Ecological signal in the size and shape of marine amniote teeth

Abstract

Amniotes have been a major component of marine trophic chains from the beginning of the Triassic to present day, with hundreds of species. However, inferences of their (palaeo)ecology have mostly been qualitative, making it difficult to track how dietary niches have changed through time and across clades. Here, we tackle this issue by applying a novel geometric morphometric protocol to three-dimensional models of tooth crowns across a wide range of raptorial marine amniotes. Our results highlight the phenomenon of dental simplification and widespread convergence in marine amniotes, limiting the range of tooth crown morphologies. Importantly, we quantitatively demonstrate that tooth crown shape and size are strongly associated with diet, whereas crown surface complexity is not. The maximal range of tooth shapes in both mammals and reptiles is seen in medium-sized taxa; large crowns are simple and restricted to a fraction of the morphospace. We recognize four principal raptorial guilds within toothed marine amniotes (durophages, generalists, flesh cutters and flesh piercers). Moreover, even though all these feeding guilds have been convergently colonized over the last 200 Myr, a series of dental morphologies are unique to the Mesozoic period, probably reflecting a distinct ecosystem structure.

1. Introduction

Since the Permian, more than 60 amniote lineages (predominantly diapsid reptiles and placental mammals) independently transitioned from terrestrial to aquatic environments [1,2]. The strong constraints of the aquatic medium channelled phenotypic evolution and forced widespread convergences, notably in body shape [3], physiology [4,5] and feeding strategies [6,7]. No matter the ancestral complexity, the teeth of most marine amniotes appear simplified towards a conical or bulbous shape, which makes their functional interpretation fairly straightforward [6,8]. As such, marine tetrapod teeth have been extensively used as a proxy for the ecological niche of their bearers, yielding important insights into the composition of ancient marine ecosystems [6,9].

However, the dominant frameworks linking tooth shape with diet do not take tooth size into account [6,8], despite appearing strongly linked to the diet of extant marine amniotes (e.g. cetaceans [10]). Being qualitative [6], or essentially based on a few discrete features [8], these frameworks do not specifically use crown morphology such as flat surfaces (e.g. the subtrihedral teeth of thalassophonean pliosaurids [11,12]), carinae, apical cusplets (e.g. the mosasaurine mosasauroid Globidens [9]) and the direction of curvature. More recent analyses of marine reptile teeth have started incorporating size [13,14] or ornamentation, but discretized [12,15]. This study presents a new, quantitative and almost fully automated protocol that addresses all these issues at once. The method uses geometric morphometrics [16–18], with an element of automatic pseudolandmarking [19], designed to accurately sample tooth shape after placing minimal homologous landmarks per tooth. We apply this protocol to a new dataset of three-dimensional (3D) tooth crown models of 54 taxa of obligate marine amniotes, both extinct and extant. We then use these data to: (i) analyse interplay between tooth shape, tooth size, gut content and dental surface complexity in marine tetrapods and (ii) propose a new framework to classify these organisms into feeding guilds and infer their trophic role. Furthermore, our new protocol can easily be applied to explore shape variation and disparity in a variety of simple structures, opening new research avenues.

2. Material and methods

(a) . Morphological data and sampling

We sampled a total 54 taxa of extinct and extant raptorial, fully toothed, aquatic amniotes, representing cetaceans, sauropterygians, mosasaurid squamates, archosaurians and ichthyosaurians, and covering most of the marginal (i.e. not palatal or pterygoid) tooth shapes present in these groups. We did not incorporate crowns from taxa with incomplete dentition such as the sperm whale Physeter macrocephalus, nor from dental plates, where the tight fit of palatal teeth results in polygonal crowns such as in placodont sauropterygians (marginal teeth from placodonts are, however, incorporated in our dataset). Nevertheless, our protocol (see below) works very well on these shapes, as well as other conical objects (figure 1; electronic supplementary material, figure S1).

Figure 1.

High-density shape sampling protocol. (a) Steps of the high-density geometric morphometrics procedure; the part in the grey background is fully automated. The crowns under ‘Feeding guilds’ have been generated by thin-plate spline of PC1 and PC2 extremes (figure 2). (b) Example of 3D crown meshes with their fixed landmarks (5, in pinkish red colour and large size) and their surface semi-landmarks (2000, in green sea colour and small size), not to scale. The silhouettes are clade-specific, not species-specific. GPA, generalized procrustes superimposition; PCA, principal component analysis. (Online version in colour.)

We sampled the best-preserved, unworn teeth within a narrow region located in the middle of the snout (i.e. half of pre-orbital snout length), where the teeth are usually large and worn [20,21], indicating these were often used in food procurement. Our sampling strategy allowed teeth with minor breaks in the enamel to be incorporated; see electronic supplementary material, table S1 for all details. For isolated teeth, we selected teeth matching the tooth shape expected in that region of the snout. One exception is the highly heterodont mosasaurid Globidens, for which we also selected one distal tooth in addition to a tooth from the mesial third of the mandible. Two worn teeth have also been selected (one belonging to Orcinus orca and one belonging to Pliosaurus sp.) to visualize how tooth wear affects position in morphospaces. By convention, we sampled right dentary teeth. In specimens where the teeth of this region were not well preserved, the strong left–right symmetry in dental elements [22] permitted the sampling of left dentary or maxillary teeth; a mirroring algorithm was applied wherever necessary. This resulted in a total of 56 tooth crowns. Most were digitized using a laser scanner (Creaform Handyscan 300), at a 0.2 mm resolution. Others have been obtained as CT scans or photogrammetric models from published electronic supplementary material, MorphoSource or colleagues (see electronic supplementary material, tables S1 and S2 for metadata).

(b) . Size and dietary categories

We gathered two measurements from the same specimens sampled for their tooth crowns, when possible: mandible length (anterior tip of the symphysis to posterior margin of the glenoid cavity) and interglenoid distance (distance between medial surfaces of the mandibular glenoids, or the corresponding cranial elements of the jaw joint when needed), both rounded to the nearest millimetre, using the software Meshlab [23]. The latter represents a proxy for the diameter of the gullet, i.e. the upper bound of the prey items (whole or fractionated) that can be swallowed. We also calculated crown height as the Euclidean distance between the apex (fixed landmark 1) and the centre of the base of the crown (taken at the mean point between fixed landmarks 4 (distal base of the crown) and 5 (mesial base of the crown)).

Gastric content reported in a specimen is generalized to the generic level if the species in question have similar mandible lengths and crown morphologies. For example, the gastric content reported for Mosasaurus missouriensis [24] is generalized to Mosasaurus hoffmanni and Mosasaurus lemmonieri but the gastric content reported for Prognathodon overtoni [24] is not generalized to Prognathodon currii (tooth crowns clearly different in shape) or Prognathodon solvayi (notably smaller skull size). The dietary categories we establish (‘flesh, large’, ‘flesh, medium’, ‘flesh, small’, ‘shelled, large’ and ‘shelled, small’; see electronic supplementary material, table S3 for details) take the variety of prey items into account, focusing on their sizes, hardness and trophic positions rather than phylogenetic relatedness (see electronic supplementary material, table S4 for the data).

(c) . High-density geometric morphometrics and morphospace occupation

The tooth crowns of raptorial marine amniotes are essentially conical (except in carnivoran mammals), leaving few homologous points that can be landmarked. Moreover, a high number of surface landmarks is needed to fully capture peculiar, functionally important traits such as carinae and flat surfaces. In this paper, we establish a new protocol mixing pseudolandmarking [25] into high-density landmarking procedures. This procedure requires placement of just five fixed 3D landmarks on a 3D model of each tooth crown, and then automatically samples thousands of points on the crown surface in R [26]. The procedure is summarized in figure 1a detailed step-by-step guide is provided in electronic supplementary material. The surface of a simple shape (here a 3D dome) is landmarked automatically (2000 surface semi-landmarks) by sampling the coordinates of the triangles composing this shape; this part of the protocol is borrowed from pseudolandmarking techniques. Then, using the geomorph v.4.0.3 [27] and Morpho v.2.9 [28] packages, an atlas is created and is used to patch the 2000 surface semi-landmarks onto each crown model. A generalized procrustes superimposition (GPA) is then called to eliminate the size and positioning factors and then a principal component analysis (PCA) is applied on the GPA coordinates to produce morphospaces. Density-based macroevolutionary landscape are then produced using the method explained by Fischer et al. [29]. Newly digitized models are deposited on Morphosource (https://www.morphosource.org/projects/000435369). All 3D tooth crown models and their fixed landmark coordinates, as well as the R script (including automatic cropping of crown models for OPCR analyses) are openly available in electronic supplementary material and ORBi (https://hdl.handle.net/2268/293921).

(d) . Statistical tests

Correlations were tested for by regressing log-transformed measurements against one another. The significance of crown size differences in terms of dietary categories and shape was tested using a Wilcoxon–Mann–Whitney test. A multivariate analysis of variance (MANOVA) was used in a pairwise manner to test for morphological differences between feeding guilds initially established by Massare [6] (see also [9,13,30] for detailed explanations of these guilds); separate MANOVAs analysed the shape differences between ‘small’ and ‘large’ crowns (defined by fossilized gut content; threshold approx. 20 mm, see Results). MANOVAs were performed using the PC axes accounting for greater than 1% of the total variance (i.e. the first four, which together explain greater than 95% of the total variance).

(e) . OPCR

Orientation patch count rotated (OPCR) quantifies surface complexity by counting the number of patches with a given orientation in a 3D model (e.g. [31]) and has been successfully applied to amniote dentition (e.g. [32]). To ensure that the OPCR and high-density geometric morphometric analyses sample the exact same portion of the 3D models, we wrote a short script that automatically crops each 3D model according to a plane defined by three of the fixed landmarks located at the base of the crown and exports this new model as 3D mesh. Each resultant mesh was then simplified to 1000 triangles using MeshLab v.2020.07 [23], with the apex aligned along the z-axis. We used the molaR_Batch function (OPCr_minimum_faces = 3 and 5) from the package molaR v.4.5 [31] to compute surface complexity.

3. Results

(a) . Morphospace occupation and crown complexity

Our novel procedure can sample marine amniote crown shape extensively (figure 1; electronic supplementary material, figure S1), and can notably capture morphological details such as some of the apicobasal ridges of pliosaurids (electronic supplementary material, figure S2). Reduction of dimensionality through PCA captures 90.79% of the variance with just two axes, the first axis accounting for 86.73% (figure 2; electronic supplementary material, figure S3). The first principal component axis (PC1) describes the aspect ratio of the tooth crown, going from the bulbous teeth of globidensine mosasaurids and placodonts (highest positive PC1) to the recurved, needle-like teeth of long-necked plesiosaurians and early thalattosuchians (highly negative PC1); morphospace occupation per species is detailed in electronic supplementary material, figure S3 and can also be visualized interactively (see R script in electronic supplementary material). The second principal component axis (PC2) describes the labiolingual compression of the tooth crown and the direction of tooth curvature: taxa with labiolingually flattened and distally recurved crowns such as mosasaurids and archaeocetes occupy the maximal positive values whereas taxa with circular crowns directed lingually (such as delphinoids) occupy the maximal negative values. In our sample, the flatness of the crown often associates with the presence of carinae (i.e. sharp cutting edges), so this axis carries an important functional signal. The third principal component axis (PC3) only accounts for 2.77% of the variance and captures slight variations of the direction of curvature as well (mesial versus mesiodistal). All the other PC axes (53 out of 56) account for less than 2% of the variance (electronic supplementary material, table S5) and will not be described. A large majority of teeth (≈75%), encompassing all main clades, cluster in a region between −0.25 and 0.0 on PC1 and close to 0.0 on PC2, representing conical and slightly labiolingually compressed crowns (figure 2a,b; electronic supplementary material, figure S3). This numerically confirms that most marine amniotes have fairly simple and similarly shaped tooth crowns. This clustering also results in unexplored dental morphologies, notably in high positive values along PC2, which contain crowns which are strongly labiolingually compressed. The region in between clear durophages (placodont sauropterygians and globidensine mosasaurids) and ‘conical’ toothed taxa is also sparsely occupied, with few taxa exhibiting straight, high dome-shaped teeth. Only three specimens occupy this region, and two of them are very large: the mosasaurine P. currii, which evolved a very large size and a crushing diet from small, flesh-eating ancestors [9] and the gigantic physeteroid Livyatan melvillei, which possesses the largest skull ever recorded for a raptorial amniote [33]. The third specimen in that region is a worn tooth of the delphinid O. orca. Our results show that the effect of apical wear can be strong, transforming the piercing conical crowns of O. orca into a morphology close to that of some shell-crushing mosasaurids. Despite having active tooth replacement [34], a similar trend is observed in Pliosaurus, travelling about 1/5th of the length of PC1 towards positive values because of apical wear (figure 2a).

Figure 2.

Dental morphospace occupation by raptorial marine amniotes. (a) Morphospace occupation visualized by the first two axes of the PCA. The diameter of each dot is directly proportional to centroid size. The colour of each dot is the corresponding OPCR value for this 3D mesh. See also the effect of apical wear on Pliosaurus and Orcinus orca (white and grey arrows, respectively). Density of morphospace occupation is visualized by shades of grey (darker = higher density; see also (b). We visualized the morphological features captured by the axes of the PCA by predicting the shape at the extremes of each axis multiplied by 1.2, thus providing a slightly more exaggerated shape than the taxa analysed. We then warping the 3D mesh of Mosasaurus hoffmanni (ULg PA 25119a) to match the four sets of predicted landmarks using a thin-plate spline function. (b) Density of morphospace occupation visualized as a 3D object, showing a clear majority of raptorial marine amniote teeth have a similar morphology. (c) Morphospace occupation (PC1, PC2) per clade. (d) Morphospace occupation (PC1, PC2) per gastric content. (Online version in colour.)

We tested how the feeding guilds first established by Massare [6] correlate with our new shape data. The ‘Crush’ guild appears clearly separated from the others (electronic supplementary material, figure S4). If only the first four axes of the PCA are used (accounting for greater than 95% of the total variance), the ‘Smash’ and ‘Crunch’ guilds are recovered as significantly distinct (MANOVA p-value < 0.05), being separated along PC2, as are ‘Cut’ and ‘Pierce’ guilds (MANOVA p-value < 0.05). The main issue lies in the ‘General’ guild(s), which cannot be distinguished from the ‘Pierce’ guild (MANOVA p-value = 0.05623) nor from the ‘Cut’ guild (MANOVA p-value = 0.05137).

Our OPCR results pair well with the density of morphospace occupation (figure 2a,b) in indicating a strong tendency towards simple conical crown morphologies. The range of OPCR values is generally low, similar to non-herbivorous, non-multicusped lizards [22,35]. Only a few taxa exhibit high values, although this is probably due to damaged enamel (e.g. L. melvillei) or variations in tooth size and resolution of scans (e.g. very large ridged teeth of Kronosaurus queenslandicus/Eiectus longmani). Values for larger taxa are thus relatively high compared to smaller species, because the resolution of laser and CT-scanning is scale-dependent. Nevertheless, the important morphological variation in marine amniote tooth shapes pairs with very little difference in their surface complexity, irrespective of scale.

(b) . Distribution of size and gastric contents

Crown size is not distributed randomly: both the centroid size and the apicobasal height of the crown are positively correlated with mandible length (centroid size: R² = 0.5424, p-value < 0.0001; crown height: R² = 0.7068, p-value < 0.0001) and interglenoid distance (a proxy for gullet diameter) (centroid size: R² = 0.6828, p-value < 0.0001; crown height: R² = 0.4832, p-value < 0.001) (electronic supplementary material, figure S5). The distribution of crown sizes along PC1 and PC2 is not random either: most teeth are clustered close to the origin of both axes, indicating that a majority of teeth have unspecialized, ‘common’ overall morphologies. This is especially evident for large teeth (i.e. crown height greater than 20 mm); the range of PC1 and PC2 values occupied by these teeth is much smaller (53.4% and 58.9% of the total spread, respectively) and their disparity is significantly smaller than that of small crowns (Wilcoxon–Mann–Whitney p-value < 0.0001; electronic supplementary material, figure S6). This indicates that large teeth are usually simple in morphology; more shape variation is seen for crown sizes between 5 and 20 mm, i.e. small to medium-sized taxa (figure 3; electronic supplementary material, figure S5).

Figure 3.

Distribution of size and shape. (a) Distribution of crown height (log) against PC1. (b) Distribution of crown height (log) against PC2. The dot size is directly proportional to crown height. In (a) and (b), the range of PC values covered by large and small crowns is indicated by a grey background. This range is much smaller for large (greater than 20 mm, above the dotted line) crowns. (c) Distribution of PC values, centroid sizes, crown heights and OPCR values per gastric content category as box plots. (Online version in colour.)

The dietary categories deduced from fossilized gut content can be discriminated using a combination of aspect ratio (PC1), curvature (PC2) and crown size (figure 3). Only two data points are available for the categories ‘flesh, medium’ and ‘shelled, large’, and each contain taxa already present in other dietary categories and will not be discussed here. The aspect ratio alone isolates durophages from large fleshy prey eaters (Wilcoxon–Mann–Whitney p-value < 0.05) and durophages from small fish/squid specialists (Wilcoxon–Mann–Whitney p-value < 0.05), [6] but not large fleshy prey from small fish/squid specialists (Wilcoxon–Mann–Whitney p-value = 0.3097). Tooth crown curvature (PC2) separates small fish/squid specialists from the rest of the dataset, whereas crown size separates large fleshy prey eaters from fish/squid specialists (figure 3; Wilcoxon–Mann–Whitney p-value < 0.001). OPCR values do not correlate with dietary categories, as would be expected given the drivers of that signal in our dataset (see above). Because crown size correlates with gullet diameter and both crown size and shape correlate with diet (figure 3; electronic supplementary material, figure S5), we posit that the combination of crown shape (PC1 and PC2) and crown size forms a solid basis to define feeding guilds, and to analyse trophic diversity in raptorial marine amniotes.

4. Discussion

(a) . Large predators have simple teeth

For more than 30 years, teeth have been intensively used to infer the palaeoecology of marine amniotes [6,8,9,13,36], based largely on the seminal paper of Massare which established a canvas linking tooth shape with prey preference [6]. However, this essentially qualitative canvas mixes diet with behaviour (e.g. the difference between ‘smash’ and ‘crunch’ guilds) and relies—at times—on data that is difficult to generalize to most marine amniotes [37]. For example, the tooth shape of P. macrocephalus is seen as indicative of a squid-rich diet in that canvas, despite evidence that the lower dentition does not play an active role in feeding [38], alongside an edentulous upper jaw and a diet including the largest marine invertebrate that has ever evolved (e.g. [39]). A series of tooth morphologies do not fit within ‘Massare's triangle’ (even if modifications have been attempted [9] as well as thorough attempts to retrofit Massare's guilds into quantitative frameworks [13]), and—perhaps most importantly—crucial functional features such as tooth size and the direction of curvature of the tooth crown (distally versus lingually) are not considered. As an example in this study, the teeth of the Middle Jurassic pliosaurid Liopleurodon ferox—regarded as an apex predator/hypercarnivore [40]—are not that different in shape from those of the near-obligately piscivorous gharial (Gavialis gangeticus) (figure 2a). However, the teeth of Liopleurodon are nearly six times larger than those of Gavialis, resulting in teeth that would be very likely to behave differently when used against prey, recalling the concept of ‘functional heterodonty’ [41]. Similarly, whereas the Early Jurassic ichthyosaurians Ichthyosaurus (≈2–3 m long) and Temnodontosaurus trigonodon (≈8 m long) have relatively similar tooth morphologies [42,43], their gut contents are distinct: fish scales and cephalopod hooklets in Ichthyosaurus [6,44] and marine reptiles and cephalopod hooklets in T. trigonodon [45]. Understanding diet through the prism of size instead of just crown shape reconciles records of extant orcas still able to hunt large vertebrate prey despite having rounded, worn (but still large) teeth [46,47].

Our gastric content data confirms that diet correlates with crown size (figure 3), which itself correlates with gullet diameter (electronic supplementary material, figure S5). A threshold separating animals able to consume large, fleshy, vertebrate prey (the so-called ‘top predators’) seems to occur for crown size greater than 20 mm in height (figure 3). This size boundary also coincides with a strong reduction in the range of crown morphologies (figure 3); above this threshold, all crowns are conical and weakly recurved, often lingually or distolingually, and carinae gradually disappear as crown size increases. From a functional point of view, we posit that the range of possible prey items correlates with crown size; once a fairly straight crown reaches 30–50 mm in height, it will probably be able to damage any kind of possible prey item if enough bite force is applied; this is why it is unsurprising to find cephalopod hooklets alongside marine reptiles in the gut of Temnodontosaurus [45]. In parallel, larger body size in active raptorial predators often requires a larger energy intake (e.g. [48]), which makes hunting small-prey items with low energetic reward less beneficial for these taxa, even though their tooth shape is not necessarily ill-suited for this task. We hypothesize that this increase in dietary possibilities with increasing tooth size (and, to a certain extent, body size) relaxes specialization pressures on crown shape, resulting in less deviation from simple conical morphologies (figure 3; electronic supplementary material, figure S6). These results also suggest that marine amniotes could potentially colonize new niches by changing their tooth (and body) size and not necessarily the shape of their teeth.

Deviations from a simple conical crown (i.e. larger range of shapes along PC axes, as well as carinae and serrations) appear especially frequent for crowns ranging from 5 to 20 mm in height (figure 3; electronic supplementary material, figure S6), suggesting that stronger evolutionary pressures might drive dental differentiation in smaller taxa [49]. Our results make it clear that an explicit incorporation of absolute tooth size (or ‘the size of killing/grasping device relative to prey size’) is crucial to infer the palaeocology of extinct marine amniotes and needs to have an influence in guild definitions.

(b) . Four main raptorial feeding guilds

We propose four discernible guilds for raptorial aquatic amniotes: durophages, generalists, flesh piercers and flesh cutters. Generalists have the widest range of food options and the breadth of the generalist guild varies with tooth size. While restricted to a subset of the positive values of PC1 (i.e. conical, slightly recurved teeth with crown height/basal diameter ratio often ranging from 1.2 to 2.7) for small and medium-sized teeth, we consider that most taxa with conical crowns larger than 20 mm high belong to the generalist guild (orcas, Temnodontosaurus, pliosaurids, many mosasauroids, etc.). As shown above (figure 3), the range of shapes above that threshold is low anyway. Durophages have low, bulbous teeth (aspect ratio less than 1.4 and often less than 0.7) which resist strong apicobasal compression, ideal for breaking the external protection of hard-shelled animals [6]. Small-prey flesh piercers and flesh cutters both have dental adaptations that presumably help them to enter flesh such as pointed apices, high aspect ratio and apicobasal ridges. A key difference lies in ornamentation, as flesh cutters usually have carinae and distally recurved crowns which are ideal for eating prey items larger than the gullet by cutting prey into consumable pieces. These recurved teeth potentially aid in processing prey; the Australian fur seal (Arctocephalus pusillus doriferus) and the leopard seal (Hydrurga leptonyx) possess recurved canines that help these predators to grip prey, which they then shake against the water surface [50,51]. The pterygoid teeth of mosasaurids might have also been adapted for gripping, although to serve in the intraoral transport of prey [52], and are absent in durophagous taxa such as Globidens. Small-prey flesh eaters generally have small, conical teeth with a high aspect ratio (2–3 times high as wide) which are well suited for dispatching small soft-bodied teleosts and coleoids or, if relatively long and procumbent, can be used as a trapping device as in some long-necked plesiosaurians [53].

Even if our high-density sampling protocol is able to capture dental ornamentation such as ridges and carinae (electronic supplementary material, figure S2), the presence or absence of such structures will result in minor displacement of some surface semi-landmarks in the 3D space. In turn, because they do not account for much of crown shape variation across our entire sample, these structures will thus not be captured by the first axes of a PCA despite having a functional signal [54]. This is a limitation of our method when applied to datasets with large shape variation such as this one, and we recommend systematically pairing quantitative palaeobiological analyses with thorough, first-hand anatomical data when discussing finer patterns of niche partitioning, such as that of Foffa et al. [13].

(c) . Convergence and constraints

The very high density of morphospace occupation on slightly negative values on PC1 and our OPCR analyses powerfully illustrate the strength of a long-known phenomenon in aquatic amniotes: convergent dental simplification [6,55]. This phenomenon transcends size and species-relatedness (figures 2 and 3; electronic supplementary material, figure S6), and canalized the dental evolution of most raptorial marine amniotes towards simple conical teeth. Yet, many Cenozoic taxa fall within the morphospace and guilds previously evolved by Mesozoic marine reptiles (figure 2; electronic supplementary material, figure S7) and their maximal occupation densities are close as well (electronic supplementary material, figure S7), indicating convergent evolution of an array of (fairly simple) shapes, which we show are related to diet. In addition to convergent simplification [55] and convergent evolution of crown shapes [6,54], the patterns of morphospace occupation are also driven by the existence of wide unoccupied regions, which can notably be explained in a functional framework. For example, strong labiolingual flattening (high positive values on PC2) combined with a narrow cross-section and strong curvature (negative values on PC1) would make the teeth unable to resist sufficient apicobasal stress to function in prey capture or mastication. Similarly, the region in between clear durophages like placodonts and globidensine mosasaurids and ‘conical’ toothed taxa is sparsely occupied only by two or three gigantic taxa (P. currii, L. melvillei and probably Machimosaurus, which we did not analyse). This suggests that these intermediate morphologies (i.e. high domes) are possibly suboptimal for either crushing or piercing prey items, as shelled sea food is often protected by a thick mineralized armour since the Mesozoic marine revolution [56]. These tooth morphologies are, however, frequent in small, arthropod-eating terrestrial squamates [35] (see also [57]). With that said, it is also important to note that factors beyond functional constraints could influence the patterns of morphospace occupation observed in our results. Teeth can develop into an amazing array of shapes, but conical teeth are still limited in form by developmental constraints that prevent certain extreme areas of morphospace being occupied [58]. Similarly, other factors such as phylogenetic inertia could drive patterns observed in the results. Nevertheless, our new data on the ecological signal in the shape and size of marine amniote teeth combines with the existence of unique Mesozoic morphologies, such as carinated, trihedral and ‘trapping’ teeth (figure 2; electronic supplementary material, figure S7) [53], in illustrating how macroecological changes in ecosystems [59] shape their predators over hundreds of millions of years.

Data accessibility

Newly digitized models are deposited on Morphosource (https://www.morphosource.org/projects/000435369). All 3D tooth crown models and their fixed landmark coordinates, as well as the R script (including automatic cropping of crown models for OPCR analyses), are openly available in the electronic supplementary material [60] and in Université de Liège's Institutional Repository ORBi (https://hdl.handle.net/2268/293921).

Authors' contributions

V.F.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, software, visualization, writing—original draft, writing—review and editing; R.F.B.: data curation, funding acquisition, resources, writing—review and editing; D.F.: data curation, investigation, methodology, software, validation, writing---review and editing; J.A.M.: data curation, formal analysis, investigation, resources, software, writing—review and editing; M.R.M.: data curation, methodology, resources, validation, writing—review and editing; K.M.M.: data curation, formal analysis, investigation, resources, software, writing—review and editing; N.B.: conceptualization, data curation, investigation, resources, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

We benefited from funding from a Fonds pour la Recherche Scientifique FNRS (MIS F.4511.19) grant to V.F. and J.A.M.), a FRIA fellowship (FC 23645) to R.F.B. and a PHC Tournesol grant to V.F. and N.B.