The last surviving Thalassochelydia—A new turtle cranium from the Early Cretaceous of the Purbeck Group (Dorset, UK)

Background. The mostly Berriasian (Early Cretaceous) Purbeck Group of southern England has produced a rich turtle fauna dominated by the freshwater paracryptodires Pleurosternon bullockii and Dorsetochelys typocardium. Each of these species is known by numerous relatively complete shells and by a single cranium. The two other turtles found in the Purbeck Group (Hylaeochelys belli, a species of uncertain affinities, and the terrestrial helochelydrid "Helochelydra" anglica) are known only from shell remains. Methods. In the present contribution, we describe a new turtle cranium from the Purbeck Group of Swanage, Dorset (southern England). We also explore the phylogenetic relationships of this new cranium and of Hylaeochelys belli in the context of a recently published global turtle matrix. Results. Before complete preparation, the new Purbeck cranium was provisionally referred to Dorsetochelys typocardium, but our analysis clearly contradicts a referral to this species in particular and to paracryptodires in general. In contrast, the new cranium shares a number of features with the Late Jurassic, coastal marine Thalassochelydia, including a posterolaterally open foramenposterius palatinum, a strong ridge on the posterior surface of the processus articularis of the quadrate, a strong posterior orientation of the processus articularis in ventral view, and a processus trochlearis oticum limited to the medial part of the otic chamber and bordered by a deep recess laterally. Our phylogenetic analysis confirms a placement of the new Purbeck cranium within the clade Thalassochelydia. Discussion. In terms of morphology, the new Purbeck cranium does not correspond to any known taxon. However, we refrain from naming a new species based on it because there is a good chance that this cranium actually belongs to the shell-based species Hylaeochelys belli (also recovered as a thalassochelydian in our phylogenetic analysis). Unfortunately, we lack any objective evidence to support this conclusion for themoment. In any case, the new Purbeck cranium confirms what others have previously suggested based on Hylaeochelys belli: thalassochelydian turtles survived the Jurassic-Cretaceous transition.

The Purbeck Group of Dorset (southern England) has been extensively exploited since the nineteenth cen-2 tury, both for building stones and more directly for fossil material. Turtles are among the most common 3 vertebrate remains found in these layers and are mostly represented by relatively complete shells. The is no strong morphological argument to support or discredit this hypothesis, Hylaeochelys belli has been 23 repeatedly suggested to have affinities with Plesiochelyidae (Lydekker, 1889;Nopcsa, 1928 Dorchester (United Kingdom) the following year. According to the label, the specimen was subsequently 44 prepared in the Department of Zoology of the University of Oxford, but this preparation was only super-45 ficial. The skull remained embedded in the matrix and only observable in palatal view (Fig. 1). Apart from 46 a mention in Milner (2004), this skull was never described or figured in the literature. 47 The specimen was borrowed from the DORCM by JA in 2010 and fully prepared in 2016 by Renaud  (Batten, 2002). DORCM G.10715 was indeed 60 found in a hard limestone rich in shell fragments. 61 Durlston Bay represents one of the largest exposures of the Purbeck Group along the Dorset coast. 62 The Purbeck Group consists of a succession of limestones and calcareous shales and marls with occa-63 sional evaporites that were deposited in a marginal setting (Allen and Wimbledon, 1991;Clements, 64 1993). The climate during this period is interpreted to have been semi-arid and becoming wetter to-65 ward the top of the sequence (Allen, 1998;Batten, 2002). The Purbeck Group represents a transition 66 between the marine succession of the underlying Portland Limestone Group and the non-marine sedi-67 ments of the overlying Wealden Group, reflecting a major regression occurring at the end of the Jurassic 68 and onset of the Cretaceous (e.g., Rawson, 2006). 69 For long, the Purbeck Group was subdivided into Lower, Middle and Upper Purbeck. This is the only 70 stratigraphical data we have for most of the fossils collected in the nineteenth century (e.g., Milner, 2004). 71 The Purbeck Group is now divided into two formations: the oldest Lulworth Formation and the youngest 72 Durlston Formation (Townson, 1975). Ensom (1985) at Worbarrow Tout and Clements ( members, each of them consisting of several beds (Fig. 3). Later, Westhead and Mather (1996) argued 75 that these so-called members were difficult to follow at the regional scale. Therefore, these authors pro-76 posed to divise the Purbeck Group into five members and to relegate the traditional members to the bed 77 status, but the latter continue to be used at least locally for the level of stratigraphical detail they provide 78 (Ensom, 2002 (Gaffney, 1975(Gaffney, , 1976Rieppel, 1980 using New Technology algorithms in TNT with default settings, tree drifting (Goloboff, 1999), parsimony 116 ratchet (Nixon, 1999), and constraints enforced. The initial level of the driven search and the number 117 of times the minimum tree length should be found were both set to 30. The most parsimonious trees 118 (MPTs) were subjected to an additional round of tree bisection and reconnection (TBR    nial roof contacts the palate, and the entire internal anatomy is unfortunately lost to us. The squamosals 169 as well as most of the jugals and quadratojugals are missing (Fig. 4). The ventral aspect of the skull was 170 exposed when discovered and subjected to weathering ( Fig. 1) Laterally, they form the dorsal margin of the orbits. The damage of the medial part of the left prefrontal 208 reveals that the frontals extend anteriorly beneath the prefrontals to meet the nasals. As noted above, a 209 similar morphology was described in some, but not all, thalassochelydians (Gaffney, 1976 The premaxillae are poorly preserved in DORCM G.10715. Only the midline suture between the two 256 premaxillae can be seen clearly ( Fig. 4C-D). However, it is apparent that the premaxillae contact the 257 maxillae laterally and the vomer posteriorly. Foramina praepalatinum were likely present, but this cannot 258 be confirmed due to poor preservation. Dorsally, the premaxillae form the ventral margin of the apertura 259 narium externa (Fig. 4A-B). On the left side, a smaller, but well defined foramen opens slightly anterior to the foramen alveolaris 286 anterior on the floor of the orbit (Fig. 4A-B). The same foramen was probably also present on the right 287 side, but the area is damaged. The nature of this foramen remains uncertain, but a similar foramen was 288 described in several thalassochelydians as well as in the stem-group chelonioid Toxochelys, although the 289 distribution and homology of this feature remains to be investigated (Gaffney, 1976; Anquetin et al., .

291
Vomer 292 As usual the vomer forms the medial septum of the apertura narium interna (Fig. 4C-D). palatines that can be seen relatively easily in some areas. Anteroventrally, the vomer widens and contacts 295 the premaxillae and maxillae, but this area is poorly preserved (e.g., the foramina praepalatinum cannot 296 be seen). Posterodorsally, the vomer flattens and widens to form part of the palatal roof. There, the 297 vomer appears to contact the pterygoids as well as the palatines.

299
In ventral view, the palatines contact the vomer medially, the maxillae anterolaterally, and the pterygoids 300 posteriorly (Fig. 4C-D). The suture with the pterygoids is poorly preserved but visible by intermittence. 301 Although this area is not very well preserved, it appears that the palatines are fully separated from one 302 another in ventral view as a result of a contact between the vomer and the pterygoids. This is also the 303 condition observed in most thalassochelydians, except Jurassichelon oleronensis (Rieppel, 1980) and per-304 haps Plesiochelys planiceps (Gaffney, 1975(Gaffney, , 1976  The ventral aspect of the pterygoids is much eroded. The pterygoid contacts the vomer (probably) and 346 palatine anteriorly, the quadrate posterolaterally (suture not preserved), the basioccipital posteromedi-347 ally, the parabasisphenoid medially, and the other pterygoid anteromedially (Fig. 4C-D). It was reported 348 that Pleurosternon bullockii and Glyptops ornatus lack a midline contact of the pterygoids due to a long ven-349 tral exposure of the parabasisphenoid that would reach the vomer anteriorly (Evans and Kemp, 1975; The supraoccipital contacts the prootic anteroventrally, the quadrate laterally (short contact), and the 375 opisthotic posterolaterally (Fig. 4A-B The exoccipitals are severely affected by the dorsoventral compression of the cranium that resulted in 383 a contact between the crista supraoccipitalis and condylus occipitalis. The exoccipital apparently con-384 tacted the supraoccipital mediodorsally, the opisthotic dorsolaterally, and the basioccipital ventromedi-385 ally (Fig. 4). The only preserved suture is the one with the opisthotic in the roof of the cavum acustico-386 jugulare. Two foramina nervi hypoglossi are visible on the left side, but only one is preserved on the right 387 side. Below this point, the preservation of the exoccipitals is extremely poor. Therefore, it is unknown 388 whether the exoccipital contacted the pterygoid as in many cryptodires (Gaffney, 1979a). The extent to 389 which the exoccipitals contribute to the condylus occipitalis is also uncertain.

391
The basioccipital contacts the parabasisphenoid anteriorly along a seemingly transverse suture (poorly 392 preserved), the pterygoid anterolaterally along a broad and oblique suture, and the exoccipital dorsally 393 (Fig. 4C-D). In DORCM G.10715, the prootic can only be observed on the dorsal surface of the otic chamber ( Fig. 4A-B). 397 The rest of the bone, notably in the ethmoid region, is not accessible. As preserved, the prootic contacts 398 the parietal anteromedially, the quadrate laterally and anteroventrally, and the supraoccipital posteriorly. 399 As noted above, a brief contact of the quadrate and supraoccipital posterior to the foramen stapedio-400 temporale prevents the prootic and opithotic to meet on the dorsal surface of the otic chamber. As 401 usual, the prootic forms the medial half of the foramen stapedio-temporale and processus trochlearis 402 oticum. The latter structure is modest in development, remarkable in being oriented quite obliquely 403 relative to the transverse axis, and limited to the medial part of the otic chamber (see Quadrate).

405
The opisthotic contacts the supraoccipital medially, the quadrate anterolaterally, and the exoccipital pos-406 teriorly (suture not preserved dorsally). At first it may appear that the opisthotic as an extensive dorsal 407 exposure, but this is due to the disarticulation and loss of the squamosal on both sides ( Fig. 4A-B). As pterygoid. As noted above, the morphology of the internal carotid sytem in DORCM G.10715 clearly re-425 minds the condition observed in some thalassochelydians, such as Plesiochelys etalloni, Plesiochelys bigleri, 426 and Jurassichelon oleronensis (Gaffney, 1976;Rieppel, 1980 is exactly the same as the one obtained during our first analysis (Fig. 5B). The merged DORCM G.10715 + 447 Hylaeochelys belli is recovered at the same position as the isolated cranium DORCM G.10715 before, that the fenestra caroticus and is thus ventrally exposed (ch. 147.1). It should be noted for the record that 478 thalassochelydians are particularly difficult to score for the latter character due to intraspecific variability 479 and preservation bias (Anquetin et al., 2015).

481
Beta taxonomy 482 When it was still partly embedded in matrix (Fig. 1) We believe the interrelationships of Thalassochelydia to be poorly understood for the moment. In 523 our phylogenetic analysis, the sister group relationships between DORCM G.10715 (or DORCM G.10715 524 + Hylaeochelys belli) and Jurassichelon oleronensis is only supported by three "reversals" to plesiomorphic 525 conditions, namely a midline contact of the pterygoids that is equal or shorter than the length of the 526 parabasisphenoid, a poorly developed crista supraoccipitalis, and a partly embedded internal carotid 527 system with the split into cerebral and palatine branches still exposed ventrally (see above). Therefore, 528 we will refrain from any comment on these relationships herein and only retain the clear placement of  (Fig. 4). However, we intentionally refrain from naming a new species based on this specimen be-535 cause we believe there is a great chance that it actually represents the cranium of the shell-based species 536 Hylaeochelys belli, which is found in the same layers. This is why DORCM G.  can make a few pertinent comments here. The morphology of the triturating surface is remarkable and 570 indicates that the two mandibles belong to the same taxon (Fig. 6). The anterior outline of the mandibles 571 is rather rounded. The triturating surface is relatively short and broad, with low lingual and labial ridges. 572 The lingual ridges become more pronounced anteriorly and do not meet one another on the symphysis, 573 rather bordering a narrow sagittal gutter. This morphology matches fairly well that of DORCM G.10715 574 (Fig. 4) and we are relatively confident that they can all be assigned to the same taxon, which is pos-575 sibly Hylaeochelys belli. Interestingly, the more complete of these two mandibles (NHMUK OR44815) 576 possesses a well-developed processus coronoideus projecting posteriorly, an angolachelonian feature 577 (non-exclusive) according to Evers and Benson (2019).

579
The configuration of the internal carotid arterial system and more specifically the position of the fora-580 men posterius canalis carotici interni has been given a great significance in terms of turtle systematics 581 (Albrecht, 1976;Gaffney, 1979a). Paracryptodires are usually depicted as having a foramen posterius 582 canalis carotici interni located halfway along the parabasisphenoid-pterygoid suture and facing ventrally. 583 The split between the two branches of the internal carotid artery (that is when the palatine branch is

586
The previous description applies fairly well to baenids, but less so to non-baenid paracryptodires. 587 This became plainly apparent to us when we revised the latter while describing DORCM G.10715. In 588 Pleurosternon bullockii and Glyptops ornatus, the internal carotid artery penetrates the cranium at the 589 level of the fused and still visible basipterygoid process, which lies in a shallow depression along the 590 parabasisphenoid-pterygoid suture. In both of these species, the authors describe and illustrate a short 591 sinus that extends posterolaterally from the foramen posterius canalis carotici cerebralis and crosses 592 the medial part of the basipterygoid process (Evans and Kemp, 1975;Gaffney, 1979b into a palatine and cerebral branches before really entering the cranium through the two corresponding 598 foramina. The split of the internal carotid artery is therefore exposed in ventral view in these two species 599 and there is no proper foramen posterius canalis carotici interni, because the internal carotid artery does 600 not travels through a canal before splitting into its cerebral and palatine branches, only through a short The above discussion highlights that the morphology of the internal carotid arterial system in non-632 baenid paracryptodires is much more variable than usually considered. We do not believe that the simili-633 tude between the condition in the paracryptodire Dorsetochelys typocardium (possibly also Uluops uluops) 634 and some thalassochelydians indicates close relationships. It more probably represents either a more 635 generalized ancestral state or a similar transitional step toward more derived arrangements of the in-636 ternal carotid arterial systems in both groups. For simplicity, we omitted to mention xinjianchelyids, 637 sinemydids and macrobaenids here, but they also represent transitional forms with their own specific 638 arrangement of the internal carotid arterial system (Rabi et al., 2013). A more comprehensive revision of 639 the existing variation in these early turtles is needed and will probably greatly impact the way characters