 0.05) are indicated in bold.

Uncorrected raw scores
 Bilateral stress (Pa)0.88654.206< 0.0010.1381.1220.325
 Lateral stress (Pa)0.82633.184< 0.0010.1261.0060.350
Phylogenetic branch lengths
 Bilateral stress (Pa)0.57110.6520.0080.0140.1170.372
 Lateral stress (Pa)0.4747.2100.0180.0150.1250.368
Equal branch lengths
 Bilateral stress (Pa)0.59511.7350.0075.0410.0040.476
 Lateral stress (Pa)0.4927.7620.0161.1629.3010.488
Table 3.    (a) Results of univariate anovas on the uncorrected raw scores and phylogenetic anovas of the first two PC axes and mean Von Mises stress through the snout in teleosaurids, using teleosaurid feeding guild as a categorical variable. (b) Results of Tukey HSD post hoc tests on the uncorrected raw scores for significant variables in each teleosaurid feeding guild comparison (Crunch/General, Crunch/Pierce I, Crunch/Pierce II, General/Pierce I, General/Pierce II, Pierce I/Pierce II).
(a)R2PMean
Uncorrected raw dataPhylogenetic branch lengthEqual branch lengthCrunchGeneralPierce IPierce II
PC10.9140.0040.074*0.058−0.165−0.0550.0790.047
PC20.0850.9220.4880.4080.0080.003−0.011−0.017
Bilateral stress (Pa)0.7110.082*0.0420.0337.7E + 41.1E + 51.5E + 51.3E + 5
Lateral stress (Pa)0.6280.1470.0450.0346.2E + 31.0E + 41.7E + 41.4E + 4
(b)Crunch/ GeneralCrunch/ Pierce ICrunch/ Pierce IIGeneral/ Pierce IGeneral/ Pierce IIPierce I/ Pierce II
 PDPDPDPDPDPD
  1. Significant differences are indicated in bold (< 0.05) or with an asterisk (< 0.10) and D is the direction of significance. A relaxed P-value was used to compensate for small sample size and low power of the test.

PC10.125ns0.004 C < P10.011 C < P20.049G < P10.160ns0.815ns 
Bilateral stress (Pa)0.629ns0.069*C < P1*0.249ns0.298ns0.792ns0.762ns

A post hoc test on the uncorrected raw scores (Table 3b) of PC1 reveals that the Crunch guild has a significantly shorter snout (and by association larger supratemporal fenestrae and a smaller frontal bone) than the Pierce I (= 0.004) and Pierce II guilds (= 0.011), and that the General guild has a significantly shorter snout than the Pierce I guild (= 0.049). However, all other guild pairwise comparisons show a non-significant difference in snout length ( 0.125). A post hoc test on the uncorrected raw scores (Table 3b) of bilateral stress reveals that only the Crunch and Pierce I guilds are significantly different with respect to stress levels (using a relaxed P-value of 0.1), with Crunch experiencing far less stress through the snout than Pierce I (= 0.069).

Metriorhynchids

Stress patterns

Colour-coded stress distribution plots for the extreme PC end-points (PC1 = −0.24 and 0.12; PC2 = −0.04 and 0.08) of morphospace illustrate the pattern of peak stress in metriorhynchid skulls during bilateral biting (Fig. 6A), unilateral biting (Fig. 6B), and lateral loading of the snout (Fig. 6C). There is no qualitative difference between stress distribution and magnitude in the bilateral and unilateral models. As such, the unilateral bite does not induce a marked torsional response in the 2D models, although there is a slight increase in stress and an asymmetry in the patterning that indicates the model is registering a unilateral load. The pattern of stress distribution is very similar between the four PC end-point models (within the confines of each loading condition), although the magnitude of stress changes.

Figure 6.

 Stress magnitude and distribution in metriorhynchid skulls representing the extreme shapes of PC1 (−0.24 and 0.12) and PC2 (−0.04 and 0.08). (A) LC 1: bilateral bite; (B) LC 2: unilateral bite; (C) LC 3: lateral load applied to the snout. Units are Pa. Dashed lines indicate the general area where the prefrontal bones are located. See text for details.

Stress patterns during a bilateral and unilateral bite (Fig. 6A,B) show that stress increases through the skull in an anteroposterior direction during biting, peaking around the supratemporal fenestrae and dissipating towards the tips of the enlarged, triangular, prefrontal bones. Examination of stress magnitude (along the midline) through the snout during a bilateral bite (Fig. 7A) shows that PC1neg (extreme short snout) is the strongest snout shape, with an average snout stress of 26 700 Pa, followed by PC2pos (extreme narrow snout) at 60 300 Pa, PC2neg (extreme broad snout) at 73 900 Pa, and finally PC1pos (extreme long snout) at 101 000 Pa. In terms of stress distribution through the snout, all four models show an increase in stress posteriorly through the snout (Fig. 7A).

Figure 7.

 Stress magnitude and distribution profiles through the metriorhynchid snout for the extreme shapes of PC1 (−0.24 and 0.12) and PC2 (−0.04 and 0.08), and the average shape in morphospace. (A) LC 1: bilateral bite. (B) LC 3: lateral load applied to the snout. Stars on metriorhynchid skull images indicate where the measurements were taken (i.e. along the midline for LC 1 or along the lateral margin for LC 3).

Conversely, during lateral loading of the snout (Fig. 6C) the pattern of stress distribution increases posteriorly along the lateral margin of the snout, peaking at the posterior constriction of the snout (i.e. the area where the snout starts to broaden) and along the lateral bar of the supratemporal fenestrae, and dissipating towards the tips of the enlarged, triangular, prefrontal bones. Examination of stress magnitude (along the lateral margin) through the snout during lateral loading (Fig. 7B) shows that PC1neg (extreme short snout) is the strongest snout shape with an average stress of 668 Pa, followed by PC2pos (extreme narrow snout) at 2240 Pa, PC2neg (extreme broad snout) at 4080 Pa and PC1pos (extreme long snout) at 8060 Pa. In terms of stress distribution through the snout, PC1neg and PC2neg show an initial increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbits, whereas PC1pos and PC2pos show a decrease in stress throughout the snout (Fig. 7B).

Calculating stress through the snouts of all 14 metriorhynchid species during a bilateral bite and the application of a lateral load (Fig. 8) shows that Dakosaurus andiniensis has the strongest snout shape and Metriorhynchus acutus has the weakest. During a bilateral bite (Fig. 8A), all species increase in stress posteriorly through the snout, except for Geosaurus araucanensis, which shows an initial increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbit. During lateral loading of the snout (Fig. 8B), all species show an initial slight increase in stress through the snout and then a decrease in stress towards the anterior limit of the orbit, except for Geosaurus gracilis, Metriorhynchus leedsi, and Metriorhynchus superciliosus, which show a stepwise decrease in stress posteriorly throughout the snout. Although the majority of species maintain their ‘ranking’ on the stress spectrum for both load types, a few species increase or decrease in stress more noticeably. For example, G. araucanensis, M. superciliosus, and Metriorhynchus casamiquelai experience a more marked decrease in stress posteriorly through the snout during a bilateral bite as compared to other species, whereas Metriorhynchus hastifer and Teleidosaurus calvadosi experience a more marked increase in stress. In addition, M. hastifer experiences a more marked decrease in stress posteriorly through the snout during lateral loading compared to other species, whereas M. casamiquelai experiences a more marked increase in stress.

Figure 8.

 Von Mises Stress profiles through the snout for the mean shapes of the 14 metriorhynchid species delineated by Pierce et al. (2009). (A) LC 1: bilateral bite. (B) LC 2: lateral load applied to the snout. Measurements were taken in the same places as in Fig. 7.

In contrast to the situation within teleosaurids, taking the mean stress value through the snout and ranking each metriorhynchid species demonstrates a close correspondence between snout strength and the distribution of species within morphospace. Results of the univariate anovas and phylogenetic anovas (Table 4a) indicate that there is a significant relationship between morphotype and stress ( 0.048), with post hoc tests (Table 4b) showing that short- and long-snouted morphotypes are significantly different from each other in terms of bilateral and lateral stress (i.e. narrow- and broad-snouted morphologies within the same length category are not significantly different). Overall, species falling within the short/narrow quadrant of morphospace tend to be the strongest during a bilateral bite and lateral loading to the snout; these are followed by species in the short/broad quadrant, the long/broad quadrant, and the long/narrow quadrant.

Table 4.    (a) Results of univariate anovas on the uncorrected raw scores and phylogenetic anovas of the mean Von Mises stress through the snout in teleosaurids, using metriorhynchid morphotype as a categorical variable. (b) Results of Tukey HSD post hoc tests on the uncorrected raw scores for significant variables in each morphotype comparisons (short-broad/long-broad, short-broad/long-broad, short-broad/long-narrow, short-narrow/long-broad, short-narrow/long-narrow, long-broad/long-narrow).
(a)R2PMean
Uncorrected raw dataPhylogenetic branch lengthEqual branch length Short/ broad Short/ narrow Long/ broad Long/ narrow
Bilateral stress (Pa)0.7870.0010.0020.0025.0E + 44.6E + 48.7E + 49.9E + 4
Lateral stress (Pa)0.7430.0030.0480.0131.8E + 31.5E + 36.2E + 37.5E + 3
(b)Short-broad/ short-narrowShort-broad/ long-broadShort-broad/ long-narrowShort-narrow/ long-broadShort-narrow/ long-narrowLong-broad/ long-narrow
PDPDPDPDPDPD
  1. Significant differences are indicated in bold (< 0.05) or with an asterisk (< 0.10) and D is the direction of significance. A relaxed P-value was used to compensate for small sample size and low power of the test.

Bilateral stress (Pa)0.985ns0.026S < L0.005S < L0.015S < L0.003S < L0.642ns
Lateral stress (Pa)0.997ns0.041S < L0.010S < L0.029S < L0.007S < L0.769ns

Diet, mechanics, and shape

Regressions of the raw data PC scores representing metriorhynchid morphospace (Fig. 1B) against mean stress through the snout (Table 5) show a strong and significant linear relationship between snout length (PC1) and bilateral (< 0.0001) and lateral stress (< 0.0001). However, snout width (PC2) and stress do not show a significant relationship ( 0.151). When the raw data are corrected for phylogeny using independent contrasts (Table 5) snout length (PC1) still shows a significant relationship, but the relationship between snout width (PC2) also becomes significant ( 0.065). Univariate anovas on the uncorrected raw scores (Table 6a) indicate a significant relationship between ‘feeding guild’ and all shape and stress variables ( 0.021). However, when the data are corrected for phylogenetic relationships, feeding guild is only significantly related to PC2 or relative snout width ( 0.065), with all other variables not significant, even at a relaxed significance level of 0.1 (Table 6a).

Table 5.   Results of linear regressions of uncorrected raw scores and phylogenetically independent contrasts between mean Von Mises stress along the snout in metriorhynchids and PC1 and PC2.
 PC1PC2
R2FPR2FP
  1. Significant differences are indicated in bold (< 0.05) or with an asterisk (< 0.10). A relaxed P-value was used to compensate for small sample size and low power of the test.

Uncorrected raw scores
 Bilateral stress (Pa)0.85470.2750.0010.0160.1870.671
 Lateral stress (Pa)0.74735.5030.0010.0030.0730.792
Phylogenetic branch lengths
 Bilateral stress (Pa)0.76746.0230.0010.1592.6410.065*
 Lateral stress (Pa)0.53215.8890.0010.2143.8100.037
Equal branch lengths
 Bilateral stress (Pa)0.54917.0090.0010.0290.4150.266
 Lateral stress (Pa)0.3437.3100.0100.0050.0690.399
Table 6.    (a) Results of univariate anovas on the uncorrected raw scores and phylogenetic anovas of the first two PC axes and mean Von Mises stress through the snout in metriorhynchids, using metriorhynchid feeding guild as a categorical variable. (b) Results of Tukey HSD post hoc tests on the uncorrected raw scores for significant variables in each metriorhynchid feeding guild comparison (Crunch/General, Crunch/Pierce I, Crunch/Pierce II, General/Pierce I, General/Pierce II, Pierce I/Pierce II).
(a)R2PMean
Uncorrected raw dataPhylogenetic branch lengthEqual branch lengthCrunchGeneralPierce IPierce II
PC10.8970.0010.9660.967−0.209−0.0750.0400.062
PC20.6080.0210.0450.065*0.042−0.0230.030−0.013
Bilateral stress (Pa)0.7610.0020.5760.5643.54E + 45.0E + 48.9E + 49.5E + 4
Lateral stress (Pa)0.6540.0110.3020.3189.3E + 21.8E + 36.0E + 37.7E + 3
(b)Cut/ SmashCut/ GeneralCut/ Pierce IISmash/ GeneralSmash/ Pierce IIGeneral/ Pierce II
PDPDPDPDPDPD
  1. Significant differences are indicated in bold (< 0.05) or with an asterisk (< 0.10) and D is the direction of significance. A relaxed P-value was used to compensate for small sample size and low power of the test.

PC10.001C < S<0.001C < G<0.001C < P0.005S < G0.011S < P0.885ns
PC20.059*C < S0.927ns0.166ns0.039S < G0.967ns0.117ns
Bilateral stress (Pa)0.731ns0.006C < G0.012C < P0.016S < G0.034S < P0.957ns
Lateral stress (Pa)0.965ns0.053*C < G0.039C < P0.065*S < G0.050S < P0.762ns

A post hoc test on the uncorrected raw scores (Table 6b) of PC1 reveals that all feeding guilds are significantly different with respect to relative snout length except the General/Pierce II guilds (= 0.885). The Cut guild has a significantly shorter snout (and by association larger supratemporal fenestrae and a smaller frontal bone) than the Smash (< 0.0001), General (< 0.0001), and Pierce II guilds (< 0.0001), while the Smash guild has a significantly shorter snout than the General (= 0.005) and Pierce II guilds (= 0.011). A post hoc test on the uncorrected raw scores (Table 6b) of PC2 reveals that the Cut guild has a significantly narrower snout (and by association shorter nasal bones, longer premaxillae, and more dorsally positioned external nares) than the Smash guild (= 0.059) and the Smash guild has a significantly broader snout than the General guild (= 0.039). Post hoc tests on the uncorrected raw scores (Table 6b) of bilateral and lateral stress reveal that all guilds experience significantly different amounts of stress during a bilateral bite and lateral loading of the snout (0.006 ≤  0.065), except Cut/Smash ( 0.731) and General/Pierce II (≥0.762). Both the Cut and Smash guilds experience less stress in the snout than the General and Pierce II guilds, with the Cut/General and Smash/General pairwise comparisons of lateral stress evaluated at a relaxed significance level of 0.1.

Discussion

Biomechanical consequences of shape

Snout constriction and displacement

The posterior constriction of the snout (i.e. the area where the snout starts to broaden to accommodate the orbits posteriorly) creates a point of weakness during lateral loading (Figs 3 and 6) in thalattosuchians. When considering skull shape in three dimensions, the mechanical significance of this feature can be explained using second moment of area (I), which is a measure of a structure’s resistance to bending (Bird & Ross, 2002). In theory, a relative decrease in rostral depth (Iy = distribution of bone about the dorsoventral axis) results in exponential increases in dorsoventral bending stress, whereas reducing rostral width (Ix = distribution of bone about the mediolateral axis) results in exponential increases in mediolateral bending stress (Busbey, 1995). In relation to thalattosuchians, this suggests that the narrower portion of the snout anterior to the constriction point experiences relatively greater mediolateral stress than the broader posterior portion of the snout during lateral loading. Accordingly, the anterior portion of the snout will fail during lateral loading before the broader posterior section, creating a point of weakness. The posterior constriction of the snout is most noticeable in teleosaurids, especially long-snouted forms. Interestingly, in the fossil record, teleosaurid snouts are often found isolated from the posterior region of the skull (for examples see Steel, 1973), confirming the presence of a natural point of weakness in the snout.

Supratemporal fenestrae and muscle development

The observed high stress levels in the supratemporal area were expected, as stress is a measure of force per unit area (Bird & Ross, 2002) and thalattosuchians have greatly enlarged supratemporal fenestrae. The reduction in skeletal mass at the back of the skull could potentially create a weak design and begs the question why these animals modified their skulls in this manner. One explanation is that enlargement of the supratemporal fenestrae would have facilitated expansion of the adductor musculature. A recent reconstruction of the adductor musculature in the teleosaur S. bollensis illustrates that this species had strongly developed M. adductor mandibulae externus profundus (MAMEP) and M. adductor mandibulae internus pseudotemporalis (MPS) muscles (Mueller-Töwe, 2006). A similar relationship between supratemporal fenestrae size and adductor muscle elaboration can be observed in extant longirostrine crocodilians (i.e. Gavialis gangeticus, Tomistoma schlegelii, and Crocodylus cataphractus), but to a far lesser extent (Endo et al. 2002). This suggests that S. bollensis potentially had a stronger bite than similarly sized modern longirostrine crocodilians because bite force is proportional to the physiological cross-sectional area of the muscles driving jaw closure (Herrel & Aerts, 2004).

Among extant crocodiles the size of the MAMEP and supratemporal fenestrae has been shown to be inversely correlated with the size of the M. pterygoideus (MPT) and pterygoid bone (Gadow, 1901; Iordansky, 1964). As noted above, the MAMEP is most strongly developed in long-snouted crocodiles and is situated within their expanded supratemporal fenestrae, but the MPT and pterygoid bone are relatively reduced. Conversely, in short-snouted crocodiles the MPT and pterygoid bone are extensively developed and the MAMEP and supratemporal fenestrae are comparatively small. The cause of this correlation is in the different roles of these two muscles: the MAMEP is an ‘adroit’ muscle because its line of action passes closer to the jaw joint, causing it to produce a more rapid but relative weak movement; the MPT (particularly the anterior belly) is a ‘strong’ muscle because its line of action is the most distant from the jaw joint, causing it to have a large mechanical advantage and greater force production (Iordansky, 1964). Combining this with knowledge of simple lever systems [which illustrates that as the snout (or outlever) becomes shorter, adduction speed is exchanged for greater force generation (Herrel & Aerts, 2004)] means that short-snouted modern crocodilians produce a stronger but slower bite than long-snouted forms.

In thalattosuchians, however, a different pattern is observed: the short-snouted forms exhibit larger supratemporal fenestrae and by association larger MAMEP muscles than long-snouted forms (Figs 1, 3, 6). On the other hand, all thalattosuchian species have relatively small pterygoid bones because the secondary palate is not fully formed [i.e. they are of ‘mesosuchian’ grade (see Langston, 1973)], thus limiting the area for muscle attachment and constraining MPT development. As short-snouted thalattosuchians were feeding on larger, tougher prey than their long-snouted counter parts (Fig. 2), bite force would have been crucial. Considering that the MPT (the ‘strong’ muscle in the system) is comparatively small in thalattosuchians, it is suggested that the MAMEP muscle became exaggerated in short-snouted forms to increase overall muscle mass and by association bite force. In combination with a short snout (or short outlever), this muscle arrangement would have created a strong, yet relatively fast bite. As a consequence, it appears that thalattosuchians adopted a trade-off between mechanical integrity of the skull and force production in order to feed on large, tough prey such as bony fish and marine reptiles. These results also demonstrate that different crocodile lineages have employed varied functional solutions to the problem of feeding on resistant food sources.

Function of prefrontal bone

Metriorhynchids are characterized by conspicuous prefrontal bones that are located just anterodorsal to the orbits, and which project laterally. The prefrontal increases in size roughly isometrically with the posterior portion of the skull, but allometrically relative to the length of the snout, so that short-snouted forms with large temporal regions also possess proportionally larger prefrontals (Fig. 1B; Pierce et al. 2009). The expansion of the prefrontal has been interpreted as protection for the orbit (Buffetaut, 1982), or as having a hydrodynamic function (Hua & Buffetaut, 1997). The metriorhynchid prefrontal, along with the lacrimal, also has been suggested to house a hypertrophied salt excretion gland to assist in osmoregulation (Fernández & Gasparini, 2000, 2008; Gandola et al. 2006). However, the mechanical significance of the prefrontal shape has never been addressed.

The FE results demonstrate that the region occupied by the prefrontal bone in metriorhynchids experiences reduced stress during the examined loading conditions (Fig. 6), indicating that the prefrontal bone might have served a function other than resisting feeding-induced stresses. For example, protection of a ‘salt gland’ and/or the eye would have been beneficial under unusual loading regimes that could have occurred during feeding or competitive interaction. Given that short-snouted forms presumably fed on larger, more dangerous prey, and used more aggressive behaviours, than long-snouted forms, the likelihood that they would encounter such unusual loading conditions is increased. Further 3D FEA could provide insight into whether the observed prefrontal shape served to reduce stress in this sensitive area under those conditions. Alternatively, the shape of the prefrontal bone may have merely served to increase the hydrodynamic efficiency of the skull and, thus, provided no mechanical advantage. An investigation into the hydrodynamic properties of the prefrontal bone may help to resolve this issue.

Does form follow function?

Extreme vs. species models

Mechanical first principles of cantilever beams dictate that structures of a specified length and shape will be stronger during bending if they are relatively wide as opposed to narrow. Likewise, structures of a specified width and shape will be stronger during bending if they are relatively short as opposed to long (Bird & Ross, 2002). Based on this concept, it is expected that thalattosuchians with short-snouted morphologies will experience less stress through their snouts than long-snouted morphologies, and that broad-snouted morphologies will experience less stress through their snouts than narrow-snouted morphologies under comparable loads. FE models of the extreme PC end-point shapes in morphospace (Figs 3 and 6) indicate that teleosaurids conform to this principle (Fig. 4), but that among metriorhynchids the extreme narrow-snouted shape (PC2pos) experiences less stress than the extreme broad-snouted shape (PC2neg) under all loading conditions (Fig. 7). This discrepancy can be explained by the extreme PC2pos end-point morphology in metriorhynchid morphospace: PC2pos has a thicker supratemporal bar and less noticeable posterior constriction of the rostrum than PC2neg, which results in lower peak stress at these highly susceptible areas.

Accordingly, based on the analysis of extreme PC end-points in metriorhynchid morphospace it was predicted that the long/narrow morphotype would be stronger than the long/broad morphotype; however, the long/narrow morphotype is actually the weakest quadrant of morphospace based on the analysis of real snout morphologies. This appears to be the result of differences between real thalattosuchian snout morphologies in the long/narrow quadrant of morphospace and the idealized shape of PC2pos. The snout shape of PC2pos is relatively broad anteriorly and maintains its thickness from the anterior tip of the snout to the prefrontal bones (Figs 1B and 6). A similar snout configuration is present in M. hastifer (Deslongchamps, 1863–1869), which experiences less stress through the snout than the other long/narrow species, as well as long/broad species (Fig. 8). However, the remaining species in the long/narrow quadrant of morphospace (Fig. 1B), especially the geosaurs, possess a more gracile, anteriorly narrow, snout morphology than PC2pos (e.g. Gasparini & Dellapé, 1976; Frey et al. 2002) and, as a result, experience higher stress than what is predicted from looking exclusively at the PC end-points. Therefore, although using extreme PC end-point morphologies to examine heuristically the relationship between morphospace occupation and mechanical performance can give a good general approximation of the relationship between form and function, it cannot provide as complete a picture as an examination of the actual species morphologies.

Morphotype

Investigating how useful ‘morphotypes’ are at describing the biomechanical behaviour of real thalattosuchian skull morphologies shows that teleosaurid skull strength can not be predicted by its morphotype category unless phylogenetic relationships are controlled for (Table 1). This result implies that not only do closely related species sit in divergent areas of morphospace (consistent with Pierce et al. 2009) they are also more ‘functionally’ divergent than expected given their evolutionary proximity. Therefore, accounting for phylogenetic relationships in this case allows us to recognize an association between morphotype and skull strength. Although this discrepancy potentially highlights a more complex relationship between snout shape and mechanical strength than can be captured by a simple four-part classification system, it may also represent an artefact of low power of the test due to small sample size.

On the other hand, metriorhynchid skull strength can be predicted by morphotype category before and after phylogeny is controlled for; however, only short- and long-snouted forms experience different magnitudes of stress (Table 4). As a consequence, it appears that shape changes associated with relative snout width do not have a great impact on overall skull strength and a two-part classification system (short vs. long) is sufficient to predict skull strength from morphology in this clade (however, see below). A similar short/long morphotype mechanical divide was recovered for extant crocodilians (Pierce et al. 2008), implying that selection for skull strength in these two clades (and perhaps teleosaurids if sample size was increased) was constrained to shape changes related to relative snout length.

Determinants of skull shape

Skull shape and mechanical performance

The regression analyses (Tables 2 and 5) demonstrate that skull function, in terms of biomechanical stress, has a strong and significant linear relationship with the distribution of species along PC1 (i.e. relative snout length) in both teleosaurid and metriorhynchid morphospace, but not PC2 (i.e. relative skull width). The relationship between skull length and stress is maintained even when the effects of phylogeny are taken into account. Therefore, there is good evidence for a real, causal relationship between mechanical function and skull length in thalattosuchians, independent of phylogeny. A surprising finding is the fact that skull width in metriorhynchids shows a significant relationship with biomechanical stress when the effects of phylogeny are taken into account, but not before. This suggests that closely related species often possess skull morphologies that are more divergent than would be expected given their phylogenetic relationship, and when variation related to phylogenetic proximity is removed it actually unmasks similarities that were previously hidden. Such a pattern might be expected if there was strong selection for ecological differentiation among sibling species, increasing the likelihood that closely related species possessed divergent morphologies. In any case, this significant result suggests that some causal relationship between mechanical function and snout width does exist within the clade.

A similar linear relationship between relative skull length and biomechanical stress (as well as hydrodynamic efficiency) is observed in modern crocodilians (Pierce et al. 2008) and has been suggested to be caused by ecological pressures with respect to feeding and foraging strategies. For instance, the long and narrow snouts of G. gangeticus, T. schlegelii, and C. cataphractus are mechanically weaker, but offer less resistance in the water (as reduced surface area minimizes drag) and increase speed of attack (through a long outlever), allowing longirostrine crocodiles to feed successfully on small, fast, agile prey. Conversely, a short and wide snout (e.g. A. mississippiensis) is slower and increases water displacement during jaw closure, but provides increased mechanical strength and bite force (through a short outlever) for seizing a range of prey sizes, permitting brevirostrine crocodiles to feed on fish, amphibians, and large terrestrial animals (Taylor, 1987; Busbey, 1995; Cleuren & Vree, 2000; Herrel & Aerts, 2004; McHenry et al. 2006). The similarity between relative skull length and mechanical performance in thalattosuchians and extant crocodiles is intriguing considering they are distantly related (Benton & Clark, 1988; Clark, 1994), have rather different skull morphologies, and have potentially very different habitat preferences and lifestyles (Hua & Buffrenil, 1996). Therefore, it is likely that skull length and mechanical performance in thalattosuchians are influenced by a similar combination of ecological and functional pressures, and that the link between skull length, mechanical performance, and ecology is a consistent feature in the evolutionary history of the Crocodylomorpha.

Impact of diet on shape and mechanics

The univariate and phylogenetic anovas demonstrate a clear relationship between inferred diet, skull shape, and mechanical performance in thalattosuchians, although teleosaurids and metriorhynchids seem to follow somewhat different pathways. In teleosaurids, diet or ‘feeding guild’ (Table 3a) is significantly correlated with relative snout length (PC1) and stress; this is not surprising considering the intimate relationship that exists between snout length and mechanical strength. Conversely, in metriorhynchids, diet is correlated with all shape and stress variables, but only if the uncorrected raw scores are assessed (Table 6a). Removing the effects of metriorhynchid phylogeny results in only relative snout width (PC2) maintaining a significant relationship with diet, an outcome consistent with the results of the independent contrast regressions. Therefore, selection on snout shape related to dietary preference worked in orthogonal directions in the two families: diet is associated with snout length in teleosaurids and snout width in metriorhynchids. Considering that length and width of the snout have dramatic effects on speed and force generation (Busbey, 1995; Herrel & Aerts, 2004), these results suggest that skull modifications among teleosaurids were directed towards increasing speed of attack to pierce soft, agile fish and cephalopods, and that changes in metriorhynchid skull morphology were focused on increasing power to smash hard belemnoids and cut through large bony fish and reptiles.

Niche partitioning and feeding behaviour

Preferred diet and character displacement

Post hoc tests of the factor ‘feeding guild’ in teleosaurids (Table 3b) reveal that the Crunch guild has a significantly shorter snout than the Pierce I and Pierce II guilds, with Crunch also experiencing less stress through the snout than Pierce I. In addition, the General guild has a significantly shorter snout than the Pierce I guild. Thus, dietary specializations in terms of hard vs. soft prey represent a potential cause of character displacement with respect to snout length in teleosaurids. Nevertheless, the Crunch/General, General/Pierce II, and Pierce I/Pierce II guilds show no morphological or mechanical differences. This may signify that changes in skull shape and/or strength between these guild pairs were not necessary for the exploitation of their preferred diet, but it also could be the result of small sample size (n = 9) and low statistical power causing Type II error. Spatiotemporal data show that the Crunch and General guilds lived during different time periods (Pierce et al. 2009) and, therefore, were not in direct competition. On the other hand, the General/Pierce I and Pierce I/Pierce II guilds overlap in time and space (Pierce et al. 2009) and would have been in direct competition. As such, the small differences in tooth form between these feeding guilds were presumably sufficient to promote niche partitioning, allowing morphologically and functionally similar species to co-exist.

Post hoc tests of the factor ‘feeding guild’ in metriorhynchids (Table 6b) reveal that all guilds, except the General/Pierce II guilds, are morphologically and/or functionally different from each other. All guilds are significantly different with respect to snout length; in terms of skull width, however, only the Cut, Smash, and General guilds show significant differences. In addition, all statistically significant guild comparisons, apart from the Cut/Smash guilds, show variation in stress through the snout during both a bilateral bite and lateral loading of the snout. As a result, in metriorhynchids, dietary specializations led to broad-scale changes in skull shape and mechanical performance – a pattern consistent with niche partitioning. Although the non-significant result between the General/Pierce II guilds may be due to small sample size (n = 14) and low power of the test, spatio-temporal data show that the General/Pierce II guilds only overlap in time and space during the Kimmeridgian (Pierce et al. 2009). Therefore, most species were not in direct competition and selection on morphological and functional displacement would have been minimal. As with teleosaurids, it is presumed that the small differences in tooth form between the General/Pierce II guilds during the Kimmeridgian may have been adequate for morphologically and functionally similar species to co-exist within a restricted geographic area and environment.

Feeding and foraging strategies

Comparison of stress patterns in teleosaurids with those in modern crocodilians (Fig. 9) shows that short-snouted teleosaurids (e.g. M. mosae) experience stress magnitudes through their snouts that are similar to extreme long-snouted modern species (e.g. C. cataphractus and G. gangeticus), but that ‘medium-’ and long-snouted teleosaurids (e.g. P. multicrobiculatus and S. gracilirostris, respectively) experience considerably more stress through their snouts than any extant crocodile. The Indian gharial (G. gangeticus) displays the proportionally longest snout of all modern forms (Pierce et al. 2008) and has the highest fishing success rate (Thorbjarnarson, 1990). Consequently, extreme snout elongation in teleosaurids (e.g. S. gracilirostris) can be hypothesized to be principally oriented towards increasing the efficiency of capturing small, mobile prey. During foraging, gharials submerge, rest on the substrate, and attack fish by rapid lateral and somewhat vertical strikes of the head and neck (Thorbjarnarson, 1990). This type of sit-and-wait ambush feeding behaviour can be envisioned for teleosaurids, as their long, thin snouts would have minimized drag during lateral attacks (Busbey, 1995; McHenry et al. 2006) and reduced water displacement during jaw closure. Likewise, their increased skeletal mass (i.e. low pneumaticity and development of thick dorsal and ventral osteoderms) would have caused high body inertia, allowing them to remain passively submerged for long periods of time (Hua & Buffrenil, 1996). A high body mass would have also permitted the body to remain stationary during rapid and precise movements of the head and neck.

Figure 9.

 Comparison of stress magnitude and distribution profiles through the snout for selected extant crocodilian, teleosaurid, and metriorhynchid species. (A) LC 1: bilateral bite. (B) LC 3: lateral load applied to the snout. Measurements were taken in the same place as in Figs 4 and 7.

Conversely, comparison of stress patterns in metriorhynchids with modern crocodilians (Fig. 9) shows that short-snouted metriorhynchids (e.g. D. andiniensis) experience similar stress magnitudes through their snouts as ‘medium’-snouted modern crocodilians (e.g. Crocodylus niloticus), and that ‘medium’-snouted metriorhynchids (e.g. T. calvadosi) experience stress magnitudes through their snouts that are comparable to long-snouted modern forms (e.g. C. cataphractus). Stress magnitudes in long-snouted metriorhynchids (e.g. M. acutus) are comparatively higher than for short- or medium-snouted metriorhynchids, and are much more similar to ‘medium’-snouted teleosaurids (e.g. P. multicrobiculatus) than to any extant taxa. Interestingly, during lateral loading of the snout, all metriorhynchids show a dramatic decrease in stress magnitude toward the anterior limit of the orbit (or prefrontal bone), with short- and ‘medium’-snouted forms falling below the stress values reported for modern species. At first glance, this would seem to imply that metriorhynchids were better able to resist lateral loading of the snout than extant crocodiles. However, one feature that is not reflected in this analysis is the increased level of bone porosity that has been reported as an aquatic adaptation for the metriorhynchid snout (Hua, 1994; Hua & Buffrenil, 1996). Such osteoporotic lightening in the cranium may have reduced the mechanical integrity of the snout by decreasing the mass of bone. Consequently, under equivalent load, a metriorhynchid snout may have experienced more stress than a modern crocodile snout of similar shape and proportion. However, bone porosity has only ever been detailed in the long-snouted species M. superciliosus and, as a result, it is unknown to what extent other species and snout-morphotypes display this histological feature.

Nevertheless, the broad range in stress patterns through the snout suggests that metriorhynchids exploited a variety of feeding and foraging strategies. The morphological and histological characteristics of metriorhynchids indicate that these crocodiles were most likely pelagic stalking predators that foraged just beneath the water surface (Hua, 1994; Hua & Buffrenil, 1996). The hunting behaviour of long-snouted metriorhynchids may have consisted of: (i) passively searching for small prey from just below the surface using small beats of the caudal fin; (ii) sudden acceleration through rapid flicks of the caudal fin and quick diving or lunging through ventral curvature of the dorsal region; (iii) attack of prey through a fast lateral strike of the head and neck; and (iv) passive ascent after capture due to skull buoyancy (Hua, 1994). Additional feeding behaviours might have been employed by extreme short-snouted metriorhynchids, such as D. andiniensis, as their increased skull strength and large ziphodont teeth indicate that they were slashing and cutting through large tough prey (Gasparini et al. 2006).

Modern crocodilians, such as the American alligator and Nile crocodile, often feed on extremely large animals that must be torn into smaller pieces for consumption. Dismemberment of prey can be accomplished by either: (i) a grasp and shake behaviour, which entails grasping large items in the mouth and shaking the head and body, thereby tearing off smaller pieces of flesh; or (ii) a rotational or spin behaviour (i.e. ‘death roll’), which entails grasping larger prey items in the mouth and spinning on the long axis, thereby tearing larger prey items into smaller pieces (Pooley & Gans, 1976; Busbey, 1995; McHenry et al. 2006). In a grasp and shake, the rostrum is subjected to lateral bending moments and dorsoventral bending moments, while during twisting the rostrum is exposed to significant dorsoventral bending and torsional forces (Busbey, 1995; McHenry et al. 2006). As short-snouted metriorhynchids possess a mechanical competency during the application of vertical and lateral loads that is similar to the American alligator and Nile crocodile (Fig. 9), use of the grasp and shake behaviour to dismember prey was probably feasible. Twist feeding, however, is much more difficult to establish using a 2D analysis as snout height has a dramatic influence on the ability to resist torsional forces (McHenry et al. 2006). Nonetheless, considering that snout height in metriorhynchids increases with decreasing snout length, with some species evolving exceedingly deep or orenirostral skulls (Gasparini et al. 2006), the application of a ‘death roll’ to tear through flesh and bone can not be ruled out.

Limitations of study

Although 2D FE models can be used as a first approximation for biomechanical investigations (Rayfield, 2004; Pierce et al. 2008), they are not entirely reflective of the morphology of the thalattosuchian skull, as species range in overall skull size and degree of dorsoventral flattening. Consequently, incorporating skull size and height into the analysis may alter the relationship between morphospace occupation and mechanical performance. For instance, it has been shown that skull size greatly impacts bite-force performance in modern crocodiles, with larger individuals having relatively stronger bites (Erickson et al. 2003). In addition, skull height influences overall skull strength by decreasing bending moments in both the dorsoventral and mediolateral direction, and alters muscle architecture and jaw biomechanics by changing the line of action of major jaw closing muscles and length of muscle fibres (Herrel et al. 2002, 2005; Verwaijen et al. 2002; McHenry et al. 2006). Considering thalattosuchian skulls range in size and height, a more in-depth 3D assessment of the connection between morphospace occupation and mechanical function may help to confirm the link between form and function. Nonetheless, the 2D analysis presented here provides insight into thalattosuchian skull mechanics and offers the necessary quantitative framework to extend and compare future studies on skull mechanics in thalattosuchians and other crocodylomorph lineages.

Conclusions

The pattern of stress distribution through the skull is very similar between teleosaurids and metriorhynchids, with stress peaking at the posterior constriction of the snout and around the enlarged supratemporal fenestrae. However, the magnitude of stress is different, with metriorhynchids usually having stronger snout morphologies than teleosaurids. In general, short-snouted morphologies experience the least amount of stress during loading, while long-snouted morphologies experience the greatest amount of stress. Nonetheless, some species performed better or worse than expected based on their position in morphospace, demonstrating that predictions based on idealized morphologies are not always accurate. The morphological diversification of thalattosuchian skull shape is complex, with teleosaurids and metriorhynchids seeming to follow somewhat independent pathways. Although both groups show a strong linear relationship between skull length and skull strength, selection on snout shape related to dietary preference is associated with snout length in teleosaurids and snout width in metriorhynchids. This result indicates that teleosaurid skulls were adapted for speed of attack, whereas metriorhynchid skulls were adapted for force generation. Niche partitioning within each family likely was attained through character displacement of skull morphology and/or function in terms of biomechanical integrity, allowing sympatric species to exploit a limited environment. In terms of feeding and foraging strategies, teleosaurids and long-snouted metriorhynchids were probably restricted to lateral attacks of the head and neck, as they have comparatively weak snout shapes. Conversely, short-snouted metriorhynchids have a relatively stronger snout shape and may have been able to employ the grasp and shake and/or ‘death roll’ feeding and foraging behaviours.

Author contributions

The core data collection, analysis and manuscript development was conducted by S. Pierce as part of her Ph.D. thesis through the Department of Earth Sciences at the University of Bristol. K. Angielczyk and E. Rayfield were Ph.D. co-supervisors and aided in concept development and critical revision of the manuscript. In addition, K. Angielczyk assisted with geometric morphometric procedures, phylogenetic comparative methods, and correlation statistics and E. Rayfield facilitated the construction and analysis of the finite element models.

Acknowledgements

This research was supported by the UK Overseas Scholarship, University of Bristol Postgraduate Scholarship, Natural Sciences and Engineering Research Council of Canada Doctoral Scholarship, Sir James Lougheed Award of Distinction, and the University of Bristol Alumni Postgraduate Scholarship, awarded to S. Pierce. K. Angielczyk’s contribution to the work was made possible by a Royal Society USA/Canada Research Fellowship. The authors thank C. Stayton and one anonymous reviewer for their constructive feedback during the development and refinement of this manuscript. For providing access to specimens in their care, we thank: C. McCarthy and S. Chapman (BMNH London), R. Smith (Manchester Museum), P. Jeffrey (Oxford University Museum), S. Jouve (Museum National d’Histoire Naturelle), M. Williams (Bath Royal Literary and Scientific Institute), M. Lowe (Sedgwick Museum), S. Ogilvy (York Museum), T. Sharpe (National Museum Cardiff), and R. Reynolds (NMNH Smithsonian Institution).

Ancillary