The vast majority of small-bodied turtles discussed herein (i.e. NHMUK PV OR 48262, 48263, 48263a, 48263c, 48263e, 48343, 48344, 48347 and 48354), including all specimens unambiguously referred to Tongemys enigmatica gen. et sp. nov., were explicitly listed by Lydekker (1889b) as originating from the collection of Samuel H. Beckles. The overall similarity of the remaining material referred herein (i.e. NHMUK PV OR 48264, 48345, 48351 and 48355) in their preservation and the adjacency of their catalog numbers make it highly plausible that these originate from this collection as well. This collection of fossil turtles was the by-product of an extensive excavation that had been carried out in search of mammalian remains under the supervision of Beckles in 1856 at Durlston Bay (Kingsley 1857) at a locality now known as “Beckles’ Pit” (Milner 2004). To reach the target of this excavation, a single layer known as the mammal bed (bed DB 83 in the terminology of Clements 1993), Beckles had up to 16 m of overburden removed from an area of more than 650 m² (Kingsley 1857). It is unclear, however, if the resulting small-bodied turtles were collected from the mammal bed per se or from the overburden. The latter conclusion is supported by the heterogeneity of the matrix in which non-turtle specimens are preserved (Sweetman et al. 2017). We, therefore, conclude that these specimens either originated from the Marly Freshwater Member, which includes the mammal bed or the overlying Cherty Freshwater Member, the two members exposed at Beckles’ Pit (Sweetman et al. 2017). These two Members are generally interpreted as representing lacustrine environments without marine influence. Among turtles, they otherwise yielded fragmentary helochelydrid remains (Barrett et al. 2002).

The best documented specimens of Pleurosternon bullockii, Dorsetochelys typocardium and Hylaeochelys belli, in contrast, originate from the overlying Intermarine Member (Milner 2004). This member is interpreted as ranging from lacustrine to lagoonal.

Although the exact location of the Jurassic-Cretaceous boundary within the Purbeck Group is still contentious, ammonite zonation, sequence stratigraphy and magnetostratigraphy suggest that the Marly and Cherty Freshwater Members are early Berriasian, while the Intermarine Member is middle to late Berriasian in age (Ogg et al. 1994).

The best-preserved small-bodied turtles from Beckles’ Pit were recently figured by Guerrero and Pérez-García (2021b). Although we here disagree with their interpretation of this material representing juveniles of Pleurosternon bullockii, we fully agree with their anatomical observations. To aid the reader, we, nevertheless, re-figure the two most telling specimens, NHMUK PV OR 48262/48265, the holotype (Fig. 1) and NHMUK PV OR 48343, the specimen with the best-preserved plastron (Fig. 2). In the rare instance where referred material provides important insights, we refer the reader to the figures of Guerrero and Pérez-García (2021b).

Figure 1. 

NHMUK PV OR 48262/48265, Tongemys enigmatica, holotype, Early Cretaceous (Barremian) of England: A. Photograph in dorsal view; B. Interpretive line drawing in dorsal view; C. 3D model of plastron in dorsal view; D. 3D model of plastron in ventral view. Abbreviations: co, costal; ent, entoplastron; epi, epiplastron; hyo, hyoplastron; hyp, hypoplastron; Ma, marginal scute; mes, mesoplastron; per, peripheral; Pl, pleural scute; Ve, vertebral scute; xi, xiphiplastron.

Figure 2. 

NHMUK PV OR 48343, Tongemys enigmatica, referred specimen, Early Cretaceous (Barremian) of England: A. Photograph in ventral view; B. Interpretive line drawing in ventral view. Abbreviations: AB, abdominal scute; EG, extragular scute; ent, entoplastron; epi, epiplastron; FE, femoral scute; GU, gular scute; HU, humeral scute; hyo, hyoplastron; hyp, hypoplastron; mes, mesoplastron; PE, pectoral scute; xi, xiphiplastron.

In the hope of recovering anatomical information from the ventral side of the holotype, which is covered in matrix, we subjected this specimen to high-resolution X-ray micro-computed tomography using a Nikon Metrology XTH 225 ST scanner at the NHMUK Imaging and Analysis Centre. The system set-up consisted of: a tungsten rotating reflection target; X-ray source set to 215 kV and 660 µA; source filtered with 1.5 mm of copper; detector gain of 24 dB; source-object distance of 358.7 mm and object-detector distance of 619.4 mm generating data with isotropic voxels with an edge length of 55.00 µm. Both parts of the specimen (NHMUK PV OR 48262 and 48265) were put together for the acquisition, which consisted of 4476 projections over a 360° rotation of the object, using the minimize ring artefact option of the instrument; each projection had a total integration time of 1 second resulting from 4 frame averaging of 250 msec exposure each. 3D mesh models were generated using the software Amira 2020.2 (Thermo Fisher Scientific, Hillsborough, USA). Data segmentation was performed combining manual masking every sixth slice with the interpolation tool in the z-axis. The 3D mesh models were generated and exported as .ply-files. The images used in Fig. 1 were taken as screen shots from SPIERSview 3.1.0. The original set of coronal slices and the 3D mesh models are available at MorphoSource (https://www.morphosource.org/projects/000434697).

Our primary comparative sources are as follows: Compsemys (a.k.a. Berruchelus) russelli (Pérez-García, 2012) from the Paleocene of France, as described by Pérez-García (2012); Calissounemys matheroni Tong et al., 2022 from the Campanian of France, as described by Tong et al. (2022); Compsemys victa Leidy, 1856 from the Paleocene of New Mexico, as described by Gilmore (1919); Peltochelys duchastelii Dollo, 1884 from the Early Cretaceous (middle Barremian to early Aptian) of Belgium, as described by Joyce and Rollot (2020); Pleurosternon bullockii from the Early Cretaceous (Berriasian) of England, as described by Guerrero and Pérez-García (2021a); and Selenemys lusitanica Pérez-García & Ortega, 2011 from the Late Jurassic (Kimmeridgian) of Portugal, as described by Pérez-García and Ortega (2011).

To investigate the phylogenetic relationships of compsemydids, we integrated Tongemys enigmatica gen. et sp. nov. into the paracryptodire character/taxon matrix of Rollot et al. (2021). The new turtle could be scored for 35 of 107 characters. The matrix was otherwise adjusted, by re-scoring the early compsemydids Peltochelys duchastelii and Selenemys lusitanica as “0,” not “1,” for characters 86 and 87, as they clearly do not exhibit extensive contacts between the inguinal and axillary buttresses with the costals, in contrast to the derived compsemydids Compsemys russelli, Compsemys victa and Kallokibotion bajazidi. The updated character taxon matrix is provided in the Suppl. material 1.

The expanded matrix was subjected to a parsimony analysis using TNT (Goloboff et al. 2008). Twenty-one characters (i.e. characters 6, 14, 16, 18, 27, 28, 31, 34, 39, 40, 41, 46, 48, 60, 63, 80, 88, 95, 97, 98 and 101) were run ordered because they form morphoclines. Proganochelys quenstedti served as the outgroup. One thousand random addition sequences were followed by a round of tree bisection reconnection. Trees suboptimal by 10 steps and with a relative fit difference of 0.1 were retained as part of the first search. A tree-collapsing rule was implemented with a minimum length of 0. Our analysis under equal-weighting resulted in 48 most parsimonious trees with 301 character-state transitions. The Pruned Trees function of TNT identified Riodevemys inumbragigas and Scabremys ornata as rogue taxa, which were subsequently removed. The 50% majority-rule tree is shown in Fig. 3. Tongemys enigmatica gen. et sp. nov. is retrieved as the most basal branching compsemydid.

Figure 3. 

Time-calibrated 50% majority-rule tree obtained from the phylogenetic analysis under equal-weighting, without rogue taxa. Unless otherwise stated, all nodes were found in 100% of all trees.

We use phylogenetic nomenclature throughout this manuscript. All names of taxa, including clades, are therefore highlighted using italics.

Compsemydidae Pérez-García et al., 2015

The taxon represented by the small-sized turtle material recovered from Beckles’ Pit has a somewhat unique taxonomic history, as authors have noted its morphological distinction ever since it was first reported in the 19^th century, but either interpreted it as the juvenile morph of Pleurosternon bullockii nevertheless (Lydekker 1889a, b; Guerrero and Pérez-García 2021b) or as a new species of compsemydid (Lapparent de Broin and Murelaga 1999).

Although we here conclude that the turtles from Beckles’ Pit represent a new species of compsemydid turtle, Tongemys enigmatica gen. et sp. nov., we agree that a number of characters are present that unite it with the coeval Pleurosternon bullockii, including a broadly similar surface texture, the number and arrangement of bones in the shell, a complete neural series, a four-sided neural I, the common fusion of neural VIII with suprapygal I, peripherals that are V-shaped anteriorly and posteriorly to the bridge, the absence of a cervical scute and a straight mid-line plastral sulcus. As our phylogenetic analysis cannot resolve the interrelationships of compsemydids, pleurosternids and baenids, it is unclear if these characters represent symplesiomorphies or homoplasies, although biogeographic arguments provide support for the former hypothesis, as early pleurosternids and compsemydids are best known from Europe.

In spite of the above listed similarities, we find considerable character evidence that distinguishes Tongemys enigmatica from Pleurosternon bullockii: much smaller size (a carapace length of ca. 15 cm versus ca. 55 cm); the development of a distinct nuchal notch that is framed by anterior protrusions of the first peripherals, instead of a rounded anterior margin; the near complete retraction of the nuchal from the anterior margin of the shell (ca. 20% of nuchal width contributes to the margin, instead of 50%), a nuchal that is laterally expanded to the approximate width of costal I resulting in a near contact of the nuchal with peripheral II which, in turn, hinders a clear contact between peripheral I and costal I, in contrast to a narrow nuchal and a broad contact between peripheral I and costal I; presence of neurals that are about as long as wide, not longer than wide; notably narrow peripherals; presence of a costal VIII that is significantly smaller than costal VII, not similar in size; a straight medial margin of costal VIII resulting in a triangular space for the suprapygals, not an angular margin; restriction of vertebral I on to the costals, instead of a clear overlap on to peripheral I; a near contact of vertebral I with marginal III, resulting in a short, not expanded contact between marginal II and pleural I; marginals that are restricted to the peripherals, broad pleurals and narrow vertebrals, instead of wide marginals that lap on to the costals, narrow pleurals and broad vertebrals; a distance between the axillary and inguinal notch that is less, not greater than the length of the plastral lobes; plastral lobes with parallel, not evenly converging sides; absence, not presence of a deep anal notch; absence, not presence of broad scute overlap on the dorsal side of the plastral hind-lobe; laterally contracting mesoplastra, instead of rectangular elements; hypoplastron much shorter than hyoplastron, not equal in anteroposterior length; extragular-humeral sulcus not orientated transversely, but rather sloping posterolaterally to nearly contact the epiplastral-hyoplastral suture; and restriction of the gulars to the epiplastron, instead of overlapping the entoplastron.

The vast majority of characters listed above were noted by Guerrero and Pérez-García (2021b), but attributed to ontogeny, as they felt that most differences pertain to proportions. While we agree with Guerrero and Pérez-García (2021b) that proportions change during the ontogeny of turtles, we disagree that this has been documented in literature for the shell of turtles beyond matter-of-fact statements. Indeed, we are only aware of very few morphometric studies that rigorously document shell growth for turtles beyond simple plots of length, width or height against age (e.g. Chiari and Claude 2011; Casale et al. 2017).

As part of their study, Guerrero and Pérez-García (2021b) gathered novel morphometric data from unambiguous Pleurosternon bullockii versus Tongemys enigmatica, in particular 2D landmarks that approximate the outlines of the nuchal, entoplastron, vertebral III and the anterior plastral lobe. In our estimation, the resulting principal component plots are insufficient to serve as taxonomy evidence for two primary reasons. First and foremost, the use of morphometrics of single shell elements for taxonomic purposes has never been used before and has not been tested by reference to known examples. Is it possible to distinguish extant species using these measurements, even closely related ones? If so, what patterns should we look for? Is this source of data sufficient to distinguish between taxonomic versus ontogenetic effects? These basic questions remain unanswered. As is, the small-bodied turtles variously plot within, outside or beyond the range of the large-bodied ones, which does not correspond to any particular hypothesis in a self-apparent way. Secondly, the nuchal, entoplastron, vertebral III and the anterior plastral lobe of the vast majority of turtles globally have a similar shape, being trapezoidal, lozenge-shaped, hexagonal and tongue-shaped, respectively. Therefore, we do not expect these elements by themselves to yield useful taxonomic data, in contrast perhaps to the morphometrics of all shell elements combined (e.g. Claude et al. 2003). Guerrero and Pérez-García (2021b) also plot regression scores of these elements against their centroid size, but here also we are unable to interpret the resulting graphs for the same reasons as listed above.

The available morphometric studies do highlight the tendency for the vertebral scutes to become relatively narrower during ontogeny, at least in the testudinid Chelonoidis nigra (Chiari and Claude 2011) and the cheloniid Caretta caretta (Casale et al. 2017). To explore if this is universally true among turtles, we collected linear morphometric measurements from a series of eight extant turtles representing all major lineages (see Suppl. material 1 for data and plots). Although the available data are not sufficient to undertake meaningful statistical analyses, we find a notable gradual decrease in vertebral width is found in all species. In contrast to the morphometric data mentioned above, this observed morphometric trend explicitly favors the taxonomic distinctness of Tongemys enigmatica and Pleurosternon bullockii, as the largest individuals of the smaller species have proportionally narrower vertebrals than the smallest individual of the latter species.

A number of studies exist that document ontogenetic variation in the shell of some extinct turtles (e.g. Brinkman 2003; Lichtig and Lucas 2015; Joyce et al. 2019; Garbin et al. 2019; Limaverde et al. 2020), but none observes ontogenetic trends among discrete characters beyond the closure of fontanelles or the loss of keels. Indeed, an underlying principle in fossil turtle taxonomy is that the shell of turtles more or less faithfully represents their species and are not influenced by ontogeny beyond changes to the level of ossification. Although we agree that this axiom has perhaps not been tested sufficiently, the very observation that no significant literature is available that might test this tenet is indirect confirmation that is seems to hold true. This is relevant for the taxonomic case at hand: if the turtle material from Beckles’ Pit were juveniles of Pleurosternon bullockii, it would imply ontogenetic changes that far outpace what is normally observed between species, such as the relative contraction of the nuchal and vertebral I, which would cause the loss of contacts and the creation of others during ontogeny. Incidentally, all of the affected characters were used in the more recent literature to diagnose other compsemydid species as being distinct, including Compsemys russelli, Peltochelys duchastelii and Selenemys lusitanica (Pérez-García and Ortega 2011; Pérez-García 2012; Joyce and Rollot 2020). Therefore, we find it inconsistent to use these characters on the one side to diagnose new species, but to push them aside as an ontogenetic nuisance at other times. The turtle material from Beckles’ Pit is easily distinguished from all other named compsemydids, particularly in the topological relationships of the nuchal and vertebral I relative to the surrounding elements and we feel justified in naming a new species, Tongemys enigmatica.

Tongemys enigmatica is intermediate in morphology between Selenemys lusitanica and Peltochelys duchastelii. This is supported by our phylogenetic hypothesis, which depicts them as a paraphyletic grade (see below). A notable difference with these species is the incomplete retraction of the nuchal from the carapacial margin and the retention of a mesoplastron, respectively. As such, we feel further justified in assigning our new species to a new genus as well.

Our phylogeny broadly corresponds to previous hypotheses (e.g. Rollot et al. 2021) by covering Compsemys russelli, Compsemys victa, Kallokibotion bajazidi, Peltochelys duchastelii and Selenemys lusitanica as members of Compsemydidae, in addition to the newly-added Tongemys enigmatica, which is recovered as the most basal representative of the clade (Fig. 3). The synapomorphies that unite this clade are the absence of cervicals, placement of marginal I mostly over peripheral I, absence of a contact between peripheral I and costal I and the presence of an entoplastron that is broader than long. All compsemydids more derived than Tongemys enigmatica are united by the absence of the nuchal from the anterior margin of the carapace and the presence of a sinuous mid-line sulcus. The novel placement of Kallokibotion bajazidi as sister to Compsemys russelli and Compsemys victa is supported by the novel recognition that only these compsemydids possess extensive contacts of the axillary and inguinal buttresses with the overlying costals. This topology further supports the notion that the clade originated in Europe during the Late Jurassic, but secondarily dispersed to North America during an uncertain time (Joyce and Rollot 2020).

Guerrero and Pérez-García (2021b) noted that the material we refer to Tongemys enigmatica originates from a more lacustrine facies, while classic Pleurosternon bullockii originate from a setting with marine influence (see Geological settings above). As they concluded all material to represent the same taxon, they reasonably inferred that the juveniles of the species (i.e. Tongemys enigmatica) inhabited inland areas, while the adults (i.e. Pleurosternon bullockii) were more towards the shore. We conclude, instead, that two taxa are present and that these turtles occupied separate ecological niches. A more ponded, inland habitat for Tongemys enigmatica is consistent with the inferred habitat preferences of other basal compsemydids, as both Selenemys lusitanica and Peltochelys duchastelii were collected from formations otherwise known for their dinosaurs (Baele et al. 2012; Mateus et al. 2017).

At present, compsemydids are only known from three localities throughout the Late Jurassic and Early Cretaceous of Europe, despite a rich record of coeval dinosaurs, particularly in France, Germany and Spain. We suspect this is a taphonomic filter, as the three available forms are notably small in size (carapace length less than 20 cm) and, therefore, likely easily overlooked. We, nevertheless, suspect that careful study of existing collections will yield additional remains from across the continent.