The skull of Sanajeh indicus, a Cretaceous snake with an upper temporal bar, and the origin of ophidian wide-gaped feeding

Abstract

Recent phylogenetic analyses differ in their interpretations of the origin and interrelationships of snakes, resulting in polarized views of snake ecology, habit and acquisition of features associated with wide-gaped feeding (macrostomy). Here, we report a new specimen of the Late Cretaceous nest predator Sanajeh indicus that helps to resolve the origin of macrostomy. The new specimen preserves an ossified upper temporal bar and a posteriorly expanded otooccipital region that lacks a free-ending supratemporal bone and retains a lizard-like palatomaxillary arch that allows limited movements during swallowing. Phylogenetic analyses of a large-scale total evidence dataset resolve Sanajeh near the base of Pan-Serpentes, as the sister group of Najash, Dinilysia and crown-group Serpentes. The Cretaceous Tetrapodophis and Coniophis represent the earliest-diverging members of Pan-Serpentes. The Cretaceous hindlimbed pachyophiids and Cenozoic Australian ‘madtsoiids’ are inside crown Alethinophidia, whereas mosasaurs are recovered invariably within anguimorphs. Our results suggest that the wide-gape condition in mosasaurs and snakes might have evolved independently, as functionally distinct mechanisms of prey ingestion. The intermediate morphology preserved in Sanajeh indicates that ingestion of large prey items (macrophagy) preceded wide-gaped, unilateral feeding (macrostomy), which appeared 35 Myr later, in the common ancestor of pachyophiids, Cenozoic Australian ‘madtsoiids’ and alethinophidians.

macrostomy, Pan, Serpentes, total evidence phylogeny, Toxicofera

INTRODUCTION

Snakes are known for their highly specialized morphology that departs from the condition in other squamates (Bellairs, 1950; Bellairs & Underwood, 1951; Underwood, 1967). In addition to body elongation and the reduction and loss of girdles and limbs, snakes underwent major remodelling of their cranial skeleton that allowed for jaw transport of whole prey and, in some alethinophidian snakes, a wide-gaped condition (macrostomy) characterized by unilateral jaw movements during swallowing (Cundall, 1995; Kley, 2001). Although vertebral repatterning and limb loss are now better understood through palaeontological (Lee & Caldwell, 1998; Rage & Escuillié, 2000; Rieppel et al., 2003; Zaher et al., 2009) and developmental discoveries (Cohn & Tickle, 1999; Gomez et al., 2008; Werneburg & Sánchez-Villagra, 2015; Leal & Cohn, 2016), the origin of wide-gaped macrostomy remains poorly resolved owing to the rarity of complete skulls in the early fossil record of snakes and conflicting phylogenetic reconstructions of their diversification.

Central to this disagreement are conflicting interpretations of the anatomy and phylogenetic affinities of several fossil snakes with well-preserved skulls, including the terrestrial Cenozoic Australian ‘madtsoiids’ Wonambi Smith, 1976 and Yurlunggur Scanlon, 1992 (Scanlon & Lee, 2000; Scanlon, 2006), the Cretaceous marine hindlimbed pachyophiids Eupodophis Rage & Escuillié, 2000, Haasiophis Tchernov et al., 2000 and Pachyrhachis Haas, 1979 (Lee & Caldwell, 1998; Rieppel et al., 2003; Rieppel & Head, 2004) and the terrestrial Cretaceous snakes Dinilysia Woodward, 1901, Najash Apesteguía & Zaher, 2006 and Sanajeh Wilson et al., 2010 (Estes et al., 1970; Apesteguía & Zaher, 2006; Zaher et al., 2009; Wilson et al., 2010; Zaher & Scanferla, 2012; Garberoglio et al., 2019a). Among these fossil taxa, pachyophiids and Australian ‘madtsoiids’ were hypothesized to be wide-gaped snakes (Zaher, 1998; Lee et al., 1999; Scanlon & Lee, 2000; Scanlon, 2006) based on a supratemporal with a free-ending posterior process extending backward beyond the craniovertebral joint, a quadrate rotated backward (only in pachyophiids), and an elongated mandible with a free symphysis and posteriorly extended dentigerous process on the dentary. As a result, interpretation of the evolution of macrostomy pivots on the hypothesis of phylogenetic relationships for these fossil snakes relative to extant ‘scolecophidians’ and anilioids, which lack macrostomy, and those extant alethinophidian species that possess it.

In one set of analyses, it is hypothesized that snakes originated in a marine setting from a macrophagous common ancestor with mosasaur lizards (Caldwell & Lee, 1997; Lee et al., 1999; Rage & Escuillié, 2000; Scanlon & Lee, 2000; Scanlon, 2006). These analyses resolve pachyophiids and Australian ‘madtsoiids’ at the base of the snake radiation, implying that macrostomy in snakes was inherited from the most recent common ancestor of mosasaurs and snakes and secondarily reduced or lost in ‘scolecophidians’ and anilioids, which, consequently, were referred to as ‘regressed macrostomatans’ (Rieppel, 2012).

In another set of analyses (Zaher, 1998; Rieppel & Zaher, 2000; Tchernov et al., 2000; Rieppel & Head, 2004; Conrad, 2008; Zaher & Rieppel, 1999a; Zaher et al., 2009; Wilson et al., 2010; Gauthier et al., 2012; Zaher & Scanferla, 2012), it is hypothesized that snakes originated from terrestrial lizards, nesting the pachyophiids and Australian ‘madtsoiids’ within or near crown-group Alethinophidia and placing the terrestrial Cretaceous Dinilysia and Najash as the earliest-branching snakes. This hypothesis of relationships implies that macrostomy was not inherited from a common ancestor with mosasaurs, but evolved within an advanced subgroup of fossil and extant snakes (Zaher & Rieppel, 1999a; Rieppel & Zaher, 2000).

Observations of feeding in extant snakes provide crucial functional context for these divergent views on the origin of macrostomy. All extant snakes consume whole prey, but mechanisms of intraoral prey transport differ substantially within the group (Cundall, 1995; Cundall & Greene, 2000; Kley, 2001). The secretive, burrowing ‘scolecophidians’ transport prey within the mouth using limited asynchronous or bilaterally synchronous raking movements of the upper or lower jaws (Kley & Brainerd, 1999; Kley, 2001), whereas wide-gaped alethinophidian snakes have acquired a highly kinetic intraoral capability that allows them to transport prey using alternating unilateral movements of the upper jaws (Cundall, 1995; Kley, 2001). The origin of macrostomy, therefore, rests on the sequence of acquisition of intraoral kinesis in basally diverging snakes, which must be inferred based on anatomical correlates of kinesis present in key bones in the upper jaws, palate and mandible.

Here, we provide new morphological data on the Late Cretaceous snake Sanajeh indicus Wilson et al., 2010 that help to establish the sequence of changes involved in intraoral cranial kinesis in snakes. The holotypic specimen of Sanajeh indicus, named for its inferred ‘ancient gape’, was preserved within a titanosaur nest, in close association with a clutch of eggs and a hatchling (~0.5 m), in latest Cretaceous-aged sediments in Gujarat, western India (Wilson et al., 2010). The osteology and unique preservational setting of Sanajeh suggest that it was a macrophagous predator that lacked key features related to macrostomy. Further preparation of this original specimen reveals previously unavailable details of the premaxilla, right septomaxilla, right upper jaw, circumorbital bones, palate and right mandible. A second specimen of Sanajeh was recently recovered at the same site, in association with a hatchling or juvenile turtle. This new Sanajeh specimen includes a partial skull preserving portions of the braincase, skull roof, palate and mandible, along with an associated palatine, a compound bone and an articulated series of precloacal vertebrae. Importantly, this new specimen preserves a complete upper temporal bar, previously unknown in snakes, a lizard-like palatopterygoid arch and an expanded suspensorium with a plesiomorphically long otooccipital that retains a lateral contact with the quadrate below the supratemporal. As detailed below, Sanajeh corroborates the hypothesis of the terrestrial origin of snakes and provides crucial information about the transition from the relatively akinetic jaw apparatus of lizards to the highly kinetic system of wide-gaped alethinophidian snakes, which involved: (1) enclosure of the posterior braincase; (2) posterior expansion of the supratemporal as the main suspensorial element; and (3) the development of kinesis within the palatomaxillary arch and between the snout and skull roof.

Tetrapod Diversity at the dholi dungri locality

The two specimens of Sanajeh indicus described here come from Late Cretaceous-aged sediments cropping out near the village of Dholi Dungri in Gujarat, western India. Although many bones and eggs are preserved in the Late Cretaceous of India, Dholi Dungri is the only reported site where eggs and bones co-occur. The initial report detailing discovery of eggs and bones together at Dholi Dungri attributed both of them to sauropod dinosaurs on the basis of the microstructure of the eggshell, which was made up of ‘short spheroliths with highly arched growth lines and narrow subvertical caniliculae’ (Mohabey, 1987). The complete egg collected in association with bones at Dholi Dungri was later selected as the type of the oospecies Megaloolithus dhoridungriensis Mohabey, 1998. However, there appears to have been an error in the number listed: Mohabey (1998) listed OGF 121’ as the type, but the egg found with bones at Dholi Dungri is unambiguously numbered ‘OGF 105’, later formalized as GSI/GC/2905 (Wilson et al., 2010). The bones in that first-described specimen (Mohabey, 1987) pertain to a hatchling sauropod and to the holotype of Sanajeh indicus (see below in Systematic Palaeontology).

Further work at the site uncovered numerous other bones and eggs, including representatives of at least two reptile species not previously reported in Dholi Dungri. These include at least three notosuchian crocodylomorph specimens and two turtle specimens. The notosuchian remains include a partial skeleton that is currently undergoing preparation, a small, isolated scute found in the same block as the referred specimen of Sanajeh indicus and a third crocodylomorph specimen that has not yet been collected. Turtle remains include the partial shell of an apparently adult individual that was collected but has not yet been prepared and the bones of what appears to be a young individual. There is no evidence for a second species of snake at Dholi Dungri. Both snake specimens share considerable morphological similarity, as described below, and they were both associated with young, possibly hatchling, reptile individuals at the same site.

MATERIAL AND METHODS

Preparation of fossil specimens

All fossil specimens of Sanajeh indicus were prepared at the University of Michigan Museum of Paleontology using a combination of chemical and mechanical techniques. Previously collected blocks were coated with a lacquer preservative that was removed using Zip-Strip; blocks were then subjected to 3% formic acid for ~2–3 h, which weakened calcareous cement (Wilson et al., 2010). All blocks were mechanically prepared using a micro-airscribe and needles to uncover the ‘up’ surface of the bones. The blocks were fitted together as they were found in the field and then moulded and cast. Specific elements of interest (e.g. braincase, dentary) were then extracted fully from the matrix. All specimens are housed in the Geological Survey of India Palaeontology Division, Central Region, Nagpur, India. Casts of some of the holotypic and referred elements are available at the University of Michigan and Universidade de São Paulo.

Computed tomographic (CT) scans of the referred specimen were performed at the University of Michigan Dental School in Ann Arbor, MI, USA, using a Scanco Medical μCT100 with a 0.5-mm-thick aluminium filter. Scan settings were 90 kVp, 155 μA and exposure time 500 ms. Uniform cubic voxels were 40 μm on a side. Three-dimensional (3D) visualization, segmentation and analysis of the reconstructed data were performed using VGStudio MAX 2.2.3 64 bit (Volume Graphics, Heidelberg, Germany).

Phylogenetic analyses

Our phylogenetic analyses were designed to test the affinities of Sanajeh indicus within Toxicofera (iguanians, anguimorphs and snakes), a clade of squamates that is well supported by molecular data (Zheng & Wiens, 2016; Pyron, 2017; Burbrink et al., 2020). We built three different datasets to evaluate the phylogenetic position of Sanajeh indicus: (1) a morphological matrix; (2) a concatenated molecular matrix; and (3) a total evidence dataset that combines morphological and molecular matrices (available at: https://doi.org/10.6084/m9.figshare.14618217.v1). We analysed these three matrices using maximum parsimony (MP) and maximum likelihood (ML) methods (Supporting Information, Fig. S1). Here, we only provide results of the analyses performed on the total evidence dataset. The total evidence dataset was also analysed using Bayesian inference (BI) to provide time-calibrated divergence estimates (Supporting Information, Fig. S2). Comparisons among topologies resulting from the three different methods are provided as Supporting Information (Figs S3–S5).

Morphological dataset

Our morphological dataset is based on the matrix in the study by Hsiang et al. (2015), which was built from multiple sources, including 605 characters from Gauthier et al. (2012), 43 characters from Zaher & Scanferla (2012), 30 characters from Longrich et al. (2012) and 88 characters novel to that analysis (see also Zaher & Smith, 2020). To these 766 characters, we added 19 new characters, for a total of 785 morphological characters (Supporting Information, File S1; Table S1). Given that many of the characters in the study by Gauthier et al. (2012) were modified by Hsiang et al. (2015), in many cases without description or rationale, the character numbers listed in those two studies do not match. For this reason, we have listed each of the deletions, rescorings and renumberings (Supporting Information, File S1). Hsiang et al. (2015) modified the dataset from Gauthier et al. (2012) further by reducing the number of terminal taxa to nearly one-third of the original (from 192 to only 73 terminal taxa), retaining only anguimorph taxa as outgroups of snakes. This reduction in terminal taxa was not accompanied by a reduction in the number of character scorings, which resulted in a dataset that contained many uninformative characters (Supporting Information, File S1). Although this surplus of uninformative character data did not impact results of the phylogenetic analysis, the decision by Hsiang et al. (2015) to retain most of the characters listed by Gauthier et al. (2012) in their dataset allowed us to follow the changes and additions that were made (Supporting Information, File S1). In order to provide the same foundation for future researchers, we retained all characters from the study by Hsiang et al. (2015) in our dataset, even those that were determined to be phylogenetically uninformative by their analysis.

Pyron (2017) performed a total evidence analysis of squamate phylogenetic relationships that combined two morphological datasets (Conrad, 2008; Gauthier et al., 2012) and a large molecular dataset (Pyron et al., 2013). The calibrated combined-data analysis consistently recovered the clade Toxicofera, formed by extant anguimorphs, iguanians and Serpentes. This clade also included the extinct Mosasauroidea (i.e. aigialosaurids and mosasaurids), which was retrieved either as the sister group of Serpentes or nested in Anguimorpha, far from Serpentes (Pyron, 2017: fig. 3b, d). The legless and serpentiform taxa (i.e. dibamids, amphisbaenians, pygopodids, Anniella Gray, 1852, Feylinia Gray, 1845 and Ophisaurus Daudin, 1803) were either positioned far from Toxicofera (e.g. near the base of the squamatan tree or as sister group of Lacertidae) or nested in Anguidae rather than in or near Serpentes. Based on these results, we chose to use as our outgroup sampling a representation of Iguania and Anguimorpha, including basally diverging mosasauroids (i.e. aigialosaurs and coniasaurs), but excluding legless forms used in the analysis by Gauthier et al. (2012). Accordingly, we expanded the outgroup taxon sampling of Hsiang et al. (2015) to include 11 iguanian lizards, 21 anguimorphs (including four basally diverging mosasauroids) and two rhynchocephalians, Sphenodon punctatus (Gray, 1842) and Gephyrosaurus bridensis Evans, 1980, totalling 90 extant and extinct terminal taxa.

We revised all scorings in the dataset of Hsiang et al. (2015) based on observations made directly on specimens, photographs or 3D video reconstructions of specimens. Several of these were scanned as part of the Digimorph Project and used originally by Gauthier et al. (2012) to construct their dataset. Full details on taxon sampling and character scorings are provided as Supporting Information (File S1). Rescorings of Eupodophis descouensi Rage & Escuillié, 2000 were based mainly on observations of the holotype after additional preparation in 2015. We also used as a second source of information the descriptions provided by Rieppel & Head (2004) and Palci et al. (2013). Rescorings of Haasiophis terrasanctus Tchernov et al., 2000 were based on new observations made on the holotype after careful removal of a thin film of adhesive and microscopic fragments of tissue. We also made first-hand observations from micro-CT images of the skulls of Sanajeh indicus (referred specimen only), Dinilysia patagonica Woodward, 1901, Wonambi naracoortensis Smith, 1976, Haasiophis terrasanctus, Eupodophis descouensi and Pachyrhachis problematicus Haas, 1979. We relied on 3D reconstructions based on synchrotron and/or micro-CT data of pachyophiids (Zaher et al., 2021), Dinilysia patagonica (available on Morphosource) and Sanajeh indicus (available on UMORF at: https://umorf.ummp.lsa.umich.edu/wp/). Scorings of Najash rionegrina Apesteguía & Zaher, 2006 were based on observations made on the holotypic and referred specimens (Zaher et al., 2009) and on new data from the literature (Garberoglio et al., 2019a, b). Scorings of Gephyrosaurus bridensis were made using available literature. We also deleted Kataria anisodonta Scanferla, Zaher, Novas, de Muizon, and Céspedes, 2013 from our analysis because it represents a rogue taxon that does not add significant information to the present work.

Molecular dataset, alignment, partition schemes and model selection

Our molecular matrix incorporated the alignments provided by Zheng & Wiens (2016) for 52 gene partitions (representing 47 loci sampled from the nuclear genome and five genes representing one locus from the mitochondrial genome) and Burbrink et al. (2020) for 394 nuclear anonymous loci (available at: https://doi.org/10.6084/m9.figshare.14618217.v1). We reduced the original alignments of both studies to 70 extant terminals, focusing our analysis on the relationships within Toxicofera. We selected terminals that could be integrated easily with the morphological matrix (see above). For the total evidence dataset, we also retained the more distant outgroup Crocodylus porosus Schneider, 1801. Inclusion of this more ancient node provided a hard-bound calibration point for the root of our phylogenetic tree.

We followed the same partition schemes used by Zheng & Wiens (2016) and Burbrink et al. (2020). For both the ML and BI analyses, we used the general time reversible (GTR) nucleotide substitution model with Γ correction for each molecular partition. The choice of GTR+Γ follows recommendations to avoid the combined use of Γ and a proportion of invariable sites (Yang, 2006; Stamatakis, 2016).

Maximum parsimony analysis

We used TNT v.1.1 (Goloboff et al., 2008) to analyse all three datasets using equally weighted parsimony. All gaps were treated as missing data, and a heuristic tree search was conducted using NEW TECHNOLOGY (Goloboff, 1999) until the consensus was stabilized five times (command xmult = consense 5). The best trees obtained at the end of the replicates were subjected to a final round of tree bisection with reconnection branch swapping (command bb). Zero-length branches were collapsed if they lacked support under any of the most parsimonious reconstructions. Node supports were calculated using 1000 bootstrap pseudoreplicates, and the supports were expressed as absolute frequencies.

Maximum likelihood analysis

We used RAxML v.8.2.3 (Stamatakis, 2014) to perform the ML analyses. We conducted a rapid bootstrap analysis (1000 pseudoreplicates) and searched for the best-scoring ML tree in the same run (option -f a) for all three datasets. We used the MK model (Lewis, 2001) for morphological character evolution in both the morphological and combined matrices; we used option -m ASC_MULTIGAMMA -K MK) with Lewis’s ascertainment bias correction (option asccorr = lewis).

Bayesian inference analysis and time-calibrated tree estimation

Given that a full Bayesian uncorrelated relaxed clock method is not computationally viable for our total evidence dataset (524 471 characters for 91 terminals), we randomly subsampled 50 partitions from our concatenated molecular matrix and combined them with the morphological partition. We repeated the partition subsampling scheme five times, and for each resulting matrix we used MrBayes v.3.2.6 (Ronquist et al., 2012) to implement the BI analyses. For each analysis, we combined the resulting five tree distribution files into a consensus tree. By following this procedure, we were able to assess different phylogenetic signals from each partition subsampling and include the phylogenetic uncertainty of 250 loci in the time-calibrated tree estimation.

For each subsampled matrix, we performed a set of four different Bayesian analyses: (1) uncalibrated phylogenetic estimation (BUE); (2) node-dating calibrated analysis (BND); (3) tip-dating calibrated analysis (BTD); and (4) node- and tip-dating calibrated analysis (BNTD).

The BUE was performed to provide an initial tree height for the calibrated analyses. We also used the results from BUE to define a topological constraint to accelerate the calibrated analyses. The topological constraint was built based only on nodes presenting posterior probabilities ≥ 0.85.

For all calibrated analyses, we used the fossilized birth–death tree model (Zhang et al., 2016) in a Bayesian total evidence-dating approach. We combined node- and tip-dating approaches in a BNTD analysis to calibrate the tree, because it has been demonstrated that using tip dates alone can contribute to unrealistically older divergence time estimates for some clades (O’Reilly et al., 2015; O’Reilly & Donoghue, 2016). However, to assess the effects of these approaches (tip-dating and node-dating), we also conducted independent analyses using both methods (BND and BTD). We used BUE, BND and BTD only as preliminary approaches to estimate and compare parameters for the BNTD analysis; we discuss further below only the Bayesian topology provided by the BNTD analysis (Supporting Information, Fig. S2).

We ran each analysis for 20 million generations, sampling trees and parameters every 1000 generations. For the molecular partitions, we used GTR+Γ, as used for the ML analysis. For the morphological partition, we used the Mkv model with variable coding and Γ-distributed rate heterogeneity. The speciation prior was set to exp(10), and the extinction and fossilization priors were set as beta(1,1). The sampling strategy was set to diversity, with a sampling probability of 0.00345.

To accommodate both morphological and molecular partitions, we set the clock rate prior as a log-normal distribution, with an estimated mean based on the length of the tree resulting from the uncalibrated Bayesian analysis. For each log-normal distribution, we set a large standard deviation, calculated as the exponent of the mean. We set the tree root age to a uniform prior from 255 to 299.8 Mya, based on the stratigraphic occurrence of Protorosaurus von Meyer, 1830 sp. (Jones et al., 2013). We used the uncorrelated relaxed independent gamma rates (IGR) clock model, with the parameter IGRvar (the amount of rate variance across branches) set to the default exp(10). This approach attempts to combine the methods and suggestions of Pyron (2017) and Simões et al. (2018). Like Simões et al. (2018), we based the clock rate prior on the estimated parameters of the uncalibrated tree, although we attempted to keep the prior density distribution closer to that used by Pyron (2017). Simões et al. (2018), in contrast, used a base clock rate of 0.0773 substitutions/Myr, which is 19 times larger than the 0.004103 substitutions/Myr used by Pyron (2017) and represents a rate of ~7%/Myr. The clock rate prior from the work of Simões et al. (2018) seems unreasonable to us, even for a rapidly evolving locus, such as mitochondrial DNA (usually evolving at ~1%/Myr). Therefore, instead of calculating the base clock rate using the ‘median value for tree height in substitutions from the entire posterior trees sample’, as in the study by Simões et al. (2018), we based our estimation on the tree length, and we set the mean clock rate as the average number of substitutions per site from root to tips divided by the mean root age.

The ages of occurrence of 20 lepidosaurian fossils were used as uniform priors for tip-dating, whereas 13 other fossil stratigraphic ranges were used as uniform priors to calibrate specific clades (node-dating) of our phylogeny (Supporting Information, File S1).

For all Bayesian analyses, convergence of independent runs was assessed using an average standard deviation of split frequencies of ~0.01, potential scale reduction factors of approximately one for all parameters, and an effective sample size > 200 for each parameter. We also used TRACER v.1.6.1 (Rambaut et al., 2014) to check for convergence and to define the burn-in visually. We used MrBayes (command sumt) to combine and summarize the estimated parameters and tree distributions generated by each analysis and to combine the results of all replicates for each independent Bayesian approach, incorporating together the uncertainty of topologies and parameters of five runs and 250 resampled partitions.

Most of the analytical procedures described above were executed using scripts in bash or R, applying functions from packages such as strap, ape (Paradis et al., 2004) or phytools (Revell, 2012). All the scripts used to analyse the data and generate the figures are available as Supporting Information (Files S2–S4).

RESULTS
SYSTEMATIC PALAEONTOLOGY

Squamata Oppel, 1811
Pan-Serpentes Head, De Queiroz & greene, 2020
Ophidia Brongniart, 1800
SanajehWilson, Mohabey, Peters & Head, 2010,
Sanajeh indicusWilson, Mohabey, Peters & Head, 2010

Hypodigm:

Here, we restrict the holotype of Sanajeh indicus to GSI/GC/2901, GSI/GC/2902, GSI/GC/2903 and GSI/GC/2906. We exclude from the holotype specimens GSI/GC/2904 and GSI/GC/2905, which were inadvertently included in the holotype (Wilson et al., 2010) but pertain to a titanosaur hatchling and egg, respectively. We refer to Sanajeh indicus (Figs 1-15) a second specimen (GSI/GC/DD4) from the Dholi Dungri locality that includes the posterior portion of the skull, comprising a partial braincase and skull roof (parietal, basioccipital, otooccipital, prootic, left postorbital, left squamosal, left supratemporal and left stapedial footplate), a partial right mandible (complete splenial, fragmentary dentary and compound bone) and right pterygoid. Found in close association with this partial skull was the posterior portion of the left compound bone, a right palatine, several precloacal vertebrae and ribs.

Figure 1.

Sanajeh indicus holotype block GSI/GC/2903. Photograph of cranial elements preserved in near-anatomical articulation atop a chain of dorsal vertebrae (anteriormost vertebra at left).

Referral of GSI/GC/DD4 to Sanajeh indicus:

Specimen GSI/GC/DD4 was found 51 m from the holotype at Dholi Dungri, in close association with elements of a turtle shell pertaining to a young individual and a single, small (1.5 cm × 2.5 cm) crocodylomorph osteoderm. The elements we attribute to specimen GSI/GC/DD4 (i.e. skull fragment, right palatine, partial compound bone and two articulated partial vertebral series) were found within a radius of < 39 cm. The palatine was found 14.3 cm from the skull fragment, and the shorter and longer articulated vertebral series were found 8.4 and 38.4 cm from the skull fragment, respectively. The proximity, size and non-duplication of these elements are all consistent with the hypothesis that they pertain to a single individual, which is corroborated further by the near-perfect articulation of the right palatine and pterygoid (Fig. 14). Referral of GSI/GC/DD4 to Sanajeh indicus is based on shared morphological similarities present in numerous overlapping cranial and postcranial elements, including the parietal, prootic, otooccipital, basioccipital, palatine, stapedial footplate, mandible, precloacal vertebrae and ribs, several of which are diagnostic (see ‘Emended diagnosis’ below).

The holotypic skull (GSI/GC/2903) includes a partial right palatine comprising a fragment of the lateral edge of the dentigerous process and an incomplete maxillary process, which is pierced by the foramen for the subocular ramus of the trigeminal nerve posteriorly. The referred specimen (GSI/GC/DD4) includes a nearly complete right palatine that preserves its four main processes and nine teeth. The holotypic and referred palatines are slightly different in size, but they bear a close resemblance to one another morphologically. Uniquely, they share an especially high posterior wall of the maxillary process of the palatine, and both have a dentigerous process. The holotypic and referred specimens also both preserve the basioccipital, which bears a wide and dorsally concave posterolateral process. The footplate of the stapes is also preserved in both specimens, which share the same general outline in lateral view. Precloacal vertebrae preserved with the holotypic and referred specimens are similar in size and possess small parazygantral foramina and posterodorsally angled neural spines.

Emended diagnosis:

Wilson et al. (2010) listed four features that together distinguish Sanajeh indicus from other snakes: (1) juxtastapedial recess rectangular; (2) basioccipital posterolateral process wide and dorsally concave; (3) precloacal vertebrae with small parazygantral foramina; and (4) precloacal vertebrae with thin, posterodorsally angled neural spines.

Wilson et al. (2010) also listed a ‘broad and squared-off supratemporal’, but we have reinterpreted this bone to be the stapedial footplate (see below in Additional description of Sanajeh indicus). Further preparation of the holotypic specimen (GSI/GC/2903) and the referred individual (GSI/GC/DD4) suggests the following additional diagnostic features of Sanajeh indicus: (5) premaxillary teeth positioned laterally, on the maxillary processes (as in pythons); (6) maxilla with a high ascending process; (7) maxilla with palatine process expanded medially to form a large, triangular shelf; (8) palatine with an anteriorly elongated dentigerous process; (9) pterygoid and palatine bones are toothed; (10) palatine and pterygoid with tongue-and-groove articulation; (11) palatine maxillary process with especially high posterior wall; (12) supratemporal elongated and tightly appressed to the braincase; (13) supratemporal lacking a free-ending process; (14) stapedial footplate enlarged; (15) stapes elongate and posteriorly curved; (16) prootic with elongate posterior process; and (17) ossified upper temporal bar formed by postorbital and squamosal.

Additional description of Sanajeh indicus

In the description of the preserved bones in Sanajeh presented below, we combine the anatomical information present in the holotypic and referred specimens. There is considerable anatomical overlap between the two specimens, which allows us to determine that they are conspecific, but even among these overlapping elements there are differences in preservation and completeness that require us to rely on both in the description. Overlapping elements include the otooccipital, supraoccipital, prootic, basioccipital, parietal, palatine, stapes, dentary, splenial, angular, coronoid and compound bone. In addition to these, there are also non-overlapping elements that are preserved in only one of the two specimens. The holotype is the only Sanajeh specimen that preserves a parabasisphenoid, septomaxilla, premaxilla, maxilla, frontal and postfrontal, whereas the referred specimen is the only one that preserves a postorbital, squamosal, supratemporal and pterygoid.

Reinterpretation of previously described elements of the holotype

We have reinterpreted the identifications of three cranial elements preserved with the holotype of Sanajeh indicus based on additional preparation of that specimen (Figs 1, 2). The new interpretations and rationale for each is briefly noted below. Scorings in the character–taxon matrix reflect these new interpretations.

Figure 2.

Sanajeh indicus holotype block GSI/GC/2903. Line drawing of cranial elements preserved in near-anatomical articulation atop a chain of dorsal vertebrae (anteriormost vertebra at left). Hatched lines indicate broken bone surface; orange tone indicates matrix; and dashed lines indicate reconstructed portions of incomplete elements. Abbreviations: bc, braincase; l, left; m, maxilla; md, mandible; pl, palatine; pm, premaxilla; pof, postfrontal; r, right; sm, septomaxilla.

• 1. ‘Supratemporal’: we interpret the element identified by Wilson et al. (2010) as the supratemporal to be the stapedial footplate, based on its shape and on the presence of a broken region near its centre that represents the base of the columella.

• 2. ‘Palatine’: the element identified by Wilson et al. (2010) as the palatine is here interpreted to be the septomaxilla (Figs 1, 2), based on its connections with other bones and comparisons with the palatine preserved with the second specimen, which bears teeth and can be identified definitively as that element (Figs 7, 14).

• 3. ‘Left mandible’: further preparation has revealed that the element identified as an anterior fragment of the left mandible in the description by Wilson et al. (2010) is the premaxilla (Figs 1–3). Its alveolar margin is almost completely preserved, and its preserved dentition matches those in the premaxillae of other snakes (Fig. 3B).

Braincases of the holotypic and referred specimens

The holotypic specimen of Sanajeh preserves a partial braincase that is best preserved on its posteroventral and right ventrolateral sides. The left posterior and dorsal portions of the holotypic braincase are not well preserved. The referred braincase is much better preserved in these respects and provides more detailed information on the anatomy of the prootic, parietal, supratemporal and otooccipital regions (Figs 8–11, 15).

The holotypic specimen retains a partial parietal, left and right prootics, right stapes, parabasisphenoid, basioccipital and fragmentary left and right otooccipitals. Except for the basioccipital, which has a well-preserved outer surface on its right side, all remaining bones show different levels of abrasion of their surface and losses of significant parts. The holotypic skull does not preserve the posteriorly directed paroccipital processes of the otoocipital or their dorsolaterally adpressed supratemporals. The referred specimen, in contrast, retains a partial parietal, left prootic, left otooccipital, left squamosal, complete left stapes, supratemporal and postorbital.

The posterior extension of the prootic that roofs the otic capsule is present in the right side of the holotype, but it is badly weathered and difficult to interpret. However, the paroccipital process of the otooccipital, supratemporal and posterior part of the prootic are nicely preserved on the left side of the referred specimen (Figs 8–11). The posterior half of the left prootic and left stapes are not preserved in the holotype, revealing the inner surface of the otic capsule at the level of the lateral semicircular canal. The ventrolaterally projected process of the otooccipital that forms the crista tuberalis is almost completely preserved on the left side of the holotype and of the referred specimen. The basioccipital is mostly preserved on the right side of the holotype and only preserved in part on the left side of the referred specimen.

Snout complex

The snout complex in snakes includes the premaxilla, nasal, septomaxilla and vomer. Neither specimen preserves the nasal or vomer, and the premaxilla and septomaxilla are preserved only in the holotypic skull of Sanajeh indicus.

Premaxilla:

Further preparation of the holotype revealed a partially complete premaxilla that was identified previously by Wilson et al. (2010) as the anterior tip of the left mandible (Fig. 3A, B). The premaxilla is preserved in near-life position, near the end (as preserved) of the right maxilla and perpendicular to its long axis. Much of the alveolar region of the premaxilla remains embedded in matrix, but the remainder is completely exposed, including two teeth on the left maxillary process. The premaxillary teeth possess a smooth surface, with no sign of either carinae or a fluted base. The nasal and vomerine processes of the premaxilla are not preserved, and the dorsal portion of the left maxillary ramus is damaged, revealing the inner surface of the premaxillary channels.

Figure 3.

Premaxilla and maxilla of Sanajeh indicus holotype GSU/GC/2903. Photographs and line drawings of right premaxilla in anterodorsal (A) and posterior (B) views and the right maxilla in lateral view (C). Hatched lines indicate broken bone surface; parallel double-line pattern indicates matrix; and dashed lines indicate reconstructed portions of incomplete elements. Abbreviations: fo, foramen; fos, fossa; mp, maxillary process; np, nasal process; pm3, premaxillary tooth 3; pm4, premaxillary tooth 4; prf, prefrontal; t, tooth.

The premaxilla is a transversely oriented structure formed by two lobe-like maxillary processes that are separated, in part, ventrally by a median embayment. Consequenntly, the anteroventral margin of the premaxilla has an undulating surface. There is no indication that the lateral aspect of the premaxilla was sutured to the maxillae. Instead, its rounded, smooth lateral margins, as best preserved on the left side, indicate a loose contact between the premaxilla and maxilla. However, the degree of mobility between these two bones is difficult to establish. Breakage has exposed premaxillary channels on each side of the nasal process, but their posterior openings cannot be defined.

Two premaxillary teeth are exposed in posterior (lingual) view (Fig. 3B). They are recurved, needle-like and have a smooth basal surface. The two teeth are positioned at the lateral extreme of the maxillary process and are tightly appressed, diverging only slightly apically owing to the narrowing of tooth diameter. Both teeth seem to be ankylosed to the rim of shallow thecae, with only limited deposition of bone of attachment at their base. Resorption pits are not visible.

Septomaxilla:

A portion of the right septomaxilla is preserved in the holotypic skull of Sanajeh indicus nearly in the natural position, located posterolateral to the right maxillary process of the premaxilla and medial to the ascending process of the maxilla (Figs 1, 2). The body of the septomaxilla is mostly missing, but the lateral flange (conchal process; sensuCundall & Irish, 2008) is almost completely preserved. It is horizontally directed, but more laterally the flange curves abruptly to form a vertically directed lamina that bears a long, partly preserved posterior spine. Its anterior edge is broken; therefore, we cannot ascertain whether it possessed an anteriorly directed spine.

Palatomaxillary complex

Four bones are associated with the palatomaxillary complex in snakes, namely the maxilla, palatine, pterygoid and ectopterygoid. The ectopterygoid is not preserved in either of the two partial skulls attributed to Sanajeh indicus.

Maxilla:

The holotype of Sanajeh indicus preserves a partial right maxilla in contact with palatal elements (septomaxilla, palatine) and an element adjacent to the posterior end of the maxilla that we identify as the postfrontal (Figs 1, 2). The body of the maxilla is preserved in part, and it is not clear how much of its anterior and posterior ends are missing. Much of the middle portion of the maxilla is not preserved, including the orbital margin and the ascending process. Part of the alveolar margin of the maxilla is still embedded in matrix, but the remainder has been prepared completely. Six maxillary teeth are exposed; four are visible in internal (lingual) view, and two are exposed in external (labial) view (Fig. 3C). The preserved teeth span the preserved length of the maxilla, and the posteriormost tooth is on a fragment of the maxilla that has been dislodged from the main body. The maxillary teeth are circular in cross-section at their base, as indicated by the last maxillary tooth. The teeth are tall, curved and cylindrical, with a pointed distal crown and an expanded base. Their curvature indicates that they are not aligned with the maxillary axis, but have their tips directed more medially. The outer layer of enamel is eroded in all observed teeth, preventing any detailed analysis of the surface ornamentation. Internally, the teeth lack plicidentine, as shown by the posteriormost tooth on the maxilla, which is broken transversely near its proximal end (Figs 3C, 15C). The tooth attachment is clearly pleurodont, with the labial part of the crown ankylosed to a high, obliquely sloping pleura (Fig. 15C). The lingual side of the tooth base is ankylosed to a poorly developed basal plate. The longest preserved tooth is ~0.6 cm.

The maxilla is a low, anteroposteriorly elongate element measuring ~7 cm. It is slightly deeper dorsoventrally in its middle third, which is poorly preserved. In lateral view, the anterior fragment of maxilla includes an elevated ascending process that meets the prefrontal and sets off a dorsally concave and medially curved anterior process (Fig. 3C). Although the ascending process is poorly preserved, its anterior and posterior ends can be recognized easily by the sharp break in slope that they make from the body of the maxilla. Additionally, the presence of the canal for the suborbital branch of the maxillary ramus of the trigeminal nerve (cranial nerve V₂) identifies definitively the position of the contact with the palatine, which in squamates provides a reliable indicator of the anterior border of the orbit. The ascending process in Sanajeh is long, extending for approximately half the estimated length of the maxilla (Fig. 3C). In this regard, it is distinct from Dinilysia and more advanced snakes, such as Wonambi, which retain a well-defined but short ascending process. Immediately posterior to the ascending process is the orbital margin, which in Sanajeh is short. The posterior portion of the maxilla is not preserved completely, but based on comparisons with Dinilysia, which it otherwise resembles, in addition to comparisons with its own dentary, the maxilla of Sanajeh would not have extended far beyond the position of the isolated fragment preserved in the block (Fig. 1).

Palatine:

Both the holotypic and referred specimens of Sanajeh indicus preserve the right palatine, which facilitates comparison of their relative size and shape (Figs 1, 2, 7). The palatine in the holotype is preserved only in part, found a near-life position (based on the relative positions of the openings for the suborbital branch of the maxillary nerve). The preserved part of the right palatine of the holotype corresponds to part of the suborbital palatine foramen, including its ventral and lateral walls and part of the dorsal wall, in addition to a small portion of the dentigerous process corresponding to its lateral margin (Figs 1, 2). The fragmented parts of the right palatine of the holotype are almost identical to the same structures present in the much better-preserved palatine of the referred specimen, allowing a direct comparison. The palatine of the referred specimen, in contrast, is nearly complete (Figs 7, 14). It was found nearly 17 cm from the posterior portion of the skull, which includes the braincase, skull roof, right pterygoid and partial right lower jaw. Segmentation of the right palatine and pterygoid of the referred specimen allowed for their digital re-articulation and identified surfaces of mutual contact (Fig. 14). The referred right palatine is well preserved and lacks only part of the anteromedial portion of the choanal process. In the description below, we rely on the referred specimen for most of the salient anatomical features; we refer to the holotypic specimen for its relationship to the maxilla.

The palatine is an anteroposteriorly elongate element (~2.4 cm) that contacts the maxilla laterally, the pterygoid posteriorly and the vomer anteromedially. In dorsal view, three processes of the palatine are visible: a broad, anterolaterally directed maxillary process, a posteriorly directed pterygoid process and a partly preserved choanal process (Fig. 14).

Like Dinilysia, Sanajeh lacks a typical dentigerous process of the palatine, but unlike the former, it has a conspicuous row of palatine teeth that reaches the anterolateral tip of the maxillary process, running along the lateral margin of the pterygoid process and anteromedial margin of the maxillary process (Fig. 14A, C, E). In contrast, Dinilysia retains a row of teeth on the palatine that is restricted to the lateral margin of the pterygoid process and, like Lanthanotus Steindachner, 1878 (Fig. 14B, D, F), does not extend to the maxillary process anteriorly.

The palatine bears a total of 13 complete alveoli, where at least eight small (~0.2 cm), recurved teeth are still preserved. The three posteriormost preserved teeth are positioned lateral to the pterygoid articulation facet on the ventral surface of the palatine. Palatine and pterygoid teeth are the same size and collinear.

The choanal process of the palatine is preserved only in part, but it is notably thick dorsoventrally (0.3 cm), especially in comparison to the more laminar pterygoid process (Fig. 14). In this way, the palatine of Sanajeh resembles that of Dinilysia and probably retained a broad contact with the vomer.

The maxillary process of the palatine is long anteroposteriorly (~1.5 cm) and broad transversely (~1.0 cm). It is longer than the incompletely preserved choanal, which, when complete, would have been likely to exceed it in length (Fig. 14). The posterior portion of the palatine is dorsoventrally high and traversed by the suborbital foramen, which enters its posterior wall and exits posterolaterally near its posterior contact with the maxilla (Fig. 14E). The posterior entrance of the foramen is twice as broad transversely as it is tall dorsoventrally (0.4 cm × 0.2 cm). The holotypic palatine matches the referred specimen, but it is incompletely preserved. The suborbital foramen is open medially in the holotype, owing to damage to its dorsal wall. The palatine overlaps the maxilla dorsally in a broad contact (Figs 1, 2).

The pterygoid process of the palatine is a broad flange of bone that bears the articular surface for the pterygoid on its ventral surface. The main part of the pterygoid articulation in Sanajeh is planar, rectangular (0.4 cm × 0.5 cm) and almost horizontally oriented (Fig. 7B). Posterior and slightly lateral to this planar contact is a narrow, V-shaped slit that receives a corresponding projection from the pterygoid (Figs 8, 10, 14A). In this way, the condition in Sanajeh resembles that in Dinilysia and in non-ophidian lizards (Fig. 14B), in which the palatine broadly overlaps the pterygoid. It differs from that of Wonambi, Yurlunggur, pachyophiids and more advanced extant snakes, which have a U-shaped groove on the palatine that indicates a vertically oriented, complex tongue-and-groove suture with the pterygoid. Sanajeh also differs from extant snakes in the position of the palatopterygoid tooth row with respect to the axis of the palatine process of the pterygoid. In extant snakes, the palatal tooth row is aligned with this axis and positioned in the middle of the palatine process, but in Sanajeh the tooth row is shifted more laterally (Fig. 14C).

Pterygoid:

The right pterygoid is preserved only in the referred specimen of Sanajeh. It is nearly complete, lacking only the distal portion of its quadrate ramus (Figs 10, 14). The pterygoid has been displaced anteriorly < 1 cm and rotated medially relative to its original position. It remains partly embedded in matrix, but most of the crucial details of its anatomy can be observed directly on the specimen (Figs 8–11) and on the digitally segmented CT scan images (Fig. 14).

The pterygoid is an elongate element (~5 cm) that has an anteriorly directed palatine ramus, a laterally directed ectopterygoid articulation, a medially directed basipterygoid articulation and a posteriorly directed quadrate ramus. The pterygoid is broadest near mid-length (1.1 cm), where the basipterygoid and ectopterygoid articulations extend medially and laterally. Posterior to this point extends the quadrate ramus, which has the form of a flattened rod (0.4 cm × 0.2 cm). The quadrate ramus is incomplete, and it is unclear exactly how far it would have projected from its preserved end. The palatine ramus bears a broad articular surface for the palatine at its distal end (Fig. 8). This articular surface faces dorsally and is 0.7 cm long and nearly as broad. There is a small, elevated ridge positioned lateral to the main articular facet that fits into a corresponding notch in the palatine (Fig. 14A). The anteriormost pterygoid tooth ends slightly posterolateral to the broad articular facet, with the pterygoid tooth row lining up with the palatine tooth row (Fig. 14C). The pterygoid tooth row of Sanajeh follows the medially concave curvature of the bone, as in Lanthanotus (Fig. 13D) and Dinilysia (Estes et al., 1970).

There are 11 teeth preserved and evidence of four additional alveoli, indicating that there were 15 teeth in the pterygoid. The teeth are small (0.3 cm long), conical and recurved.

Circumorbital complex

Based on the arguments and observations presented in the Discussion section below, we suggest the following primary homology statements for the dorsal, posterodorsal and posteroventral orbital elements in fossil and extant snakes: (1) the ‘supraorbital’ is the neomorphic dorsal orbital element known to occur in Calabaria Gray, 1858, Loxocemus Cope, 1861 and pythonids; (2) the postfrontal is the posterodorsal (pachyophiids, Dinilysia, Najash and Yurlunggur) or posterior (extant snakes) orbital element intimately associated with the skull roof at the frontoparietal suture; (3) the jugal is the posteroventral element known to occur in the fossil pachyophiids, Dinilysia, Najash and Yurlunggur, inferred to be present in Sanajeh (see postorbital description), but considered to be absent in extant alethinophidian snakes; and (4) the postorbital bone is currently known only in Sanajeh among extinct and extant snakes.

Thus, up to six bones can be associated with the orbital margin in snakes: prefrontal, postfrontal, supraorbital, jugal, postorbital and maxilla. The prefrontal, postfrontal and maxilla are present in both extant and fossil snakes. The supraorbital is known to occur only in some extant snakes, whereas the jugal occurs only in the known fossil pachyophiids, Dinilysia, Najash and Yurlunggur. Below, we describe the bones bordering the upper part of the orbit as preserved in Sanajeh, which include the postfrontal and postorbital (see above for description of the maxilla).

Postfrontal:

The holotypic skull preserves a nearly complete right postfrontal (Fig. 4). The postfrontal was not found in its natural articulation with other skull roof bones (i.e. frontal and parietal). It was found in a vertical orientation appressed against the lateral surface of the maxilla, at the level of the posterior border of its palatine process (Figs 1, 2). There is a small, rod-like element adhered to its dorsal surface (Fig. 4).

Figure 4.

Postfrontal of Sanajeh indicus holotype GSU/GC/2903. Photographs and line drawings of right postfrontal in dorsal (A), ventral (B) and posteromedial (C) views. Hatched lines indicate broken bone surface, and parallel double-line pattern indicates matrix. Abbreviations: as f, articular surface for frontal; as p, articular surface for parietal; orm, orbital margin; vpr, ventral process; ?, unidentified bone.

The postfrontal is a relatively small, subtriangular element with an anteromedially directed apex and a posteriorly directed base. The medial edge, which would have contacted the frontal, is straight, and the lateral edge, which would have bordered the orbit, is convex. The posteriorly directed base is slightly angled and provides the articular surface for the parietal. The postfrontal is slightly longer anteroposteriorly than it is broad at its base (1.1 cm × 0.9 cm) and bears a slightly dished dorsal surface; that is, it is slightly concave dorsally and convex ventrally. This dished surface contains the small, rod-like element, which we interpret to be a fragment of the right postorbital (see below in Postorbital and squamosal).

In dorsal view, the orbital margin of the postfrontal is convex laterally and measures 1.1 cm. Its ventral surface tapers toward its lateral edge, ending in a paper-thin, textured orbital margin. The postfrontal is anteroposteriorly short (1.1 cm), implying that the orbit was similarly abbreviate. This assessment is consistent with the short (0.6 cm) orbital margin identified on the maxilla.

The frontal articulation, which has an approximately anteroposterior orientation, is straight and occupies much of the length of the postfrontal. This articulation forms an angle of slightly > 90° with the parietal articulation, which itself consists of two segments that meet at an oblique angle. Near the point at which the frontal and parietal articular surfaces converge is a distinctive C-shaped articular surface extending below the plane of these other articulations. This C-shaped facet received a small process of the parietal, which can be seen in the well-preserved left parietal on the referred skull. There does not appear to be a flange that underlapped the parietal and frontal, as there is in Yurlunggur.

The frontal articulation and orbital margin meet anteromedially to form the apex of the postfrontal. There is no sign of a more transversely oriented articular surface anteriorly, which suggests that the prefrontal did not contact the postfrontal, nor is there a ventrally directed process that would have contacted the jugal.

Postorbital and squamosal:

The referred specimen of Sanajeh possesses a complete upper temporal bar formed by the postorbital and squamosal bones, which are nearly complete and preserved in what appears to be close to their natural position (Figs 8–11, 13, 15). We describe these two bones together. The postorbital appears to be complete, or nearly so, but the squamosal is broken posteriorly and might have been slightly longer. The holotypic skull preserves only an eroded fragment of the left half of the skull roof, which bears no evidence of temporal arcade bones. Nevertheless, the concave dorsal surface of the postfrontal is overlain by a small rod-shaped element that might represent a fragment of the anterior postorbital (Fig. 4). This fragment is appropriate in size and shape and is in the position expected for the postorbital.

The two upper temporal arch elements are unequal in size and different in shape. The postorbital is shorter and more strap-like than is the squamosal, which is longer and more rod-like. The postorbital is 2.2 cm long, 0.3 cm tall and < 0.1 cm transversely; the squamosal is 3.2 cm long and 0.2 cm tall and wide. They overlap one another by ~0.8 cm as preserved. The postorbital and squamosal are not tightly articulated, but there is evidence for their mutual articulation based on facets on both elements that are best viewed in 3D reconstructions based on micro-CT scan images (Fig. 13). The two bones retain their surface of contact adjacent to each other, showing that they are not individually displaced anteriorly or posteriorly from their natural positions. The CT scan sections in the horizontal and transversal planes also help to clarify the very distinct nature of the postorbital and squamosal elements (Fig. 15A–C). The curvature of the two bones individually and in contact follows that of the temporal region, which supports our interpretation of their identity. The anterior end of the postorbital slightly expands laterally to form a blunt, dorsoventrally tapering projection that is likely to have been clasped dorsally by the postfrontal and ventrally by the jugal (not preserved), as suggested by the presence of flat areas of contact on both dorsal and ventral surfaces (Fig. 13A). There is no evidence of a direct bony connection between the posterior end of the squamosal, which is not preserved, and the paroccipital region of the skull. Instead, the squamosal was probably seated loosely near the dorsal head of the quadrate, held in place by soft tissue connections in the shallow gutter formed on the dorsal surface of the suspensorium along the parietal–otooccipital contact. Extant anilioids and ‘scolecophidians’ retain an upper temporal tendon that seems to represent the remnant of the upper temporal bar of Sanajeh, in both cases receiving the more lateral adductor mandibulae muscle bundle (Rieppel, 1980; Zaher, 1994a).

Braincase and suspensorium

We describe the braincase and suspensorium together below because they form a functional complex in snakes.

Supratemporal:

A complete left supratemporal is preserved in the referred skull (Figs 8–11, 15); no trace of the supratemporal is preserved with the holotypic skull. Although complete, the supratemporal in the referred skull is exposed mostly in lateral view.

The supratemporal is an anteroposteriorly elongate element (2.8 cm) that contacts the prootic ventrally, the parietal dorsally and the otooccipital medially (Figs 8–11, 15). It projects posteriorly and slightly laterally and, together with the parietal, prootic and otooccipital, it forms an elongate suspensorial process. The anterior half of the supratemporal is sandwiched between the prootic and parietal (Figs 8–11, 15). It is convex dorsally and flat ventrally, tapering toward a pointed anterior tip. Immediately posterior to the prootic, the supratemporal expands to twice its dorsoventral height to cover almost completely the entire lateral surface of the otooccipital. At that level, the contact between the prootic and the supratemporal is interrupted by a lateral expansion of the otooccipital, which is sandwiched between the former two bones on the ventrolateral surface of the suspensorium (Figs 10, 15A). Posterior to this point, the supratemporal gradually expands dorsally before tapering toward its posterior end. As a result, the posterior half of the supratemporal also has a convex dorsal surface. The supratemporal has a vertically oriented contact with the otooccipital posteriorly to form the distal part of the suspensorial process, extending ~1 mm beyond it posteriorly. The posterior portion of the supratemporal is slightly convex laterally and, together with the portion of the otooccipital exposed laterally on the suspensorial process, it would have received the quadrate head. In Sanajeh, a lateral contact between the prootic and otooccipital is retained below the supratemporal, with the supratemporal expanding posteriorly and ventrally to cover most of the otooccipital laterally (Fig. 10). The posterior end of the squamosal, freed from its contact with the dorsal head of the quadrate by the expansion of the supratemporal, sits on top of the suspensorium and was probably attached loosely through connective tissues to the shallow gutter formed along the parietal–otooccipital contact (Fig. 8).

Parietal:

Remnants of the parietal are preserved on both holotypic and referred skulls. A fragment of the left side of the parietal is present in the holotypic skull, but it is poorly preserved. The parietal of the referred specimen is not completely preserved, but the left side is nearly complete, hence we can reconstruct its shape with confidence (Figs 8–11).

The parietal is a dorsally arched element that is much longer than it is broad (4.9 cm × 2.0 cm), owing to its elongate supratemporal processes. The parietal contacts the frontal anteriorly, the postfrontal anterolaterally, the parabasisphenoid ventrally and the supratemporal, prootic and supraoccipital posterolaterally. The parietal has a conspicuous median sagittal crest that is elevated ~2 mm from its surface (Fig. 8). The anterior face of the parietal bears a vertically oriented articular surface that would have contacted both the frontal and postfrontal bones (Fig. 11). This vertical surface is nearly flat except for a small median depression that received a process of the frontal; paramedian depressions that received the postfrontal and a small lip immediately lateral to this would have fitted into a groove in the postfrontal. The total height of this articular surface is 0.6 cm.

The parietal contacts the parabasisphenoid and prootic anterolaterally. The parabasisphenoid articulation can be observed only in the holotypic skull, in which it is concave ventrally and smooth. The parietal and prootic contact each other on the anterior margin of the opening for cranial nerve V and stay in contact posteriorly until they are separated by the supratemporal bone (Fig. 10). The prootic forms most of the margin of the opening for cranial nerve V, with only a small portion formed by the parietal.

The parietal has an elongate supratemporal process that extends posteriorly well beyond the foramen magnum, tapering to a pointed tip positioned at the level of the posterior extreme of the prootic (Fig. 8). The length of the supratemporal processes of the parietal is unique among snakes. In this region, the parietal is sandwiched between the supratemporal laterally and the otooccipital ventrally and medially (Fig. 15D). Along the length of its contact with each of these elements, the parietal is raised to form a gentle ridge. Between these two posteriorly converging ridges is a small sulcus, which we hypothesize received the posterior end of the squamosal (Fig. 8). Note that the squamosal is displaced only a short distance (3 mm) from this sulcus.

Prootic:

The prootic is preserved in both holotypic and referred specimens. The holotypic left and right prootic are eroded; the right preserves only its posteromedial surface, which received the stapedial footplate (see below), and the left preserves only its most interior wall and lacks its superficial structures. The referred braincase preserves a fragment of the right prootic and has a nearly complete, well-preserved left prootic that lacks only part of its ventral edge, which originally contacted the parabasisphenoid (Figs 8–10).

The left prootic is roughly tetraradiate in lateral view, with two processes extending anteriorly to form the opening for the single trigeminal foramen of cranial nerve V and two processes extending posteriorly around the fenestra ovalis. The surfaces of the two anterior processes are clearly abraded, having lost their original shape. However, the longer and broader dorsal anterior (‘alar’) process retains its contact with the parietal through a broadly rounded, dorsoventrally expanded suture that overlaps the descending flange of the parietal (Fig. 10), much like in Najash rionegrina (Zaher et al., 2009). The highly abraded ventral anterior process is shorter and narrower, abutting on a small projected surface of the parietal that might also result from the process of abrasion of this bone. The two anterior processes nearly contact one another on the anterior margin of the opening for cranial nerve V, separated by only a small margin by the parietal.

The prootic expands posteriorly to embrace the fenestra ovalis, forming its anterior, dorsal and ventral margins. The posterior extension of the prootic that surrounds the fenestra ovalis ventrally is relatively short (~1.0 cm), whereas the posterodorsal process (Oelrich, 1956) surrounding it dorsally is much longer (1.7 cm) and robust (Fig. 10). The posteroventral extension of the prootic that articulates with the basioccipital and forms the part of the lower margin of the fenestra ovalis is abraded and has lost its surface. The surface within which opens the foramen for the facial nerve, posterior to the trigeminal foramen, is also damaged, and its opening occurs in the abraded surface of the prootic. A smooth surface contact and posterior opening of the Vidian canal are not visible either, because this region of the skull was originally exposed on the surface of the block. The posterodorsal process of the prootic and its contacts are mostly intact, extending well beyond the fenestra ovalis and angling slightly dorsal direction toward its sharply blunted end, which abuts the supratemporal and receives a small part of the otooccipital (Fig 10). Dorsal and anterior to the stapes, the prootic forms a well-developed crista prootica that overhangs the edge of the stapedial footplate, projecting lateral to the latter (Fig. 10).

Otooccipital:

The otooccipitals are fragmentary in the holotype, and only part of the left element is preserved in the referred specimen. Nevertheless, this bone reveals an unexpected morphology that provides new insight into the evolution of the suspensorium in snakes. The otooccipital contacts the parietal, supratemporal, prootic and quadrate. In life, it would also have contacted the squamosal, which is now slightly displaced from its original position and lacks its posterior tip, and the supraoccipital, which is eroded in the referred specimen.

The otooccipital is an elongate element that projects posteriorly and slightly laterally, forming an expanded, posteriorly concave paroccipital process that receives the prootic ventrolaterally and the supratemporal dorsolaterally (Figs 8–10, 15). The left otooccipital of the referred specimen is visible laterally, sandwiched by the prootic and supratemporal (Figs 10, 15A). This small surface of the otooccipital must have contacted the quadrate in life (as confirmed by articular surfaces preserved in the region), retaining the condition present in lizards, in which the crista parotica receives the dorsal condyle of the quadrate (Bellairs & Kamal, 1981; McDowell, 2008). The left otooccipital of the referred specimen lacks the proximal portion, eroded along with the supraoccipital and its right counterpart surrounding the foramen magnum. Only the medial, inner braincase surface of the left otooccipital remains from that region of the bone, where the apertures for the vagus and glossopharyngeal nerves are visible.

Basioccipital:

Only a small portion of the left part of the basioccipital is present in the referred specimen. It is completely worn on its external surface, but its medial surface is preserved (Fig. 9). The apertura medialis recessus scalae tympani is visible as an expanded, slightly U-shaped commissure pressed to the sutural contact with the ventral margin of the otooccipital.

Supraoccipital:

The external surface of the supraoccipital is completely lost, and the bone is visible only in internal (medial) view (Fig. 8). It is triangular, tapering ventrally, with anterior and posterior margins contacting the prootic and otooccipital, respectively, and dorsally enclosing the posterior semicircular canal. Its concave internal surface bears a conspicuous foramen endolymphaticum.

Stapes:

Stapes are preserved with both holotypic and referred specimens. In both cases, a large footplate is present. The footplate is preserved only in part in the holotypic skull (incorrectly identified as a supratemporal by Wilson et al., 2010; see above), but both the footplate and columella (stapedial shaft) are completely preserved in the referred specimen (Figs 9, 10). The referred stapes was found in near-perfect anatomical position within the fenestra ovalis; the holotypic stapes was displaced posteriorly from its original position (Zaher et al., 2017).

The footplate of the referred specimen is relatively large (1.1 cm × 0.8 cm) and elliptical in shape. The columella is nearly 2 cm long and extremely narrow at mid-length (~2 mm in diameter). The base of the columella is massive, and its shaft projects posterolaterally from the posterior half of the footplate, strongly curving immediately after it passes the posterior edge of the fenestra ovalis to take a posteriorly directed orientation. The columella expands to nearly twice the diameter of its shaft toward its distal end, terminating in a broad, tongue-shaped process.

Mandible

The holotypic specimen of Sanajeh preserves elements of both mandibles. The left mandible is heavily damaged and has not been re-prepared since it was described originally. The right mandible is almost complete and, although damaged, it was removed for further preparation (Figs 5, 6). It is composed of the dentary, coronoid, angular and anterior portion of the compound bone. A small fragment of the splenial is present along the Meckelian groove in the midline of the medial surface of the dentary (Fig. 5B). The referred specimen also preserves a portion of both mandibles. The right mandible was found in close association with the braincase and skull roof (Figs 8–11), and a partial left mandible was found nearby (Fig. 12). All bones of the mandible are preserved to varying degrees, including the dentary, splenial, angular and compound bone. Two fragmentary teeth are preserved in the holotypic right dentary.

Figure 5.

Mandible of Sanajeh indicus holotype GSU/GC/2903. Photographs and line drawings of right mandible in lateral (A) and medial (B) views. Hatched lines indicate broken bone surface; parallel double-line pattern indicates matrix; and dashed lines indicate reconstructed portions of incomplete elements. Abbreviations: a, angular; al5, alveolus for tooth 5; asfo, anterior surangular foramen; c, coronoid; cb, compound bone; c fa, coronoid facet; d, dentary; d4, dentary tooth 4; d6, dentary tooth 6; d11, dentary tooth 11; Mc, Meckelian groove; mfo, mental foramen; pmfo, posterior mylohyoid foramen; sp, splenial; sp fr, splenial fragment; ?, unidentified bone.

Dentary:

A nearly complete dentary is preserved with the holotypic right mandible (Figs 5, 6), and the posteroventral process and compound bone articulation are preserved with the right mandible of the referred specimen (Figs 8–12). Together, these two specimens account for most, if not all, of the salient morphology of the dentary. Owing to coverage by matrix and adjacent bones, portions of the dentition and the medial aspect of the dentary could not be observed. The dentary is an elongate element (~64 mm long × 12 mm tall × 7 mm wide) that is gently bowed laterally and ventrally. The preserved curved anterior tip of the dentary bears an expanded, slightly curved, anteroposteriorly directed facet of contact for the opposite dentary (Fig. 6). The surface is rugose, suggesting a tight ligamentous contact conferring limited mobility between the dentaries. The Meckelian groove is open throughout the preserved length of the dentary, extending ventrally from the tip to the level of the anterior one-third of the bone (Figs 5B, 6). From that point, it attains a more ventromedial position and deepens dorsoventrally. The splenial articulates within the Meckelian groove, where a few fragments attributable to it are preserved. The contact for the compound bone is located on the posterolateral surface of the dentary, in an elongate fossa framed by its posterodorsal and posteroventral processes. The posterodorsal process is preserved in the holotype and the posteroventral process in the referred specimen. A contact with the coronoid was probably present on the medial aspect of the dentary immediately behind the tooth row, but corresponding portions of the dentary and coronoid are not preserved well enough in the two specimens to determine. A prominent mental foramen opens at approximately mid-height in the holotypic dentary at the level of the third alveolus (Fig. 5A). A total of 11 teeth are definitively present on the dentary, although one or more additional smaller alveoli might be present distal to the last preserved 11^th dentary tooth.

Figure 6.

Dentary of Sanajeh indicus holotype GSU/GC/2903. Anterior portion of the right dentary in lateral (A), medial (B), ventral (C) and dorsal (D) views. Abbreviations: bp, basal plate; d, dentary; Mc, Meckelian groove; pl, pleura; sd, subdental shelf; sy, symphyseal facet; t, tooth.

Figure 7.

Palatine of Sanajeh indicus referred specimen GSI/GC/DD4. Photographs and line drawings of right palatine in dorsal (A) and ventromedial (B) views. Hatched lines indicate broken bone surface, and parallel double-line pattern indicates matrix. Abbreviations: as m, articular surface for the maxilla; as pt, articular surface for the pterygoid; ch pr, choanal process; pl1, palatine tooth 1; pl6, palatine tooth 6; pl13, palatine tooth 13; sbofo, suborbital foramen.

There are three nearly complete teeth and portions of several others preserved in the dentary of the holotype (Fig. 5). One complete tooth is preserved under the sediment in the right dentary of the referred specimen (Fig. 15C). The teeth are slightly recumbent and gently recurved, and they taper distally to a point from a base that is ~1 mm wide mesiodistally. Tooth implantation in Sanajeh is pleurodont and resembles the ‘modified alethinophidian type’ of implantation present in Najash (sensuZaher & Rieppel, 1999b), with teeth implanted in distinct tooth sockets formed by prominent interdental ridges that extend onto the lingual wall of the subdental shelf. The orientation of the sockets with respect to the pleura is typical of a labial pleurodont condition, in which the pleura is high and the lingual dental ridge is vertically directed. These ridges provide bony support for the ankylosis of the labial part of the crown to the high, obliquely sloping pleura; the lingual part of the crown is ankylosed to a poorly developed basal plate on the dorsal surface of the well-developed subdental shelf (Fig. 15C).

Compound bone:

The right compound bone is preserved in the holotypic specimen of Sanajeh (Fig. 5), and parts of both right and left compound bone are preserved in the referred specimen (Figs 8–12). Individually, these elements are incomplete, but together they preserve the complete length of the compound bone and many of its main features. However, owing to matrix cover and breakage, contacts with adjacent lower jaw elements are not apparent. The compound bone is a composite element comprising the fused prearticular, surangular and articular. In both holotypic and referred individuals, there is no trace of sutures among constituent elements. The compound bone forms the majority of the posterior portion of the mandible, contacting the dentary anteriorly, the angular medially and the coronoid anterodorsally, and articulating with the quadrate at its posterior end.

Figure 8.

Cranial bones of Sanajeh indicus referred specimen GSI/GC/DD4. Photograph (A) and line drawing (B) of partial skull, including braincase, skull roof and posterior portions of right palate and mandible in dorsal view. Hatched lines indicate broken bone surface; parallel double-line pattern indicates matrix; and dashed lines indicate reconstructed portions of incomplete elements. Abbreviations: as ect, articular surface for the ectopterygoid; as pl, articular surface for the palatine; as sq, articular surface for the squamosal; bo, basioccipital; c, coronoid; cb, compound bone; d, dentary; oto, otoccipital; p, parietal; po, postorbital; pr, prootic; pt, pterygoid; so, supraoccipital; sp, splenial; sq, squamosal; st, stapes; sut, supratemporal; t, tooth; ?, unidentified bone.

The anterior half of the compound bone is preserved in the holotype (Fig. 5). The ventral margin of the anterior surangular foramen is preserved as a sulcus on its dorsolateral surface, which is eroded in that region (Fig. 5A). The articulation between the compound bone and dentary is visible on the right mandible of the referred individual. The dorsoventral height of the compound bone diminishes gradually toward its anterior end, terminating in a blunted tip that fits within the space formed by the posterodorsal and posteroventral processes of the dentary. The posterior one-third of the right compound bone is preserved in the referred individual and in the holotype. The contact for the angular could not be identified, and a coronoid eminence of the surangular is absent. A well-preserved posterior portion of the left compound bone was found with the referred specimen (Fig. 12). This portion of the compound bone is rod-like, with a prominent medial flange immediately in front of the articular surface for the quadrate that corresponds to the expanded medial flange of the surangular, located between the posterior margin of the adductor fossa and the anterior margin of the articular surface for the quadrate. A similarly expanded, slightly concave horizontal surface of the surangular is present in Dinilysia, Varanus Merrem, 1820 and Wonambi (Scanlon, 2005; Zaher et al., 2012). In dorsal or ventral views, this flange gradually expands medially before tapering again toward its broken anterior margin. The articular facet for the quadrate is exposed in dorsal and medial views. The articular surface is broad and saddle shaped, convex transversely and concave anteroposteriorly (Fig. 12). Immediately posterior to the quadrate articular surface is a small retroarticular process that extends posteriorly. Both the anterior tapering surface, which would have been located at the posterior border of the mandibular fossa, and the posterior retroarticular process are eroded and incomplete. A prominent opening for the chorda tympani is present on the ventromedial aspect of the compound bone (Fig. 12D). A smaller opening, the posterior surangular foramen, is present on the lateral surface of the compound bone, slightly anterior to the articular surface (Fig. 12A).

Coronoid:

Portions of the coronoid are preserved in both the holotypic and referred specimens. The holotypic mandible preserves the most complete and informative fragment of the coronoid, which has been displaced a little posteriorly and medially out of position (Fig. 5). The referred specimen preserves a comparably poorer coronoid that is preserved near its original position atop the anterior portion of the compound bone, only contacting the posterodorsal edge of the splenial via its anterior process (Fig. 10). When complete, it would have extended slightly further anteriorly to reach the posterior edge of the dentary tooth row. The more complete, holotypic coronoid is roughly triradiate in medial or lateral view, with a prominent dorsal peak, an anterior facet that expands anteroventrally to form a triangular surface for insertion of the adductor externus superficialis musculature (Haas, 1973; Zaher, 1994a) and an expanded, slightly concave medial surface for the adductor externus profundus musculature (Fig. 5). The posterior margin is missing.

Splenial:

Fragments of the splenial were preserved in the more complete, intact right mandible of the holotype. The referred specimen of Sanajeh preserves an almost complete right splenial, which is preserved against the medial surface of the posteroventral process of the dentary (Figs 9, 10). As such, the splenial has been slightly rotated and translated ventrally from its presumed original position on the medial aspect of the dentary. It is preserved in contact with the coronoid posterodorsally and with the compound bone posterolaterally. This preserved part of the splenial includes its articular surface for the angular, which is a fairly low, blunt process that is flush with the ventral edge of the splenial. The vertically oriented articular margin with the angular is broad and forms a shallow, concave articular surface for the angular. Medially, the splenial is excavated by a deep notch that probably received an anterior medial prong projecting from the anterior medioventral margin of the angular, forming a complex articular contact similar to the one described in extant snakes (Rieppel & Zaher, 2000). There is a large anterior mylohyoid foramen, which is enclosed within an elongate fossa that extends anteriorly ~6 mm. Immediately anterior to this fossa, near the preserved anterior edge of the splenial, is a small, partly preserved anterior inferior alveolar foramen enclosed in a shallow fossa. The Meckelian groove was probably covered by the splenial.

Figure 9.

Cranial bones of Sanajeh indicus referred specimen GSI/GC/DD4. Photograph (A) and line drawing (B) of partial skull, including braincase, skull roof and posterior portions of right palate and mandible in ventral view. Hatched lines indicate broken bone surface, and parallel double-line pattern indicates matrix. Abbreviations: as bp, articular surface for the basipterygoid process; bo, basioccipital; c, coronoid; cb, compound bone; d, dentary; oto, otoccipital; p, parietal; po, postorbital; pr, prootic; pt, pterygoid; sp, splenial; st, stapes; sut, supratemporal; t, tooth; ?, unidentified bone.

Angular:

The angular is absent in the referred specimen but almost completely preserved in the holotypic right mandible, where it is partly obscured by a broken and displaced large fragment of the coronoid (Fig. 5B). It lacks its anterior articular surface. The angular is an elongate, strap-like element that measures 27 mm long by 4 mm deep. It is displaced posterodorsally from its original position within a subtriangular fossa on the ventromedial surface of the compound bone. A large foramen for exit of the posterior branch of the mylohyoid nerve is present toward the anterior end of the angular, set within a dorsoventrally shallow but anteroposteriorly elongate fossa. Apart from its articulation with the splenial, the angular maintains its dorsoventral height throughout its length. Its transverse thickness appears to taper posteriorly, such that its posterior end is approximately one-half the thickness at the splenial articulation.

Phylogenetic analyses

We investigated the affinities of Sanajeh within Toxicofera in analyses of our total evidence dataset using three different phylogenetic methods (MP, ML and BI). The three methods converged on a similar general topology, but there were important differences concentrated in five areas of the tree: (1) Gephyrosaurus Evans, 1980, Sphenodon Gray, 1831 and remaining squamates; (2) Coniophis Marsh, 1892, Tetrapodophis Martill et al., 2015 and remaining snakes; and more inclusive nodes within (3) Alethinophidia, (4) Anguimorpha and (5) Pleurodonta. Areas 1–3 were retrieved as unresolved polytomies in the MP consensus tree (Fig. 16), whereas areas 4 and 5 are represented by three fully resolved but discordant topologies involving the rogue taxa Gobiderma Borsuk-Białynicka, 1984, Polychrus Cuvier, 1817, Proplatynotia Borsuk-Białynicka, 1984, Pseudopus Merrem, 1820 and Shinisaurus Ahl, 1930. In the discussion below, we focus on the topology in the strict consensus tree of our MP analysis (Fig. 16) as a scaffold for discussion of interrelationships within Ophidia and character support for particular clades (see Supporting Information, Files S1, S5 and S6), noting major differences from ML and BI results. Numbers in parentheses represent bootstrap support values for the MP and ML trees and posterior probability support values for the BI tree. Bootstrap and posterior probabilities < 70% and 0.80, respectively, are not shown (–). For detailed visualization and comparisons of topological results from the three analyses, please see the Supporting Information (Figs S3–S5).

Toxicofera is retrieved in all three analyses with Pan-Serpentes as the sister group of the clade formed by Iguania and Anguimorpha (100/85/–). All three analyses recover a monophyletic Iguania (100/100/1.0) supported by 15 unambiguous synapomorphies, with both Acrodonta and Pleurodonta robustly supported as sister groups (Supporting Information, File S1).

Anguimorpha is a statistically robust clade in both MP and ML analyses (100/83/0.9), supported by ≥ 17 unambiguous synapomorphies (Supporting Information, File S1). Marine mosasaurians are unambiguously nested within Anguimorpha (contraReeder et al., 2015), either as the sister group of remaining anguimorphs (BI analysis) or nested within them (ML and MP analyses). This robust result indicates that the mosasaurian ‘wide gape’ evolved independently from snakes. Anguimorpha includes four monophyletic groups whose interrelationships differ among analyses: Mosasauria (87/99/1.0), Varanidae (100/89/0.99), Anguioidea (99/100/–; including extant Anguidae and Xenosauridae) and Diploglossa (88/–/–; including Anguioidea and Helodermatidae). Additionally, hypothesized affinities of Gobiderma, Proplatynotia and Shinisaurus differ among analyses. A palaeoanguimorphan clade composed by Shinisaurus and varanids, often retrieved in molecular studies (Vidal & Hedges, 2005), is rejected by the MP analysis and recovered with no statistical support in both ML and BI analyses. Both MP and ML trees recover a clade formed by mosasaurians and varanids (Platynota sensuCamp, 1923), with statistical support only in the latter analysis (Fig. 16; Supporting Information, Fig. S1); mosasaurians are resolved as the sister group of all remaining anguimorphs in the BI tree (Supporting Information, Fig. S2). Monstersauria (Conrad et al., 2011), including Gobiderma, Proplatynotia and helodermatids, is recovered as a clade in both MP and ML analyses, but in the BI analysis Gobiderma and Proplatynotia are recovered as stem varanids (Conrad, 2008). Shinisaurus is resolved alternatively as the sister group of Diploglossa (MP analysis; Fig. 16), in a clade formed by mosasaurs and varanids (ML analysis; Supporting Information, Fig. S1) and in a clade formed by Gobiderma, Proplatynotia and varanids (BI analysis; Supporting Information, Fig. S2). Anguimorpha is supported by 17 unambiguous synapomorphies in our MP analysis (Supporting Information, File S1).

Pan-Serpentes is retrieved as a statistically robust clade (100/99/1.0) supported by 17 unambiguous morphological synapomorphies. The Early Cretaceous Tetrapodophis amplectus Martill et al., 2015 and Late Cretaceous Coniophis precedens Marsh, 1892 (Longrich et al., 2012) are recovered as the earliest-diverging members of Pan-Serpentes (sensuHead et al., 2020), but their interrelationships with other snakes cannot be resolved unambiguously. Tetrapodophis and Coniophis form a moderately supported clade (79/0.96) in the ML and BI analyses (Supporting Information, Figs S1, S2), but are in a polytomy at the base of Pan-Serpentes in the MP tree (Fig. 16). Sanajeh and the Late Cretaceous snakes Najash and Dinilysia are successive sister taxa to crown-group Serpentes (sensuHead et al., 2020), recovered with robust statistical support (100/97/1.0) based on 20 unambiguous synapomorphies.

Within crown-group Serpentes, ‘scolecophidians’ are recovered as a paraphyletic assemblage, with anomalepidids clustering as the sister group of alethinophidian snakes. This arrangement finds statistical support only in the ML and BI trees (96/1.0), being supported by one unambiguous synapomorphy in the MP analysis (Supporting Information, File S1). Alethinophidia is robustly supported (100/97/1.0), but relationships within it are not fully resolved (Fig. 16). All three analyses retrieve the following clades: the Australian Cenozoic ‘madtsoiids’ Yurlunggur and Wonambi, the Cretaceous marine Pachyophiidae and the extant Aniliidae, Tropidophiidae, Uropeltoidea, Pythonoidea, Booidea, Bolyeriidae and Caenophidia (including Acrochordoidea and Colubroides). Pachyophiids and the Australian ‘madtsoiids’ form a statistically supported clade resolved as the sister group of the remaining alethinophidians in both ML and BI analyses (78/1.0). Amerophidia (sensuVidal et al., 2007), which includes extant Anilius Oken, 1816 and tropidophiids, is retrieved in both ML and BI analyses as the sister group of all remaining extant alethinophidians, but with ambiguous support (–/1.0). The monophyly of Amerophidia is not supported in the MP analysis, which leaves Anilius and tropidophiids in a basal alethinophidian polytomy that includes Australian ‘madtsoiids’, pachyophiids and a statistically unsupported clade including uropeltoids, pythonoids, booids, bolyeriids and caenophidians (Fig. 16). Interrelationships between these clades show little concordance among the three analyses. The clade Constrictores, recently defined by Georgalis & Smith (2020) as encompassing pythonoids, booids and bolyerioids, is retrieved in the ML and BI trees with ambiguous support (92/–; Supporting Information, Figs S1, S2). The MP tree recovers pythonoids, booids and bolyeriids as successive sister groups of Caenophidia with no significant statistical support (< 70%). Uropeltoidea, in contrast, is recovered as the sister group of the clade formed by Pythonoidea, Booidea, Bolyeriidae and Caenophidia in the MP tree; of Constrictores (i.e. Pythonoidea, Booidea and Bolyeriidae) in the BI tree; and of Caenophidia in the ML tree. None of these hypotheses receives minimum statistical support. The validity of ‘Constrictores’ as a natural group remains uncertain, although weak support has been found in previous phylogenetic analyses (Burbrink et al., 2020), in the ML and BI analyses presented here, and in previous morphological descriptions (Scanferla et al., 2016; Smith & Scanferla, 2016, 2021; Scanferla & Smith, 2020; Zaher & Smith, 2020; Georgalis et al., 2021). Constrictores is independently supported by at least one morphological character not in our dataset: the presence of a specialized slip of the superficial parallel fibres of the adductor medialis muscle that attaches via a fascia of its own to the lateral surface of the compound bone (Zaher, 1994a, b). This muscular arrangement is unique to Constrictores, being present in the enigmatic bolyeriids and ecophenotypically cryptozoic and fossorial constrictores (e.g. Calabaria, charinaids, erycids and ungaliophiids; Frazzetta, 1971; Zaher, 1994a, b). Aniliids, uropeltoids, tropidophiids and caenophidians do not possess this muscular specialization (Zaher, 1994a, b; Georgalis & Smith, 2020; Zaher & Smith, 2020).

DISCUSSION

The intermediate morphology of the skull of Sanajeh indicus revealed by the new specimen and re-preparation of the holotype has important implications for the interrelationships of basally diverging snakes, the evolution of wide-gaped feeding and the identity of the bones bordering the orbit in snakes.

Phylogenetic interrelationships of stem snakes

Snakes were traditionally considered to be close relatives of platynotan anguimorphs (Camp, 1923; McDowell & Bogert, 1954), but recent molecular studies have recovered them in a clade with anguimorphs and iguanians (Toxicofera; Vidal & Hedges, 2005). An analysis by Reeder et al. (2015), which used a morphological and molecular dataset combining the matrices from Gauthier et al. (2012) and Wiens et al. (2012), found six unambiguous morphological synapomorphies supporting Toxicofera. Although this clade is statistically robust in all molecular analyses, interrelationships among its three major groups are not resolved unambiguously by molecular evidence, with anguimorphs tending to form a poorly supported clade with iguanians, instead of clustering with snakes (Vidal & Hedges, 2005; Wiens et al., 2010; Reeder et al., 2015; Zheng & Wiens, 2016; Streicher & Wiens, 2017; Burbrink et al., 2020). It is noteworthy that Northcutt (1978) identified at least six characters in the forebrain and midbrain shared by varanids and iguanians that corroborate the molecular signal supporting that clade. These unexpected morphological similarities led the author to propose the name Dracomorpha for this group (which also included teiids). One possible explanation is that divergences between iguanians, anguimorphs and snakes are much older than previously thought, and key early fossils representing these deep splits are still lacking, as suggested by the long ghost lineage interval of almost 50 Myr separating extant Serpentes from the other two toxicoferan clades in a recent phylogenomic tree (Burbrink et al., 2020).

Among extinct toxicoferans, mosasaurs have been hypothesized to be closely related to snakes, either as their sister group or as a paraphyletic lineage leading to snakes (Lee, 1997; Palci & Caldwell, 2010). However, these hypotheses have been contested in view of conflicting evidence (Rieppel & Kearney, 2001) and are weakened by the absence of a confirmed record of mosasaurs in the Jurassic (Polcyn et al., 2014). Recently, Caldwell et al. (2015) argued that ParviraptorEvans, 1994 (Evans, 1996; Broschinski, 2000) and ‘parviraptorids’ (Caldwell, 2020) represent at least four distinct lineages that extend the fossil record of snakes ~70 Myr, well into the Jurassic, thus appearing to corroborate the deep molecular divergence dates for the group (Table 1). ‘Parviraporids’ have had a complex taxonomic history, attributable, in part, to differential inclusion of disarticulated and associated materials collected from the Middle and Upper Jurassic of England, Portugal and the USA. Parviraptor was recovered within Gekkonomorpha by Conrad (2008), who scored both the associated cranial and postcranial elements originally attributed to the known specimens in an expanded phylogenetic analysis of squamate relationships. The postcranial remains were excluded in the analysis by Caldwell et al. (2015), and thus far there has been no reconciliation of the previous hypothesis of a mosasaurian–snake clade with the parviraptorid–snake hypothesis (Caldwell et al., 2015; Simões et al., 2018; Garberoglio et al., 2019a; Caldwell, 2020). New associated materials from the Middle Jurassic Kilmaluag Formation of Skye, currently under study (Panciroli et al., 2020), might offer an opportunity to resolve the affinities of ‘parviraptorids’.

Table 1.

Estimated dates of divergence (in millions of years) for the main nodes in Toxicofera from Jones et al. (2013), Pyron (2017), Zheng & Wiens (2016) and the Bayesian total evidence tip and node calibrated analysis from the present study

Node	Jones et al. (2013)	Pyron (2017)	Zheng & Wiens (2016)	Present study
Lepidosauria	242 (238–249.5)	244.5 (239.0–249.5)	277.6	244.06 (238.73–249.59)
Squamata	193 (176–213.2)	186.8 (172.3–202.7)	205.1	–
Toxicofera	–	–	184.6	180.30 (168.38–191.28)
Iguania	135.8 (116.7–152.0)	129.0 (115.3–141.4)	168.2	136.66 (127.31–146.89)
Anguimorpha	129.5 (128.1–134.2)	113.4 (101.4–126.1)	114.1	148.78 (133.72–165.75)
Pan-Serpentes	–	–	–	160.45 (146.82–171.84)
*Ophidia		121.8 (118.4–142.4)*	–	139.46 (129.79–147.92)
Serpentes	109.6 (81.1–137.0)	–	128.1	123.88 (116.73–131.62)
Alethinophidia	–	–	92.7	111.75 (106.68–123.15)
Caenophidia	–	–	80.6	67.62 (62.16–72.91)
Colubroides	–	–	75.2	59.74 (54.86–66.77)

Values in parentheses represent the confidence interval as estimated in each study.

*Ophidia is equivalent to Serpentes in the study by Pyron (2017).

The four-legged Cretaceous serpentiform squamate Tetrapodophis amplectus has emerged as possibly the earliest snake (Martill et al., 2015; Zaher & Smith, 2020). Recently, Paparella et al. (2018) and Caldwell (2020: 62) hypothesized that Tetrapodophis is not closely related to snakes, suggesting that it is a ‘non-ophidian pythonomorph’ nested within a mosasaur + snake clade. More recently, Caldwell et al. (2021) provided a fuller revision and phylogenetic scoring of the anatomy of Tetrapodophis, concluding that it is a dolichosaurid pythonomorph. Our scorings of Tetrapodophis were based on first-hand observations of the original specimen, and our interpretations of the anatomy and attendant scoring disagree significantly with those of Caldwell et al. (2021), as do our phylogenetic results. Tetrapodophis was recovered here as a basal member of Pan-Serpentes, having no special affinities with mosasaurs (Fig. 16), as recently suggested (Zaher & Smith, 2020). Inclusion of Tetrapodophis within Pan-Serpentes is statistically robust (100/99/1.0) and supported by 17 unambiguous morphological synapomorphies, including the presence of recurved, needle-like teeth ankylosed to shallow sockets formed by lingually expanded interdental ridges (Martill et al., 2015) and zygosphenes projecting distinctly above the prezygapophyseal articulation (Supporting Information, File S1). Tetrapodophis appears to fill, in part, the temporal and morphological gaps that persist between four-legged squamates and the clade that includes hindlimbed stem snakes and the crown clade. This latter group, formed by Dinilysia, Najash, Sanajeh and crown-group Serpentes, is here called Ophidia and supported in our analysis by 13 unambiguous synapomorphies, including a premaxilla with an arched alveolar margin, a suborbital ramus of the maxilla extending posteriorly to the orbit, subdivided synapophyses and the complete loss of forelimbs and pectoral girdle (Supporting Information, File S1).

The evolution of macrostomy in snakes

Sanajeh indicus is the first known ophidian to preserve two key transitional osteological features: an upper temporal bar formed by the postorbital and squamosal; and a paroccipital process contacting the quadrate below the supratemporal. This intermediate morphology is crucial information for understanding the major modifications related to the origin of the wide-gaped condition (macrostomy) in extant snakes.

Enclosure of the posterior braincase wall in snakes

The posterior skull of the referred specimen of Sanajeh illustrates a key transition that led to the enclosure of the posterior braincase wall, loss of the upper temporal bar and expansion of the supratemporal as the main suspensorium element in snakes.

In typical snake outgroups, such as Varanus and Iguana Laurenti, 1768 (Evans, 2008; Werneburg et al., 2015), the posttemporal and supratemporal fenestrae separate the braincase from the temporal region. Near the junction of these two fenestrae, the squamosal meets the supratemporal and the paroccipital process of the otoccipital in a tripartite joint that receives the head of the quadrate in a position posterolateral to the occipital condyle (Fig. 16). The supratemporal bone in lizards is typically a relatively narrow element appressed against the supratemporal process of the parietal and sandwiched between the otooccipital and squamosal. The enclosure of the posterior braincase in snakes involved collapse of the posttemporal fenestra and loss of the metakinetic joint (Versluys, 1910; Frazzetta, 1962) between the parietal and supraoccipital. This transformation resulted in the reduction of the upper temporal bar and incorporation of the supratemporal and parietal into the lateral braincase wall, flanking the paroccipital process of the otooccipital along the metakinetic axis (Frazzetta, 1962).

Although Sanajeh lacks the metakinetic joint, it retains a complete upper temporal bar formed by loosely connected postorbital and squamosal bones (Fig. 16). The strap-like postorbital is short but reaches the orbital region, where it was probably clasped by the jugal and postfrontal. In contrast, the rod-like squamosal lacks tight contact with the quadrate and was probably held in place by soft tissues well medial to the supratemporal, which supplanted it as the main suspensorial element of the skull. Reduction of the upper temporal bar in Sanajeh allowed this portion of the skull to project posteriorly as a compact, multipartite suspensorial process (Fig. 16). In Sanajeh, the otooccipital remains plesiomorphically elongated, extending alongside the supratemporal (Fig. 15A, B). Only a small portion of the otooccipital is exposed on the lateral surface of the braincase, which contacts the quadrate, as it does in lizards (Figs 10, 15A).

Figure 10.

Cranial bones of Sanajeh indicus referred specimen GSI/GC/DD4. Photograph (A) and line drawing (B) of partial skull, including braincase, skull roof and posterior portions of right palate and mandible in left lateral view. Hatched lines indicate broken bone surface; parallel double-line stippling indicates matrix; and dashed lines indicate reconstructed portions of incomplete elements. Abbreviations: amfo, anterior mylohyoid foramen; as bp, articular surface for the basipterygoid process; as pl, articular surface for the palatine; as sut, articular surface for the supratemporal; bo, basioccipital; c, coronoid; cb, compound bone; d, dentary; oto, otoccipital; p, parietal; p cr, parietal crest; po, postorbital; pr, prootic; pt, pterygoid; sp, splenial; sq, squamosal; st, stapes; sut, supratemporal; VII, opening for seventh cranial nerve; ?, unidentified bone.

Figure 11.

Cranial bones of Sanajeh indicus referred specimen GSI/GC/DD4. Photograph (A) and line drawing (B) of partial skull, including braincase, skull roof and posterior portions of right palate and mandible in anterior view. Hatched lines indicate broken bone surface, and parallel double-line pattern indicates matrix. Abbreviations: as pl, articular surface for the palatine; c, coronoid; cb, compound bone; d, dentary; oto, otoccipital; p, parietal; p cr, parietal crest; po, postorbital; pr, prootic; pt, pterygoid; sp, splenial; sq, squamosal; st, stapes; sut, supratemporal; t, tooth; ?, unidentified bone.

Figure 12.

Compound bone of Sanajeh indicus referred specimen GSI/GC/DD4. Posterior portion of left compound bone in lateral (A), medial (B), dorsal (C) and ventral (D) views. Abbreviations: ar, articular surface for the quadrate; ctfo, foramen for the chorda tympani; psfo, posterior surangular foramen; su cp, surangular portion of the compound bone.

The unique morphology of the lateral braincase bones in Sanajeh is intermediate between the plesiomorphic suspensorium of toxicoferan lizards, in which the otooccipital articulates with the quadrate ventral to the supratemporal and lateral to the prootic, and the derived condition in the remaining snakes, where the otooccipital is excluded from a lateral contact with the quadrate by the contact between the supratemporal and prootic. Ventral to this contact, the robust posterior process of the prootic extends well beyond the fenestra ovalis, also resembling the condition in lizards. Although the stapes of Sanajeh has a large footplate, like Najash and Dinilysia, its shaft is much longer than the diameter of the footplate, consistent with the posteriorly expanded otico-occipital region (Fig. 10). Dorsally, the anterior portion of the supratemporal is sandwiched between the prootic and parietal, as it is in Anilius, Cylindrophis, Wagler, 1828, Dinilysia and Najash. Unlike those taxa, however, in Sanajeh the elongate supratemporal process of the parietal contacts the otooccipital in a broad suture that excludes the prootic from the dorsal braincase (Fig. 8). In Dinilysia and Najash, the supratemporal process of the parietal and the paroccipital process of the otooccipital are reduced, exposing the prootic dorsally on the braincase (Fig. 16).

An even more accentuated shortening of the multipartite suspensorial process occurs in Anilius and Cylindrophis, in which a dorsal exposure of the prootic is also known, further resulting in loss of the supratemporal bone in the more specialized uropeltids (Fig. 16). The condition seen in Anilius and Cylindrophis is morphologically intermediate between that in Sanajeh and the remaining crown-group snakes, which modified this process toward different extremes. In the strictly fossorial ‘scolecophidians’, these elements were also reduced to the point of losing the suspensorial process altogether and, in many cases, the supratemporal bone itself (convergent with uropeltids). In contrast, wide-gaped alethinophidians (with the exception of ‘anilioids’) evolved an elongate, unipartite suspensorial process composed of the free-ending supratemporal that projects away from the reduced parietal, prootic and otooccipital (Fig. 16).

Palatomaxillary kinesis in extinct and extant snakes

Most lizards use a hyolingual or cranioinertial transport mechanism to ingest their prey (Gans, 1969; Schwenk, 2000). Extant snakes are unique within squamates in that they developed a gnathic (jaw-based) transport mechanism to ratchet their mouth over prey using kinetic elements of the palatomaxillary arch and snout (Cundall, 1995; Kley, 2001). Stem snakes, such as Dinilysia and Sanajeh, combined a derived snake-like dentition and premaxilla with a plesiomorphic lizard-like, relatively akinetic palatomaxillary arch, illustrating a key transitional step from a cranioinertial to a gnathic prey transport mechanism. Sanajeh possessed a snake-like premaxilla, with an arched alveolar margin for passage of the tongue, recurved and needle-like teeth and rounded, smooth lateral margins, indicating a loose contact between the premaxilla and maxilla suggestive of mobility between these two bones (Figs 1–3). The septomaxilla resembles that of snakes in its nearly complete lateral vertical flange. In contrast, the maxilla is more lizard-like, including an unusually long and high ascending process (Fig. 3C). Maxillary and dentary tooth attachments in Sanajeh are clearly pleurodont, as in most lizards (Fig. 15C). The labial part of the crown is ankylosed to a high, obliquely sloping pleura, and the lingual side of the tooth base is ankylosed to a poorly developed basal plate. The absence of plicidentine can be verified in the posteriormost maxillary tooth, which is broken transversely near its proximal end.

The palatine–pterygoid contact of Sanajeh resembles the condition in lizards and in Dinilysia, in which the palatine overlaps the pterygoid in a broad, planar contact that is oriented almost horizontally (Fig. 14). Posterior and slightly lateral to this planar palatine–pterygoid contact is a narrow, V-shaped slit that receives a corresponding projection from the pterygoid (Fig. 14A). This condition differs from that of Wonambi, Yurlunggur, pachyophiids and more advanced extant snakes, all of which have a vertically oriented, complex tongue-and-groove palatine–pterygoid articulation. Interestingly, the palatine of Sanajeh also resembles that of Dinilysia and lizards in lacking a distinct dentigerous process. However, unlike Dinilysia, the palatine of Sanajeh has a row of palatine teeth that extends beyond the level of the choanal process to reach the anterolateral tip of the maxillary process (Fig. 14C). This intermediate condition anticipates the differentiation of the anteriorly projecting dentigerous process of the palatine in Wonambi, Yurlunggur, pachyophiids and extant alethinophidian snakes.

The palatine teeth in Sanajeh are recurved but much smaller than the maxillary and dentary teeth. The choanal process of the palatine is partly preserved, being notably thick dorsoventrally, much like the condition in Dinilysia and lizards (Fig. 14A, B), suggesting a tight interlocking contact or suture with the vomer. The dorsomedial surface of the pterygoid of Sanajeh bears a deep, smooth facet that would have received the basipterygoid process (Figs 9, 14). This synovial joint between the braincase and palate is also present in lizards, Dinilysia and Najash, as indicated by a notch in the pterygoid that receives the basipterygoid process. This joint is absent in Wonambi, Yurlunggur, marine pachyophiids and all extant snakes, which have an extended, flat surface on the medial edge of the pterygoid in place of the aforementioned notch. This flat surface contacts the basipterygoid process (when present) via tendinous tissues that allow limited movement. The presence of pitted and slightly worn surfaces on the pterygoid and basipterygoid process articular surfaces in Wonambi has been interpreted as evidence for a synovial joint (Scanlon, 2005), but a similar pitted and worn surface is also observed in some species of pythonids (H.Z., pers. obs.).

The combination of these three features (a thick choanal process, a broad planar contact between the palatine and pterygoid, and a synovial articulation between the basipterygoid process and pterygoid) suggests that the palatomaxillary arch of Sanajeh would have been capable of only limited movements. As in Dinilysia and Najash, the palatomaxillary arch of Sanajeh could not act as an effective ‘macrostomatan’ upper jaw ratchet. Thus, we infer that these stem snakes were not capable of the unilateral feeding movements present in alethinophidians (Kley, 2001). Restricted palatomaxillary mobility seems to be correlated with the absence of a movable prokinetic joint that confers rotational movements to the snout (Cundall, 1995; Kley, 2001). The large overlap between the nasals and frontals on the skull roof present in Dinilysia and Najash (Frazzetta, 1970; Zaher & Scanferla, 2012; Garberoglio et al., 2019a) was probably also shared by Sanajeh. Despite its limited palatomaxillary kinesis, Sanajeh possessed a well-developed intramandibular joint that would have allowed considerable bending of the mandibles, suggesting a movable connection at the tip of the dentaries, as in Dinilysia, Najash and all extant snakes.

The origin of macrophagy in snakes

The basally branching position of Dinilysia, Najash and Sanajeh suggests that the evolution of macrophagy in snakes occurred initially through structural modifications of the lower jaws toward increasing gape, including liberation of the dentary tips and acquisition of an intramandibular joint. Although Sanajeh retained an upper temporal bar, its gape size was expanded further by posterior displacement of the jaw joint beyond the occipital condyle via the posterior projection of the multipartite suspensorial process, which included the otooccipital, parietal, supratemporal and prootic (Fig. 15D). Our phylogenetic analysis predicts that macrophagous habits in snakes evolved by the Late Jurassic (Table 1; Supporting Information, Fig. S2) and involved mandibular ratcheting but very limited mobility of the palatomaxillary and snout regions.

This pattern was retained and built upon by ‘scolecophidians’, which evolved their own specialized maxillary and mandibular raking mechanisms without any auxiliary snout mobility (Kley & Brainerd, 1999; Kley, 2001). The later evolution of unilateral feeding mechanisms in alethinophidian snakes coincided with the appearance of a nasofrontal prokinetic joint, which conferred lateral rotational mobility to the entire snout complex (Cundall, 1995). A highly kinetic snout and palatomaxillary arch, along with a free-ending unipartite suspensorial process comprising only the supratemporal, are estimated to have arisen during the Early Cretaceous in the ancestor of pachyophiids, Wonambi, Yurlunggur and all extant alethinophidians (Table 1; Fig. 16). These features probably appeared initially as a more efficient mechanism for ingesting whole prey, rather than as a way to subdue and engulf larger and heavier prey. The ability to ingest exceptionally large and heavy prey seems to be restricted to only a limited number of derived alethinophidian snakes that include Python Daudin, 1803 (e.g. Bartoszek et al., 2018; Cundall & Irish, 2021) and specialized caenophidians, such as acrochordids, viperids and Dasypeltis Wagler, 1830 (Greer, 1997; B. C. Jayne, pers. comm.). This exceptional gape observed in these lineages has led to the misleading notion that all derived alethinophidians possess a wide-gaped condition (e.g. Rieppel, 1988; Zaher, 1998; Caldwell, 2020). However, gape size capacity, which ultimately determines ingestion capacity, has been measured only rarely in the literature, preventing a more objective definition of ‘macrostomy’ (Cundall & Irish, 2021). Moreover, the ingestion of exceptionally large-diameter prey involves not only the set of traditional skull and jaw bone specializations (e.g. Rieppel, 1988), but also a number of overlooked soft tissue features in the head and neck, such as skin elasticity, oesophageal expansion and parallel-fibred adductor musculature (Cundall & Greene, 2000; Cundall et al., 2014; Cundall, 2019; Cundall & Irish, 2021; Gripshover & Jayne, 2021).

Recent efforts toward quantifying maximal gape size in two large macrostomatan snakes (Python and Boiga Fitzinger, 1826) provide crucial insight into how anatomy constrains and affects feeding performance. Jayne & Bamberger (2019) have shown that Python molurus (Linnaeus, 1758) has a significantly larger gape than Boiga irregularis (Merrem, 1802), resulting primarily from distensible soft tissues of the chin and neck region and from a much larger intermandibular ligament, which accounts by itself for 41% of the gape in Python but only 17% in Boiga.

Quantifying gape size will ultimately help to clarify the role of hard and soft tissue characters associated with ingestion of large prey in ‘macrostomatan’ taxa. Although the central role that soft tissue specializations might play is difficult to determine in fossil taxa, information on their bone anatomy and articulation are crucial to help trace the evolutionary steps that led to the origin of macrostomy in alethinophidian snakes. The intermediate cranial morphology preserved in Sanajeh and other stem snakes helps to piece together these first evolutionary steps related to the origin of the wide-gaped condition and its varied trajectories within fossil and living ophidian species.

Homology of the circumorbital bones in snakes

The homologies of the bones that frame the orbit in snakes have been the subject of debate for many years (e.g. Rieppel, 1977; Zaher & Scanferla, 2012; Palci & Caldwell, 2013). Specifically, the identity of the bone or bones forming the dorsal and posterior margins of the orbit (i.e. those bridging the skull roof and maxilla posterior to the orbit) is contentious, owing to the absence of topological landmarks that would allow their straightforward identification in fossil and extant snakes. The unique presence in Sanajeh of a complete upper temporal bar provides evidence bearing on the homology of the circumorbital bones in snakes.

In fossil snakes where these elements are preserved, there are two bones that bridge the skull roof and maxilla, a ‘posterodorsal orbital element’ identified as the postfrontal (Zaher & Scanferla, 2012; Palci & Caldwell, 2013) and a ‘posteroventral orbital element’ identified as either the postorbital (Zaher & Scanferla, 2012) or the jugal (McDowell, 2008; Palci & Caldwell, 2013). The posteroventral orbital element is known to occur in the fossil snakes Dinilysia, Eupodophis, Najash, Pachyrhachis and Yurlunggur, where it is tightly connected to the posterior end of the maxilla (Estes et al., 1970; Caldwell & Lee, 1997; Scanlon, 2006; Palci et al., 2013; Garberoglio et al., 2019a). The majority of extant snakes, in contrast, typically possess only a single ‘posterior orbital element’, traditionally identified as the postorbital (Cundall & Irish, 2008). This element is tightly fixed to the skull roof at the level of the frontoparietal suture, and it is confined to the posterodorsal portion of the orbit in the vast majority of snakes; it does not reach the maxilla ventrally. In addition to this posterior orbital element, some modern snakes (viz. Calabaria, Loxocemus and Pythonidae) possess a ‘dorsal orbital element’ that is considered traditionally to represent either a neomorphic bone called the supraorbital (Rieppel, 1977; Boughner et al., 2007; Cundall & Irish, 2008) or the postfrontal (Underwood, 1976; Bellairs & Kamal, 1981; Palci & Caldwell, 2013).

Among extant snakes, all fossorial ‘scolecophidians’ lack the dorsal and posterior orbital elements, with the exception of anomalepidids, which possess an unusual ‘postorbital element’. This element, also called the ‘suborbital’ (Dunn & Tihen, 1944), is positioned lateral to the other adjacent elements of the skull and does not contact the maxilla ventrally or the frontoparietal dorsally. It is partly embedded in the ligamentous fascia covering the Harderian gland and receives fibres of the adductor mandibulae externus superficialis muscle (Haas, 1962, 1964, 1968). In three of the four anomalepidid genera, this element extends ventral to the rudimentary eye, from the posterior region of the nasal complex to the posterior tip of the prefrontal (Haas, 1964; Cundall & Irish, 2008; Rieppel et al., 2009; Santos, 2018). In the remaining anomalepidid genus, Anomalepis Jan, 1860, the ‘postorbital element’ extends further posteriorly along the lateral surface of the anterior portion of the parietal (Haas, 1968). This element has been identified variously as a postorbital (Haas, 1964, 1968; McDowell, 1987), a jugal (Haas, 1962; McDowell, 2008; Palci & Caldwell, 2013), a fused postorbitofrontal (Dunn, 1941) and a fused postorbital and jugal (McDowell & Bogert, 1954; List, 1966). Its association with the adductor mandibulae externus superficialis muscle favours the homology with the postorbital, whereas its position on the ventral orbit is consistent with a jugal. However, the identity of this element remains enigmatic in the absence of embryological evidence or of sutural contacts with adjacent bony elements. For this reason, we prefer the neutral designation ‘postorbital element’ (e.g. Rieppel et al., 2009; Linares-Vargas et al., 2021). The discovery in Sanajeh of a definitive postorbital and squamosal implies that the homology of the ‘postorbital element’ in anomalepidids lies with the jugal or postorbital, given the evidence at hand discussed above.

Recently, Palci & Caldwell (2013) suggested that the posterior orbital element in extant snakes is homologous to the jugal of lizards, rather than the postorbital as identified traditionally. This interpretation conflicts with the identification of the jugal in its typical position as a posteroventral orbital element in the fossil snakes Dinilysia, Eupodophis, Najash, Pachyrhachis and Yurlunggur. Three lines of evidence are crucial to this hypothesis that extant snakes possess a jugal: (1) the posterior orbital element in Calabaria and in some pythonids, boids and colubroideans contacts the maxilla, closing the posterior margin of the orbit in a way that resembles the jugal of non-ophidian squamates; (2) the embryological precursor of the posterior orbital element of extant snakes develops in a position comparable to that of the jugal in lizards; and (3) the ‘quadratomaxillary ligament’ of extant snakes corresponds to the ‘quadratojugal ligament’ of lizards. By extension, these authors argued that the exclusively dorsal orbital element present in the pythonids Calabaria and Loxocemus corresponds to the postfrontal of lizards. We contest each of these lines of reasoning in turn below.

Palaeontological arguments supporting the absence of a postorbital in extant snakes

Sanajeh preserves temporal bones not typically present in fossil and extant snakes, including the postfrontal, postorbital and squamosal. The latter two elements formed an upper temporal bar that was clasped anteriorly by the postfrontal and jugal and loosely connected to the paroccipital region posteriorly. The retention of these temporal bones in Sanajeh implies that the loss of the upper temporal bar occurred after body elongation and forelimb loss in snakes.

Although the jugal was not preserved in Sanajeh, we infer that it was present and completed the posterior orbital bar, because the postorbital bears an articular facet for it on its anteroventral surface (Fig. 13). The preservation of the postfrontal and inferred presence of a jugal in Sanajeh strongly suggest that these elements were also present in the phylogenetically adjacent stem snakes, bridging the skull roof and maxilla in Dinilysia, Eupodophis, Najash, Pachyrhachis and Yurlunggur. In contrast, the consistent non-preservation of a postorbital and a squamosal in these fossil snakes indicates that these bones were truly absent (or highly reduced) in the most recent common ancestor of Dinilysia, Najash and crown Serpentes. The pachyophiid Haasiophis has only one posterior orbital element, which has been identified typically as the postorbital. However, its positional association with the maxilla, absence of an articular facet on the frontoparietal suture, and lack of any vestige of a second posterior orbital bone (e.g. Eupodophis and Pachyrhachis) suggest that this small orbital bone is the jugal, as is also preserved in Eupodophis and Pachyrhachis.

Figure 13.

Temporal arch bones of Sanajeh indicus referred specimen GSI/GC/DD4. Three-dimensional computed tomographic scan reconstruction of the left postorbital (blue) and squamosal (pink) in lateral (A), medial (B), dorsal (C) and ventral (D) views. Abbreviations: as j, articular surface for the jugal; as pof, articular surface for the postfrontal.

Figure 14.

Palatal bones of Sanajeh indicus referred specimen GSI/GC/DD4 and Lanthanotus borneensis (FMNH 148589). Three-dimensional computed tomographic scan images of right palatine (blue) and pterygoid (yellow) of Sanajeh indicus (A, C, E) and Lanthanotus borneensis (B, D, F) in dorsal (A, B), ventral (C, D) and oblique anterolateral (E, F) views. Arrows point to the broad contact area between the pterygoid and palatine. Anterior is toward the left in A, B and toward the right in C–F. Scale bar represents 10 mm for Sanajeh and 2 mm for Lanthanotus. Abbreviations: as bp, articular surface for the basipterygoid process; as ect, articular surface for the ectopterygoid; as m, articular surface for the maxilla; ch pr, choanal process; q pr, quadrate process; sbo, suborbital foramen; t, tooth.

Figure 15.

Cranial bones of Sanajeh indicus referred specimen GSI/GC/DD4. Three-dimensional computed tomographic scan cutaway views along the horizontal (A, B) and transverse (C, D) planes. Arrowheads in B indicate the position of the transverse cutaway views in C, D. Abbreviations: bp, basal plate; oto, otooccipital; p, parietal; pl, pleura; po, postorbital; pr, prootic; pt, pterygoid; sd, subdental shelf; sq, squamosal; sut, supratemporal; t, tooth.

Figure 16.

Phylogenetic relationships of Toxicofera. Strict consensus of eight most parsimonious trees (996 303 steps) based on an unweighted total evidence analysis using TNT. Unlabelled nodes have bootstrap values of 100%. Asterisks indicate bootstrap values < 70%. Images on the right are dorsal views of the left temporal region of Varanus salvator (FMNH 35144), Sanajeh indicus (GSI/GC/DD4), Dinilysia patagonica (MACN-RN 1013; reversed) and Python molurus (TNHC, uncatalogued). Note the relative positions of the occipital condyle (oc) and jaw joint (qmj) marked by the ball-and-stick symbol in the bottom left corner of each box. Images are not to scale. Abbreviations: f, frontal; oc, occipital condyle; oto, otooccipital; p, parietal; po, postorbitofrontal; pt, pterygoid; q, quadrate; qmj, quadrate–mandibular joint; sq, squamosal; sut, supratemporal.

Although these conclusions support the presence of a jugal in Pachyophiidae, Dinilysia and Najash, they do not support unequivocally the presence of a jugal closing the posterior orbital bar in extant alethinophidian snakes (Palci & Caldwell, 2013). It is more likely that this element is the postfrontal, implying that the jugal was lost in extant alethinophidians.

Embryological arguments supporting the absence of a jugal and the presence of a neomorphic supraorbital in extant snakes

Palci & Caldwell (2013) suggested that the ossification of the posterior orbital element of Python sebae (Gmelin, 1788) (Boughner et al., 2007) mirrors the developmental and topological patterns of the jugal in the scincoid Acontias meleagris (Linnaeus, 1758) (Brock, 1941). That is, both elements start their ossification from a membranous precursor that is located posterior to the orbit, free from contact with surrounding bones. The developmental pattern of the posterior orbital element in Python and the jugal of Acontias Cuvier, 1817 are fundamentally distinct. The posterior orbital element in Python is either dorsoventrally or slightly posteroventrally–anterodorsally oriented (Boughner et al., 2007: fig. 6A), whereas and the jugal of Acontias is posterodorsally–anteroventrally oriented (Brock, 1941: fig. 5). Although neither of these bones contacts surrounding bones, the spatial orientation of the posterior orbital element in Python and the jugal of Acontias is consistent with the developmental sequence of the postfrontal and jugal, respectively (Bellairs & Kamal, 1981; Werneburg et al., 2015). The position and orientation of the posterior orbital element of Python are similar to those of the postfrontal of Varanus (Werneburg et al., 2015: fig. 3E); both form the posterodorsal margin of the orbit and bear a posteroventral–anterodorsal orientation, which rules out their identification as the jugal. In addition, a broader sampling of taxa supports this assessment. Bellairs (1950: figs 4, 5E) illustrated a late embryo of another fossorial lizard, the anguioid Anniella pulchra (Gray, 1852), in which the jugal appears in a more natural position, slightly behind the posterior extremity of the maxillary precursor and in a ventrolateral position with respect to the orbit.

The posterior orbital element in extant representatives of acrochordids, viperids and elapids develops in a posterodorsal position in the orbit and has a dorsal–ventral or posteroventral–anterodorsal orientation (e.g. Kamal et al., 1970; Haluska & Alberch, 1983; Rieppel & Zaher, 2001; Polachowski & Werneburg, 2013; Khannoon & Evans, 2015; Al Mohammadi et al., 2019). Al Mohammadi et al. (2019) identified in Psammophis sibilans (Linnaeus, 1758) a ‘well-defined crescent-like condensation (unossified) […] visible in the posterodorsal margin of the orbit’. They did not identify the posterior orbital element in snakes as a jugal because it does not develop as such in any of the extant snakes studied so far: the jugal develops from a primordium located in the posteroventral aspect of the orbit, near the posterior margin of the maxilla, with which it articulates (Werneburg et al., 2015; Werneburg & Sánchez-Villagra, 2015).

There seems to be no consensus on whether the posterior orbital element in extant snakes is a postfrontal or a postorbital (Kamal et al., 1970; Bellairs & Kamal, 1981; Cundall & Irish, 2008; McDowell, 2008). Khannoon & Evans (2015) noticed that this element is never in contact with the frontal or the parietal, which contrasts with the connections of a typical postfrontal, and concluded that ‘the bone is in the position of the postorbital of all limbed squamates’ (see also Bellairs & Kamal, 1981). Three pieces of evidence favour identification of the posterior orbital element as the postfrontal instead of a postorbital: (1) loss of the upper temporal arcade (including the postorbital) before the divergence of extant alethinophidian snakes; (2) embryological development of the posterior orbital element in a posteroventral–anterodorsal orientation; and (3) the embryological origin of the postorbital, which is anteroposteriorly oriented and more posteriorly positioned in the temporal region (e.g. Rieppel, 1994; Werneburg et al., 2015).

Embryological information for the dorsal orbital element in snakes, identified as either the supraorbital or the postfrontal, is relatively scarce (Boughner et al., 2007; Werneburg & Sánchez-Villagra, 2015) compared with that of the posterior orbital element, for which a substantial amount of developmental data is available in the literature. The limited amount of early ontogenetic information indicates that the dorsal orbital element develops late relative to other bones in squamates, which is consistent with the hypothesis that it is a neomorphic ossification, like the palpebral of Varanus (Werneburg et al., 2015). According to Boughner et al. (2007), the dorsal orbital element (their supraorbital) of Python appears only around the 54^th day after oviposition, whereas the posterior orbital element (their postorbital) appears much earlier, around the 33^rd day, thus favouring the hypothesis of a neomorph ‘supraorbital’ formation.

According to Palci & Caldwell (2013), the adult condition of the dorsal orbital element in Calabaria clearly shows that it is homologous to the postfrontal, because it ‘straddles the frontoparietal suture, a feature typical of the postfrontal of lizards’. However, the dorsal orbital element is itself an exception among extant snakes, being present only in Pythonidae (~38 extant species), its sister group Loxocemus and Calabaria within extant alethinophidian snakes. Of these, Calabaria and Loxocemus are the only two exemplars in which the dorsal orbital element straddles the frontoparietal suture (Zaher & Smith, 2020). Both the dorsal and posterior orbital elements are absent in the pythonoid Xenopeltis Reinwardt, 1827, but in the extinct Messelopython Zaher & Smith, 2020 and in extant pythonids it is the posterior orbital element that articulates or clasps the skull roof at the level of the frontoparietal suture in a strong syndesmotic attachment, like a postfrontal of non-ophidian squamates (Zaher & Smith, 2020: fig. S1). Therefore, the available evidence in adults of the few extant lineages retaining a dorsal and a posterior orbital element seems to support the neomorphic origin of the supraorbital instead of its identity as the postfrontal of other squamates.

The quadratomaxillary ligament as a topological marker

Calabaria and some pythonids have a broad ventral contact between the maxilla and the posterior orbital bone. This contact is covered by a thick layer of connective tissue that receives the quadratomaxillary ligament. Palci & Caldwell (2013) suggested that this ligament is equivalent to the ‘quadratojugal ligament’ of lizards, which attaches to the posteroventral margin of the jugal. If true, this homology statement would support the hypothesis that the posterior orbital bone is a jugal in Calabaria and all extant snakes.

Although Calabaria does have a ventrally expanded posterior orbital element, it also has a posteriorly reduced maxilla that ends at the level of the posterior orbital element. The proximity of the posterior end of the maxilla and ectopterygoid to the ventral extremity of the posterior orbital element in Calabaria provides a point of attachment for the quadratomaxillary ligament. This arrangement is uncommon in snakes; in most snakes, the quadratomaxillary ligament does not reach the posterior orbital element, but attaches to the anterior end of the ectopterygoid and the posterior end of the maxilla, which projects posteriorly beyond the orbit. The lateral face of the ligament attaches to the adjacent epithelium of the corner of the mouth, bridging the gap between the latter complex and the quadratomandibular capsule (Frazzetta, 1966: fig. 25) and playing an important role in the specialized kinesis of the skull in snakes (Anthony & Serra, 1949; McDowell, 1986). Likewise, in most snakes, it is the postorbital ligament that bridges the gap between the posterior orbital element and the maxilla, not the quadratomaxillary ligament, from which it is distinct and separated (see Haas, 1973; Zaher, 1994a; McDowell, 2008). As noted by McDowell (2008), ‘It is the postorbital ligament of snakes that most closely matches the position of the jugal of lizards.’ In Calabaria, the postorbital and quadratomaxillary ligaments are likely to have fused together, producing a condition that bears a superficial resemblance to lizards.

Based on the available evidence, we find little support for the hypothesis that the quadratomaxillary ligament represents a topographical landmark corroborating the homology between the posterior orbital element of extant snakes and the jugal of lizards.

CONCLUSION

Sanajeh indicus is the first snake to preserve two key transitional osteological features: an upper temporal bar formed by the postorbital and squamosal; and a paroccipital process contacting the quadrate below the supratemporal. Sanajeh and other stem snakes, such as Dinilysia and Najash, lacked a prokinetic joint and retained a plesimorphic palatopterygoid bar, which allowed for only limited palatal kinesis. The intermediate morphology of Sanajeh and other stem snakes helps us to gain a better understanding of the major modifications related to the origin of macrostomy in alethinophidians.

The evolution of wide gape in snakes began with collapse of the posttemporal region and acquisition of an elongate, multipartite suspensorial process, which was transformed in different directions in modern snakes. ‘Scolecophidian’ snakes reduced this process autapomorphically in association with their retreat into subterranean environments and concomitant shift to feeding on small prey items; more advanced alethinophidian snakes expanded the supratemporal bone into an elongate, unipartite suspensorial process that greatly displaced the jaw joint posteriorly. ‘Scolecophidians’ evolved their own specialized maxillary and mandibular raking mechanism without any prokinetic joint mobility, whereas in the most recent common ancestor of alethinophidian snakes, including the Cretaceous marine pachyophiids, complex unilateral jaw ratcheting, characterized by a highly kinetic snout and palatomaxillary arch, along with a free-ending unipartite suspensorial process, evolved in association with the acquisition of a nasofrontal prokinetic joint.

The presence of a well-preserved upper temporal bar in Sanajeh helps to clarify the homology of the posterior orbital bones present in extant snakes. With the exception of the ‘suborbital’ bone of anomalepidid snakes, whose homology remains enigmatic, a postorbital bone is likely to be absent in all extant snakes, and the single posterior orbital bone present in the majority of crown snakes corresponds to the postfrontal, not the jugal, of lizards.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article at the publisher's web-site:

File S1. Specimens examined, character list, uninformative morphological characters, list of morphological synapomorphies in relevant nodes of the maximum parsimony (MP) consensus tree, list of coding changes compared with Gauthier et al. (2012) as implemented by Hsiang et al. (2015) and list of the tip-calibration and node-calibration points used in the Bayesian inference (BI) analysis.

File S2. Set of bash commands to handle multilocus files for RAxML analyses (scripts_bash_format.txt).

File S3. R script used for subsampling loci from the concatenated matrices (scripts_R_format.txt).

File S4. Order of terminals in the matrix. File used (‘sanajeh_terminal_order.csv’) to implement commands in R script for the definition of topological constraints in MrBayes analyses (accessory_texts_for_scripts.txt).

File S5. Character optimizations from PAUP plotted on the total evidence maximum parsimony (MP) consensus tree topology. Character numbering corresponds to characters 1–785 in the original character list.

File S6. Character optimizations from TNT plotted on the total evidence maximum parsimony (MP) consensus tree topology. Character numbering here follows TNT, starting with character 0 and ending with character 784, being equivalent to characters 1–785 in the original character list.

Figure S1. Maximum likelihood (ML) tree (lnL −15700.382659) of a total evidence analysis using RAxML, combining molecular and morphological data for 91 species. Numbers near nodes represent bootstrap values < 100% but > 70%. Asterisks indicate bootstrap values < 70%, indicating low confidence in that node. Unlabelled nodes have bootstrap values of 100%.

Figure S2. Majority-rule consensus tree of the posterior distribution of five Bayesian inference (BI) total evidence analyses combining molecular and morphological data for 91 species. Grey bars indicate the maximum and minimum ages for fossil terminal taxa; blue bars indicate the 95% highest probability density for the estimated node ages. Numbers near nodes represent posterior probabilities < 1.0 but > 0.8 based on the posterior distribution of trees from the five Bayesian total evidence analyses. Asterisks indicate posterior probabilities < 0.80, indicating low confidence in that node. Unlabelled nodes have posterior probabilities of 1.0.

Figure S3. Comparison of topologies derived from maximum likelihood (ML) and maximum parsimony (MP) analyses of the total evidence dataset.

Figure S4. Comparison of topologies derived from Bayesian inference (BI) and maximum parsimony (MP) analyses of the total evidence dataset.

Figure S5. Comparison of topologies derived from maximum likelihood (ML) and Bayesian inference (BI) analyses of the total evidence dataset.

Table S1. List of characters in the datasets from Gauthier et al. (2012; GKMRB), Hsiang et al. (2015; HG), Longrich et al. (2012; LBG) and Zaher and Scanferla (2012; ZS).

ACKNOWLEDGEMENTS

This study was carried out under a Joint Collaborative Programme through a Memorandum of Understanding for ‘Study of Late Cretaceous snake fossil from Lameta Formation of Kheda District, Gujarat’ involving the Geological Survey of India (GSI; Government of India, Ministry of Mines) and the University of Michigan Museum of Paleontology. We thank the erstwhile Director General of the GSI, Deputy Directors General Seva Das and K. S. Misra and Director K. K. K. Nair for facilitating this project. We thank J. Head, S. Peters, R. Rathore, S. Rathore and M. Wilson for their contributions to field research in 2007, which was supported by grants to J.A.W.M. from the National Geographic Society Committee for Research and Exploration (NGS 8127-06), the American Institute for Indian Studies and the National Science Foundation (NSF EAR 1736606). We are grateful to D. Cundall and B. Jayne for stimulating discussions on morphofunctional aspects of snake evolution and macrostomy. The fossil specimens of Sanajeh indicus were prepared by W. Sanders. Scientific illustrations and figure layouts were done by C. Abraczinskas (Figs 1–5, 7–11) and Alberto B. Carvalho (Figs 6, 12–15). C. Abraczinskas provided suggestions and contributed to the layout of Figure 16. We are indebted to V. Brumfeld and R. Rabinovich for coordinating and providing the CT scans of Haasiophis terrasanctus, and to A. Scanferla, M. Polcyn and M. S. Y. Lee for providing the 3D reconstructions of Dinilysia patagonica, Pachyrhachis problematicus and Wonambi naracoortensis, respectively. We are also indebted to T. Rowe and J. Maisano for providing 3D reconstructions of several squamate taxa scanned as part of the DigiMorph project funded by the National Science Foundation (DEB-0132227 and EF-0334961). We are grateful to Alberto B. Carvalho for segmenting the specimens for which we collected raw CT data. H.Z. is grateful to the following colleagues who allowed access to specimens under their care: N.-E. Jalil, J.-C. Rage, C. de Muizon (Centre de Recherche en Paléontologie – CR2P, Muséum National d’Histoire Naturelle, Paris, France); T. Smith and A. Folie (Royal Belgian Institute of Natural Sciences, Brussels, Belgium); K. Smith (Senckenberg Research Institute, Frankfurt, Germany); A. Dubois, A. Ohler and I. Ineich (Laboratoire des Reptiles et Amphibiens, Muséum National d’Histoire Naturelle, Paris, France); M. Day and M. Richter (Department of Earth Sciences, The Natural History Museum, London, UK); P. D. Campbell, D. Gower and M. Wilkinson (Department of Life Sciences, The Natural History Museum, London); M. Talanda (Department of Paleobiology and Evolution, University of Warsaw, Warsaw, Poland); J. Kobylinska (Polish Academy of Sciences, Warsaw, Poland); U. Göhlich (Natural History Museum, Vienna, Austria); I. Zorn (Geological Survey of Austria, Vienna, Austria); O. Rauhut and M. Nose (Bayerische Staatssammlung für Paläontologie und Geologie, Munich, Germany); R. Schoch (Staatliches Museum für Naturkunde, Stuttgart, Germany); O. Rieppel, A. Resetar and H. Voris (The Field Museum, Chicago, IL, USA); D. Frost, D. Kizirian, C. Myers, C. Raxworthy and F. Burbrink (Division of Vertebrate Zoology AMNH, New York, NY, USA); M. Norell (Division of Paleontology, AMNH, New York); R. Rabinovich (Museum of Paleontology, The Hebrew University of Jerusalem, Jerusalem, Israel); R. Nussbaum and D. Rabosky (Museum of Zoology, University of Michigan, Ann Arbor, MI, USA); P. Larson, R. Farrar and N. Larson (Black Hills Institute, Hill City, SD, USA); G. Puorto and F. L. Franco (Instituto Butantan, São Paulo, SP, Brazil); D. Rossman (Museum of Natural Science, Louisiana State University, Baton Rouge, LA, USA); J. Bonaparte and A. Kramarz (Museo Argentino de Ciencias Naturales Bernardino Rivadavia, Buenos Aires, Argentina); M. A. Reguero (Museo de La Plata, Universidad Nacional de La Plata, Buenos Aires, Argentina); J. Scanlon (University of New South Wales, Sydney, NSW, Australia); M. S. Y. Lee and M. Hutchinson (South Australian Museum, Adelaide, SA, Australia); and K. de Queiroz, R. McDiarmid, G. Zug and R. Heyer (National Museum of Natural History, Washington, DC, USA). The authors thank K. T. Smith, G. Georgalis and two anonymous reviewers for their thoughtful reviews.

COMPETING INTERESTS

The authors declare no competing interests.

FUNDING

This study was funded by a field grant EAR-1736606 from the National Geographic Society Committee for Research and Exploration grant 8127-06 (to JAWM); National Science Foundation to J.A.W.M. and by research grants 11/50206-9 and 2018/11902-9 from the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) to H.Z. We also thank A. Pophare, Head of the Department of Geology, RTM Nagpur University, for providing research facilities under the Ministry of Earth Sciences (Government of India) Grant MoES/PO(GEOSCI)/49/2015 to D.M.M.

AUTHOR CONTRIBUTIONS

H.Z., J.A.W.M. and D.M.M. designed this research project. D.M.M. and J.A.W.M. collected the fossil data. H.Z. developed and scored the morphological dataset. F.G.G. wrote the scripts and compiled and processed the molecular dataset. H.Z. and F.G.G. conducted the phylogenetic analyses. H.Z., J.A.W.M., F.G.G. and D.M.M. wrote the manuscript. H.Z., J.A.W.M. and F.G.G. designed the figures.

DATA ACCESSIBILITY

Files containing the combined molecular and morphological datasets (in Phylip and TNT formats) and list of DNA partitions are available for download at: https://doi.org/10.6084/m9.figshare.14618217.v1