Dietary specializations and diversity in feeding ecology of the earliest stem mammals

Differences in function and dietary ecology between Morganucodon and Kuehneotherium show that lineage splitting during the earliest stages of mammalian evolution was associated with ecomorphological specialization and niche partitioning. The very earliest mammals, living in the Late Triassic and Early Jurassic around 200 million years ago, were small and are often presumed to have been generalized insectivores. Now a close study of two iconic Early Jurassic basal mammals, Morganucodon and Kuehneotherium, shows that niche partitioning and dietary specialization were well under way even at that early date. Analysis of tooth wear and jaw biomechanics shows that whereas Morganucodon had powerful jaws, capable of crushing hard prey such as beetles, Kuehneotherium was adapted for snapping at softer prey. The origin and radiation of mammals are key events in the history of life, with fossils placing the origin at 220 million years ago, in the Late Triassic period1. The earliest mammals, representing the first 50 million years of their evolution and including the most basal taxa, are widely considered to be generalized insectivores1,2. This implies that the first phase of the mammalian radiation—associated with the appearance in the fossil record of important innovations such as heterodont dentition, diphyodonty and the dentary–squamosal jaw joint1,3—was decoupled from ecomorphological diversification2,4. Finds of exceptionally complete specimens of later Mesozoic mammals have revealed greater ecomorphological diversity than previously suspected, including adaptations for swimming, burrowing, digging and even gliding2,5,6, but such well-preserved fossils of earlier mammals do not exist1, and robust analysis of their ecomorphological diversity has previously been lacking. Here we present the results of an integrated analysis, using synchrotron X-ray tomography and analyses of biomechanics, finite element models and tooth microwear textures. We find significant differences in function and dietary ecology between two of the earliest mammaliaform taxa, Morganucodon and Kuehneotherium—taxa that are central to the debate on mammalian evolution. Morganucodon possessed comparatively more forceful and robust jaws and consumed ‘harder’ prey, comparable to extant small-bodied mammals that eat considerable amounts of coleopterans. Kuehneotherium ingested a diet comparable to extant mixed feeders and specialists on ‘soft’ prey such as lepidopterans. Our results reveal previously hidden trophic specialization at the base of the mammalian radiation; hence even the earliest mammaliaforms were beginning to diversify—morphologically, functionally and ecologically. In contrast to the prevailing view2,4, this pattern suggests that lineage splitting during the earliest stages of mammalian evolution was associated with ecomorphological specialization and niche partitioning.

The origin and radiation of mammals are key events in the history of life, with fossils placing the origin at 220 million years ago, in the Late Triassic period 1 . The earliest mammals, representing the first 50 million years of their evolution and including the most basal taxa, are widely considered to be generalized insectivores 1,2 . This implies that the first phase of the mammalian radiation-associated with the appearance in the fossil record of important innovations such as heterodont dentition, diphyodonty and the dentary-squamosal jaw joint 1,3 -was decoupled from ecomorphological diversification 2,4 . Finds of exceptionally complete specimens of later Mesozoic mammals have revealed greater ecomorphological diversity than previously suspected, including adaptations for swimming, burrowing, digging and even gliding 2,5,6 , but such well-preserved fossils of earlier mammals do not exist 1 , and robust analysis of their ecomorphological diversity has previously been lacking. Here we present the results of an integrated analysis, using synchrotron X-ray tomography and analyses of biomechanics, finite element models and tooth microwear textures. We find significant differences in function and dietary ecology between two of the earliest mammaliaform taxa, Morganucodon and Kuehneotherium-taxa that are central to the debate on mammalian evolution. Morganucodon possessed comparatively more forceful and robust jaws and consumed 'harder' prey, comparable to extant smallbodied mammals that eat considerable amounts of coleopterans. Kuehneotherium ingested a diet comparable to extant mixed feeders and specialists on 'soft' prey such as lepidopterans. Our results reveal previously hidden trophic specialization at the base of the mammalian radiation; hence even the earliest mammaliaforms were beginning to diversify-morphologically, functionally and ecologically. In contrast to the prevailing view 2,4 , this pattern suggests that lineage splitting during the earliest stages of mammalian evolution was associated with ecomorphological specialization and niche partitioning.
Recently, much progress has been made in understanding the pattern and timing of the radiation of mammals 7-9 , revealing successive waves of taxonomic and ecomorphological diversification in Middle-Late Jurassic to Palaeogene stem clades and crown groups 2,10,11 . However, understanding of early mammaliaforms and the initial radiation of mammals has lagged behind. Here we address this problem by testing the hypothesis that two of the earliest and most basal mammaliaforms were ecomorphologically distinct. Morganucodon watsoni 12 and Kuehneotherium praecursoris 13 are central to the debate on mammalian origins and are of fundamental phylogenetic importance (Extended Data Fig. 1). Morganucodon is one of the earliest (Late Triassic to Early Jurassic) and bestknown Mesozoic mammals, with a global distribution; Kuehneotherium is of a similar age and size 1,12,13 . Both taxa are thought to be generalized insectivores 1 and co-existed (see Supplementary Information for discussion of sympatry) on a small landmass present during the Early Jurassic marine transgression (Hettangian-Early Sinemurian, about 200 Myr ago), in what is now Glamorgan, South Wales, UK 1,12 (Extended Data Fig. 2). In addition to the apomorphic mammalian jaw joint, both taxa retain the plesiomorphic articular-quadrate jaw joint, as indicated by a well-developed postdentary trough (Fig. 1a, b), thus indicating that the postdentary bones still functioned as part of the jaw joint, rather than being incorporated into a definitive mammalian middle ear as in modern mammals 1,2,12-14 (sensu ref. 15). Curiously, Kuehneotherium possesses advanced molars, with cusps arranged in an obtuse-triangle pattern 13,16 (Extended Data Fig. 3b).
We tested hypotheses of functional and dietary specialization in these early mammaliaforms by generating digital mandibular reconstructions, and applying a suite of techniques: classical mechanics, finite element modelling and quantitative textural analysis of tooth microwear. The mandible is a good choice for study of feeding adaptations as it is primarily adapted for biting, and is not constrained by sensory systems such as eye or brain size 17 . Our null hypothesis was that functional performance did not differ between the two taxa.
Applying classical mechanics, we calculated the mechanical advantage for mid-molar, premolar and canine bites, reflecting the efficiency of the jaw system at transmitting force from the adductor muscles to the bite point. This revealed that Morganucodon has a notably larger mechanical advantage than Kuehneotherium (almost 50% greater during midmolar biting) ( Table 1), indicating that the mandible of Morganucodon had the potential to generate much larger bite forces than Kuehneotherium, and implying that Kuehneotherium bites were potentially faster but less forceful. We also determined jaw strength in bending and torsion during biting, treating the mandibular corpus as a beam 18 . The pattern of bending strength reveals a very different profile between the two taxa ( Fig. 1c, d). Morganucodon shows peak resistance to bending at the rear of the tooth row as might be expected, as this region serves as a structural linkage between the tooth row and posterior functional elements of the jaw, such as the jaw joint and muscles 19 . However, Kuehneotherium shows peak resistance in the region of the anterior molars. Resistance to torsion (J) shows similar patterns (Fig. 1e, f). This different biomechanical profile in Kuehneotherium may reflect the importance of resisting bending in the central tooth row, to maintain the sharp bladed triangulated molars in precise occlusion 16 .
Finite element analysis allowed us to calculate stress, strain and deformation to assess the mechanical behaviour of the jaws 20 . This analysis can provide informative comparative data in the absence of known input parameters 17 and as such the two taxa were loaded with equal adductor muscle forces and constrained at the jaw joint and bite points (Extended Data Fig. 4). Finite element analysis shows that, during a simulated bite, despite similar length and surface area, the dentary of Kuehneotherium experiences greater maximum von Mises stress and maximum principal strain than Morganucodon, regardless of bite position, and higher reaction forces at the jaw joint, despite generating consistently less bite reaction force (Fig. 1g, h and Table 1). Kuehneotherium does not possess a robust condyle as in Morganucodon (Fig. 1a, b), further reducing its ability to withstand high reaction forces at the jaw joint. We tested whether Morganucodon or Kuehneotherium could generate enough bite force to pierce 'hard' insect cuticle (where 'hard' and 'soft' refer to the ease with which prey is pierced and chewed 21 ). Estimation of bite force can circumscribe the range of potential prey, providing a measure of feeding performance and ecological partitioning 22,23 . A variety of insect prey was available at the time: the Glamorgan fissures have yielded beetle remains 24 , and soft-bodied insects, such as scorpion flies, were well established in the Early Jurassic 25 . (See Supplementary Information for discussion of potential prey.) A bite of 2 N is required to pierce the cuticle of a 'hard' insect (for example, a beetle) of appropriate prey size for Morganucodon or Kuehneotherium 26,27 . For Morganucodon, a simulated 2 N bite at midmolar m2 (see Methods) did not generate excessive stress in the jaw (maximum 54 MPa) (Fig. 1g). For Kuehneotherium, increasing muscle loadings (keeping the ratio of muscle recruitment intact), to simulate a bite of 2 N at mid-molar m3 (Fig. 1i), produced higher reaction forces at the dentary condyle (5.45 N compared with 2.38 N for Morganucodon), and maximum von Mises stress values up to 134 MPa, which is 2.5 times that of Morganucodon and close to the value of tensile stress failure for bone 28 . This suggests that Kuehneotherium was probably incapable of processing 'hard' cuticle, and further illustrates differences in the biomechanical performance of the jaws. Comparative biomechanical data therefore point to morphofunctional and dietary specialization in these two taxa.
The hypothesis that Morganucodon and Kuehneotherium consumed different prey was independently tested by comparing their tooth microwear textures with those of extant insectivores with known dietary preferences (specimens listed in Extended Data Table 1). Recent work on insectivorous bats has shown that microwear textural analysis based on threedimensional roughness parameters discriminates between insectivore species that consume different proportions of 'hard' prey (such as beetles) and 'soft' prey (such as moths) 29 . Bats provide a useful comparative data set for our work because of their well-studied dietary differences and similarity in size to Morganucodon and Kuehneotherium. We compared the fossil taxa with four species of bats: Plecotus auritus (brown long-eared bat; a specialist on 'soft' insects); Pipistrellus pipistrellus (common pipistrelle) and Pipistrellus pygmaeus (soprano pipistrelle) (more mixed diet, both specialize on Diptera (flies), but P. pipistrellus consumes insects with a wider range of cuticle 'hardness' and more 'hard' prey than P. pygmaeus); and Rhinolophus ferrumequinum (greater horseshoe bat; mixed diet, but including more beetles-prey that is among the 'hardest' of insects) (see Extended Data Table 2 for dietary details). In the bats, nine roughness parameters differ significantly between species 29 , and principal component analysis (PCA) of these parameters (Fig. 2) separates bats according to dietary preferences in a space defined by principal component axes 1 and 2 (together accounting for 88.3% of variance), with axis 1 strongly correlated with dietary preferences (r s 5 0.81, P , 0.0001). Increasingly negative values indicate higher proportions of 'hard' prey, while increasing positive values indicate increasing proportions of 'soft' prey 29 .
Projecting data for Kuehneotherium and Morganucodon onto the axes resulting from the analysis of bats produces clear separation of the two taxa. Morganucodon has negative values for principal component 1 (PC1), overlapping and extending beyond values for R. ferrumequinum. Slightly rougher textures in Morganucodon suggest that it consumed a higher proportion of 'hard' prey. Most Kuehneotherium specimens have positive values for PC1, overlapping the range of the 'soft' insect specialist Pl. auritus. Two specimens have negative PC1 values and plot into a space defined by the mixed-feeding Pipistrellus. Thirteen roughness parameters from Morganucodon and Kuehneotherium are correlated with the bat dietary axis (PC1; Extended Data Tables 3-5), including nine of the ten parameters that in bats are correlated with diet 29 , and values for PC1 differ significantly between the two fossil species (F 5 5.67; d.f. 5 6, 29; P 5 0.0005). Pairwise tests (Tukey's honestly significant difference; P , 0.05) indicate that microwear textures in Morganucodon and Kuehneotherium differ from one another, yet Morganucodon does not differ from bats with mixed or 'harder' diets, and Kuehneotherium does not differ from the 'soft' insect specialist and mixed feeders. Kuehneotherium specimens from different fissure localities do not differ from one another (see Supplementary Information for specimen and fissure details). That Kuehneotherium and Morganucodon are so clearly separated by application of PCA based on extant bats with different diets provides powerful evidence that the two fossil taxa had diets that differed significantly in terms

RESEARCH LETTER
of prey 'hardness', and provides independent validation of distinctive mechanical behaviour and function revealed through our standard beam analysis and finite element modelling. In summary, our analyses reveal previously hidden trophic diversity and niche partitioning at the base of the mammalian radiation, supporting a hypothesis of coupled lineage splitting and ecomorphological adaptation of the skull and jaws, even during the earliest stages of mammalian evolution. Our approach, combining biomechanical analyses with tooth microtextural validation of dietary differences, does not require exceptionally preserved specimens, and is applicable to fragmentary fossil remains. As such, it has the potential to provide direct evidence of ecomorphology and adaptation through a range of vertebrate radiations, using the most commonly preserved fossil elements: teeth and jaws.
Online Content Methods, along with any additional Extended Data display items and Source Data, are available in the online version of the paper; references unique to these sections appear only in the online paper.

METHODS
Digital models. Digital mandibular reconstructions (Fig. 1a, b) were generated by combining synchrotron radiation X-ray tomographic microscopy or micro-computed tomography scans from similar sized, mature individuals of each taxon (Extended Data Fig. 3 and Supplementary Information). Morganucodon mandibular specimens were scanned using synchrotron radiation X-ray tomographic microscopy at the Swiss Light Source TOMCAT beamline. The material was scanned at 18 keV and a pixel size of 1.85 mm (NHMUK PV M85507) or 22 keV and 3.7 mm (UMZC Eo.D. 45 and UMZC Eo.D.66) to match the size of the sample with the field of view of the microscope. The Kuehneotherium mandibular specimens and molar teeth, provided to illustrate the difference in occlusion, were scanned on a Bruker Skyscan 1172 micro-computed tomography system in the Department of Archaeology and Anthropology at the University of Bristol. A lower resolution was adequate for the finite element models (20 mm), although one specimen (NHMUK PV M92779) was also scanned at 3.1 mm for internal detail.
Processing was performed using Avizo (Visualization Sciences Group). For each taxon, the computed tomography data for each specimen were re-oriented and scaled slightly if required, then digitally merged and manually rendered to produce the reconstructions. As all specimens were similar sizes, very limited scaling was required. The final size of the reconstructed jaw was based on scaling all specimens to the size of the most complete specimen. Any empty alveoli were digitally infilled and tooth crowns removed to leave all tooth surfaces at an equivalent level to the tooth neck. Validation studies have shown that edentate finite element models better represent experimental strains than dentate models 31 , which obviates the need to reconstruct missing teeth in the original scan data. A two-dimensional surface mesh generated in Avizo (Visualization Sciences Group) was exported to Hypermesh (part of the Hyperworks suite from Altair), where the digital jaws were converted into finite element models. Two-dimensional mesh optimization was performed and a three-dimensional finite-element mesh generated of linear four-noded tetrahedral (C3D4) elements (115,213 for Morganucodon and 68,555 for Kuehneotherium). Four-noded tetrahedra may be stiffer than ten-noded tetrahedra and may slightly underestimate strain 32 but our models are comparative, and tetrahedral elements are useful for modelling complex and intricate morphologies. The dentary and tooth roots were created as separate parts with shared mesh boundaries. In the case of fossil material, where it is impossible to validate predictions of bite force with in vivo experimental data, it is possible to compare the relative performance of finite element models if the models are properly scaled. Scaling to equal force:surface area ratio provides a comparison of stress-strength performance based solely on shape 33 . The linear dimensions (canine to condyle) of the mandibles are 17.5 mm and 20.0 mm for Morganucodon and Kuehneotherium respectively. The surface area ratio of Morganucodon:Kuehneotherium, calculated from Avizo, is 1.02:1; as there is very slightly more coronoid process and incisors missing from the Kuehneotherium model, the models are taken to be equal in surface area for this analysis. In this case it was therefore not necessary to scale the applied muscle forces, and the models could be directly compared using the same loadings. The models were converted to SI base-unit linear dimensions (metres) when imported into Abaqus/CAE for finite element analysis.
The dentary bone was assigned isotropic and homogenous material properties: Young's modulus of 18 GPa and Poisson's ratio of 0.3. Properties for the tooth roots were as for dentine: Young's modulus of 25 GPa and Poisson's ratio of 0.3. It is possible that the actual jaw material properties were not isotropic, but in the absence of data otherwise, we assume isotropy for the sake of this analysis. Likewise, we will never know the elastic properties of Mesozoic mammaliaform jaws, yet recent nanoindentation studies on the jaws of rats, squirrels and guinea pigs have revealed a range of 10-30 GPa for the Young's modulus of bone, and 15-25 GPa for incisor dentine 34 ; our chosen values encompass this range.
The orientations (lines of action) of the mammaliaform anterior and posterior temporalis and superficial and deep masseter adductor muscles were reconstructed (Extended Data Fig. 4a, b). The medial pterygoid was omitted as, if present, it was small 35 . Adductor muscle orientation was deduced by estimating the position of the point of origination on the skull, coupled with study of the insertion areas on the mandibular fossae of the fossil specimens. The detailed description of the skull of Morganucodon 36 was used as reference for the muscle origins for Morganucodon. There is no skull material for Kuehneotherium; however, the gross morphology of the lower jaw of the slightly older Brazilian eucynodont Brasilitherium 37,38 is very similar to Kuehneotherium (P.G.G., unpublished observations), so the general skull shape of Brasilitherium was used, with caution, as a proxy for the position of the Kuehneotherium muscle origins (Extended Data Fig. 4c, d). The temporalis has its origin on the parietal bones of the skull, with development of a sagittal crest in Morganucodon, although this latter is not known for Kuehneotherium. The temporalis inserts both medially and laterally in the temporal fossae on either side of the coronoid process, with the muscles divided into anterior and posterior vectors 35 . The anterior temporalis insertions on the lateral and medial sides of the jaws were represented by a predominantly vertical vector and were coupled to a single point in space representing their origin on the skull. The posterior temporalis component was dealt with by applying a posteriorly directed load, as from the posterodorsal portion of the coronoid process. The coronoid hook of Morganucodon 12 and posterodorsal portion of the coronoid process in Kuehneotherium are missing in the reconstructions, so a means to simulate the missing portions of the coronoid processes for the muscle attachment was devised. A multi-point rigid body was created, from the broken edge of the coronoid process to the position of the coronoid hook, in effect completing the coronoid process and producing a point in space for the posterior temporalis origin (Extended Data Fig. 4 insert). The masseter in mammals is divided into superficial and deep components, with their origins on the zygomatic arch 35 : the superficial masseter at the anterior end on the jugal and the deep masseter posteriorly on the squamosal 36 . They insert laterally on the jaw behind the tooth row. In Morganucodon, the two mandibular fossae are distinct and that of the superficial masseter is adjacent to the angular process of the jaw. There is no angular process in Kuehneotherium, but the masseteric fossa is well developed. The lines of action of muscles defined here were used to calculate mechanical advantage, and for the finite element analysis. Classical mechanics. The digital models were used to calculate mechanical advantage and beam theory to measure strength in bending. Mechanical advantage, as a measure of the efficiency of the jaw system to transfer input muscle force from the adductor musculature to the point of biting, is frequently used as a metric of jaw function 39 and correlates with prey choice and feeding ecology in organisms such as fish 40,41 . Mechanical advantage is calculated here as the length of the in-lever (the moment arm of the adductor musculature) divided by the length of the out-lever (the moment arm of the bite: the distance from the jaw joint to the bite point). The moment arms were calculated by taking scaled screen images of the jaws in Abaqus, both in lateral and dorsal orientations to calculate the in-lever arm for each of the four muscles (anterior temporalis, posterior temporalis, deep masseter and superficial masseter). A central point was chosen within the muscle insertion area, in each case, to give the line of action of each muscle. The four muscle vectors were resolved to give a single adductor muscle vector, and the in-lever arm calculated.
Beam theory has been applied to the mandibular corpus in a number of studies 18,22,42 and is related to dietary specialization in small mammals such as bats 43 . A measure of strength in bending is estimated from the section modulus Z, calculated at specific intervals along the jaw. The section modulus Z is the second moment of area (I) divided by the distance from the neutral axis to the outer edge, in the plane of bending, so the orientation considered affects the value of the bending strength. In this case the section modulus was measured in the dorsoventral (Z x ) and mediolateral planes (Z y ). We follow ref. 18 in measuring Z at interdental gaps along the tooth row to the canine, plus a further section just posterior to the ultimate molar. The reconstructed hemimandibles were digitally sliced perpendicular to the long axis of the mandible to produce cross-sectional images at the interdental gaps. Each cross-sectional image was loaded into ImageJ 44 , and Z x and Z y calculated using the plugin MomentMacro. We measured all interdental gaps, with postcanine numbers of 8 and 12 for Morganucodon and Kuehneotherium respectively. The polar moment of inertia (J), the beam's ability to resist torsion, is calculated from the addition of the second moment of area, I, in the dorsoventral (I x ) and mediolateral (I y ) planes. Finite element analysis. Boundary constraints were applied to the condyle and at three bite points: canine, final premolar and mid-molar. The anterior incisor bite point could not be included, as there are no specimens of this portion of the mandible in Kuehneotherium. The ultimate premolar is the largest of the premolars in both taxa. In Morganucodon, the largest (second) molar was chosen as the 'mid molar' and in Kuehneotherium a mid-row (third) molar was used. Both single node 45 and distributed area (stiff beam elements) constraints 46 have been used to estimate bite forces (discussion in ref. 17), but we used multipoint constraints, with master and slave nodes to minimize artificial stress concentrations 47 . There were approximately 22 nodes constrained for each tooth and 30 nodes constrained at the jaw joint. Boundary conditions should never restrict deformations allowed by the represented environment 47 , so the bite points were appropriately constrained in four degrees of freedom (U 1 5 U 2 5 U R2 5 U R3 5 0), and the dentary condyle in four degrees of freedom (U 1 5 U 2 5 U 3 5 U R3 5 0). (NB U 1 is the mesiodistal axis, U 2 is the dorsoventral axis and U 3 is the axis along the length of the jaw; U refers to translational movement, U R refers to rotational movement.) As the true muscle loadings for Morganucodon and Kuehneotherium are not known, we used a comparative approach, assuming an equal muscle load applied to each taxon 48,49 . Given the equivalence of surface area between the two models, this was appropriate 33 . For the initial models, the relative contribution of each muscle to overall bite force was based on muscle ratios assigned to Morganucodon in ref. 35. The authors of this reference assign unit values of muscle forces to the jaw of Morganucodon: anterior temporalis, 10; posterior temporalis, 8; superficial masseter, 8; deep masseter, 8. The values are arbitrary but reflect the relative proportions each RESEARCH LETTER muscle contributes to bite force production. The actual loading forces we used were as follows: anterior temporalis, 2 N; posterior temporalis, 1.6 N; superficial masseter, 1.6 N; deep masseter, 1.6 N. We call this the standard loading regime. Contracting together at 100% activation, these muscles generate a bite reaction force of 2 N at the mid-molar of the Morganucodon finite element model, sufficient to pierce insect cuticle (see below). The calculations in ref. 35 are based on a unilateral bite, but with adductors active on both sides, so allowance is made for the balancing side. Both Morganucodon and Kuehneotherium have a mobile symphysis 12,30 and, as this current study is a comparative one of single lower jaws, it does not make assumptions about the forces on the balancing side, and the loads above are applied to the individual mandibular rami.
The reaction forces were queried in Abaqus at the bite points and condyle and the maximum von Mises stress patterns recorded. Von Mises stress is calculated as it indicates regional deformation as a function of the three principal stresses s 1 , s 2 and s 3 (ref. 50). Maximum principal strain values were also recorded (Table 1). Muscle loadings were then manipulated to obtain a bite reaction force sufficient to pierce appropriately sized beetle carapace. Myotis bats are a similar size to Morganucodon and Kuehneotherium (skull length approximately 14 mm), and beetles in their stomachs range in length from 1 to 10 mm, with a 10 mm beetle requiring 2-3 N of force to pierce the insect 27 . This is corroborated by Myotis velifer bats recorded as having a bite force of 2.2 N (ref. 26). The initial muscle loads in Morganucodon produced a bite reaction force of 2 N at the molar bite, but in Kuehneotherium it was necessary to increase the muscle loadings by 1.75 times to give a 2 N reaction force at the constrained molar tooth. Microtextural analysis of tooth microwear. Recent work has shown that quantitative microtextural analysis of tooth wear is a powerful tool for dietary discrimination and investigation of trophic resource exploitation in a range of extant and fossil vertebrates [51][52][53][54][55][56] . Our analysis compared the values for ISO 25178-2 areal texture parameters 57 for worn tooth surfaces in Morganucodon and Kuehneotherium with the results of analysis of the relationship between texture and diet in extant insectivorous bats 29 . Three-dimensional microtextural analysis is entirely independent of our other functional and biomechanical analyses and thus provides effective validation of our results.
Material used for microwear analysis is listed in Extended Data Table 1. Fossil specimens were prepared at the School of Earth Sciences, University of Bristol, by immersing dried blocks of matrix in hot water, with the addition of dilute hydrogen peroxide only if required. The exception was two Pontalun 3 Kuehneotherium molars, prepared at University College London in the 1970s, with added sodium hexametaphosphate (Calgon) to aid dissolution of the matrix. No acetic acid was used on any specimens. Bat specimens were all wild-found, acquired from UK sources (Extended Data Table 1) and assumed to be natural deaths. Specimens were fixed in either ethanol (University of Bristol and North Lancashire Bat Group) or 10% formalin solution (Veterinary Laboratory Agencies). Taxa were selected to include insectivores with well-constrained differences in their diets 58,59 . See Extended Data Table 2 and ref. 29 for details.
Our methods for capture of three-dimensional microwear data follow those developed in ref. 60 (for full details see ref. 29). Three-dimensional surface data were captured from tooth wear facets (distal protoconid facet of m 2 for bats, distobuccal wear facet of the main cusp of m 2 for Morganucodon and a mid-row molar for Kuehneotherium) using an Alicona Infinite Focus microscope G4b (IFM; software version 2.1.2, field of view 145 mm 3 110 mm, lateral optical resolution 0.35-0.4 mm, vertical resolution 20 mm; lateral resolution factor for the IFM set at 0.3) (see ref. 29). All three-dimensional data were edited to delete dirt and dust particles from the surface (using Alicona IFM software) and exported as .sur files. All subsequent processing of data used SurfStand (version 5.0.0). Data were levelled, and a fifth-order robust polynomial and a robust Gaussian wavelength filter (l c 5 0.025 mm) applied to remove gross surface form and long-wavelength features of the tooth surface. This generated a scale-limited roughness surface (Fig. 2) from which we derived ISO 25178-2 standard roughness parameters 57 . Sample sizes used in this study are relatively small; however, as demonstrated in ref. 29, this does not prevent detection of dietary signals through microtextural analysis. Data were explored using analysis of variance, correlations and principal components (on correlations; PCA). All statistical analysis of microtextural data used JMP 9. The results of Shapiro-Wilk tests indicated that some roughness parameters were non-normally distributed (P . 0.05), but for almost all parameters we were unable to reject the null hypothesis of normality for logtransformed data, so log-transformed data were used for analysis. Where homogeneity of variance tests revealed evidence of unequal variances, Welch analysis of variance was used.