The oldest known snakes from the Middle Jurassic-Lower Cretaceous provide insights on snake evolution

The previous oldest known fossil snakes date from ~100 million year old sediments (Upper Cretaceous) and are both morphologically and phylogenetically diverse, indicating that snakes underwent a much earlier origin and adaptive radiation. We report here on snake fossils that extend the record backwards in time by an additional ~70 million years (Middle Jurassic-Lower Cretaceous). These ancient snakes share features with fossil and modern snakes (for example, recurved teeth with labial and lingual carinae, long toothed suborbital ramus of maxillae) and with lizards (for example, pronounced subdental shelf/gutter). The paleobiogeography of these early snakes is diverse and complex, suggesting that snakes had undergone habitat differentiation and geographic radiation by the mid-Jurassic. Phylogenetic analysis of squamates recovers these early snakes in a basal polytomy with other fossil and modern snakes, where Najash rionegrina is sister to this clade. Ingroup analysis finds them in a basal position to all other snakes including Najash. The origin and evolution of snakes remain poorly understood. Here, the authors show that fossils previously described as anguimorph lizards are ancient snakes and demonstrate that they share features with snakes and lizards, which suggests great diversity of snakes by the Jurassic period.

T he previous understanding of the fossil record of early snake evolution (Late Mesozoic) relies on isolated vertebrae from Africa 1 (100 Myr ago), isolated jaws and vertebrae from North America 2,3 (98-65 Myr ago), a number of nearly complete snakes, some with rear limbs [4][5][6][7] , from the circum-Mediterranean region (98-95 Myr ago), and two taxa of relatively complete snakes, one with rear limbs, from Argentina (94-92 Myr ago 8,9 and 86-80 Myr ago 10,11 ). This morphologically, ecologically and phylogenetically diverse assemblage of snakes appears in the fossil record around the world almost simultaneously (B100-94 Myr ago). The long standing questions in snake palaeontology have centred on when, where and how snakes evolved and radiated from within Squamata before the early part of the Late Cretaceous.
Here we report on four new species of significantly older fossil snakes (167-143 Myr ago) recognized from cranial and postcranial remains found in the United Kingdom, Portugal and the United States. These new data extend the known geological range of snakes by nearly 70 million years into the mid-Mesozoic, indicating that their origin was coincident with the known radiation of most other major groups of squamates in the mid-Jurassic 12,13 during the final stages of the break-up of Pangaea into Laurasia and Gondwana. It is also important to note that this new record for early snakes, co-occurring with early anguimorphs such as Dorsetisaurus 12,13 , fills a major chronological gap predicted by molecular phylogenetics 14 .
In stratigraphic order, the oldest snake recognized here is found in rocks dated as Bathonian (B167 Myr ago), Middle Jurassic, from Southern England 15 , followed by a North American record dated as Kimmeridgian (B155 Myr ago, Upper Jurassic, Colorado, USA) 16 , which appears to be a contemporary of another taxon from the Kimmerdigian (Upper Jurassic, Guimarota, Portugal) 17 . The youngest snake taxon and materials recognized here are found in rocks dated as Tithonian (B150 Myr ago; Upper Jurassic) to Berriasian (B140 Myr ago; Lower Cretaceous) outcropping near Swanage, Dorset, Southern England 15 .

Systematic palaeontology.
Order Squamata Opell, 1811 Suborder Serpentes Linnaeus, 1758 Parviraptor estesi gen. et sp. Holotype. Left maxilla on block NHMUK R48388, Natural History Museum, London. Locality horizon and age. Durlston Bay, Swanage, Dorset, England; Purbeck Limestone Formation (Upper Jurassic; Tithonian/Lower Cretaceous; Berriasian). Emended Diagnosis. Long low, ascending process of maxilla; premaxillary process turned medially; narrow prefrontal facet on ascending process of maxilla. Differs from maxilla of Coniophis precedens in lacking medial process at anterior end, in having greater degree of recurvature of teeth. Differs from Dinilysia patagonica in having gracile maxilla, relatively smaller teeth, less pronounced medial deviation at anterior tip and relatively smaller palatine process. Differs from Portugalophis lignites in having narrower premaxillary process. Differs from Diablophis gilmorei in having larger palatine process and in lacking medial curvature of anterior end of maxilla.
Revised Description. Maxilla exposed in medial and dorsal views; long, 24 tooth positions preserved, bearing low ascending process extending from tooth position 4 to tooth position 16/17; anterior superior alveolar foramen large, positioned at front of ascending process; prominent premaxillary process, narrow and turned medially; supradental shelf narrow and thin with prominent palatine process adjacent to posterior end of ascending process; margin of palatine process damaged; long, tooth-bearing suborbital ramus with 10 tooth positions posterior to prefrontal facet and palatine process; maxillary tooth positions defined by three-sided alveolus with no medial border; preserved teeth attached to rims of alveoli; presence of tooth in alveolus closes small posterolingual notch forming basal nutrient foramen; teeth conical, circular cross sections, recurved, with labial and lingual carinae. Description. Right frontal (NHMUK R8551) in lateral view with well-developed prefrontal and postfrontal facets and deep, medially curved descensus frontalis with well-developed suboptic shelves on posteroventral margin forming posterior portion of optic nerve foramen (Cranial Nerve II); closely associated vertebrae with tall neural spines, massive, vertical synapophyses with distinct, posteriorly expanded parapophyseal and diapophyseal facets; complete neural arch with short zygosphenes; posterior margin lacks incised median notch; elevated, round condyles with weakly constricted necks; centrum narrow posteriorly and wide anteriorly in ventral view; inferior margin of centrum, no development of haemal keel; possible sacral vertebra with robust transverse processes; left neural arch from an atlas with narrow, posteriorly directed neural spine; no notochordal canal. Remarks. Because of the loss of the provenance of this block with respect to the block NHMUK 48388, the lack of articulation of the elements, and the polytaxonomic composition of the Purbeck blocks, we do not feel that there is sufficient evidence to support referral of the frontal to type maxilla of Parviraptor estesi, although we do recognize that such a referral is a strong possibility. process and having strong medial deviation of anterior end of maxilla. Differs from P. lignites and from Eophis underwoodi in having more fully developed subdental lamina relative to dentary size. Differs from Coniophis precedens in the absence of medial process on anterior tip of maxilla; greater degree of recurvature of teeth; alveoli primarily on lateral parapet of dentary and maxilla; presence of subdental lamina and multiple mental foramina on dentary; and in having tall neural spines and small zygosphenes and zygantra.
Description. Right maxilla (LACM 4684/140572) small, preserves 10-11 tooth spaces with broken crowns; ascending process low and long with small notch in posterior margin of apex; supradental shelf broken, thin anteriorly and thickened posteriorly; lateral surface smooth with four nutrient foramina; alveoli/interdental ridges with incised medial walls forming nutrient notch; tooth crowns conical, recurved, sharp; holotype right dentary missing symphysis and postdentary articulation region; straight in dorsal view with medial turn anteriorly and 11 tooth spaces; well-developed subdental lamina forming medial border of subdental gutter; lateral wall with seven mental foramina; dentary teeth attached to three-sided alveoli; teeth circular in cross-section, conical and strongly recurved; neural spine tall, synapophyses massive and vertical, and condyle elevated above bottom of centrum; neural canals show 'trefoil' organization often present in snakes 3,8 ; condyle round, with weakly constricted neck and synapophysis with posteriorly expanded parapophyseal and diapophyseal facets; centrum narrow posteriorly, wide anteriorly; inferior margin of centrum without haemal keel; paralymphatic fossae present; no notochordal canal; small zygosphene-zygantrum articulations.  Etymology. Eos, dawn (Greek); ophis, snake (Greek); underwoodi (Surname); in recognition of being oldest known snake material, and lifelong impact/contributions of Garth Underwood to study of snakes.
Comparative anatomy of early snakes. These new snake taxa are based on specimens that were either previously described and named, or referred to, various species or groups of anguimorph lizards [15][16][17] . The problems associated with identifying any of the Parviraptor specimens as the remains of snakes were due to the fact that the original taxon 15 (Supplementary Fig. 3). The original characterization of Parviraptor cf. P. estesi 15 was based on isolated specimens found more than 150 km away at Kirtlington, Oxfordshire, the United Kingdom (Fig. 1n), in rocks that are B30 million years older than the type and referred block (Fig. 1p). Reinterpretation of these specimens finds that the original type and referred specimens of Parviraptor estesi 15 and Parviraptor cf. P. estesi 15 include the remains of possibly three separate taxa of snakes, a large number of gekkotan as well as other indeterminate lizard elements, and also a number of non-squamate elements and one invertebrate fragment ( Supplementary Figs 12-23). The identity and affinities of Parviraptor were further confounded by recent phylogenetic investigations that, based on the original descriptions 15 , found the taxon to have gekkotan affinities 18,19 . These hypotheses arose from the assignment to Parviraptor cf. P. estesi 15 Fig. 13) are identified here as gekkotan, which explains the character scorings for Parviraptor as a gekkotan 18,19 , in contrast to the original description 15 . The original descriptions of Diablophis gilmorei 16 and Portugalophis lignites 17 avoided the chimaera problem of Parviraptor estesi/aff. Parviraptor 15 as the materials selected from the microvertebrate assemblages were only compared with the maxilla and teeth of P. estesi.
Identifying the affinities of disarticulated vertebrate remains is difficult and relies on detailed comparisons of specialized features that characterize one vertebrate over another. However, snakes possess a number of detailed cranial and postcranial anatomies that are definitively characteristic of the group, distinguishing them from other squamates, and that are present in these new Middle Jurassic-Lower Cretaceous forms. The unique features of snake cranial, dental and axial skeletal elements make it possible to identify snakes in the fossil record from isolated or even fragmentary elements 2,3,20 .
Skull, jaw and dental features characterizing fossil and modern snakes 21 that are present in the oldest snakes described here (Figs 1a-l and 2a-t; Table 1; Supplementary Figs 1-11; Supplementary Videos 1-6 and 11-14), recognizing of course the limits imposed by incompleteness and preservation, include (1) long, tooth-bearing sub-orbital ramus of maxilla, unique to snakes with exception of some gekkos (Fig. 2a,c- Figs 7g and 8c; Supplementary Videos 13, 14); (6) dorsal position and superior exposure of anterior opening of superior alveolar canal (anterior superior alveolar foramen) along anterodorsal margin of maxilla (in lizards, this feature is variably projecting medially, and is covered by nasals, septomaxilla, or overhanging ascending process of maxilla) (Fig. 2a-f; Supplementary Figs 7e and 9a); (7) smooth and rounded anteromedial process of maxilla for ligamentous connection with premaxilla ( Fig. 2a-f; Supplementary Videos 10-12); (8) well-developed descensus frontalis and posteriorly, suboptic shelves ( Fig. 2s-t; Supplementary  Fig. 3c,d); (9) strongly recurved maxillary and dentary teeth where maxillary teeth are more recurved than the dentary teeth; a sigmoidal curvature cannot be observed; however, such curvature is not present in all snakes and can also disappear ontogenetically (see for example, Fig. 2a Comparison of the alveolar walls of Eophis underwoodi with those of the modern snake Xenopeltis unicolor (Fig. 2o,p;  Supplementary Fig. 6a-d) shows a striking similarity in the construction between the two taxa. In both taxa the interdental ridge is continuous posteriorly with the lateral parapet to form a J-shaped boundary to the alveolus. An obvious difference between the jaws of the oldest known snakes and geologically younger snakes is the presence of a maxillary supradental shelf (Fig. 2a-c) and distinct dentary subdental shelf and sulcus dentalis that runs the length of the tooth-bearing element (Fig. 2g-k). In modern snakes, inclusive of scolecophidians, and geologically younger snakes such as Yurlunggur, the medial bony tissues of the maxilla and dentary grade smoothly 20,21 into the tooth row without forming a shelf ( Fig. 2d-f,l-m) though a small shelf persists on the dentary immediately superior to the articulation with the posterior portion of the splenials (Fig. 2l-m). Fossil snakes such as Dinilysia are similar to the oldest known snakes, while Coniophis is more similar to Yurlunggur.
Associated with the tooth bearing elements of Diablophis gilmorei, is a small elongate element that appears to be the right surangular bone (Supplementary Fig. 10i-k); the element was not included in the parsimony analysis as the association is weak. However, this disarticulated surangular is long, bears a large snake-like adductor fossa and laterally directed coronoid process, but is not fused into a single compound bone (thus, the articular surface for the quadrate is only partly preserved, as the articular bone is missing). It may well be that B140 Myr ago the compound bone characteristic of all other fossil and modern snakes 20,21 had not yet appeared in snake evolution.
Vertebral features that are present in two of the new fossil snakes (aff. Parviraptor estesi; Diablophis gilmorei) and are shared with fossil snakes such as Najash rionegrina and Coniophis precedens 2,3 (Fig. 3a- The features shared by aff. Parviraptor estesi and Diablophis gilmorei with Najash rionegrina and the much younger (B100 Myr difference) Coniophis precedens are interesting and important, despite the much weaker development of the zygosphenezygantral articulations and width of neural arch platform in the first two snakes (Fig. 3a-e; Supplementary Figs 4a-g and 11a-n). However, in contrast to most known snakes 8,9 , there is one potential sacral vertebra present on the aff. Parviraptor estesi (NHMUK R8551) slab (Fig. 3a) and another in the vertebral specimens referred to Diablophis gilmorei (4684/140572). Each vertebra is far more robust than the materials described for Najash rionegrina 8,9 (Fig. 3d), suggesting the possible presence of robust pelvic girdles and hind limbs. The referred postcloacal vertebra of D. gilmorei (LACM 4684/120472) possesses a large transverse process that is directed laterally, not anteriorly as in modern snakes, and is more similar in size and orientation to postcloacal transverse processes of Najash. As this report extends the fossil record of snakes by B70 million years, it is likely that these early snakes shared with younger fossil snakes 4,7,8 the presence of at least rear limbs. However, we remain conservative on this point and have not coded this information for phylogenetic analysis ( Supplementary Information Section 3); presence or absence of forelimbs remains mere speculation as it cannot be verified by reference to current specimens.
Phylogeny. At the alpha taxonomic level, based on comparative anatomy, Parviraptor, Diablophis, Eophis and Portugalophis are confidently identified as snakes [1][2][3]20,[22][23][24] . However, to address questions of phylogeny, we conducted two separate analyses using two vastly different data matrices. To test the possibility that these new snakes might fall outside of the snake clade in an overall analysis of squamate phylogeny, we included them in a recent large matrix of all squamates 25 . We also examined the more specific characters related to snakes, by including them in a data matrix originally developed to test snake ingroup relationships 3,7,8,[26][27][28] (Supplementary Methods; Supplementary  Dataset 1). Both phylogenetic analyses were run using the software programs PAUP 29 and TNT 30 , the latter was also used to calculate bootstrap values of statistical support. The matrix 25 analysed in our examination of overall squamate phylogeny included the addition of the maxillae of Parviraptor, Portugalophis and Diablophis as terminal taxa (the maxillae were selected because, of all the isolated elements, they are those that allow for the largest number of character states that can be coded; other elements associated with these maxillae were not included to provide the strictest test of phylogenetic relationships); the goal of this analysis was to test whether or not these partial skeletal remains would be recovered as snakes, or fall outside that clade. In the strict consensus tree of 108,981 shortest trees retrieved in PAUP 29 (tree length: 5290 steps), all three taxa were reconstructed within a polytomy comprised of Mesozoic snakes (Dinilysia, Pachyrhachis, Haasiophis and so on), scoleocophidians and alethinophidians, with Najash as the most basal snake. The strict consensus of 200 trees retrieved using TNT 30   in PAUP 29 , and the only differences consist in the repositioning of some basal macrostomatans (for example, Calabaria, Eryx and Lichanura) and in the addition to the basal polytomy of the taxa Casarea, Xenophidion, Loxocemus and Xenopeltis. The clade Ophidia (that is, all extant and fossil snakes) was retrieved with strong bootstrap support (BS ¼ 85) (Fig. 4a; Supplementary  Methods). These results are supportive of our identification of these specimens as the remains of snakes, and that derived features of snakes were present among squamates as early as the Middle to Upper Jurassic.
In our parsimony analyses of snake ingroup phylogeny, we submitted a taxon-character matrix of 27 terminal taxa and 237 characters, derived in large part from a recent study 3 that in turn had adapted characters and taxa from several other works 8,11,24,26,27,31 (Supplementary Methods; Supplementary Dataset 1). Our characterization of the Upper Cretaceous snake, Coniophis precedens 3 is more conservative than that given in a recent report, and is restricted to the vertebral assemblage, a single maxillary fragment (the other maxillary fragments are identified here as lizards, not snakes) and a fragmentary right dentary 3,24 . We also followed the more restrictive concept of Najash 8 in coding that taxon as was recently suggested in a conservative revision of the materials assigned to the type of that taxon 9 (Supplementary Dataset 1). The results of our ingroup phylogenetic analysis suggest that the new Jurassic/Lower Cretaceous fossil snakes either form a clade in the sistergroup position to all other known snakes, or form a paraphyletic assemblage at the base of the radiation of all other snakes (Fig. 4b). The new data, and the results of our analysis of snake phylogeny (see also Supplementary Figs 24 and 25), agree with previous ingroup analyses of fossil snakes 3,4,8,[26][27][28]31 that hypothesized that the most basal snakes are not the extant scolecophidians 7,32 . The new early snakes add to those hypotheses by extending the fossil record of snakes by 70 million years (Fig. 1m-p) coincident with the radiation of other mid-Jurassic squamate groups 12,13 and with molecular clock predictions 14 .
Paleobiogeography. An extension of the fossil record of snakes to the Middle Jurassic is not surprising, as it coincides with the radiation of other major groups of squamates 12,13 , even though these new records of diverse and broadly distributed mid-Mesozoic (B167-145 Myr ago) snake assemblages (Fig. 1m-p) create a significant temporal gap with the previous early records of snakes (B100 Myr ago) from Africa 1 , North America 2 , Brazil 32 and Europe 33 . An unexpected result is the paleoecological and paleobiogeographical diversity and distribution of these Middle Jurassic-Early Cretaceous snakes (Fig. 1m-p). These earliest snakes are found preserved in rocks deposited in coal swamps (Portugalophis) 17 , mixed coastal lake and pond systems (Eophis, Parviraptor, aff. Parviraptor), river systems (Diablophis) 16 and on epicontinental islands located up to several hundred kilometers off the coastline of the closest large landmass (Fig. 1n-p). Subsequent snakes appear in rocks deposited in fluvial [1][2][3][8][9][10][11] and marine environments 4-7 , making the paleogeography and paleoecology of the earliest snakes into important reference points for later evolutionary radiations.
It is possible that these early snakes were isolated on various islands and continents during the break-up of Pangaea in the mid-Mesozoic (Fig. 1n). Such a scenario finds support based on the similarities of their cranial features to those of Dinilysia 10,11 and later madtsoiids 22,[26][27][28] (Fig. 2a-p), which certainly appear to be restricted to Gondwana continents during the Mesozoic and Cenozoic. It is also possible that snakes forming these island assemblages arrived as secondarily aquatic invaders. Secondary invasions of marine environments certainly characterize the subsequent evolutionary histories of numerous clades of fossil and modern snakes [4][5][6]31 . In fact, there is no evidence to suggest that these radiation scenarios are mutually exclusive, and so it is possible that both Pangaean-derived vicariance and multiple adaptive radiations (marine and terrestrial) influenced the Jurassic radiation of snakes and the subsequent Gondwanan and Laurasian evolution of these animals through the Mesozoic and Cenozoic 4-11 .

Discussion
These earliest snakes provide insights on aspects of snake origins, adaptive radiations, phylogeny and evolution. A recent study on the Late Cretaceous snake Coniophis precedens concluded that the snake body evolved before the snake head 3 ; this conclusion was based on supposed lizard-like features observed in isolated and disassociated skull elements linked to similarly isolated vertebrae from numerous individuals collected at two disparate localities. Observations of living groups of lizards and snakes suggest the opposite transformation of morphology from that proposed in a recent study of material referred to Coniophis 3 , that is, that the snake skull evolved before the elongate and limb-reduced to limbless postcranial skeleton. For example, scincid lizards, whether fully limbed and short-bodied, or limb-reduced to limbless and elongate, are diagnosable as scincid lizards by their distinctive skull anatomy. This is also true for all limb-reduced to limbless anguids, cordylids, gerrhosaurids, gymnophthalmids and pygopodids 34,35 . Because snakes are lizards, nested within that monophyletic assemblage 31 , it is possible to extend this analogy as a test of the conclusions presented in the aforementioned Coniophis study 3 . The observation that 'skinkness' or 'anguidness' is diagnosed not by elongation and limb reduction to limblessness, but rather by the shared possession of anatomical features of the head, and the recognition that snakes are lizards, leads to the prediction that the fossil record of snake evolution will likely reveal four legged, short bodied, 'stem-snakes' that possess 'snake' skull anatomies. Such a prediction is logically consistent with the pattern seen in all groups of living limbless to limb-reduced lizards; the recent claim to the contrary 3 violates that prediction and would be extraordinary if supported by fossil evidence.
While we cannot ascertain the shape, length, form and so on of any aspect of the body of the earliest snakes (B167 Myr ago) reported herein, these animals are, however, identifiable as snakes by their cranial features. If 'snakeness' is recognized in the numerous cranial features of the B167-143 Myr ago snakes described herein, nearly 100 million years before Coniophis precedens 3 , and if cranial features diagnose squamate groups regardless of postcranial evolution, then the prediction given here is that evolution of the snake skull was the key innovation of the clade, and not elongation and limb reduction, contrary to previous claims 3 . By the Middle Jurassic-Lower Cretaceous, and amidst the break-up of the old supercontinent Pangaea, snakes were distinguished from their lizard relatives by the possession of a skull with snake cranial and dental features that are certainly present by the Cenomanian 4-7 and are retained to the present. Similar to many lineages of squamates 34,35 , it seems likely that they then subsequently evolved elongate and limb-reduced to limbless bodies, evolving and adapting through more than 167 million years of earth history to become the extremely diverse group they are today.

Methods
Materials examined. The specimens of fossil (13 taxa) and living snakes (76 taxa) examined in this study belong to the collections of the following institutions: American Museum of Natural History, New York, USA (AMNH); Field Museum NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6996 ARTICLE Phylogenetic analysis. The sistergroup and ingroup relationships of fossil and modern snakes were approximated using parsimony methods in both PAUP and TNT. The sistergroup relationships of Parviraptor, Portugalophis and Diablophis, and a large sample of other fossil and modern snake terminal taxa, were tested using the matrix and methods (for example, ordering, outgroup selections and so on) of a recent study testing overall squamate relationships 25 . Sixteen characters were coded in that matrix 25 for the maxillary elements of the three fossil snake taxa described here (see Supplementary Methods for details). The ingroup relationships of snakes, including the fossil taxa described here, were tested using a significantly modified data matrix (see Supplementary Methods; Supplementary Dataset 1) from a recent study reporting on the phylogenetic relationships of the Mesozoic snake taxon, Coniophis 3 . The same outgroup taxon was used as was employed in that previous study 3 , although with some modifications to the coding for the 'anguimorph root' 3 . Contra the previous study 3 , all characters were analysed unordered and unweighted in both matrix tests. The data were analysed using the Heuristic Search option in PAUP* 4.0b (ref. 29) and character transformations were optimized using the ACCTRAN assumption. The data set was also tested using two different search routines in TNT 30 ; the Traditional search and the Drift search option from the New Technology searches. The Traditional search was run using 100 Wagner seed trees, 100 replications, but no swapping algorithims were utilized (for example, SPR, TBR) as the data set is not excessively large. The Drift search was chosen because unlike the Ratchet option it does not reweight characters during the analysis. This analysis was run using the default settings for the search.