Geography and geology of the holotype ofRhinochelys amaberti

The holotype ofRhinochelys amaberti Moret (1935) (UJF-ID.11167) was found in the Vallon de la Fauge, near Villard-de-Lans in the Isère department (France) (Fig. 1) (Moret, 1935). The Vallon de la Fauge area belongs to a larger geological and geographical unit called the Vercors. In the northeastern Vercors, where the Vallon de la Fauge is located, the thick carbonate successions of the Barremian-Early Aptian are surimposed by bioclastic limestone from the Late Aptian (about 15 m thick), followed by several meters of Albian greenish sandstones and glauconitic marls in decimetric beds (Fig. 1). These Albian deposits are known locally as ‘Béton phosphaté’ because of their abundant phosphatized ammonites and bioclasts (Arnaud, 1988). The sandy and marly deposits overlying the carbonates belong to the Marnes Bleues Formation, which extends from the Aptian to the Cenomanian (Bréhéret, 1997) (Fig. 1). The outcrop in the Vallon de la Fauge solely consists of these glauconitic and sandy marls and has been dated from theStoliczkaia dispar Zone (Late Albian, late Early Cretaceous;Moret, 1935). Thus,Rhinochelys amaberti is roughly contemporaneous with the specimens from the Cambridge Greensand Member (Fig. 2) (West Melbury Marly Chalk Formation, UK;Hopson, 2005;Martill & Unwin, 2012) and bears the same coloration as the specimens (seeFig. S1–S8) from Cambridge Greensand Member (Fischer et al., 2014).

Geography and geology of the Cambridge Greensand Member specimens

We describe a series of previously unreported testudine fossils originating from the Cambridge Greensand Member in the Cambridgeshire (Fig. 2) (UK) (seeFigs. S1–S8). The Cambridge Greensand Member (Fig. 2) is a Cenomanian remanié deposit, corresponding to theMantelliceras mantelli Zone and theNeostlingoceras carcitanense Subzone (Hopson, 2005). The Cambridge Greensand Member must have formed in a shallow epeiric sea environment (Unwin, 2001); its fauna comprises vertebrate and invertebrate remains (e.g.,: pterosaurs, ichthyosaurs, turtles, plesiosaurs, etc.Benton & Spencer, 1995) either dating from Cenomanian or Late Albian (Unwin, 2001). The latter have a dark brown color due to their phosphatised nature (Benton & Spencer, 1995) and originate from the underlaying Gault Formation (Fig. 2) (Benton & Spencer, 1995;Hopson, 2005;Martill & Unwin, 2012;Fischer et al., 2014).

Phylogenetic analyses

There is a considerable disagreement on the intra-generic diversity ofRhinochelys. As much as 16 distinct species have been proposed in the past (Seeley, 1869;Lydekker, 1889a;Moret, 1935;Collins, 1970), but more recent studies have considered one or two valid species ofRhinochelys:R. pulchriceps andR. nammourensis (Tong et al., 2006;Hooks, 1998;Hirayama, 1994), notably due to a lack of characters to differentiate them in existing datasets (Hooks, 1998). Nevertheless, there is a notable intrageneric disparity of skull morphologiesRhinochelys (Collins, 1970), which we confirmed by our study of a series of unreported skulls from the Cambridge Greensand Member (seeSupplementary Information). To test this, we established a series of morphotypes (Rhinochelys morphotype ‘cantabrigiensis’,Rhinochelys morphotype ‘elegans’, and ‘GS63 GS67’), based on the data fromOwen (1851),Lydekker (1889b),Lydekker (1889a) andCollins (1970) and our own interpretations (seeTable 1), and we assessed their phylogenetic position. The rationale for each morphotype referal can be found in theSupplementary Information.

We modified the dataset ofCadena & Parham (2015). This matrix was chosen as the most recent dataset available for Testudines with a wide character sampling. We added four new taxa to this dataset: (i)Rhinochelys amaberti, using the holotype skull (UJF-ID.11167) and the associated mandible; (ii)Rhinochelys morphotype ‘elegans’Lydekker (1889a), using the specimens IRSNB GS64 and IRSNB GS65 (seeFigs. S1 andS2; seeTable 1), as well as the description and drawings ofLydekker (1889a) andCollins (1970); (iii)Rhinochelys morphotype ‘cantabrigiensis’Lydekker (1889a) based on the description and pictures of the holotype fromLydekker (1889a) andCollins (1970); and (iv) IRSNB GS63 and IRSNB GS67 (seeFigs. S7 andS8; seeTable 1) as representatives of a potentially new taxon from the Cambridge Greensand. We also revised the scores ofRhinochelys pulchriceps (Owen, 1851) using IRSNB GS68 (seeFigs. S3–S5; seeTable 1) plus the descriptions and pictures ofOwen (1851),Lydekker (1889a) andCollins (1970). Finally, we have corrected a series of erroneous scores forSyllomus aegyptiacus (Lydekker, 1889b) (character 10 is scored 1 instead of 0, and character 40 is scored 1 instead of ?) (Weems, 1980),Euclastes platyopsCope (1867) (character 1 is scored 1 instead of ?, and character 10 is scored 0 instead of ?) (Hay, 1905),Euclastes acutirostrisJalil et al. (2009) (character 10 is scored 0 instead of 1) (Jalil et al., 2009),Pacifichelys (character 10 is scored 0 instead of 1),Natator depressus (Garman, 1880) (character 10 is scored 0 instead of 1) (Brinkman et al., 2009).

The original character set ofCadena & Parham (2015) could not resolve the relationships of early sea turtles (seeFig. S9). Indeed, in this dataset, ‘middle’ Cretaceous (Aptian-Turonian) chelonioid all have identical or nearly identical character scoring, necessarily resulting in a large polytomy (Cadena & Parham, 2015). This effect was further increased with the new taxa added above. Moreover, the presence of a large number of unrelated continental (including freshwater) taxa in the original matrix likely produced far-field effects that split Pan-Chelonioidea among Cryptodira (‘full matrix’; seeFig. S9) (Myers et al., 2017). To cope with these issues, we removed all taxa that are likely non-marine, and chose the Late JurassicSolnhofia parsonsiGaffney (1975) as the outgroup.Solnhofia is a pan-cryptodiran that appears clearly outside the Pan-Chelonioidea clade, making a sensible outgroup for the present analysis, as the relationships of more derived pan-cryptodirans has not reached a consensus (e.g.,Zhou & Rabi (2015);Cadena & Parham (2015)). InCadena & Parham (2015),Solnhofia is the closest taxon to the marine turtle node without being part of it. As mentionned above, new characters are needed to recover the relationships of chelonioids. We have thus added (and illustrated) 10 new characters and altered the character #10 ofCadena & Parham (2015), using our observations of Albian cheloninoids as well as thorough analysis of the literature (seeFigs. S10–S19). These characters underwent a critical evaluation based on the recommendations ofSimoes et al. (2016). We used Mesquite v.3.10 to manage the nexus file. The resulting dataset (‘cheloninoid matrix’) contains 48 taxa and 266 characters.

We also assess the phylogenetic relationships of chelonioids using a different but widely used dataset: the matrix fromBardet et al. (2013), which stems fromKear & Lee (2006). Five new taxa were added to this dataset (called ‘Bardet matrix full’): (i) Rhinochelys nammourensis using descriptions and pictures fromTong et al. (2006); (ii) Rhinochelys amaberti; (iii)Rhinochelys morphotype ‘elegans’; (iv)Rhinochelys morphotype ‘cantabrigiensis’ and (v) ‘IRSNB GS63 IRSNB GS67’ (using the same data sources as in the chelonioid matrix discussed above). No new characters were added in this matrix in order to maintain two widely distinct datasets.

All phylogenetic datasets were analyzed in maximum parsimony using the New Technology Search (parsimony ratchet) of TNT v.1.5 (Goloboff & Catalano, 2016) (drift and ratchet activated; 200 ratchet iterations) to identify a series of ‘best candidate’ trees that were then subjected to tree branch reconnections. Bremer Decay indices were computed using TNT v.1.5. Because the strict consensuses of both datasets were largely unresolved due to the inclusion of poorly known OTUs, we used the a posterior method ‘Iterative Positional Congruence Reduced’ (IterPCR) established byPol & Escapa (2009), and recently implemented within TNT v.1.5 (Goloboff & Catalano, 2016), to identify wildcard taxa. For the ‘Bardet matrix’, the removal ofRhinochelys morphotype ‘cantabrigiensis’ was sufficient to recover a well-resolved consensus (called ‘Bardet matrix’ throughout). All strict consensus cladograms were computed and time calibrated (ages obtained from paleobiology database) in R using thestrap package (Bell & Lloyd, 2015). Branch lengths were reconstructed using both the ‘basic’ and ‘equal’ methods. The stratigraphic congruence of the most parsimonious trees was evaluated using theape (Paradis, Claude & Strimmer, 2004),geoscale (Bell & Lloyd, 2015), andpaleotree (Bapst, 2012), andstrap (Bell & Lloyd, 2015) packages.

By all means, our phylogenetic analyses should be regarded as preliminary, as it appears clear from our results that much more data are needed to recover robust relationships among chelonioids. To further test the robustness of some of our results, we used Templeton’s nonparametric test implemented in PAUP* (Templeton, 1983). This method establishes the statistical difference of length between topologies. The trees were constrained using the force function in TNT.

Systematic palaeontology

Description of the holotype ofR. amaberti

Thecranium is 5.3 cm long and 2.4 cm high. It is dorsoventrally compressed (Fig. 3) and is wider than it is long. This general shape of the cranium thus differs from those ofR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970);Rhinochelys morphotype ‘elegans’ (Lydekker, 1889a;Moret, 1935;Collins, 1970) andR. nammourensis (Tong et al., 2006), but appears similar to that ofRhinochelys morphotype ‘cantabrigiensis’ (Lydekker, 1889a;Moret, 1935;Collins, 1970) in having a skull that is wider than long. The anterior surface of the skull is vertically oriented in lateral view (Fig. 4), resulting in an anteroposteriorly short skull lacking a rostrum or a beak unlike inArchelon (Hay, 1908),Ocepechelon (Bardet et al., 2013) andProtostega (Hay, 1908), but is similar toAshleychelys (Weems, 2014). Posteriorly to the level of the naris, the skull rapidly reaches its maximum height, at the level of the first half of the anterior margin of the orbit (Figs. 4 and5). Such a skull shape is also seen inR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970). In dorsal view (Fig. 6), the skull expands laterally just posterior to the orbits. This lateral expansion of the skull extends up to the postorbital-jugal-quadratojugal suture, as inR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970) and unlike inRhinochelys morphotype ‘elegans’ orRhinochelys morphotype ‘cantabrigiensis’ (Lydekker, 1889a;Moret, 1935;Collins, 1970).

Themaxilla possesses a shallow sinusoidal sulcus (Figs. 4 and5) which connects the nasal cavity and the orbit. The maxilla forms an important bulge, dorsally to this sulcus. This bulge is here larger than in any other species currently referred to asRhinochelys (Lydekker, 1889a;Moret, 1935;Collins, 1970). This feature does not seem to be age-related in protostegids (Collins, 1970;Tong et al., 2006). As a consequence, the maxillary bone is hardly visible in dorsal view whereas in ventral view it is predominant and conceals a great portion of the prefrontal (Fig. 6). In all other species currently referred to asRhinochelys (exceptR. pulchriceps, seeFigs. S3 andS6), the bulge is shallow and cannot be seen in ventral view but only in anterior and dorsal views (Lydekker, 1889a;Collins, 1970) (also seeFigs. S1,S2,S7 andS8). This bulge has, however, been regarded as a common trait amongRhinochelys (Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). The maxilla also contributes to the orbital rim (Fig. 5) up to the three-fourths of its ventral margin (infraorbital margin), and up to the mid-height of its anterior margin (as in the other species currently referred to asRhinochelys) (Lydekker, 1889a;Moret, 1935;Collins, 1970). The maxilla contacts the nasal at the dorsal and lateral extremities of the nasal fossa. The maxilla also contacts the premaxilla and the prefrontal.

Theparieto-frontal dome is parallel to the ventral surface of the maxilla, but is not completely flat (Figs. 4 and5). Such an extended flattened parieto-frontal dome is also found inR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970). There are several cranial sulci running across the frontal and parietal (Fig. 6). These sulci contain the parietal-postorbital and frontal-postorbital sutures, and form a ‘Y’ structure on the frontals which then continues on the nasals (the stem of the ‘Y’ coincides with the median suture of theses bones). Theparietal is short and wide, smaller than that ofR. pulchriceps (Owen, 1851;Collins, 1970) andR. nammourensis (Tong et al., 2006).

In profile, the‘snout’ (formed by the premaxillae and extremity of maxillae;Figs. 4 and5) ofR. amaberti is shorter and narrower than that ofR. pulchriceps. Indeed, the ventral view reveals that the snout ofR. amaberti forms a ‘V’ shape (Fig. 6) rather than the ‘U’ shape seen inR. pulchriceps (Owen, 1851;Lydekker, 1889a;Collins, 1970).

Thenasal cavity is oval in shape, with its long axis oriented horizontally (Fig. 3). In profile, the ventral edge of the nasal fossa, which is formed by the maxilla, forms a right angle (Figs. 4 and5). This region is rounded rather than angular in the other species currently attributed toRhinochelys (i.e.,‘R. cantabrigiensis’, ‘R. jessoni’ (Lydekker, 1889a;Moret, 1935;Collins, 1970)) or clearly forms an obtuse angle (i.e.,R. pulchriceps, andRhinochelys morphotype ‘elegans’ (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970)) (see alsoFigs. S1–S3,S6–S8).

In comparison to other chelonioids, theorbit is large compared to the size of the skull: in profile, its height almost makes up for the total height of the skull (Figs. 4 and5), which is a common feature of the genusRhinochelys (see alsoFig. S4) (Owen, 1851;Lydekker, 1889a;Collins, 1970). The orbit ofR. amaberti is oval in shape, as inR. pulchriceps (Owen, 1851;Lydekker, 1889a;Collins, 1970), and unlike the rounded shape seen inRhinochelys morphotype ‘elegans’,Rhinochelys morphotype ‘cantabrigiensis’, (Lydekker, 1889a;Moret, 1935;Collins, 1970). The dorsal opening of the orbit seems greater than that of otherRhinochelys species (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970) with the exception ofR. nammourensis (Tong et al., 2006). The orbit is still facing laterally like in the otherRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006), and unlike inEuclastes (Hay, 1908;Lynch & Parham, 2003;Hirayama & Tong, 2003;Jalil et al., 2009),Ocepechelon (Bardet et al., 2013),Alienochelys (De Lapparent de Broin et al., 2014), andAllopleuron (Ubaghs, 1875). The orbital rim is formed by the maxilla (the most important contribution), jugal, postorbital, frontal (the smallest contribution) and prefontal.

Thefrontal is large and makes up an important portion of the upper orbital rim (Fig. 6). The anterior expansion meeting the nasal bone is extremely small, preventing the prefrontals from meeting medially. However it possesses an important dorsal expansion. The sulcus forming a ‘V’ shape medially (see above) is narrower than the one ofRhinochelys morphotype ‘cantabrigiensis’ andRhinochelys morphotype ‘elegans’ (Lydekker, 1889a;Moret, 1935;Collins, 1970), but similar to that ofR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970). The frontal contacts the nasal bone, prefontal, postorbital and parietal. The surface defined by the nasal and frontal (Figs. 4 and5) is inclined dorsally with respect to the ventral surface of the maxilla by an angle of 25°. As inR. pulchriceps, the participation of the frontal to the supraorbital rim is small (Owen, 1851;Lydekker, 1889a;Collins, 1970), this frontal participation is clearly larger inRhinochelys morphotype ‘elegans’ andRhinochelys morphotype ‘cantabrigiensis’ (Lydekker, 1889a;Moret, 1935;Collins, 1970).

Anasal bone is present (Figs. 3 and6), which is a feature found in all protostegids (Zangerl, 1953). It is wider than it is long (6 mm × 1.6 mm), and shows a 50° angle with respect to the ventral surface of the maxilla (Figs. 4 and5). The nasal meets the maxilla dorsally to the upper margin of nasal fossa (Figs. 4–6); their suture extends up to the dorsal border of the nasal fossa (Figs. 4 and5). Nasal and frontal also share a suture which prevents the prefrontals from meeting medially. This suture is extremely reduced, compared to other species currently assigned to the genusRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). The nasal bone meets the frontal, prefrontal and the maxillary bone.

Theprefontal (Fig. 6) is large and possesses an important dorsal expansion (greater than in other species currently attributed toRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006)) which is however smaller than the dorsal expansion of the frontal (like in other species currently assigned toRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006)). The frontals almost meet medially but are separated by the thin nasal-frontal suture (Fig. 6). Thus the frontal only shares a suture with the maxilla, nasal and frontal. The prefrontal contributes greatly to the supraorbital margin, especially in comparison to the frontal. This prefrontal contribution is more important than in otherRhinochelys species (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). The prefrontal also accounts for the anterior bulge of the skull, but to a lesser degree than the maxilla.

Thepostorbital reaches ventrally to the mid-height of the orbits where it meets the jugal (Figs. 4 and5). Posteriorly, it shares a suture with the quadratojugal and the squamosal, but does not contact the quadrate. Dorsally, the postorbital also starts at the mid-length of the orbits where it meets the frontal and the parietal. The dorsal expansion of the postorbital is well developed as the dorsal width of the skull extends as far as the postorbital-jugal-quadratojugal suture (Fig. 6). The covering of the temporal region by the postorbital, squamosal and parietal is a protostegid feature (Zangerl, 1953).

Theventral border of both maxilla and premaxilla is entirely flat; there is no anterior notch or lifting forming a beak (Figs. 4 and5). This differs from the other species ofRhinochelys found in the UK (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970).‘Rhinochelys elegans’ possesses a rather flat surface since the ventral margin of its premaxilla slightly rises dorsally (Lydekker, 1889a;Moret, 1935;Collins, 1970).

Thepremaxillary profile is straight (Figs. 4 and5) likeRhinochelys morphotype ‘elegans’ (Lydekker, 1889a;Collins, 1970) (while it is convex in the other species ofRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970)). The premaxilla is rather small, and its height is slightly greater (7 mm) than its width (5 mm).

Thequadrate is circular in shape (Figs. 4 and5) but is dorso-ventrally crushed and not complete (posteriorly broken). The quadrate entirely comprises the cavum tympani, like in the other species currently assigned toRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). It shares a suture with the quadratojugal, squamosal and pterygoid. The quadrate also bear the incisura columella auris and the incomplete processus trochlearis oticum. These can be seen in posterior view of the skull (Fig. 6). The ventral surface of the quadrate condyle is almost flat, and is twice as wide than long.

Thequadratojugal is semi-circular in shape and partially wraps around the quadrate (Figs. 4 and5): at the dorsal-most point of the quadrate, the quadratojugal contacts the squamosal which then wraps around the other half of the quadrate. These morphologies are also found in the other species ofRhinochelys (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). The quadratojugal contacts the jugal, postorbital, squamosal, and the quadrate. Thesquamosal is not complete and meets the quadrate, quadratojugal and the postorbital (Figs. 4 and5).

Thejugal is crescent-shaped, as is the quadratojugal, but its concavity is anteriorly oriented (Figs. 4 and5). It meets the maxilla at the last posterior fourth of the orbit, the postorbital at the posterior mid-height of the orbits, and the quadratojugal posteriorly. This jugal morphology is similar to that ofR. pulchriceps (Owen, 1851;Lydekker, 1889a;Moret, 1935;Collins, 1970;Tong et al., 2006). However, the jugal is of similar size and expansion in all of the other species currently referred to asRhinochelys, and also possesses the same sutures (which differ in size and position).

Thebasisphenoid andpterygoid are damaged and only visible in ventral view (Fig. 6). The basisphenoid is triangular in shape, shows at least three foramina, and separates the pterygoids posteriorly. Several arterial patterns have been highlighted among living and extinct turtles, which are based on one of the two broad forms: (A) either the bifurcation between the palatal and cerebral branches of the carotid artery is floored by bones, (B) or this split occurs outside the skull (Sterli & Fuente, 2010;Sterli et al., 2010). The posterior extension of the pterygoid in ventral view also influences the position of the carotid foramina (Myers et al., 2017). These foramina might not be the cerebral and palatal branches of the carotid artery because a single foramen should not be paired in derived turtles (D Brinkman, pers. com., 2018). Therefore the identity of the remaining foramina remains ambiguous at the moment. The pterygoids meet medially, but their anterior portion is not preserved and thus we cannot observe any foramina to compare with the basisphenoid. The pterygoid contacts the quadrate laterally.

Theassociated mandible (Fig. 7) is not complete and only presents partial dentaries. In dorsal view, a thin crest is visible on the anterior part of the symphysis with two small notches on each side. In profile, the dentary reaches its maximal thickness rapidly; it does not thicken posteriorly. The dentary forms a subtle beak: its anterior surface faces mostly anteriorly. In ventral view the symphysis presents a sulcus. If this mandible belongs to the holotype ofRhinochelys amaberti, then the symphysis length would be less than one-third of the total jaw length as estimated from the length of the skull.

Phylogenetic analyses

Our analysis of the chelonioid dataset yielded in 42 most parsimonious trees, each with a length of 399 steps. The Consistency Index is 0.421 and the Retention Index is 0.631, giving a Rescaled Consistency Index of 0.266. The strict consensus of the most parsimonious trees (Fig. 8) recovers the three traditional chelonioid families (Protostegidae in green, Cheloniidae in blue, Dermochelyidae in red) but with slightly different compositions than in previous attempts (e.g.,Hirayama, 1998;Kear & Lee, 2006). Because of our focus is on Protostegidae andRhinochelys in particular, only some of these differences will be discussed here; others are discussed in theSupplementary Information. The deep nodes of the Pan-Chelonioidea clade are not well resolved, forming a polytomy that includes several other taxa previously regarded as cheloniids (Gaffney & Meylan, 1988;Hirayama, 1994;Hirayama, 1997;Weems, 2014;Brinkman et al., 2015), such asCorsochelys haliniches, formerly regarded as dermochelyid (Hirayama, 1994;Hirayama, 1997;Hooks, 1998) or a cheloniid (Bardet, 1995;Kear & Lee, 2006) andNichollsemys baieri, previously regarded as a cheloniid (Cadena & Parham, 2015) or close toToxochelys and cheloniids (Brinkman et al., 2015), which form here part of the Pan-Chelonioidea basal polytomy. This topology resembles that ofBrinkman et al. (2006), who also recoveredNichollsemys as an early chelonioid.Ctenochelys groups withAllopleuron andLophochelys, which were formerly regarded as cheloniids (Hirayama, 1994;Hirayama, 1997;Weems, 2014). This clade is also included in the general polytomy. As a consequence,Toxochelys is here isolated outside Pan-Chelonioidea.

Protostegidae + Dermochelyidae do not form a clade, unlike in some previous studies of chelonioid relationships (the ‘Dermochelyoidea’ ofGaffney & Meylan (1988);Lehman & Tomlinson (2004);Bardet et al. (2013);De Lapparent de Broin et al. (2014);Cadena & Parham (2015) or the ‘Pandermochelys’ ofKear & Lee (2006)). Basal protostegids includeCalcarichelys and, unexpectedly, the Neogene taxonSyllomus aegyptiacus. This taxon was previously regarded as a cheloniid of uncertain affinities (Lynch & Parham, 2003;Parham & Pyenson, 2010) or a dermochelyid (Cadena & Parham, 2015). The following state changes support the clade unitingSyllomus with other protostegids: character 10.0; character 88.0; character 116.3 and character 263.1. The other characters (45, 57, 58, 66, 75, 137, 153, 197, 198, 247, 251) that support the node unitingSyllomus with protostegids (i.e., node 73 → 64 from the apomorphy list inSupplementary Information) are unknown forSyllomus. We tested the robusteness of the placement ofSyllomus as a protostegid instead of a cheloniid. In TNT, we constrained tree search to recover a clade unitingSyllomus + cheloniids (i.e.,Ashleychelys palmeri, Euclastes wielandi, Euclastes acutirostris, Euclastes platyops, Erquelinnesia gosseleti, Puppigerus camperi, Itilochelys rasstrigin, Argillochelys africana, Argillochelys cuneiceps, Mexichelys coahuilaensis, Terlinguachelys fischbecki, Pacifichelys spp., Eochelone brabantica, Tasbacka ouledabdounensis, Tasbacka aldabergeni, Chelonia mydas, Lepidochelys kempii, Lepidochelys olivacea, Caretta caretta andNatator depressus). This yielded most parsimonious trees that have a length of 413 steps, which is 14 additional steps than in the unconstrained analysis. Still, Templeton’s test (Templeton, 1983) indicates that this solution is not statistically distinguishable (p = 0.0668) from the most parsimonious solution (i.e.,Syllomus as a protostegid).

WhileChelosphargis andCalcarichelys are still recovered as protostegids, they are not closely related, unlike inZangerl (1960),Gaffney & Meylan (1988),Hirayama (1994),Bardet (1995) andHirayama (1997).Calcarichelys is, however, extremely fragmentary, which might affect its position. Derived protostegids separate into two clades (‘a’ and ‘b’ inFig. 8).Santanachelys gaffneyi (clade ‘b’) is recovered in a more derived position than inHirayama (1998),Kear & Lee (2006),De Lapparent de Broin et al. (2014) andCadena & Parham (2015) and is grouped withDesmatochelys padillai. This ‘intermediate’ position ofSantanachelys gaffneyi andBouliachelys suteri in between the two species ofDesmatochelys suggests that the Barremian taxonD. padillai does not belong toDesmatochelys, otherwise known from the Turonian-Santonian interval (Cadena & Parham, 2015). This clustering is coherent sinceSantanachelys and the species ofDesmatochelys also bear several primitive protostegid features (Zangerl, 1960;Elliot, Irby & Hutchison, 1997).Chelosphargis remains a protostegid (Zangerl, 1960;Gaffney & Meylan, 1988;Bardet, 1995), and is placed at the most primitive position in clade ‘b’.

Our main result of the analysis of the ‘Chelonioid’ dataset is that the genusRhinochelys is no longer recovered as a basal protostegid and is split in two clades (Fig. 8),Rhinochelys morphotype ‘elegans’,Rhinochelys morphotype ‘cantabrigiensis’ and the unnamed taxon represented by the specimens IRSNB GS63 and IRSNB GS67 are placed among clade ‘b’ (close toD. padillai andSantanachelys), whereasR. pulchriceps, R. nammourensis andR. amaberti are placed within clade ‘a’. ARhinochelys-Desmatochelys affiliation has been proposed before (Bardet, 1995) and is here recovered for a subset of this genus only (Rhinochelys morphotype ‘elegans’,Rhinochelys morphotype ‘cantabrigiensis’, IRSNB GS63-IRSNB GS67). Clade ‘a’ presents two sister lineages (Fig. 8): the clade containingR. amaberti, R. pulchriceps, R. nammourensis, and the clade containing bothArchelon-Protostega andOcepechelon-Alienochelys. This position of theOcepechelon-Alienochelys clade differs from previous studies where they were considered dermochelyoids, and thus sister taxa to dermochelyids and protostegids (alongsideBouliachelys) (Bardet et al., 2013;De Lapparent de Broin et al., 2014). The clustering ofOcepechelon andAlienochelys is possibly attributable to their peculiar morphology. Indeed, these poorly known and aberrant taxa have barely no character available to distinguish them in the dataset we used.

We tested the polyphyly hypothesis ofRhinochelys using a Templeton test. In TNT we forced the strict monophyly ofRhinochelys; this produced most parsimonious trees with a length of 400 steps, which is just one step longer than the most parsimonious solution. Logically, the Templeton test yielded a highp-value (p = 0.7389), indicating that the polyphyly and monophyly ofRhinochelys are statistically indistinguishable in this dataset.

Our analysis of the ‘Bardet matrix full’ resulted in 99 most parsimonious trees, each with a length of 245 steps (seeFig. S20). We re-run the analyses after deleting the wildcard taxon identified by the IterPCR method (Rhinochelys morphotype ‘cantabrigiensis’). This resulted in 27 most parsimonious trees, each with a length of 245 steps as well (Fig. 8). The strict consensus cladogram presented inFig. 8 also recovers a non-monophyleticRhinochelys:R. amaberti, R. pulchriceps andRhinochelys morphotype ‘elegans’ are recovered as closedly related, forming a polytomy close to the base of the clade Protostegidae.Ocepechelon bouyai is also included within this polytomy. The morphotype ‘GS63-GS67’ is more closely related toSantanachelys than the otherRhinochelys species.R. nammourensis is recovered as a more derived protostegid, and forms a polytomy withProtostega spp.,Chelosphargis andDesmatochelys lowi. Other aspects of the topology differ from the results obtained from the analysis of the chelonioid dataset (Fig. 8). This is possibly due to the narrower sampling of taxa and different character set. Both dataset agree that the generaDesmatochelys andRhinochelys, as presently defined, might not be monophyletic.

Stratigraphic congruence

Stratigraphic congruence metrics were computed for the ‘chelonioid’ matrix (Fig. 9). For the two metrics considered (the Gap Excess Ratio, GER, and the Stratigraphic Congruence Index, SCI), the scores obtained for the input trees are at most marginally better than that of randomly generated trees (with slightly better results for the GER than the SCI). These indicate weak stratigraphic congruence for the most parsimonious trees, whose topologies force long ghost lineages. This suggests that either the morphological data presently available severely biases the phylogeny of fossil chelonioids and/or that the sampling of fossil chelonioids is poor.