Dietary constraints of phytosaurian reptiles revealed by dental microwear textural analysis

Phytosaurs are a group of large, semi‐aquatic archosaurian reptiles from the Middle–Late Triassic. They have often been interpreted as carnivorous or piscivorous due to their large size, morphological similarity to extant crocodilians and preservation in fluvial, lacustrine and coastal deposits. However, these dietary hypotheses are difficult to test, meaning that phytosaur ecologies and their roles in Triassic food webs remain incompletely constrained. Here, we apply dental microwear textural analysis to the three‐dimensional sub‐micrometre scale tooth surface textures that form during food consumption to provide the first quantitative dietary constraints for five species of phytosaur. We furthermore explore the impacts of tooth position and cranial robustness on phytosaur microwear textures. We find subtle systematic texture differences between teeth from different positions along phytosaur tooth rows, which we interpret to be the result of different loading pressures experienced during food consumption, rather than functional partitioning of food processing along tooth rows. We find rougher microwear textures in morphologically robust taxa. This may be the result of seizing and processing larger prey items compared to those captured by gracile taxa, rather than dietary differences per se. We reveal relatively low dietary diversity between our study phytosaurs and that individual species show a lack of dietary specialization. Species are predominantly carnivorous and/or piscivorous, with two taxa exhibiting slight preferences for ‘harder’ invertebrates. Our results provide strong evidence for higher degrees of ecological convergence between phytosaurs and extant crocodilians than previously appreciated, furthering our understanding of the functioning and evolution of Triassic ecosystems.

However, despite this high consensus on diet, the evidence supporting many phytosaur dietary hypotheses is, in fact, relatively weak. For example, hypotheses that are based on simple analogies between tooth shape and function in extant crocodilians and phytosaurs (Chatterjee 1978;Hunt 1989;Hungerb€ uhler 1998Hungerb€ uhler , 2000 have rarely been subjected to explicit functional testing. Only a few crocodilians almost exclusively consume fish or tetrapods (e.g. the piscivorous gharial, Gavialis gangeticus; Hussain 1991). Most crocodilians are more generalistic and/or prefer other foods such as crustaceans (e.g. the American crocodile, Crocodylus acutus; Thorbjarnarson 1988). This dietary diversity is known from observational studies and stomach content analyses (Grigg & Kirshner 2015;Bestwick et al. 2019) which, in some cases, contrasts with hypotheses of crocodilian diets that are based solely on tooth shape. Only rarely has this dietary diversity in extant crocodilians been considered in reconstructions of phytosaur palaeobiology. Fossilized stomach contents and bite marks on dinosaur bones have provided evidence of carnivory or piscivory in some phytosaurs (Chatterjee 1978;Hungerb€ uhler 1998;Li et al. 2012;Drumheller et al. 2014) but these are limited to a handful of specimens. Furthermore, content fossils only provide 'snapshots' of entire diets and may be biased towards preservation of indigestible items (Bestwick et al. 2018). Robust reconstructions of phytosaur diets are therefore vital not only for understanding their ecological roles within Triassic food webs, but for providing broader insight into Triassic ecosystem functioning and ecological convergence with crocodilians (Jacobs & Murry 1980;Roopnarine et al. 2007;Drumheller et al. 2014).
A more robust approach to hypothesis testing involves dental microwear texture analysis (DMTA); quantitative analysis of the sub-micrometre scale three-dimensional textures that form on tooth surfaces during food consumption (Daegling et al. 2013;Bestwick et al. 2019). The relative difficulty experienced by consumers in piercing and/or processing food items determines the microwear patterns that form on teeth, which consequently provides direct evidence of diet (Daegling et al. 2013;Bestwick et al. 2019). The technique uses standardized texture parameters (Ungar et al. 2003; International Organization for Standardization 2012) to quantify microwear, and thus dietary, differences between species and/or populations, and therefore does not assume direct relationships between the morphology and inferred functions of teeth (Daegling et al. 2013;Purnell & Darras 2016). Non-occlusal (non-chewing) tooth texture differences between extant reptiles, including archosaurs, have been shown to exhibit dietary signals that reflect their diet over the last few weeks to months of life, even with low sample sizes (Bestwick et al. 2019;Winkler et al. 2019). The known relationships between microwear texture and diet in extant reptiles therefore provides us with a robust multivariate framework with which to quantitatively test and constrain phytosaur dietary hypotheses using DMTA (Bestwick et al. 2020a).
However, several endogenous non-dietary variables of phytosaur anatomy and functional morphology, and their potential roles in microwear texture formation, need to be considered. One variable is the position of sampled teeth within tooth rows. Across all major vertebrate groups, teeth that are positioned closer to the jaw adductor musculature are subjected to higher bite forces during jaw closure (Kammerer et al. 2006;Santana & Dumont 2009;Santana et al. 2010;Erickson et al. 2012;Porro et al. 2013). Although mechanisms of microwear formation are not fully understood (Calandra et al. 2012;Schulz et al. 2013a, b) changes in bite force along tooth rows during feeding may be expected to have some influence on intraspecific microwear textures. Many phytosaur teeth are preserved isolated from the jaws (Meyer 1861;Parrish 1989;Hungerb€ uhler 1998;Datta et al. 2019), thus testing for microwear differences between tooth positions is important for understanding whether standardized sampling positions are needed in these extinct reptiles. Furthermore, unlike most modern reptiles, many phytosaurs exhibit heterodonty along their tooth rows, with distinctive morphotypes described from teeth positioned at the anterior tip of the rostrum, the premaxilla and the maxilla respectively (Hungerb€ uhler 2000;Datta et al. 2019). Heterodonty supposedly enabled phytosaurs to feed on a greater range of food items, with some teeth better adapted for piercing and handling different items (Hungerb€ uhler 2000). This idea, however, has yet to be explicitly tested and thus strengthens the need for investigation with DMTA.
Another variable to be considered is the morphological robustness of phytosaur skulls. In broad terms, morphologically 'gracile' phytosaurs are characterized by slender and tubular rostra and 'robust' phytosaurs are characterized by much deeper rostra, often with distinct crests along the dorsal margin (Hunt 1989;Hungerb€ uhler 2002;Witzmann et al. 2014). There are some suggestions that the robust and gracile morphotypes from contemporaneous phytosaurs from the same geological sites represent sexual dimorphs of the same species (males and females, respectively; Zeigler et al. 2003;Hunt et al. 2006;Kimmig & Spielmann 2011). These morphotypes have been interpreted as evidence of dietary differences, with deeper rostrums facilitating predation on larger animals, usually tetrapods, and slender rostrums better adapted for capturing smaller animals, usually fish (Hunt 1989;Parrish 1989;Hungerb€ uhler 2000;Heckert & Camp 2013). Hypotheses of dietary partitioning using cranial robustness and/or sexual dimorphism, however, have been subjected to little explicit testing (Wall et al. 1995;Irmis 2005). Determining whether phytosaur microwear texture differences can be explained by skull morphotype in addition to, or instead of, diet will provide more thorough understanding of how phytosaurs interacted with food items.
Here, we present the first application of DMTA to five species of phytosaur to test the hypotheses that microwear texture differs between phytosaur species and that microwear texture differences reflect dietary differences. We also use the results of microwear analysis to test hypotheses of niche partitioning between robust and gracile phytosaurs, and that morphological differences between teeth in different locations in the jaw reflect functional differentiation. Hypotheses were tested using a multivariate framework comprising microwear texture data from extant crocodilians and varanid lizards with known diets (Bestwick et al. 2019). While no direct morphological or ecological comparisons have been made between phytosaurs and varanids, previous DMTA of reptiles has shown that microwear texture differences are more strongly linked to dietary differences than they are to tooth morphology or phylogeny (Bestwick et al. 2019(Bestwick et al. , 2020aWinkler et al. 2019). The varanid species included in the analysis exhibit diets that are both similar and different to the crocodilians and thus contribute a more robust multivariate framework.

Specimen material
Tooth microwear textures were sampled from five phytosaur species: Machaeroprosopus pristinus, Mystriosuchus planirostris, Nicrosaurus kapffi, N. meyeri and 'Smilosuchus lithodendrorum'; see Bestwick et al. (2020b, table S1) for the complete specimen list. Phytosaur species were chosen to represent a range of taxonomic groups, spanning robust and gracile skull morphologies and included specimens that retained teeth from varied locations along the tooth row. Phytosaurs were designated as either robust or gracile based on previous descriptions of these morphotypes, e.g. Hunt (1989

Dietary guild assignments of extant reptiles
Extant reptiles were selected to include taxa with wellconstrained dietary differences determined from stomach content studies (Taylor 1979;Auffenberg 1981Auffenberg , 1988Greene 1986;Losos & Greene 1988;Thorbjarnarson 1988;Sah & Stuebing 1996;Delany et al. 1999;Wallace & Leslie 2008;Laverty & Dobson 2013;Dalhuijsen et al. 2014;Rahman et al. 2017). Studies were chosen that provided dietary compositions as volumetric data (or frequency data at an absolute minimum; see Bestwick et al. 2020b for the full percentage breakdown for each species). Taxa that had not been subjected to ecological studies that provided volumetric or frequency breakdowns of diet were not sampled for study. Studies were also chosen that provided, where possible, representative sample sizes and close spatial proximity of the dietary study to the known location(s) from which the sampled specimens were collected. Extant reptiles were assigned to dietary guilds which account for the relative 'intractability' (roughly equivalent to hardness; Evans & Sanson 2005) of prey as food, based on Bestwick et al. (2019). Guilds include: carnivores (tetrapod consumers); piscivores (fish consumers); 'harder' invertebrate consumers (invertebrates with hard exoskeletons, e.g. beetles, crustaceans and shelled gastropods); 'softer' invertebrate consumers (invertebrates with less hard exoskeletons, e.g. crickets, grasshoppers, dragonflies, damselflies and ants); 'softest' invertebrate consumers (invertebrates with soft exoskeletons, e.g. invertebrate larvae, butterflies, moths, arachnids and millipedes); omnivores (combination of plant and animal matter).
Where possible, specimens of the same ontogenetic stage, usually adults, were sampled to minimize unconstrained variability introduced from potential ontogenetic dietary differences and more rapid tooth shedding rates in younger individuals (Berkovitz 2000). The availability of museum specimens allowed Cr. porosus to be split into adults and juveniles to reflect the known ontogenetic dietary shifts in this species (Taylor 1979;Sah & Stuebing 1996). Specimens with lower jaw lengths of < 50 cm were classified as juveniles; specimens with jaw lengths exceeding 50 cm were considered to be adults (Bestwick et al.
Caiman crocodilus exhibits seasonal dietary shifts, consuming more fish in the wet season (March-May) and more invertebrates in the dry season (August-September) (Laverty & Dobson 2013). To minimize potential seasonbased microwear variation, sampled Ca. crocodilus were specimens that had died early in the dry season and were classified as piscivores under the assumption that their tooth surface textures retained the piscivore signal accumulated during the wet season.

Sampling strategy
Phytosaur specimens were cleaned before sampling using an ethaline solvent gel, produced and applied according to Williams & Doyle (2010). Extant reptile teeth from dry skeletal specimens were cleaned using 70% ethanol-soaked cotton swabs to remove dirt and consolidant. Microwear data were acquired from non-occlusal (non-chewing) labial surfaces, as close to the tooth apex as possible. In extant reptiles, the mesial-most dentary tooth was sampled; in phytosaurs the premaxillary and maxillary teeth were sampled. Differences in sampling locations between extant and extinct reptiles are not an issue due to our experimental design. Phytosaur teeth are initially studied to test for microwear texture differences between teeth from difference positions of the tooth row and thus potential intra-jaw dietary partitioning. A lack of texture differences would indicate that tooth position is not a confounding variable in phytosaur microwear characteristics and that it is acceptable to project phytosaur teeth into multivariate texture-dietary spaces that comprise extant teeth from a single sampled position as part of phytosaur dietary reconstructions (see DMTA Statistical Analyses, below). To test for microwear differences along the tooth row, phytosaur jaws were divided into three regions: anterior teeth correspond to the mesial-most three/four teeth in the premaxilla and are morphologically characterized by greater robustness and apico-basal length relative to all other teeth in the tooth row (Bestwick et al. 2020b, fig. S1A); posterior teeth correspond to the seven to ten distal-most teeth in the maxilla (depending on species) that are morphologically characterized by an approximate phylloform ('leaf-shaped') morphology with slight to moderate apical recurvature, prominent serrated carinae and a labially convex D-shaped cross-section (Bestwick et al. 2020b, fig. S1C); middle teeth correspond to teeth located in between the anterior and posterior regions and are morphologically characterized by their conical shape with a circular cross-section and may be carinated and/or possess serrations (Bestwick et al. 2020b, fig. S1B). It was not possible to sample all three tooth positions from every taxon due to the sporadic preservation of in situ teeth in phytosaur fossils. In some taxa ('S. lithodendrorum', S. gregorii, Ma mccauleyi) an additional set of large teeth occur at the transition between the premaxilla and maxilla. These teeth are slenderer than those of the terminal rosette, and none were included in this study due to preservation quality.
High fidelity moulds were taken of all teeth using President Jet Regular Body polyvinylsiloxane (Colt ene/Whaledent Ltd., Burgess Hill, West Sussex UK). Initial moulds taken from each specimen were discarded to remove any remaining dirt and all analyses performed on respective second moulds. Casts were made from these moulds using EpoTek 320 LV Black epoxy resin mixed to manufacturer's instructions. Resin was cured for 24 h under 200 kPa (2 bar/30 psi) of pressure (Protima Pressure Tank 10 l) to improve casting quality. Small casts were mounted onto 12.7 mm SEM stubs using President Jet polyvinylsiloxane with the labial, non-occluding surfaces orientated dorsally to optimize data acquisition. All casts were sputter coated with gold for three minutes (SC650, Bio-Rad, Hercules, CA, USA) to optimize capture of surface texture data. Replicas produced using these methods are statistically indistinguishable from original tooth surfaces (Goodall et al. 2015).

Surface texture data acquisition
Surface texture data acquisition followed standard laboratory protocols (Goodall et al. 2015;Purnell & Darras 2016;Bestwick et al. 2019). Data were captured using an Alicona Infinite Focus microscope G4b (IFM; Alicona GmbH, Graz, Austria; software v. 2.1.2), using a 1009 objective lens, producing a field of view of 146 9 100 µm. Lateral and vertical resolution were set at 440 nm and 20 nm respectively. Casts were orientated so that labial surfaces were perpendicular to the axis of the objective lens.
All 3D data files were processed using Alicona IFM software (v. 2.1.2) to remove dirt particles from tooth surfaces and anomalous data points (spikes) by manual deletion. Data were levelled (subtraction of least squares plane) to remove variation caused by differences in tooth surface orientation at the time of data capture. Files were exported as .sur files and imported into Surfstand (v. 5.0.0; Centre for Precision Technologies, University of Huddersfield, UK). Scale-limited surfaces were generated by application of a fifth-order robust polynomial to remove gross tooth form and a robust Gaussian filter (wavelength k c = 0.025 mm; Schulz et al. 2013a; Purnell & Darras 2016). ISO 25178-2 areal texture parameters (International Organization for Standardization 2012) were then generated from each scale-limited surface. ISO parameter details can be found in Table 1.

DMTA statistical analyses
Log-transformed texture data were used for analyses as some of the texture parameters were non-normally distributed (Shapiro-Wilk, p > 0.05). The parameter Ssk (skewness of the height distribution of the 3D surface texture; Table 1) was excluded from analysis as it contains negative values and thus cannot be log-transformed.
To test the hypotheses that microwear textures differ between teeth from different positions along phytosaur tooth rows, and between morphologically gracile and robust phytosaurs, analysis of variance (ANOVA) with pairwise testing (Tukey HSD) were used for each texture parameter between tooth row positions and robustness categories, irrespective of species. Where homogeneity of variance tests revealed evidence of unequal variances, Welch analysis of variance was used. Texture parameters that exhibited significant differences along tooth rows were subjected to principal components analysis (PCA) and canonical variates analysis (CVA; a form of linear discriminant analysis). The separation of tooth row positions along PC and CV axes was investigated using ANO-VAs and Tukey HSD tests. Significant texture parameters which exhibit no pairwise differences under Tukey HSD comparisons were excluded from PCA and CVA. To test the hypothesis that phytosaur microwear texture differences between species reflect dietary differences, PCA and CVA were first used to explore texture parameters exhibiting significant differences between reptile dietary guilds. ANOVA with pairwise testing was used on the values of each PC and CV axis to determine whether guilds occupy different areas of multivariate space along these axes. Spearman's rank was used to independently test for correlations between each PC and CV axis and dietary characteristics (e.g. proportion of diet that comprises vertebrates). Phytosaur microwear data were then independently projected onto the PCA and CVA axes of extant reptile dietary guild microwear differences. Phytosaurs were projected onto independent datasets that comprised both crocodilians and varanids (Bestwick et al. 2019) and a dataset that comprised only crocodilians to provide more robust constraints on phytosaur diets.
A Benjamini-Hochberg (B-H) procedure was used to account for the possibility of inflated Type I error rates associated with multiple comparisons (Benjamini & Hochberg 1995). The false discovery rate was set at 0.05. The B-H procedure was not needed for the Tukey HSD tests as it already accounts for inflated Type I error rates

Phytosaur tooth position
Under ANOVA, none of the ISO parameters differ between tooth positions (Bestwick et al. 2020b, table S3). The CVA conducted on all texture parameters found significant differences between tooth position along CV axes 1 and 2 (CV 1: F = 82.96, df = 2, 14, p < 0.0001; CV 2: F = 10.58, df = 2, 14, p = 0.0016). The predictive model misclassified only 5.88% of specimens (Fig. 1). Tukey post hoc tests revealed significant differences along CV 1 between middle and posterior (p < 0.0001) and anterior and middle (p < 0.0001), and along CV 2 between anterior and posterior (p = 0.0013) and anterior and middle (p = 0.0159). The two parameters that exhibit the largest differences between tooth positions along both CVs 1 and 2 are Sdq (root mean square gradient of surface) and Sdr (developed interfacial area ratio; Fig. 1B; Table 1). In a general sense, the anterior and posterior tooth textures exhibit the steepest gradients and the middle tooth textures exhibit the greatest complexity ( Fig. 1; Table 1). Along CV 2, Sq (root mean square height of surface) exhibits large differences between tooth position; the posterior tooth textures exhibit the highest surface height ( Fig. 1; Table 1).
CVA of all 21 texture parameters produces significant separation between gracile and robust phytosaurs along CV 1 (100% of total variation; t = À66.656, p < 0.0001). Zero cases were misclassified from this model (Fig. 2C). Texture differences for the 16 parameters previously reported as non-significant (ANOVA) are very small in some instances. On average, robust phytosaurs have higher surface textures (higher Sq, higher Sa; average surface height), have a deeper core (higher Sk, core roughness depth), have larger core and core void volumes (higher Vmc, material volume of the surface core; higher Vvc, void volume of the surface core), and larger valleys and valley void volumes (higher Sv, maximum valley depth; higher Vvv, void volume of the surface valleys; higher Svk, mean valley depth below the core material) (Table 1). Robust phytosaur textures are also more uniform (lower Str, texture aspect ratio) with fewer, smaller peaks (lower Vmp, material volume of surface peaks; lower Smr1, proportion of surface that consists of peaks; lower Smr2, surface bearing area ratio) that comprise less of the surface texture (lower Sds, summit density) (Table 1). In broad terms, robust phytosaurs have rougher tooth surface textures when all ISO parameters are considered.

Phytosaur dietary reconstructions
As previously documented, four texture parameters significantly differ between dietary guilds in the crocodile and varanid dataset (Spk, mean height of peaks above core material; Sds; Vmp; and Smr1; see Bestwick et al. 2019, table 2 for full ANOVA results). See Figure 3A-E and F-J for example digital elevation models of extant reptile and phytosaur tooth surfaces, respectively, from which texture data were acquired. PCA of these four parameters separates extant reptiles into a texture-dietary space defined by PC axes 1 and 2 ( Fig. 4; Bestwick et al. 2020b,  fig. S2). PC 1 negatively correlates with proportions of total vertebrates in reptile diets and positively correlates with total invertebrates and PC 2 positively correlates with dietary proportions of 'softer' invertebrates (see Bestwick et al. 2019, table S3 for full dietary correlation results). ANOVA of the PCA results find that PCs 1 and 2 differ between guilds (PC 1, F = 4.9316, df = 4, 90, p = 0.0012; PC 2, F = 4.6676, df = 4, 90, p = 0.0018); piscivores differ from 'harder' invertebrate consumers and omnivores (PC 1, Tukey HSD); 'harder' invertebrate consumers differ from carnivores and 'softer' invertebrate consumers; 'softer' invertebrate consumers differ from piscivores (PC 2, Tukey HSD).
Projecting phytosaur data into this texture-dietary space plots them within the bounds of the extant reptiles and, as a whole, they tend to all cluster relatively close to zero along PCs 1 and 2 where most of the extant dietary guilds overlap ( Fig. 4; Bestwick et al. 2020b, fig. S2). Machaeroprosopus pristinus specimens are broadly distributed along PC 1, the vertebrate-invertebrate dietary axis, and have similar PC 2 values of around zero. Mystriosuchus planirostris specimens generally exhibit positive PC 2 values and exhibit a greater degree of overlap with 'softer' invertebrate consumers along this axis than do other phytosaurs. Nicrosaurus kapffi specimens generally exhibit more negative PC 1 values than other phytosaurs and overlap more strongly with piscivores along this axis. Nicrosaurus meyeri specimens exhibit values close to zero along PCs 1 and 2 and the centroid of these points sits centrally within the space occupied by carnivores and piscivores. 'Smilosuchus lithodendrorum' specimens exhibit the largest degree of clustering and plot very close to zero along PCs 1 and 2.
CVA of the four texture parameters separates extant reptiles in a texture-dietary space defined by CV axes 1 and 2 (together accounting for 97.2% of total variance; Fig fig. S3, table S5). ANOVA of the CVA results provides evidence of additional discriminatory power: CVs 1 and 2 differ between dietary guild (CV 1: F = 7.277, df = 4, 90, p < 0.0001; CV 2: F = 3.918, df = 4, 90, p = 0.0056); 'harder' invertebrate consumers differ from 'softer' invertebrate consumers, carnivores and piscivores; omnivores differ from carnivores and 'softer' invertebrate consumers F I G . 1 . A, CVA multivariate space of phytosaur dental microwear, constructed with 21 ISO textural parameters, grouped by tooth position. B, CVA biplot indicating the magnitude of each textural parameter on the CV axes defining maximum separation of tooth position, and the direction of maximum variation for each parameter along CVs 1 and 2. For parameter definitions, see Table 1. (CV 1, Tukey HSD); piscivores differ from carnivores and omnivores (CV 2, Tukey HSD).
Projecting phytosaur data into this texture-dietary space also plots them within the bounds of the extant reptiles. The CVA predictive model is highly unreliable, misclassifying 65.6% of cases in the extant reptile dataset. In this model, 36.8% of phytosaurs are classified as carnivores, 26.3% of phytosaurs as 'softer invertebrate consumers, 21% as piscivores, 10.5% as 'harder' invertebrate consumers and 5.3% as omnivores. Phytosaurs are more closely clustered together to other specimens of the same taxon, and together as a clade within the texture-dietary space ( Fig. 5; Bestwick et al. 2020b, fig. S3). Machaeroprosopus pristinus specimens, for example, are less broadly distributed along CV 1, the tetrapod dietary axis, and overlap with most dietary guilds along this axis. In contrast, Ma. pristinus specimens are more broadly distributed along CV 2, the vertebrate and fish-invertebrate and 'softer' invertebrate dietary axis. In this texture-dietary space My. planirostris specimens are relatively widely distributed along CV 1, occupying both positive and negative values along this axis, but also exhibit similar CV 2 values of around zero. Nicrosaurus kapffi specimens cluster close together around zero on CV 1 but are relatively distributed along CV 2. Nicrosaurus meyeri specimens are relatively widely distributed along CV 1 as two specimens overlap more strongly with carnivores while another overlaps with 'harder' invertebrate consumers. 'Smilosuchus lithodendrorum' specimens all occupy positive CV 1 values, overlapping more strongly with carnivores and 'softer' invertebrate consumers, and occupy PC 2 values of around zero.
Phytosaurs once again plot within the bounds of the texture-dietary space, although in this space they are perhaps more broadly distributed across the space and taxa do not cluster as strongly together ( Fig. 6; Bestwick et  CVA of all 21 ISO parameters for the extant crocodilian dataset separates guilds in a texture-dietary space defined by CV axes 1 and 2, which together form 100% of total variance ( Fig. 7; Bestwick et al. 2020b, fig. S5). CV 1 positively correlates with proportions of total vertebrates in crocodilian diets (r s = 0.4735, p = 0.0013) and negatively correlates with total invertebrates (r s = À0.4228, p = 0.0047). CV 2 positively correlates with total invertebrates in crocodilian diets (r s = 0.4501, p = 0.0025) and with dietary proportions of 'harder' invertebrates (r s = 0.5205, p = 0.0003), and positively correlates with proportions of total vertebrates in crocodilian diets (r s = À0.3792, p = 0.0121) and with dietary proportions of fish (r s = À0.5028, p = 0.0006; Bestwick et al. 2020b, table S7). ANOVA of the axes produces significant separation of dietary guilds (CV 1: F = 41.5073, df = 2, 40, p < 0.0001; CV 2: F = 5.5098, df = 2, 40, p = 0.0077). Tukey HSD tests reveal that all guilds are separate from each other along CV 1 (p < 0.0001), but no pairwise differences are exhibited along CV 2.
Projecting phytosaur data into this texture-dietary space plots all but one specimen (N. kapffi, TTU-P 13078, specimen no. 11) within the bounds of the extant crocodilians ( Fig. 7; Bestwick et al. 2020b,  fig. S5). Furthermore, all but two specimens (N. kapffi, SMNS 13078, specimen no. 11 and 'S. lithodendrum', TTU-P 15661, specimen no. 6) overlap with piscivores along CV 1, and specimens of the same taxa do not cluster as closely together as in other texture-dietary spaces. This model misclassified 27.9% of specimens; 57.9% of phytosaurs were classified as piscivores, 26.3% carnivores and 15.8% as 'harder' invertebrate consumers. F I G . 4 . Principal component texture-dietary space of four ISO texture parameters (Spk, Sds, Vmp and Smr1) for reptiles and phytosaurs. Texture-dietary space based on extant reptile data with phytosaurs projected onto the first two axes as unknown datum points. Specimens with associated letters represent surfaces A-J in Figure 3. Arrows show significant correlations of dietary characteristics along PC axes 1 and 2. Phytosaur specimen information corresponding to PCA plot number can found in Bestwick et al.

Microwear differences between tooth positions
Overall, our analyses show very subtle texture differences between teeth from different positions of the tooth row. In broad terms, the middle teeth have the most complex textures, the posterior teeth have the highest surface textures, and the posterior teeth and anterior teeth to a lesser extent, have the steepest slopes. Of the three ISO parameters that varied the most with tooth position (Sdq, Sdr and Sq), none differed between the extant reptile dietary guilds of Bestwick et al. (2019). These three parameters, however, did differ between similar dietary guilds of Winkler et al. (2019), although differences for Sdq and Sdr were small. This largely indicates that the subtle texture differences between tooth positions are not due to intra-jaw dietary partitioning.
The lack of texture differences largely contrasts with previous suggestions that the morphologically different teeth along the tooth rows of some phytosaurs, such as N. kapffi, were used to acquire and rudimentarily process different foods (Hungerb€ uhler 2000). The fang-like, serrated anteriormost teeth were suggested to have seized and held soft and fleshy prey, the D-shaped, bicarinate posterior-most teeth were suggested to have seized large, harder prey, and the triangular middle teeth were suggested to have dismembered food items of all sizes (Wall et al. 1995;Hungerb€ uhler 2000). While our analyses cannot refute these suggestions of different behavioural uses along the tooth row, the large degree of textural similarity between teeth is indicative of food items with similar material properties being consumed along the entire tooth row.
It has been suggested that phytosaurs employed a head-shaking technique to process large items into smaller pieces before swallowing, passively cutting food material as it moved along their serrated teeth (Chatterjee 1978;Hungerb€ uhler 2000). Such a technique has been demonstrably shown in carcharhinid sharks (Frazzetta 1988) and has been suggested in tyrannosaurid dinosaurs (Farlow & Brinkmann 1994). Alternatively, phytosaurs may have exhibited behaviours similar to the infamous 'death roll' performed by extant crocodilians, where individuals seize prey within their mouths and then spin around the long axis of their bodies in order to process large prey (Drumheller et al. 2019). Although the death roll has never been explicitly hypothesized for phytosaurs, the near universal occurrence of this behaviour among extant crocodilians (irrespective of skull ecomorphology, diet or phylogenetic relatedness; Drumheller et al. 2019) and the high degree of morphological convergence between crocodilians and phytosaurs (Chatterjee 1978;Stocker 2012;Witzmann et al. 2014) means that this F I G . 5 . Canonical variate texture-dietary space of four ISO texture parameters (Spk, Sds, Vmp and Smr1) for reptiles and phytosaurs. Texture-dietary space based on extant reptile data with phytosaurs projected onto the first two axes as unknown datum points. Specimens with associated letters represent surfaces A-J in Figure 3. Arrows show significant correlations of dietary characteristics along PC axes 1 and 2. Phytosaur specimen information corresponding to PCA plot number can found in Bestwick et al. behaviour cannot be automatically ruled out for phytosaurs. The mechanisms behind microwear formation may be poorly understood (Calandra et al. 2012;Schulz et al. 2013a, b) but it is not unreasonable to suggest that these processing behaviours, if exhibited across the entire tooth row, could have resulted in similar intra-jaw tooth microwear textures (Blateyron 2013). Higher forces experienced by posterior teeth during this behaviour, as a result of being positioned closer to the jaw adductor musculature (Kammerer et al. 2006;Erickson et al. 2012;Porro et al. 2013), may have caused the subtle texture differences between posterior and the anterior and middle teeth, although this has yet to be unequivocally tested.

Microwear differences across a spectrum of robustness
Overall, our analyses show that morphologically robust phytosaurs exhibit rougher tooth microwear textures than gracile phytosaurs. Of the five ISO parameters that differed the most (Sdq,Sdr,Sp,Sz,S5z), none differed between the extant dietary guilds from Bestwick et al. This indicates that texture differences between robust and gracile phytosaurs are minimally due to dietary differences.
At first glance, our analyses appear to corroborate previous suggestions of phytosaur dietary differences based on comparative anatomy with extant crocodilians; i.e. cranially robust taxa were more likely to have been carnivorous, and cranially gracile taxa were more likely to have been piscivorous (Hunt 1989;Parrish 1989;Hun-gerb€ uhler 2000;Heckert & Camp 2013). Morphometric analyses of the skulls of extant crocodilians and odontocete whales found extreme morphological convergence between the highly piscivorous taxa; e.g. G. gangeticus and the La Plata river dolphin, Pontoporia blainvillei, respectively (Iijima 2017;McCurry et al. 2017a). These taxa independently exhibit elongate rostra and shallow mandibles, among other morphological similarities (Iijima 2017;McCurry et al. 2017a). Since these taxa are separated by nearly 300 million years of evolution (Lee 1999), it is reasonable to assume that similar morphologies in the skulls of gracile phytosaurs were also adaptations for piscivory. However, the microwear textures of robust and gracile phytosaurs do not occupy separate areas in any of the texture-dietary spaces of extant reptiles. This suggests that the material properties of consumed foods are similar for both morphological groups and that the relationship between robustness and diet is not as straightforward as previously assumed.
When phytosaurs are grouped by robustness, microwear texture differences may be showing a signal for F I G . 6 . Canonical variate texture-dietary space of 21 International Organization for Standardization (ISO) texture parameters for extant reptiles and phytosaurs. Texture-dietary space based on extant reptile data with phytosaurs projected onto the first two axes as unknown datum points. Specimens with associated letters represent surfaces A-J in Figure 3. Arrows show significant correlations of dietary characteristics along PC axes 1 and 2. Phytosaur specimen information corresponding to PCA plot number can found in Bestwick et al. (2020b). prey size as opposed to diet per se. Larger items require more oral handling before consumption, most commonly tearing items into bite-sized pieces (Cleuren & De Vree 2000;D'Amore & Blumenschine 2009). This increases the frequency of tooth-food interactions which provides more opportunities for microwear textures to form, irrespective of the taxonomic identity of consumed foods (Bestwick et al. 2019). The skulls of consumers are also subjected to greater mechanical stresses and strains during these processing behaviours (Walmsley et al. 2013;McCurry et al. 2017c). Adductor muscle reconstructions of several phytosaurs found that the skulls of robust taxa were better suited for dealing with higher feeding-related mechanical loads and were therefore better adapted for predating larger items (Wall et al. 1995). More broadly, biomechanical modelling of several groups of extant taxa, including crocodilians and odontocetes, similarly found that taxa with elongate rostra are subjected to higher feeding-related stresses (Walmsley et al. 2013;McCurry et al. 2017c). In these taxa, cranial shape is a useful predictor of prey size, with consumption of larger items only exhibited by robust taxa (Metzger & Herrel 2005;McCurry et al. 2017b). That our analyses are consistent with anatomical comparisons and biomechanical models in both phytosaurs and extant taxa strongly indicates that morphologically robust and gracile phytosaurs may have partitioned resources by physical size, as opposed to the taxonomic identity or material properties, of consumed foods.

Reconstructions of phytosaur diets
Overall, the areas of occupied texture-dietary space in both the extant reptile and crocodilian-only datasets suggest that, as a clade, phytosaur dietary diversity was relatively low. The tendency for specimens to plot around the centre of the texture-dietary spaces, where most or even all of the extant guilds overlap, suggests that phytosaurs were likely to have been dietary generalists. The strong overlap of phytosaurs with piscivores and carnivores in the crocodilian-only texture-dietary space indicates that generalized phytosaur diets primarily consisted of vertebrates, with individual taxa exhibiting slight preferences for fish or tetrapods. This allows us to test previous dietary hypotheses and provide more refined characterization of the ecological roles that phytosaurs performed within Triassic food webs.
Our results broadly support previous inferences of diet based on comparative anatomy, content fossils and associations that phytosaurs were predominantly piscivorous and/ or carnivorous (Chatterjee 1978;Hunt 1989;Hungerb€ uhler 1998Hungerb€ uhler , 2000Li et al. 2012;Heckert & Camp 2013;Drumheller et al. 2014;Stocker et al. 2017) albeit with higher degrees of dietary generalism than previously appreciated. Many inferences simply regard phytosaurs as obligate or near-obligate members of these dietary guilds. In reality, diet is often much more complex as lines of evidence do not always agree. For example, preserved stomach contents of the gracile phytosaur Parasuchus hislopi include temnospondyl, basal archosauromorph and rhynchosaur F I G . 7 . Canonical variate texture-dietary space of 21 International Organization for Standardization (ISO) texture parameters for extant crocodilians and phytosaurs. Texture-dietary space based on extant crocodilian data with phytosaurs projected onto the first two axes as unknown datum points. Specimens with associated letters represent surfaces B, D, F-J in Figure 3. Arrows show significant correlations of dietary characteristics along PC axes 1 and 2. Phytosaur specimen information corresponding to PCA plot number can found in Bestwick et al. remains (Chatterjee 1978), which starkly contrasts with ideas of obligate piscivory based on its anatomy (Hunt 1989). More broadly, several crocodilians which have traditionally been regarded as obligate piscivores based on their cranial anatomy, such as the false gharial (Tomistoma schlegelii) and the freshwater crocodile (Crocodylus johnstoni), in fact have much more variable diets that also include mammals, birds, reptiles and amphibians (Drumheller et al. 2014;Drumheller & Wilberg 2020; and references therein). Our extant multivariate frameworks do include dietary specialists such as G. gangeticus (Hussain 1991) and the carnivorous V. komodoensis (Auffenberg 1981), which both plot in the 'high vertebrate diet' areas of texture-dietary spaces (Bestwick et al. 2019). That phytosaurs generally do not plot in the same areas as the dietary specialists further indicates that these Triassic reptiles were, in ecological terms, more similar to extant reptiles with generalist and opportunistic diets.
Despite our data indicating that the study phytosaurs were dietary generalists, DMTA nevertheless provides quantitative constraints on phytosaur diets and ecological roles. Mystriosuchus planirostris, for example, has been interpreted a piscivore based on its slender rostrum and association with fluvial palaeoenvironments (Hunt 1989;Hungerb€ uhler 1998). Though My. planirostris plots more strongly with piscivores in the crocodile-only texture-dietary space, this phytosaur shows slightly higher degrees of overlap with carnivores in the crocodilian and varanid texture-dietary spaces. This suggests that My. planirostris was predominantly a carnivore with fish probably comprising the remainder of its diet. Based on biomechanical modelling of other slender-snouted phytosaurs (Wall et al. 1995), it is not unreasonable to suggest that My. planirostris probably predated small tetrapods. Nicrosaurus kapffi has been interpreted as a carnivore based on its robust cranial morphology and its postcranial morphology that is indicative of a more terrestrial lifestyle, relative to other phytosaurs (Hunt 1989;Hungerb€ uhler 2000;Kimmig & Arp 2010). However, since N. kapffi overlaps strongly with both carnivores and piscivores in the texture-dietary spaces, this phytosaur probably consumed both tetrapods and fish. Nicrosaurus meyeri, alternatively, has been interpreted as a piscivore based on its cranial anatomy (Hungerb€ uhler & Hunt 2000). In our analysis, N. meyeri generally plots in similar areas of the texturedietary spaces as N. kapffi, suggesting that both these phytosaurs had similar mixed diets. No explicit dietary hypotheses have been made for Ma. pristinus, but other slender-snouted phytosaurs from the same deposits as Ma. pristinus from the middle Norian Chinle Formation, western USA, have been interpreted as piscivorous (Parrish 1989). While this phytosaur does show some overlap with extant piscivores in the texture-dietary spaces, Ma. pristinus also shows the strongest degree of overlap with 'harder' invertebrate consumers of all study phytosaurs. This suggests a broader dietary range than has previously been proposed for phytosaurs, and, given the likely combination of fish and 'harder' invertebrates in the Ma. pristinus diet, it is not unreasonable to suggest that this phytosaur could be an ecological analogue of Ca. crocodilus or Cr. acutus (Thorbjarnarson 1988;Laverty & Dobson 2013). Similarly, no explicit dietary hypotheses have been made for the robust taxon 'S. lithodendrorum', although morphologically similar phytosaurs have been interpreted as carnivorous (Hunt 1989;Parrish 1989;Heckert & Camp 2013). However, the overlap with carnivores and piscivores in the crocodilian and varanid texture-dietary spaces and overlap with piscivores and 'harder' invertebrate consumers in the crocodile-only texture-dietary space suggests a much more diverse diet.
More broadly, our results provide novel insight into Late Triassic ecosystem functioning. Our study phytosaurs are known from the same geological formations from the middle-late Norian of western USA and central Europe . Since DMTA indicates that phytosaurs exhibited large degrees of dietary overlap, it is reasonable to assume that there were potentially high levels of dietary competition between multiple phytosaur taxa, and with other archosaur clades, within Triassic food webs. This assumption is not unfounded as extant contemporaneous archosaurs also exhibit dietary competition, such as Ca. crocodilus and the black caiman, Melanosuchus niger, in the Amazon (Laverty & Dobson 2013). Several non-mutually exclusive factors could explain how Triassic ecosystems could support high levels of dietary competition. First, their semi-aquatic lifestyles would have enabled a degree of spatial partitioning from large, terrestrial archosaurs such as rauisuchids, which are largely regarded as carnivorous (Chatterjee 1978;Parrish 1989;Nesbitt et al. 2013;Drumheller et al. 2014). Second, dietary generalism and opportunism could have reduced intra-specific competition, as is exhibited by extant varanids (Losos & Greene 1988;Rahman et al. 2017) and some crocodilians (Rosenblatt et al. 2013(Rosenblatt et al. , 2015. As mentioned above, size-based resource partitioning may have occurred between contemporaneous phytosaurs based on total body size, cranial robustness (e.g. between N. kapffi and N. meyeri) and/or sexual dimorphism. The final factor mentioned involves archosauromorph metabolisms. Histological studies, combined with phylogenetic bracketing, of archosauromorphs have suggested that phytosaurs possessed metabolisms intermediate between those of extant reptiles (ectothermic and poikilothermic) and of extant mammals and birds (endothermic and homoeothermic) (Cubo & Jalil 2019). Lower metabolisms in phytosaurs, relative to endotherms, would have enabled individuals to consume fewer items to meet metabolic requirements, and thus lower competition levels (Grady et al. 2019;Jessop et al. 2020). This highlights the uniqueness of reptile-dominated ecosystems not only for the Late Triassic, but for the entire Mesozoic.

CONCLUSIONS
We used DMTA to provide the first quantitative evidence on the diets of phytosaurs and to explore possible impacts of tooth position and cranial robustness on tooth microwear textures. Our analyses find no evidence of dietary partitioning along phytosaur tooth rows. The very subtle textural differences found along tooth rows are interpreted as the result of different loading pressures that teeth experience during the acquisition and oral processing of food items. Despite the overall similarity of tooth texture along phytosaur tooth rows, we nevertheless recommend standardized sampling positions in future DMTA for robust dietary analyses and better understandings of phytosaur feeding behaviours. Our analyses find texture differences between cranially robust and gracile phytosaurs that are probably the result of prey size, and the higher biomechanical forces required to seize and process larger prey, rather than differences in the material properties of prey. These results, and subsequent implications for phytosaur diet, are somewhat consistent with biomechanical biting models of several clades of extant animal. However, further modelling of phytosaur biting behaviours would greatly increase our understanding on inter-specific resource partitioning. We provide the first quantitative constraints of phytosaur diets, revealing that phytosaurs were predominantly carnivorous and/or piscivorous with few preferences for 'harder' invertebrates. Overall phytosaur dietary diversity is relatively low with indications that taxa exhibited dietary generalism and opportunism, rather than strict niche partitioning. This not only contrasts with many hypotheses of phytosaur diets, but also shows higher degrees of ecological convergence with extant crocodilians than previously appreciated. Our analyses therefore support that phytosaurs were important components of Triassic food webs and reveal similarities between Triassic and modern ecosystems. Future application of DMTA to phytosaurs, particularly Middle and latest Triassic taxa, would further enhance our understanding of the ecological roles that phytosaurs performed within Triassic food webs and on the functioning and evolution of Triassic ecosystems.