Pterosaur dietary hypotheses: a review of ideas and approaches

ABSTRACT Pterosaurs are an extinct group of Mesozoic flying reptiles, whose fossil record extends from approximately 210 to 66 million years ago. They were integral components of continental and marginal marine ecosystems, yet their diets remain poorly constrained. Numerous dietary hypotheses have been proposed for different pterosaur groups, including insectivory, piscivory, carnivory, durophagy, herbivory/frugivory, filter‐feeding and generalism. These hypotheses, and subsequent interpretations of pterosaur diet, are supported by qualitative (content fossils, associations, ichnology, comparative anatomy) and/or quantitative (functional morphology, stable isotope analysis) evidence. Pterosaur dietary interpretations are scattered throughout the literature with little attention paid to the supporting evidence. Reaching a robustly supported consensus on pterosaur diets is important for understanding their dietary evolution, and their roles in Mesozoic ecosystems. A comprehensive examination of the pterosaur literature identified 314 dietary interpretations (dietary statement plus supporting evidence) from 126 published studies. Multiple alternative diets have been hypothesised for most principal taxonomic pterosaur groups. Some groups exhibit a high degree of consensus, supported by multiple lines of evidence, while others exhibit less consensus. Qualitative evidence supports 87.3% of dietary interpretations, with comparative anatomy most common (62.1% of total). More speciose groups of pterosaur tend to have a greater range of hypothesised diets. Consideration of dietary interpretations within alternative phylogenetic contexts reveals high levels of consensus between equivalent monofenestratan groups, and lower levels of consensus between equivalent non‐monofenestratan groups. Evaluating the possible non‐biological controls on apparent patterns of dietary diversity reveals that numbers of dietary interpretations through time exhibit no correlation with patterns of publication (number of peer‐reviewed publications through time). 73.8% of dietary interpretations were published in the 21st century. Overall, consensus interpretations of pterosaur diets are better accounted for by non‐biological signals, such as the impact of the respective quality of the fossil record of different pterosaur groups on research levels. That many interpretations are based on qualitative, often untestable lines of evidence adds significant noise to the data. More experiment‐led pterosaur dietary research, with greater consideration of pterosaurs as organisms with independent evolutionary histories, will lead to more robust conclusions drawn from repeatable results. This will allow greater understanding of pterosaur dietary diversity, disparity and evolution and facilitate reconstructions of Mesozoic ecosystems.

Qualitative evidence also includes associations of potential food items with pterosaur fossils. Of these, specimens with remains of other organism(s) in their gut and throat (content fossils) are interpreted as direct evidence of diet (Wild, 1984;Hone et al., 2015a;Witton, 2018). Evidence from coprolites (fossilised faeces) falls into a similar category, but determining producers of coprolites can be challenging. Other studies infer diet from general associations with taxa from the same stratigraphic interval and/or depositional environments as the preserved pterosaur specimens (Kellner, 2003;Chatterjee & Templin, 2004;Tütken & Hone, 2010). Finally, ichnofossils are used to infer foraging behaviours and habitat preferences, leading to dietary interpretations (Lockley & Wright, 2003;Mazin et al., 2003;Fiorillo et al., 2015).
During the last 30 years, dietary analysis of extinct animals has become increasingly quantitative, allowing explicit hypothesis testing and robust, repeatable conclusions (Lauder, 1995;Ferry-Graham, Bolnick & Wainwright, 2002;Veldmeijer et al., 2007;Anderson et al., 2011). Quantitative functional morphology methods do not provide direct evidence of actual food items, but of feeding and/or foraging behaviours that can be used to infer more or less plausible dietary hypotheses (Amiot et al., 2010;Bright, 2014). Finite element analysis (FEA) applies external forces to digital anatomical reconstructions and calculates resultant stress and strain distributions (Rayfield et al., 2001;Fastnacht, 2005;Anderson et al., 2011). Skull reconstructions of the pterosaur Pteranodon, for example, show that it experienced relatively low stresses and strains during jaw closure, indicating that it could effectively seize prey by rapid jaw closure (Fastnacht, 2005).
Stable isotope analyses of carbon ( 13 C/ 12 C) and oxygen ( 18 O/ 16 O) ratios from bone apatite and tooth enamel can reveal whether animals inhabited terrestrial or marine environments, allowing some limited dietary inferences to be drawn (Tütken & Hone, 2010).
We present the first comprehensive synthesis of pterosaur dietary interpretations. For each interpretation, we identify dietary statements and underpinning evidence categories and evaluate how robustly supported different dietary interpretations are within, and between, key pterosaur groups. This provides the basis for a discussion of biological and nonbiological influences on apparent pterosaur dietary diversity.
Phylogenetic analyses of Pterosauria (Kellner, 2003;Unwin, 2003;Lü et al., 2010;Andres et al., 2014), show agreement in identifying approximately 20 principal groups, each sharing consistent features of the jaws, dentition (where present), and cranial and postcranial anatomy (Figs 1 and 2). Some of these groups are clades, whilst others are grades of morphologically similar taxa. These groups form the taxonomic units used herein.
The taxonomic contents of these groups exhibit similarities between published phylogenies, but they are not universally agreed upon. Three distinct phylogenies are therefore used to examine the impact of phylogenetic uncertainty when considering dietary hypotheses. These phylogenies are referred to as 'Unwin', 'Kellner' and 'Andres' after the authors of the initial iterations of these data sets. Herein, the Unwin phylogeny represents a composite tree of Unwin (2003) and Lü et al. (2010) (Fig. 2). The Kellner phylogeny is a composite tree of Kellner (2003), Wang et al. (2009), Wang et al. (2012 and Rodrigues et al. (2015) (see online Appendix S1, Fig. S1). The Andres phylogeny is from Andres et al. (2014;see online Fig. S2). Where sets of phylogenies used to construct a tree conflict, relevant taxa are collapsed into a polytomy. Whilst the preferential selection of any one of these phylogenies is to some extent arbitrary, the Unwin phylogeny is primarily used herein because it exhibits the highest stratigraphic congruence (Andres, 2015). Labelled principal groups in the Unwin phylogeny ( Fig. 2) thus denote the order of Section IV. Further information on phylogenies is included as online Supporting Information (see online Appendix S1).

(a) Pterosaur diets
The primary data set for this study is a compilation of interpretations, each comprising a statement regarding the diet of a taxon that either is one of the recognised principal taxonomic groups, or falls within one of those groups. Each interpretation was unambiguously assigned to one dietary category and to one evidence category. Dietary statements without supporting evidence were excluded. Dietary interpretations were compiled from the literature. Publications citing previous interpretations but lacking novel data or reasoning were excluded. One hundred and twenty six publications contained at least one novel dietary interpretation. When a publication identified the same dietary and evidence categories for multiple taxa within the same principal group, this was treated as a single novel interpretation. When a publication provided a single dietary statement for a single group but supported it with more than one evidential category, this was tabulated under each evidence category.
Differences in the taxonomic content of principal groups between the phylogenies led to differences in the numbers of identified interpretations. For example, a publication assessing diet for two species from the same principal group in one phylogeny was treated as one interpretation. If these two species then fell within different groups in a different phylogeny, this interpretation was counted twice, once for each group. As a result, 314 interpretations were identified for the Unwin phylogeny, 311 for the Kellner phylogeny and 301 for the Andres phylogeny.
Details of interpretations for each phylogeny are included as online Supporting Information (see online Appendix S1).

(b) Species listing
Valid pterosaur species were derived from Dean et al. (2016) and a literature review, ending in February 2017. Eight species could not be referred to any of the principal groups and were thus excluded (see online Appendix S1). The final data set comprises 180 species (see online Appendix S1).

(c) Pterosaur publications from 1784 to 2017
A data set of all pterosaur-focused publications was compiled to estimate research effort through time, from 1784 to February 2017. The sum of all publications (1828) is likely a slight underestimate.

(4) Analyses
Numbers of interpretations for each dietary and evidence category were summed for each principal group. Total percentages of each evidence category were calculated. To investigate possible biological and non-biological drivers of the diversity of dietary interpretations, numbers of species from each principal group were compared to numbers of assigned dietary categories. More speciose groups are hypothesised to have greater dietary diversity because they exhibit more morphological variation . Ornithocheiridae was by far the most speciose group with 36 species and was removed as an outlier (the second was Ctenochasmatidae with 18 species). The number of dietary groups was non-normally distributed (Shapiro-Wilk, P < 0.05), and therefore a Spearman's rank correlation was used when comparing dietary groups and species counts.
Pterosaur publications were sorted into decadal bins from 1780-1789 to 2010-2017 and compared with dietary interpretations, sorted into the same time-bins. Publications and dietary interpretations were transformed using generalised differencing to reduce the likelihood of type I errors when performing correlations on time-series data. Correlations were then tested using Spearman's rank tests. Cumulative data on publications and dietary interpretations were plotted over the same timespan.
To assess changes in dietary consensus through time, dietary interpretations for each principal group were assigned to one of four different stages of pterosaur research history (Witton, 2013)

(1) Analysis of pterosaur dietary interpretations
Total numbers of dietary interpretations for each principal group (Fig. 3) show large differences, from 39 in Azhdarchidae to just two in basal dsungaripteroids (although the latter comprises just one genus). Marked disparity also occurs in numbers of hypothesised diets. For Nyctosauridae and basal dsungaripteroids, for example, only a single diet has been proposed, whereas for Campylognathoididae, basal ctenochasmatoids and Azhdarchidae there are six different diets (Fig. 3). For most groups, a single diet is supported by multiple evidential categories, with one or more other diets suggested by one or a few interpretations. For some groups, different diets are hypothesised for the same taxa within these groups, and in others different diets are associated with particular taxa (see online Appendix S1). Some groups, such as Istiodactylidae and Azhdarchidae, exhibit roughly even splits between the two most common diets, while others, such as basal ctenochasmatoids, exhibit small disparities between most-supported and lesser-supported diets (Fig. 3).

(2) Analysis of categories of evidence
There is a large disparity in the numbers of interpretations supported by qualitative (content fossils, associations, ichnofossils and comparative anatomy) and quantitative (functional morphology and isotope analyses) approaches, with the former supporting 87.3% of all interpretations (Fig. 4). Comparative anatomy is the most common evidence category (62.1% of total), with general associations second (15%) (Fig. 4). Interpretations based on analysis of functional morphology make up 12.1% of interpretations; content fossils and ichnofossils each make up 5.1%; stable isotope analyses make up 0.6% (Fig. 4).

(4) Dietary consensus and phylogenies
The phylogenetic distributions of dietary interpretations for particular groups are broadly consistent among the three phylogenies (Fig. 3, see online Figs S3 and S4). For several groups one dietary interpretation is strongly supported (e.g. Anurognathidae, Pteranodontidae), with little support for others. Several groups exhibit some dietary disparity, although one interpretation remains dominant (e.g. Dsungaripteridae, Lonchodectidae). A few groups exhibit numerous interpretations, all of which receive some support (e.g. Azhdarchidae). High congruence in terms of group definitions and content between the Unwin and Kellner trees result in similar distributions of dietary interpretations, especially for monofenestratans (Fig. 3, see online Fig. S3). There is also high congruence for Anurognathidae and Campylognathoididae, but lower congruence for 'basal-most pterosaurs' (=Group A in the Kellner phylogeny) and Rhamphorhynchinae (=Rhamphorhynchidae in the Kellner phylogeny).
The distribution of interpretations for monofenestratan groups in the Andres phylogeny show high congruence with the Unwin and Kellner phylogenies (see online Fig. S4). However, 'basal-most pterosaurs' and Rhamphorhynchidae in the Andres phylogeny contain substantially more interpretations than their equivalents in the other phylogenies. indicates that a linear model provides a poor fit to the data (r 2 = 0.19, excluding Ornithocheiridae). Overall, more dietary hypotheses have been proposed for more speciose groups. Species assortments are included as online Supporting Information (see online Appendix S1).
exceed new numbers of dietary interpretations, except in 1991 (Fig. 6B). The earliest identified reports of hypothesised diets and evidential categories are listed in Table 1.
Istiodactylidae and Azhdarchidae by contrast are evenly affiliated with piscivory and carnivory through time ( Fig. 7B-D).

IV. PTEROSAUR DIETS (1) Non-pterosaur pterosauromorphs
Determining the plesiomorphic dietary condition for pterosaurs is problematic, because different phylogenies recover different pterosaurs at the base of Pterosauria. Lü et al. (2010) for example, considers Preondactylus buffarinii the basal-most pterosaur, while Kellner (2003) and Bennett (2007) consider Anurognathidae the basal group. This makes it difficult to infer which pterosaurs exhibit the plesiomorphic dietary condition. Understanding the diets of non-pterosaur pterosauromorphs may provide a solution. At present, Scleromochlus, which pre-dates the earliest known pterosaurs by around 15-20 million years (Dalla Vecchia, 2003;Nesbitt, 2011), is the only putative non-pterosaur pterosauromorph, but its phylogenetic position is controversial. Euparkeria capensis, from the early Middle Triassic of South Africa, represents a close outgroup of archosaurs and may approach the plesiomorphic archosaur body plan (Senter, 2003;Nesbitt, 2011;Sookias & Butler, 2013;Sookias, 2016), and has been used as an outgroup in several phylogenetic analyses of pterosaurs (Bennett, 1996;Andres et al., 2014). It could be argued that understanding diets of non-archosaur archosauriforms such as Euparkeria may help elucidate ancestral pterosaur diets (Sookias, 2016). However, Scleromochlus and Euparkeria are anatomically different in many respects from pterosaurs (Nesbitt, 2011), thus limiting their utility in inferring the plesiomorphic Pterosauromorpha and Pterosauria dietary conditions.
Scleromochlus has been interpreted as an agile insectivore based on its elongate hind limbs (Benton, 1999). This is consistent with a hypothesis of insectivory as the ancestral pterosaur diet (Padian, 1980), although this was an extrapolation from the observation that insectivory is the most common diet in modern bats.
Euparkeria possess large orbits and serrated, laterally flattened teeth which have been interpreted as indicative of carnivory (Ewer, 1965;Senter, 2003). Euparkeria has been suggested as insectivorous during its early life-cycle stages before becoming carnivorous as an adult (Senter, 2003). Sookias & Butler (2013), however, argued that Euparkeria fed on small insects and/or small tetrapods due to a lack of well-developed jaw musculature for orally processing food items.
Insectivory interpretations are based on comparative anatomy and functional morphology and associations (Fig. 3). The fang-like, widely spaced dentitions of basal-most pterosaurs have been suggested as suitable for catching insects. Buckland (1829) even speculated that the forelimbs of Dimorphodon macronyx resembled those of modern insectivorous bats. Numerous insect-like fragments are known from the same deposits as Dimorphodon in the Blue Lias Formation, UK (Buckland, 1829).  Dimorphodon and Preondactylus buffarinii, based on their quadrate-articular jaw joints. Relatively small areas for muscle attachments are consistent with low bite forces and rapid jaw closure for catching insects (Ősi, 2009, 2011).
Piscivory interpretations are based on comparative anatomy and an absence of ichnology (Fig. 3). The absence of identifiable ichnofossils has been used to suggest that these pterosaurs fished over water (Unwin, 2007). The deep, rounded snout of Dimorphodon has been compared with the rostrum of the piscivorous Atlantic puffin (Fratercula arctica) (Bakker, 1986). However, optimal Dimorphodon floating positions investigated using three-dimensional (3D) digital reconstructions from 'mathematical slices' of specimen illustrations do not support this hypothesis (Henderson, 2010;Hone & Henderson, 2014). This analysis suggests that Dimorphodon would have had most of its body submerged and therefore did not spend much time foraging in water (Hone & Henderson, 2014).
Carnivory interpretations are based on comparative anatomy and functional morphology (Fig. 3). Rapid jaw closure, suggested by adductor muscle reconstructions, may have helped in the capture of small vertebrates (Ősi, 2011;Dalla Vecchia, 2013). Morphological reassessments of Dimorphodon forelimbs and pectoral girdle suggest an erect posture which would have facilitated terrestrial foraging for small vertebrates (Witton, 2015b).
Insectivory interpretations are based on comparative anatomy and associations (Fig. 3). Carniadactylus rosenfeldi exhibits few wear facets on its teeth which suggests a preference for soft invertebrates (Ősi, 2011).
Generalism interpretations are based on comparative anatomy and functional morphology (Fig. 3). The diversity of campylognathoidid dentitions are argued to have allowed exploitation of numerous dietary items (Padian, 2008a;Witton, 2013). Functional morphological analysis subjecting 3D reconstructions (see Section IV.2) of the Eudimorphodon skull and dentition to dorso-ventrally directed forces reveal high bite forces for its skull length (Henderson, 2018). This is consistent with processing diverse food items (Henderson, 2018).
Insectivory and carnivory interpretations are based on comparative anatomy (Fig. 3). Histological thin sections of Rhamphorhynchus bones from different-sized individuals reveal that Rhamphorhynchus hatchlings had slow growth rates and were potentially unable to fly (Prondvai et al., 2012). Young Rhamphorhynchus were suggested to have fed on insects and small vertebrates before learning to fly, although the possibility of post-hatching feeding by parents cannot be ruled out (Prondvai et al., 2012).
Insectivory interpretations are based on associations and comparative anatomy (Fig. 3). Lü & Bo (2011) hypothesised Jianchangopterus zhaoianus from the Tiaojishan Formation, China, as an obligate insectivore as a result of strict niche partitioning with unrelated, potentially piscivorous, pterosaurs.
Carnivory interpretations are based on associations and comparative anatomy (Fig. 3). Initial descriptions of Darwinopterus modularis noted spike-like teeth for gripping vertebrate prey, such as gliding mammals and other pterosaurs, which are known from Tiaojishan (Lü et al., 2010).
Insectivory interpretations are based on comparative anatomy of a Liaoxipterus specimen with an elongated hyoid apparatus (collection of small bones which attach to the posterior of the tongue). The proportional lengths of these bones are reportedly similar to modern chameleons (Chameleonidae), and Liaoxipterus was thus reasoned to have caught insects with the aid of a projectile tongue (Lü, Xu & Ji, 2008b;Lü, 2015).
Piscivory interpretations are based on content fossils, associations, comparative anatomy and functional morphology (Fig. 3). Content fossils include fish remains within several Pteranodon stomachs (Brown, 1943;Hargrave, 2007). Pteranodontids are exclusively known from marine deposits and have thus been suggested as albatross analogues (Marsh, 1876;Eaton, 1910;Wellnhofer, 1991;Bennett, 1993Bennett, , 1994Unwin, 2006;Witton, 2013). Comparative anatomical analyses agree on piscivory (Fig. 3). Skim-feeding was initially proposed as Pteranodon and modern avian skimmers (Rynchops spp.) possess similarly narrow jaws (Marsh, 1876;Eaton, 1910;Zusi, 1962). Spiral-shaped joints between the quadrate and articular in the lower jaw were interpreted as attachment points for throat sacs for scooping up fish (Eaton, 1910;Hankin, 1912). However, cervical vertebrae of Pteranodon were later judged too small for scooping and skimming for fish (Witton, 2013). Functional morphological analyses have tested the feasibility of feeding behaviours. Bramwell & Whitfield (1974) placed scaled replicas of the Pteranodon skeleton in wind-tunnels and inferred slow flight and gliding speeds to help catch fish whilst on the wing. By contrast, other early flight models found Pteranodon flight profiles were more suited for skim-feeding, although their body masses were extrapolated from modern seabirds (Hazlehurst & Rayner, 1992). Later energy expenditure modelling found that capturing aquatic organisms on the wing or whilst resting on water surfaces were energetically feasible behaviours (Habib, 2015).
Piscivory interpretations are based on comparative anatomy and associations (Fig. 3). Pterodactylus dentitions have alternatively been suggested to be adapted for snatching small fishes from water surfaces (Rayner, 1989;Bennett, 2012).
Durophagy interpretations are based on comparative anatomy (Fig. 3). The Cycnorhamphus jaw and dentition has been interpreted as an adaptation for crushing crustacean exoskeletons (Witton, 2013).
Filter-feeding interpretations are based on comparative anatomy (Fig. 3). The unspecialised teeth of basal ctenochasmatoids have been suggested to have facilitated basic filter-feeding in shallow water bodies (Unwin, 2006).
Generalism interpretations are based on functional morphology (Fig. 3). A 3D model of the skull subjected to dorso-anteriorly directed forces found that Pterodactylus had intermediate bite forces and thus perhaps consumed diverse food items (Henderson, 2018).
Generalism interpretations are based on comparative anatomy and associations (Fig. 3). Lonchodectes is known from the shallow marine Cambridge Greensand Formation of the UK, which is suggested to have contained numerous aquatic organisms which could have supported generalist diets (Unwin, 1996(Unwin, , 2006. The dorso-ventrally flattened teeth of Lonchodectes were reasoned to have facilitated the handling and consumption of varied food items (Unwin, 1996(Unwin, , 2006. Piscivory interpretations are based on comparative anatomy and an absence of terrestrial ichnofossils (Fig. 3). Slender, recurved teeth in the anterior halves of their jaws (Fig. 1M) have been argued to be suitable for catching fish (Wang & Zhou, 2006;Lü et al., 2016b).
Durophagy interpretations are based on comparative anatomy, ichnofossils, associations and functional morphology (Fig. 3). Contemporaneous ichnofossils suggest that dsungaripterids foraged for hard-shelled organisms in shallow water bodies (Unwin, 2007). Associations indicate that dsungaripterids are mostly found in inland fluviolacustrine deposits (Matsukawa et al., 2006). Comparative anatomical evidence derives from jaws and teeth. Dsungaripterus weii, for example, exhibits upturned jaw tips with low-crowned, anvil-shaped teeth in the posterior half of its jaws (Young, 1964;Wellnhofer, 1991), suggesting Dsungaripterus might have picked out bivalves, gastropods and crabs with its jaw tips before cracking open their shells or exoskeletons with its teeth (Young, 1964;Wellnhofer, 1991;Lü et al., 2009). Functional morphology of 3D models of Dsungaripterus reveals awkward floating positions (see Section IV.2), indicative of a pterosaur better adapted for foraging in shallow waters (Hone & Henderson, 2014).
Durophagy interpretations are based on associations, comparative anatomy and functional morphology (Fig. 3). The fossil records of tapejarids, seeds and gymnosperm cones in the Lower Cretaceous partially correlate (Pinheiro, Liparini & Schultz, 2014). Tapejara adductor muscle reconstructions show well-developed systems for potentially consuming harder items (Pinheiro et al., 2014). This is corroborated by 3D constructs of the Tapejara skull which exhibited high bite forces for potentially cracking open seeds (Henderson, 2018).
Carnivory interpretations are based on comparative anatomy and content fossils (Fig. 3). A fossilised regurgitate pellet, tentatively identified as tapejarid in origin, contains small bird remains (Veldmeijer et al., 2012). Comparative anatomy includes a partial dentary from the La Huerguina Formation, Spain, which may have had a 'cutting-edge' rhamphotheca for catching small vertebrates (Vullo et al., 2012).
Piscivory interpretations are based on associations (Fig. 3). Brazilian tapejarids are known from lagoonal and marine deposits and are reasoned to have fed on contemporaneous fish (Unwin & Martill, 2007).
Insectivory interpretations are based on comparative anatomy and associations (Fig. 3). Nemicolopterus crypicus, from the Jiufotang Formation, China was suggested to have pursued insects along tree branches aided by curved phalanges .
Piscivory interpretations are based on associations and comparative anatomy (Fig. 3). Chaoyangopterus zhangi is known from lacustrine deposits with numerous fishes (Wang & Zhou, 2006). Comparative anatomical evidence is based on the scissor-like edentulous jaws of these pterosaurs helping to catch fish when on the wing (Wang & Zhou, 2006). Generalism interpretations are based on comparative anatomy (Fig. 3). Chaoyangopterids exhibit similar limb proportions to related groups such as azhdarchids that have been proposed as ground-based foragers (see Section IV.16) (Witton, 2013).
Piscivory interpretations are based on comparative anatomy and functional morphology (Fig. 3). Thalassodromeus sethi lower jaws were proposed to be similar to skimming Rynchops (Kellner & Campos, 2002a,b). Lower jaw replicas of Thalassodromeus and Tupuxuara cristata placed in flume tanks however, experienced levels of drag that would not have allowed stable skimming (Humphries et al., 2007).
Carnivory interpretations are based on comparative anatomy, associations, ichnology and functional morphology (Fig. 3). Comparative anatomical interpretations include long necks and jaws (Fig. 1S) for carcass probing (Lawson, 1975;Wilkinson & Ruxton, 2012) and/or aerial predation (Nessov, 1984;Chatterjee & Templin, 2004). Their necks, however, have been reinterpreted as too stiff for these roles, and azhdarchids have alternatively been suggested as ground-based predators and scavengers based on their long limbs (Witton & Naish, 2008Witton, 2013;Naish & Witton, 2017). Where multiple azhdarchids are known, niche partitioning is suggested (Witton & Naish, 2008;Naish et al., 2015). The Haţeg Basin, Romania, for example contains Hatzegopteryx thambema with a 10 m wingspan , Eurazhdarcho langendorfensis with a 3 m wingspan, and an unnamed azhdarchid with a 3 m wingspan and short, wide cervical vertebrae . Hatzegopteryx was reasoned to have consumed the largest prey, with the short-necked azhdarchid consuming larger prey than Eurazhdarcho as its neck potentially offered greater mechanical advantages .
Piscivory interpretations are based on comparative anatomy, functional morphology and ichnofossils (Fig. 3). Possible azhdarchid track-ways on mudflats suggest wading behaviours (Unwin, 2007). However azhdarchid feet have been argued to have been too small to have provided support on sandy, muddy ground (Witton & Naish, 2008). Comparative anatomy of azhdarchid necks suggests some articulation at their shoulders which may have allowed their heads rudimentarily to bend forward and seize fish on the wing (Martill et al., 1998;Chatterjee & Templin, 2004;Averianov, 2013). Functional morphology indicates azhdarchid gapes similar to Rynchops, based on jaw bone articulations, suggesting skim-feeding (Kellner & Langston, 1996;Ősi, 2004). However this was not supported by flume tank results indicating that skim-feeding was energetically unfeasible (Humphries et al., 2007). Theoretical reconstructions of possible azhdarchid throat pouches suggest that scooping would have put incredibly high strain on their necks .
Generalism interpretations are based on comparative anatomy and associations (Fig. 3). Numerous azhdarchid remains from terrestrial deposits led to the hypothesis that azhdarchids were opportunistic ground-based foragers, with larger genera consuming more animals for higher energy returns (Witton & Naish, 2008Witton, 2013).
Insectivory interpretations are based on ichnology (Fig. 3). Azhdarchid tracks on mudflats from the Unhangari Formation, South Korea, were interpreted as indicative of foraging for insects (Hwang et al., 2002).

V. DISCUSSION
(1) Why is there a lack of consensus in pterosaur diet research?
Overall, there is limited consensus on diets for most pterosaur groups. Most dietary interpretations are supported by qualitative evidence, most commonly from comparative anatomy. There is strong consensus on diets for some pterosaur groups, such as insectivory in Anurognathidae and piscivory in Ornithocheiridae and Pteranodontidae, supported by one or several evidential categories. For other groups there is little consensus: in both basal ctenochasmatoids and Azhdarchidae, for example, six distinct diets have been suggested. In general, pterosaur principal groups with more species exhibit a higher diversity of dietary hypotheses. Choice of phylogeny has some impact on consensus, with the largest differences observed between phylogenies for non-monofenestratan groups. Numbers of dietary interpretations exhibit small increases for much of pterosaur research history, with most interpretations proposed in the 21st century. Changes in the number of dietary interpretations through time do not correlate with numbers of pterosaur publications.
That the majority of dietary interpretations are qualitative is unsurprising because such interpretations can be readily proposed without rigorous experiments or analyses. Comparative anatomy uses extrapolations from observational, and occasionally experimental, studies of extant organisms (Aerts, 1990;Kellner & Campos, 2002b;Witton & Naish, 2008). Associations similarly rely on straightforward understanding of modern food webs when reconstructing ecological relationships between contemporaneous taxa, and of geological processes when interpreting depositional environments of specimens (Wellnhofer, 1991;Frey & Tischlinger, 2012). Content fossils are often interpreted as direct records of trophic interactions between extinct organisms (Hone & Faulkes, 2014;Hone et al., 2015a;Witton, 2018), although their rarity limits their utility, and there is an inbuilt bias towards preservation of food items consumed immediately prior to death, which may be atypical (Davis & Pineda-Munoz, 2016). Furthermore, the usefulness of gut contents for indicating typical diet in animals that consume a range of items is limited when small sample sizes are all that is available. Analyses based on extant predators indicate that large numbers of individuals, from tens to even hundreds, need to be sampled to capture a true picture of diet (e.g. Szczepanski & Bengtson, 2014). Ichnofossils are more common, although the poor quality and uncertainty regarding the makers of many specimens limits robust ecological interpretations (Lockley & Wright, 2003;Lockley, Harris & Mitchell, 2008).
Many quantitative methods were not developed and readily utilised until the late 20th and early 21st centuries (Aerts, 1990;Lauder, 1995;Rayfield et al., 2001), and are more difficult to use because: (i) they often require specialist technology and/or equipment; (ii) they require thorough understanding of experimental design and hypothesis testing; (iii) they can require destructive sampling of specimens (Aerts, 1990;Anderson et al., 2011). Robust quantitative methods have therefore been unavailable for much of pterosaur research history.
There are several reasons why different groups of pterosaurs exhibit different levels of consensus regarding dietary interpretations. The observed correlation between species and numbers of dietary interpretations per principal group provides one possible explanation. Pterosaur groups that exploited more new food sources in their respective palaeoenvironments potentially underwent greater eco-morphological changes which facilitated further exploitation, likely resulting in more speciation events . Zhou et al. (2017) explored this idea by morphologically quantifying pterosaur skulls, jaws and dentitions with respect to assigned diets from the literature. The regions of morphospace occupied by pterosaurs were indeed better explained by their respective assigned diet than by evolutionary relatedness . Dietary assignments by Zhou et al. (2017), however, were primarily based on qualitative evidence and/or untested hypotheses from the literature and should therefore be interpreted with caution. The disparity in levels of consensus regarding diet across Pterosauria as shown in the current study illustrates the difficulty in reliably uncovering biological signals that may explain pterosaur dietary diversity.
Furthermore, pterosaur diets are unlikely to have fallen into discrete categories as described herein and in Zhou et al. (2017). Many interpretations of pterosaur feeding behaviours and diets from morphological evidence concern only what they appear to be optimally adapted to eat. This is because pterosaur functional attributes are mostly inferred from comparisons of skeletal features with analogous structures in modern taxa, under the assumption that similarity in morphology correlates with similarity in function (Lauder, 1995). However, other pterosaur tissues (muscles, nervous system, etc.) are rarely preserved and pterosaur structures may have had other functions related or unrelated to feeding, and flexibility in feeding behaviours that cannot be inferred from hard tissues alone (Lauder, 1995). This compromises the assumption that pterosaurs were adapted to consume only one type of food. Furthermore, pterosaurs might have occasionally consumed food items outside of their normal dietary ranges (see online Appendix S1). A modern example involves fruits and seeds found in the stomachs of several crocodilian species (Platt et al., 2013a). Crocodilians are renowned predators, but fruits and seeds have been found in high enough quantities to rule out accidental consumption [see Platt et al. (2013a) and references therein]. Although incorporating potential dietary plasticity would be very useful for palaeoecosystem reconstructions, we currently do not have the techniques to do so.
The lack of consensus in pterosaur dietary interpretations is more likely explained by non-biological and historical signals. The patchy quality of the pterosaur fossil record Dean et al., 2016) is one non-biological example as some dietary interpretations are made on few and/or poorly preserved specimens, resulting in low confidence levels. Buckland (1829), for example, hypothesised insectivory in Dimorphodon based on the limited post-cranial material known at the time. Only with the description of the skull decades later was piscivory hypothesised (Seeley, 1901). Lonchodectids are interpreted as piscivores and generalists, but are mostly known from fragmentary remains (Unwin, 1996;Lü et al., 2016b). Basal dsungaripteroids are universally interpreted as durophagous but few dietary interpretations have been proposed, at least in part because their fossil record is poor (Bennett, 2006;Barrett et al., 2008). Pterosaurs from Lagerstätten and other deposits with well-preserved remains generally have higher levels of dietary agreement. Pteranodontids are known from huge specimen numbers from the Niobrara Formation (Marsh, 1876;Eaton, 1910;Brown, 1943). Anuroganthids are known from few, but extremely well preserved, specimens from the Solnhofen Limestone and Tiaojishan and Yixian formations (Rjabinin, 1948;Bennett, 2003bBennett, , 2007. These finds have allowed thorough investigations into the functional morphology of these pterosaurs to test interpretations of piscivory and insectivory respectively (Bramwell & Whitfield, 1974;Hazlehurst & Rayner, 1992;Habib & Hall, 2012;Habib & Witton, 2013;Habib, 2015). This highlights the importance of finding more, and ideally well-preserved, specimens to generate and test dietary hypotheses.
Of all pterosaurs, azhdarchids have the greatest diversity of dietary interpretations, yet they are known from relatively fragmentary remains (Lawson, 1975;Cai & Wei, 1994;Vremir et al., 2015). Azhdarchids are the largest flying organisms ever to have lived, thus it is reasonable to suggest that these pterosaurs have received disproportionate levels of scientific study (e.g. Lawson, 1975;Wellnhofer, 1991;Cai & Wei, 1994;Martill, 1997;Martill et al., 1998;Hwang et al., 2002;Chatterjee & Templin, 2004;Witton & Naish, 2008Brown, 2015;Martill & Ibrahim, 2015;Vremir et al., 2015). Higher levels of study on specimens of limited number and quality increase the likelihood of tentative, yet varied, dietary interpretations. This further indicates how levels of dietary consensus are at least partially confounded by non-biological signals.
Disputes over evolutionary relationships also mask biological signals of pterosaur diets. Placements of species within different groups make it difficult to elucidate whether dietary ranges are representative for principal groups or an artefact of the chosen phylogeny. This is especially problematic for non-monofenestratan groups because they exhibit lower morphological disparity than monofenestratan pterosaurs (Prentice et al., 2011;Zhou et al., 2017). The Andres phylogeny is an example of this as it contains fewer, more inclusive non-monofenestratan groups than the other phylogenies. Better resolution of the membership of pterosaur principal groups would therefore assist with elucidating true dietary ranges for each group. Better specimens of known taxa would assist with more confident phylogenetic placements.
The absence of correlation between numbers of pterosaur publications and dietary interpretations through time is unsurprising. Through the 19th and early 20th centuries, pterosaur publications mainly comprised species descriptions and systematics from European and North American Lagerstätten such as Solnhofen and the Niobrara Chalk (Marsh, 1876;Seeley, 1901;Eaton, 1910;Arthaber, 1921). Very few of these studies hypothesised diets as during this time period ecology was a largely unrecognised area of scientific study. The slight decrease in publications during the middle 20th century saw few new specimen and species descriptions (Rjabinin, 1948). The increase in publications in the latter half of the 20th century coincided with discoveries of new Lagerstätten, including the Santana and Crato formations in Brazil (Wellnhofer & Kellner, 1991;Unwin & Martill, 2007), and the Yixian Formation in China (Ji, Ji & Padian, 1999), as well as a broader renaissance in reptile palaeobiology. New, morphologically distinct, pterosaur groups were discovered during this time, resulting in new hypothesised diets such as filter-feeding in ctenochasmatids (Bakker, 1986), herbivory/frugivory in tapejarids (Wellnhofer & Kellner, 1991) and generalism in lonchodectids (Unwin, 1996). The last quarter of the 20th century also saw palaeontological research expand from mostly description-and systematic-based studies to focus also on the wider biology and ecology of extinct taxa (Bramwell & Whitfield, 1974;Bramble, 1978;Bakker, 1986;Witmer, 1995). Many pterosaurs described during and after this time included some ecological interpretation, including possible diet, but many of these interpretations were constructed from simple qualitative comparisons and analogies with modern biology. The 1 year in which numbers of dietary interpretations exceeded that of pterosaur publications, 1991, can be explained by the publication of a comprehensive summary and interpretation of pterosaur research up to that time, including dietary interpretations for most pterosaur groups (Wellnhofer, 1991) (see online Appendix S1). Further pterosaur specimen discoveries, e.g. the Tiaojishan and Jiufotang Formations (Wang & Zhou, 2006;Lü et al., 2010;Lü & Bo, 2011), continued the increase in publications into the 21st century.
Increases in the number and rate of publication and dietary interpretations is also explained in part by the application of new techniques and new types of evidence. These include biomechanical analyses (Bramwell & Whitfield, 1974;Fastnacht, 2005;Hone & Henderson, 2014;Henderson, 2018), and stable isotope analyses (Amiot et al., 2010;Tütken & Hone, 2010). Ichnofossil evidence was not used to inform dietary interpretations until the 21st century due to debates over whether the creators of ichnofossils were pterosaurian or crocodilian (Lockley & Wright, 2003).
Greater appreciation for pterosaurs, and extinct taxa in general, as organisms with independent evolutionary histories which faced unique selection pressures in their respective palaeoenvironments, are helping with the construction of testable hypotheses (Witton & Habib, 2010;Fiorillo et al., 2015). Many pterosaur dietary interpretations in the 19th and 20th centuries were based on simple extrapolations from modern flying vertebrates or semi-aquatic reptiles (Seeley, 1901;Padian, 1980Padian, , 1983Wellnhofer, 1991;Unwin & Henderson, 2002;Hone, 2012). These ideas over time sometimes became established as received wisdom and were generally not subject to scrutiny or testing. Noted similarities between Dimorphodon and puffin rostra, as evidence of piscivory in the pterosaur, for example (see Section IV.2) (Seeley, 1901;Bakker, 1986), were overly simplified as the Dimorphodon rostrum is formed from bone whereas the Fratercula rostrum is formed from keratin and soft tissues (Ősi, 2011;Badikova & Dzerzhynsky, 2015). More sophisticated flight models indicate that Dimorphodon did not have the flight profile of an able fisher (Witton, 2008(Witton, , 2015a, which has begun to cast doubt on this long-untested hypothesis of diet for this pterosaur. Earlier flight models directly extrapolated pterosaur weights from modern birds with little account for the structural differences between avian and pterosaurian skeletons, with interpretations of pterosaur feeding ecology, and consequently diet, sometimes constructed from these models (Padian, 1980(Padian, , 1983Rayner, 1989;Hazlehurst & Rayner, 1992). Many later quantitative studies account for independent evolutionary histories and have produced strongly supported dietary hypotheses for some pterosaurs, including anurognathids (see Section IV.3) (Habib & Hall, 2012;Habib & Witton, 2013), and pteranodontids (see Section IV.9a) (Hone & Henderson, 2014;Habib, 2015). However, in other groups they have provided evidence for new diets, such as generalism in basal ctenochasmatoids (Henderson, 2018).
Current dietary diversities and evidential categories show that we currently lack reliable evidence to support dietary interpretations for many pterosaur groups, and this prevents us from teasing out the biological signals explaining pterosaur diet diversity and disparity. Greater use of robust, experiment-led approaches, in tandem with other lines of evidence, is therefore needed to construct plausible dietary hypotheses in pterosaurs with low levels of dietary consensus and to test interpretations in pterosaurs with higher levels of consensus. This will help uncover biological signals for understanding pterosaur dietary evolution and for reconstructing Mesozoic ecosystems.
(2) How appropriate are current methods for forming dietary hypotheses?
The range of methods employed in analysis of pterosaurs has proven highly successful in generating interpretations of diet. What is less clear, however, is how well they have performed in differentiating between alternative interpretations, and this partly explains why, for most pterosaur groups, multiple interpretations have been proposed. Greater use of analytical and experiment-led methodologies (Veldmeijer et al., 2007;Padian, 2008a), along with careful experimental design and full appreciation of the methods and study taxa (Unwin & Henderson, 2002;Hutchinson, 2012) has the potential to improve both the rigour and consistency of dietary interpretations. No single method will provide complete understanding; application of multiple independent techniques is likely to yield the most robust interpretations.

(a) Comparative anatomy
Comparative anatomy is fundamental for analysis of homology and character evolution but has limitations as a tool for analysis of function and diet. Its application in this context is based on the assumption that convergence of morphological structures between extant and extinct taxa allows us to infer similar functional roles for structures in extinct taxa, including foraging and feeding ecology (Gould & Lewontin, 1979;Fisher, 1985;Purnell, 1999;Ferry-Graham et al., 2002). In essence, interpretations are based on analogy (see Thomason, 1995).
There are a number of difficulties with this approach, not least of which is that many examples of its application can be criticised as 'adaptationist' (Gould & Lewontin, 1979;Fisher, 1985;Ferry-Graham et al., 2002). Criticism is valid where structures are studied as single traits that are assumed to be optimised for their particular hypothesised function, without consideration of phylogenetic and fabricational constraints that limit organisms' capacity to adapt to selective forces (Seilacher, 1970;Gould & Lewontin, 1979). Similarly, for many structures it is probably unrealistic to assume that they are adapted to perform a single function, resulting in trade-offs and a structure that is optimally adapted for neither role. The link between form and function is not always evident: teeth of one form can be applied to the consumption of a variety of foodstuffs, for example. This is what lies behind Liem's paradox, where organisms possess feeding-related morphologies that are apparently specialised for particular food items, but whose typical diets are composed of other constituents (Robinson & Wilson, 1998;Binning, Chapman & Cosandey-Godin, 2009). Conversely, traits that appear different can perform the same functions. Categorised as many-to-one mapping of form and function (Wainwright et al., 2005), this further undermines the value of analogy in interpreting function and diet. Unfortunately, many interpretations of pterosaur diet that draw on evidence of comparative anatomy fall into these traps. Most analyses focus exclusively on skulls, jaws or even individual teeth (Fastnacht, 2005;Ősi, 2011), and while not necessarily wrong, they often fail to account for the various constraints on optimal design and the possibility of many-to-one mapping.
The main issue is that hypotheses and interpretations generated through analogy and comparative anatomical approaches in themselves provide no mechanism for testing, and in some cases fail to pass the basic scientific criterion of testability. One interpretation of diet based on an extant functional analogue might conflict with another interpretation, equally well supported with a different analogue. Without employing methods in addition to comparative anatomy, such as functional morphological analyses, the only criteria upon which to evaluate such conflicting hypotheses are the ultimately subjective assessments of plausibility [see Purnell (1999) for discussions of similar issues in a different phylogenetic context].

(b) Associations
Despite the straightforwardness of using associations, deriving dietary interpretations from simple presence and/or absence of contemporaneous fossils amounts to little more than speculation. Likewise, solely examining depositional environments does not account for organism dispersal abilities and taphonomic biases such as carcass transportation (Veldmeijer et al., 2007;Witton & Naish, 2008;Tütken & Hone, 2010). Associations are nevertheless important for pterosaur research as they can serve as starting points for forming hypotheses when using experimental approaches.

(c) Functional morphology
Quantitative methods have begun to shift the emphasis away from a largely comparative, analogy-based framework towards a more hypothesis-testing approach, but also have their own issues. For example, structural attributes, including bone strength, can never be directly measured from fossil material and must be inferred from extant analogues. This is especially problematic for taxa with no extant descendants, such as pterosaurs. Functional morphological models are therefore only as representative as the parameters set and assumptions made in the experimental design (Anderson et al., 2011). Some investigations into pterosaur biting behaviours, such as the FEA and cantilever analyses of Fastnacht (2005) (Fig. 8), used human bone strengths as substitutes. These issues can be mitigated through sensitivity analyses designed to test the impact of starting conditions, such as the mechanical properties of structures, on the outputs of biomechanical models . The mechanical properties of bone and skeletal structures for theses analyses can be measured in extant taxa using strain gauges, either in live feeding trials (in vivo data) (Porro et al., 2013), or in experimental application of external loads to the skulls of deceased organisms (ex vivo data) Rayfield, 2011). Sensitivity analyses on ex vivo data from domestic pig (Sus domesticus) crania, for example, found that models which accounted for the heterogeneous nature of material properties in bone across the cranium exhibited fewest discrepancies with experimental outputs . Bone strength data can be subsequently extrapolated into models of extinct taxa, such as pterosaurs, and will improve our understanding of extinct functional behaviours (Degrange et al., 2010;Bright, 2014).
In vivo and ex vivo validation tests have also been performed on modern archosaurs including alligators (Porro et al., 2013) and ostriches (Rayfield, 2011), but further tests are needed on other archosaurs before these data can be confidently extrapolated to pterosaurs. Potential candidates include sea birds and piscivorous crocodilians as they have been suggested as pterosaur analogues (Seeley, 1901;Hazlehurst & Rayner, 1992;Witton, 2008;Bennett, 2014), and form an extant phylogenetic bracket around pterosaurs (Bennett, 1996;Nesbitt, 2011), with size and bone thickness differences between these taxa and pterosaurs taken into account. Adductor and neck musculatures could also be incorporated further to assess stress and strain distributions (Lautenschlager et al., 2013). This could help confirm whether pterosaurs procured and processed food items as hypothesised from earlier muscle reconstructions (Ősi, 2011;Pinheiro et al., 2014). Functional impacts of pterosaur rhamphothecas should also be investigated for relevant taxa because food items do not make direct contact with bone during feeding (Lautenschlager et al., 2013).
Understanding pterosaur locomotion, including terrestrial (walking), aerial (stall speeds) and aquatic-based (diving and wading) behaviours can provide new constraints, or corroborate existing constraints, on possible foraging strategies and food-acquisition behaviours. Locomotory models, however, are heavily influenced by mass estimations (Witton, 2008(Witton, , 2015aHenderson, 2010;Witton & Habib, 2010). Greater account needs to be made for the unique pterosaur body construction to allow pterosaur flight profiles to be more confidently placed within suitable biological contexts (Fig. 9). Humerus and femur structural strength models used for anurognathids (see Section IV.3) could be applied to other pterosaurs to investigate terrestrial behaviours (Habib & Hall, 2012;Habib & Witton, 2013). Terrestrial behaviours can also be partially verified through ichnofossil interpretations. Energy consumption modelling (see Section IV.8) should be used for pterosaurs with well-constrained mass estimates to calculate energy expenditures of locomotory behaviours to assess energetic viabilities of different diets (Habib, 2015).

(d) Content fossils
Content fossils, although rare, are an important part of investigations into extinct diets (Wild, 1984;Hone & Faulkes, 2014;Hone et al., 2015a;Witton, 2018). Over-reliance on further discoveries of these fossils, however, is ill advised. Issues include alternative explanations of content fossils as post mortem artefacts, with items introduced via water flows (Tweet et al., 2008), and incorrectly inferring extents of dietary specialism through preservation of items consumed almost immediately prior to death (Platt et al., 2013b;Davis & Pineda-Munoz, 2016), and/or through small sample sizes (e.g. Szczepanski & Bengtson, 2014; see Section V.1). While careful consideration of specimen depositional environments should help account for post mortem artefacts, content fossil evidence can only ever indicate that a species sometimes consumed a particular food item and is therefore best interpreted in conjunction with other lines of evidence (Frey & Tischlinger, 2012).

(e) Ichnofossils
Well-preserved ichnofossils can indicate possible foraging and/or feeding behaviours such as swimming and sediment  Fig. 9. Ecomorphospace of modern birds (grey shading) and 19 sampled pterosaurs (black dashed area with selected genera) with respect to aspect ratio and wing loading, as deduced from mass estimations and bone strength analyses, to infer possible flight styles. Specified foraging ecologies (marine soarers, diving birds, water birds, 'poor' fliers, thermal soarers, aerial predators), refer to modern bird flight styles. Adapted from Witton (2013).
probing (Hwang et al., 2002). However, extrapolating inferred behaviours from trace fossils to contemporary pterosaurs does not guarantee those taxa were the perpetrators. Potential peck marks are yet to be verified as pterosaurian in origin. The absence of Late Triassic and Early Jurassic pterosaur ichnofossils may reflect an unknown preservation bias towards environments with softer sediments (Lockley & Wright, 2003;Witton, 2018), as opposed to air-based feeding (Unwin, 2007). Ichnology is becoming more objective through 3D photogrammetry, which produces high-resolution digital reconstructions via alignment of photographs taken from numerous angles (Fig. 10) (Falkingham, 2012;Breithaupt et al., 2015). 3D photogrammetry offers more-effective ways of assessing ichnofossil perpetrators, locomotory behaviours and depositional environments (Fig. 10) (Xing et al., 2012;Breithaupt et al., 2015), which can aid in investigating diet.

(f ) Stable isotope analysis
Analysis of stable isotopes provides an additional source of data that is independent of morphological evidence. It is commonly applied to extant vertebrates as a tool for dietary analysis, in many cases focusing on δ 15 N and/or δ 13 C. In the context of extant animals, isotopic approaches to dietary reconstruction are not without methodological limitations (Nielsen, Popp & Winder, 2015;Nielsen et al., 2017). Analysis of δ 15 N, for example, provides a measure of the relative trophic position of a species within a specific trophic web, rather than food items consumed Right; 3D photogrammetric model of artificial casts of LUGP3-001.2. Blue denotes higher topographies, red-white areas denote lower topographies. Topographical differences between the pes and manus impressions are potentially caused by variable substrate consistencies. Adapted from Xing et al. (2012). (Crawford, McDonald & Bearhop, 2008), and multiple dietary combinations can result in the same δ 13 C and δ 15 N values (Caut, Angulo & Courchamp, 2009). Using isotopic evidence to infer diets in fossils is even more problematic. Some of the most informative stable isotopic analyses for dietary reconstruction, such as δ 15 N, are based on sampling tissues and biomolecules that do not preserve in typical fossils that are millions of years old (e.g. blood, muscle, hair, bone collagen). Analysis of carbon isotopes from tooth enamel is possible, but this is primarily used to infer the relative proportions of C 3 and C 4 plants consumed (see Davis & Pineda-Munoz, 2016), so not useful for analysis of pterosaurs.
To date, there have been only a few investigations of stable isotopes in pterosaurs, primarily because sampling is destructive and well-preserved pterosaur remains are comparatively rare. The analyses looked at δ 18 O and δ 13 C (Amiot et al., 2010;Tütken & Hone, 2010), which provide evidence of habitat rather than diet per se, i.e. signals of freshwater versus marine environments, for example (Fig. 11). Future work could focus on well-preserved fragmentary, and thus expendable, material such as isolated teeth, to provide new dietary constraints for pterosaurs.

(g) Quantitative 3D microtextural analysis of tooth microwear
Only one study has examined dental wear patterns on pterosaur teeth and used this to infer aspects of diet and the material properties of food (Ősi, 2011). Dietary hypotheses from this study were constructed from observations of gross pterosaur tooth and wear pattern morphology using two-dimensional (2D) images from scanning electron   (Ősi, 2011). Identifying and interpreting features in this way however, can be problematic because inter-observer error rates are high and identification of wear features is influenced by the orientation of specimens relative to the electron beam and detector within the SEM (Ungar et al., 2003;Scott et al., 2005Scott et al., , 2006. A more robust approach involves quantitative analysis of the sub-micron scale 3D surface textures of teeth, known as dental microwear texture analysis (DMTA; Ungar et al., 2003;Scott et al., 2006). Microwear is produced on teeth when an organism feeds, as interactions with food items cause microscopic chipping and scratching on tooth surfaces (Rensberger, 1978;Walker, Hoeck & Perez, 1978;Teaford, 1988). As microwear formation is determined by the material properties of food, the technique provides direct evidence of the nature of what has been consumed, and analysis does not rely on assumptions of a close relationship between the morphology and inferred functions of teeth (Purnell, Seehausen & Galis, 2012;Daegling et al., 2013;Purnell & Darras, 2016). Because data acquisition is not operator dependent, DMTA avoids the inherent observer bias of 2D analyses. Most applications of DMTA focus on placental mammals with occlusal dentitions, but it is applicable to dietary analysis of older, more basal material, and to taxa with non-occlusal dentitions (Purnell et al., 2012;Gill et al., 2014;Purnell & Darras, 2016). Its applicability to pterosaurs is the subject of on-going investigation. Analysis of tooth surface textures in pterosaurs, when validated against those of extant taxa with known dietary differences, has the potential to provide robust tests of dietary hypotheses.
(2) Most pterosaur dietary interpretations are supported by qualitative evidence including comparative anatomy, associations, content fossils and ichnofossils with a minority supported by quantitative evidence from functional morphology and isotope analyses.
(3) Some pterosaur principal groups exhibit high levels of consensus regarding diet, supported by several evidential categories; others exhibit lower levels of consensus, with different interpretations inferred from conflicting evidence of the same categorical type, and typically poorly constrained analogy drawn from comparative anatomy.
(4) More speciose pterosaur groups exhibit higher diversity of hypothesised diets. Whilst this may reflect biological signals such as ecological radiations or niche partitioning, non-biological causes, such as historical biases in the data, quality of the fossil record and research intensity are more likely. These biases mean it is currently difficult to reliably test the hypothesis that the apparent patterns of pterosaur dietary diversity and disparity are biologically controlled.
(5) Examining patterns of dietary diversity using different phylogenies reveals higher consensus among monofenestratan groups and lower consensus in non-monofenestratan groups. Better resolution on the membership of pterosaur groups would assist in uncovering true diets for respective groups.
(6) Numbers of pterosaur publications per year and dietary interpretations do not correlate through time. The majority of interpretations were proposed in the 21st century. The almost exponential rise in the number of publications containing dietary interpretations since the 1980s coincides with discoveries of new Lagerstätten and other exceptionally preserved sites, as well as applications of new techniques to pterosaurs.
(7) Many dietary interpretations are based on simple extrapolations and comparisons with modern biology with little scope for testing. Qualitative methods can serve as starting points for generating hypotheses, but quantitative tests provide more robust analyses and insights into dietary diversity, evolution and the ecological roles of pterosaurs. Improvements to current methods and application of novel methods to pterosaurs will provide better constraints on diets in pterosaurs with low levels of consensus, and better tests of dietary hypotheses in pterosaurs with high levels of consensus. This will allow reliable investigations into possible biological signals behind pterosaur dietary diversity and disparity, which will allow greater understanding of pterosaur dietary evolution and facilitate reconstructions of Mesozoic ecosystems.

VII. ACKNOWLEDGEMENTS
All authors contributed to the conception of the study and analytical design. J.B. and D.M.U. collected the data. J.B. analysed the data. J.B. wrote the manuscript with contributions from all authors. Thanks to Romain Amiot, Mark Witton and Li-Da Xing for permissions for Figs 11, 9 and 10, respectively. Thanks to Mark Witton for access to useful papers and discussions on pterosaur diets. Thanks to The Pterosaur Database, and Polyglot Palaeontologist (http://www.paleoglot.org) for access to useful papers. We thank two anonymous reviewers for helpful comments on the manuscript. J.B. is funded by a NERC studentship awarded through the Central England NERC Training Alliance (CENTA; grant reference NE/L002493/1) and by the University of Leicester.