A new large tortoise from the early Oligocene (Arikareean NALMA) of Oaxaca, southern Mexico and its phylogenetic position within Pan-Testudinidae

ABSTRACT Pan-Testudinidae is the total clade of extant terrestrial tortoises, which includes extinct fossil members of their stem lineage. Members of this clade have a rather scarce fossil record in Mexico, and the few specimens known in scientific collections are poorly studied. Here, we described a new species of basal testudinid turtle, based on a single specimen from the early Oligocene deposits exposed in the marginal facies of the Chilapa Formation in Oaxaca, southern Mexico. The new taxon exhibits osteological characteristics that support its insertion as a basal Testudinidae. The phylogenetic relationships of the new turtle were assessed using a total evidence approach (morphological + molecular) in a global Pan-Testudinidae context using Implied Weighted Maximum Parsimony (IWMP), Standard Maximum Parsimony (SMP) and Bayesian Inference (BI). Although the BI consensus tree is not well resolved, the results obtained by IWMP and SMP retrieved its branching close to the root of Testudinidae. The differences between the topologies of the three phylogenetic analyses show that the position of several taxa within Testudinidae is affected by the phylogenetic analyses performed. The new taxon from Oaxaca, here reported, represents the first Palaeogene and the southernmost tortoise described from Mexico and the oldest Testudinidae known in the country.


Introduction
Pan-Testudinidae is the total clade of extant terrestrial tortoises, which includes extinct fossil members of their stem lineage (Joyce et al. 2021). The clade had a wider geographic range than extant tortoises. Molecular clock analysis suggests that Pan-Testudinidae split from Pan-Geoemydidae in the late Cretaceous, about 73.8 Ma (Pereira et al. 2017).
Peñasco and La Colorada, Sonora (Miller 1980;Carranza-Castañeda 2006;White et al. 2010;Carbot-Chanona et al. 2020). Jiménez-Hidalgo et al. (2015) described the Iniyoo Local Fauna, from Santiago Yolomécatl, in northwestern Oaxaca. The faunal assemblage was originally assigned to the Late Eocene. Testudinids cf. Hadrianus sp. and Stylemys sp. were reported from this fauna (Jiménez-Hidalgo et al. 2015). A recent reexamination of that material allowed the recognition of a new species of Testudinidae tortoise herein described. We also present a revision of the phylogenetic placement as inferred under different phylogenetic approaches, and we comment on their distribution in a paleobiogeographic context.

Geological setting
The specimen here described is part of the Iniyoo Local Fauna and was collected in a fluviolacustrine fossiliferous sequence of 325 m thick of red-ochre to cream-coloured fine-grained sand, clay, silt beds, with some conglomerate and silcrete beds that crop out in the surroundings of the town of Santiago Yolomécatl (Figure 1), in the northwestern part of the State of Oaxaca, southern Mexico (Jiménez-Hidalgo et al. 2015. This sequence was originally considered part of the Chilapa Formation with an Eocene-early Oligocene age (Santamaría-Díaz et al. 2008). Later, a late Eocene age was assigned to the fossiliferous beds based on radiometric dates (35.7 ± 1.0 and 32.9 ± 0.9 Ma) previously published for the Cañada María Andesite, nearby the study area (Jiménez-Hidalgo et al. 2015;Jiménez-Hidalgo et al. 2018). Guerrero-Arenas et al. (2020) provided a new age for fossiliferous beds based on U-Pb detrital zircon geochronology estimated at 30.6 Ma, which corresponds to the Rupelian age (Oligocene). These authors also informally subdivided the fossiliferous beds into three sections: the 'lower beds'' conformed by limestone, silts, sands, and conglomerate intercalations; the 'middle beds' consist of clayey silt and silty sand, with sand and conglomerate interbedding; and the 'upper beds' conformed by clayey siltstone, silty sand, silcretes, sand and conglomerate strata (Guerrero-Arenas et al. 2020). The described specimen UMPE 443 was collected from the 'middle beds'.
Although some authors have considered that the fossiliferous beds are part of the Yolomécatl Formation (e.g., Ferrusquía-Villafranca et al. 2016;Ferrusquía-Villafranca and Wang 2021), the most up-to-date geological work asserts that the unit is a marginal facies of the Chilapa Formation; also, an additional U-Pb maximum depositional age of 30.62 ± 0.67 Ma based on detrital zircon crystals collected from a sandstone that is approximately 9.2 m above the fossil vertebrate-bearing interval was also provided (see details in Jiménez-Hidalgo et al. 2021).

Preparation
The specimen UMPE 443 was originally embedded in a reddish silt matrix. The specimen was cleaned using dental needles, and soft bristle brushes. The broken parts were joined with waterproof commercial glue and hardened with cellulose acetate diluted in acetone.

Phylogenetic analysis
To explore the phylogenetic relationships of the specimen UMPE 443, a total evidence analysis was carried out under three different methods (Implied Weighted Maximum Parsimony, Standard Maximum Parsimony, and Bayesian Inference), using a total evidence data matrix (morphological + molecular) modified by Vlachos and Rabi (2018), with additional information from Vlachos (2018). To this data matrix we added the scoring of UMPE 443 using Mesquite 3.61 (Maddison and Maddison 2019). The scoring of Paleotestudo canetotiana, Pelorochelon soriana, 'Testudo' antiqua, and 'Testudo' eocaenica from Vlachos and Rabi (2018) was updated following Pérez-García et al. (2020) and Vlachos et al. (2020). In total, 43 extinct and 27 extant taxa and 20,170 characters (1-170, morphological characters;171-20,170, nuclear and mitochondrial DNA) were used in the Standard Maximum Parsimony and Implied Weighted Maximum Parsimony analysis (see Supplementary material).

Standard Maximum Parsimony
The Standard Maximum Parsimony analysis ran using the New Technology tree search, following the same parameters and outgroup of the IWMP but considering all characters with equal weight. Statistical supports (Bremer and bootstrap support, Consistency Index and Retention Index) were also calculated as in the IWMP analysis mentioned above.

Bayesian inference
For Bayesian inference, the molecular data matrix from Guillon et al. (2012) included in Vlachos and Rabi (2018), was modified. This matrix originally contained 20,000 molecular characters from the complete mitochondrial genome and four nuclear genes. The annotation of each gene and its length were allocated after introducing segments of the matrix in Blast®/BlasN suite website (https://blast.ncbi.nlm.nih.gov/Blast.cgi) (Zhang et al. 2000). Then, to confirm their length, each gene was aligned with the corresponding accession gene as downloaded from GenBank® (www.ncbi.nlm.nih.gov/genbank/) (Clark et al. 2016) using the SeaView 5.0.4 (Gouy et al. 2010) interface and the Clustal Omega programme (Sievers et al. 2011) for multiple sequence alignment sets with default options. In Mesquite 3.61 (Maddison and Maddison 2019) the molecular sequences were divided into data blocks, the protein coding genes were defined by codon position, and input files for model selection software were generated. Also, sites not assigned to any of the specified data blocks were deleted producing a molecular data set of 19,266 sites. The model selection process was carried out in PartitionFinder2 (Lanfear et al. 2016). The best-fit partitioning scheme and the best model for each molecular subset were selected according to the Bayesian Information Criterion (BIC). The analysis estimated a Maximum-likelihood tree with RaxML v.8 (Stamatakis 2014), the reason why the unassigned sites were removed (see PartitionFinder2 manual) and ran with the rcluster algorithm (Lanfear et al. 2014) for large datasets. Settings for the phylogenetic analysis with the molecular blocks were obtained from the best scheme output file.
The morphologic dataset was the same used in the Maximum Parsimony analysis. Both matrices were merged using the Mousepad plain text editor, producing a combined matrix of 19,436 characters (see Appendix 5 of Supplementary data). The Bayesian inference analysis was performed in MrBayes 3.2.7a  for Linux systems and ran for 10,000,000 generations with two parallel runs (20,000,0000 total generations), 3 heated chains and 1 cold chain, discarding 20% of the samples from the beginning of the chains. The morphological block was set to use gamma-shaped rate variations (Brusatte and Carr 2016;Diaz-Cruz et al. 2019). The designed outgroup taxon was the same as in the parsimony analysis. To optimise the computing time for phylogenetic analysis, we also used the Beagle library . Convergence was assessed in tracer v1.7.1 (Rambaut et al. 2018). We calculated a 50% majority-rule consensus (MRC) on our combined data set, given that maximum clade credibility (MCC) and maximum a posteriori (MAP) methods to obtain consensus trees from Markov chain Monte Carlo sampling often include incorrect trees (see O'Reilly and Donoghue 2018). The implementation of MRC followed its potential for retrieving 'correct clades' in our analysis.

Systematic palaeontology
For suprageneric ranks, we use the categorical clade arrangement, which is independent of categorical rank (PhyloCode 2020, Art. 3.1; Cantino and De Queiroz 2020). The names of turtle clades are based on those proposed by Joyce et al. (2021), who follow the recommendations of PhyloCode 2020 (Cantino and De Queiroz 2020). Diagnosis. aff. 'Stylemys' gisellae sp. nov. can be attributed to Testudinidae considering the following character combination: the alternating rectangular and non-rectangular arrangement of neurals; a coincidence between the costo-peripheral suture and the pleuro-marginal sulci; thin medial ends and wide lateral ends of costals, alternating with wide medial ends and thin lateral ends of costals; and pleural 1 not contacting the nuchal bone and wide vertebrals, almost equal to pleurals.
Remarks. aff. 'Stylemys' gisellae sp. nov. differentiates from other North American members of Pan-Testudinidae based on a unique combination of morphological characters.
aff. 'Stylemys' gisellae sp. nov. differs from Hadrianus spp. in that the first has eight neurals and the first pleural does not overlap the nuchal, while Hadrianus has nine neurals and the first pleural overlapping the posterolateral corner of the nuchal. aff. 'Stylemys' gisellae sp. nov. differs from Floridemys nana in that the nuchal is wider than long, and the neural 1 is rectangular in shape, with all sides almost equal in size while in F. nana the nuchal is longer than wide, and the neural 1 is hexagonal in shape, with the anterolateral sides longer than the posterolateral sides. In aff. 'Stylemys' gisellae sp. nov. the neural 1 is rectangular, the neurals 2 and 4 are octagonal, and the seventh and eighth costal bones do not contact medially, and differs from Oligopherus laticuneus in that the neurals 1, 2 and 4 are hexagonal, and the seventh and eighth costal bones have medial contact. aff. 'Stylemys' gisellae sp. nov. differs from 'Testudo' brontops in that the nuchal is wider than long, the fifth neural is hexagonal in shape, the width of the vertebrals is almost equal to the pleurals, and has short pectoral scutes medially ( Figure 2); while in 'T'. brontops the nuchal is about 1/3 times wider than long, the fifth neural is rectangular in shape, the vertebrals are narrower than the pleurals, and have long pectoral scutes medially. aff. 'Stylemys' gisellae sp. nov. differs from Stylemys capax mainly in that the vertebrals are almost equal to the pleurals; while in S. capax the vertebrals are narrower than the pleurals. aff. 'Stylemys' gisellae sp. nov. differs from Stylemys nebrascensis and S. inuscitata in its size of about 92 cm and that the neural 2 is octagonal with the anterolateral and posterolateral sides shorter than lateral sides, the neural 3 is quadrangular, and the neural 4 is octagonal, while the size of Stylemys nebrascensis and S. inuscitata is about 20-30 cm, the neural 2 is hexagonal with the short sides positioned posterolaterally, and the neurals 3 and 4 are hexagonal in shape. aff. 'Stylemys' gisellae sp. nov. differs from Hesperotestudo osborniana and He. williamsi, and remaining Hesperotestudo species, in that the fifth neural is hexagonal in shape while in Hesperotestudo spp. it is rectangular. Finally, aff. 'Stylemys' gisellae sp. nov. differs from Gopherus edae and G. hexagonatus, and the remaining Gopherus species, in that the posterior margin of the first vertebral scute is significantly narrower than its anterior margin, while in Gopherus, the first vertebral scute is subequal in size to the anterior margin.
Derivation of name. gisellae, in honour to Giselle Carbot-Hernández, the firstborn daughter of GC-C, who is a constant source of inspiration.
Type locality and horizon. La Tortugota locality, Chilapa Formation, near Yolomécatl town, state of Oaxaca, southern Mexico. Early Oligocene, Rupelian (earliest Arikareean [Ar1] NALMA). The locality was registered in the Dirección de Registro Público de Monumentos y Zonas Arqueológicos e Históricos of Instituto Nacional de Antropología e Historia (INAH). The exact details of the locality could be provided on request; these data are not published to prevent looting of the fossil site.

Description and comparison
General shape and preservation of the shell. The specimen UMPE 443 is a large tortoise (total length ~920 mm) with a domed shell . The nuchal and all the neural series are preserved. On the left side, the peripheral 1 and 2, costal 1-3, and 6-8 are partially conserved. On the right side, the peripheral 1 and costal 1-4 are conserved, and the costal 5-8 are partially conserved. The sulci between the scales have sharp-raised ridges, as in some species of Chelonoidis.
Nuchal. Hexagonal in shape and wider than long, with the anterolateral borders contacting the peripheral 1, while the posterolateral border contact costal 1 (Figure 2). This morphology is similar to that of Hadrianus corsoni, Stylemys, Hesperotestudo and Gopherus. Only Floridemys nana has a nuchal longer than wide.
The neural 2 in aff. 'Stylemys' gisellae sp. nov. is octagonal, with the anterolateral and posterolateral sides shorter than the lateral sides. This shape is similar to that of S. capax, Hesperotestudo and Gopherus (Hutchison 1996;Vlachos 2018). On the contrary, in Stylemys nebrascensis and S. inuscitata, and in the basal forms Hadrianus corsoni, O. laticuneus and F. nana, the neural 2 is hexagonal in shape with the short sides positioned posterolaterally (Hay 1908).
The neural 4 is not clearly visible in aff. 'Stylemys' gisellae sp. nov., because this part is slightly damaged (Figure 2). However, its shape can be inferred as octagonal by the configuration of the costal surrounding it. In Fl. nana, 'Testudo' brontops, Hesperotestudo and Gopherus the neural 4 is octagonal in shape too (Hay 1908;Hutchison 1996) Costals. The costals in aff. 'Stylemys' gisellae sp. nov. have an alternating arrangement where the costals 1 and 3 are wider at the medial end than at the lateral end, while the costals 2 and 4 are thinner at the medial end than at the lateral end. The width of the lateral ends in the costals 2 and 4, is more noticeable than in Oligopherus, Stylemys and Hesperotestudo.
The costal 1 contacts the nuchal, neurals 1 and 2, and peripherals 1-3. The costal 2 contacts medially the neural 2, but the contact with the peripherals is not visible (Figure 3). The costal 3 contacts medially the neurals 2-4. The costal 4 contacts medially the neural 4 and the anterolateral side of the neural 5. The costal 5 is narrow and contacts medially the posterolateral side of the neural 5 and the anterolateral side of the neural 6. The costal 7 contacts medially the posterolateral side of the neural 6 and the anterolateral side of the neural 7. Finally, the costal 8 contacts medially the posterolateral side of the neural 7, the anterolateral side of the neural 8, and the anterolateral side of the suprapygal 1. The contact of costals 5, 6, 7 and 8 with the peripherals is not visible because of the carapace damage.
Peripherals. Only the right and left peripheral 1 and the left peripheral 2 are partially preserved. Peripheral 1 reaches the anterolateral border of the nuchal through a straight suture. Posteriorly, the peripheral 1 meets the costal 1. The peripheral 2 is shorter than peripheral 1, and it contacts medially the costal 1. The suture between the peripheral 1 and 2 with the costal 1 is coincident with the sulci between pleural 1 and marginal 2.
Suprapygal. Two suprapygals are present. The suprapygal 2 is larger than the suprapygal 1, but due to poor preservation it is not possible to observe if the contact between both suprapygals is straight and perpendicular to the axial plane or if the first suprapygal embraces a lenticular second one (Figure 4).

Pygal.
It is not possible to determine the general shape of the pygal, but it seems quadrangular in shape and contacts laterally with the peripheral 11. The pygal notch is not present (Figure 4).

Cervical scute.
The left sulcus of the cervical scute is only visible, suggesting the presence of the cervical, but its shape and size cannot be deduced (Figure 3).
Vertebral scutes. All five vertebral scutes are preserved (Figure 2). The vertebrals are as wide as the pleurals, like in He. osborniana, He. orthopygia, S. inuscitata, S. nebrascensis and Gopherus agassizii. Vertebral 1 is wider than long and overlaps the posterior half of the nuchal, the anterior half of neural 1, and the medial side of the costal 1.  The vertebral 2 is sub-squared in shape, and overlaps the posterior half of the neural 1 covering the neural 2 entirely and the anterior half of the neural 3. Anterolaterally, it overlaps the posteromedial side of the costal 1; laterally the medial side of the costal 2, and posterolaterally the anteromedial side of costal 3.
The vertebral 3 is slightly wider than long and overlaps the posterior half of the neural 3, it covers entirely the neural 4, and overlaps the anterior half of the neural 5. Anterolaterally, it overlaps the posteromedial side of the costal 3, laterally the medial side of the costal 4, and posterolaterally the anteromedial side of the costal 5.
The vertebral 4 is longer than wide. Its anterior sulcus is on the anterior half of the fifth neural, and the posterior sulcus is on the middle part of the neural 8. Anterolaterally, it overlaps the posteromedial part of the costal 5, laterally the medial side of the costals 6 and 7, and posterolaterally the anteromedial corner of the costal 8.
The vertebral 5 is not completely visible, but it seems to cover the pygal and suprapygal. Its anterior sulcus is in the middle part of the neural 8 and anterolaterally overlaps part of the costal 8.
Marginal scutes. Only the right marginals 1 and 2 are visible. The marginal 1 overlaps the anterolateral margin of the nuchal, the medial part of the peripheral 1, and the anterior corner of the costal 1 (Figures 2 and 3). This last morphological character is not present in any other Testudinidae. The marginal 2 does not contact the lateral margin of the nuchal, but it overlaps parts of the costals 1 and 2.
Pleural scutes. The sulcus between the marginal 2 and the pleural 1 are coincident with the suture between the peripheral 1 and the costal 1. The pleural 1 does not overlap with the nuchal. Its medial sulcus contacts the vertebrals 1 and 2.
General shape and preservation of the plastron. The plastron is partially preserved. Only the right and left epiplastron, part of the entoplastron, left hyoplastron, left hypoplastron and part of the right xiphiplastron are visible ( Figure 5). The sulcus is not visible as a consequence of preservation.
Epiplastra. Only part of the right epiplastron is conserved, but its poor preservation does not allow observing its shape or relationship with the other plastral bones.
Entoplastron. The entoplastron is partially conserved, and only the right side is visible. Its general shape is not well appreciated, but presumably it was pentagonal.
Hyoplastra. Almost all the right hyoplastron and part of the left hyoplastron are conserved. The right hyoplastron contacts the right epiplastron throughout a perpendicular suture, as in other Testudinidae. The contact of the hyoplastron with the hypoplastron is not visible.

Hypoplastra.
Only two fragments of the right hypoplastron are preserved. Their poor preservation prevents observing the relationship with other bones.
Xiphiplastra. Only part of the left xiphiplastron is conserved. Its shape and relationships with other bones are not clear.

Phylogenetic and character analysis
The Implied Weighted Maximum Parsimony analysis (IWMP) retained three trees with a TBR best score of 400.60. The consensus tree has a Tree Length with an adjusted homoplasy of 12,930, a Consistency Index (CI) of 0.337 and a Retention Index (RI) of 0.190. The topology of the IWMP differs from the topology obtained with the SMP (see supplemental Fig. S1). In general, the IWMP tree is well resolved (Figure 6) although most resolved clades have low support values (bootstrap group present/contradicted values; Goloboff et al. 2003). The clade Pan-Testudinidae is supported by three synapomorphies (characters and states in parenthesis): first neural hexagonal with short sides facing posteriorly (character 85:0), thin and elongated costal rib head (character 93:1), and fused trochanter of the femur (character 155:1). Consistent with Vlachos and Rabi (2018), Fontainechelon cassouleti (Claude and Tong, 2004) was retrieved as the most basal member of Pan-Testudinidae supported by a rectangular fourth neural (character 88:0), posterior carapace border posteriorly flared (character 114:1), the gulars and the entoplastron not in contact (character 122:0), angle formed by gularo-humeral sulci of about >90° (character 123:0), and the lateral portion of the humero-pectoral sulcus perpendicular to the axial plane with convex anteriorly lateral parts (character 126:2). Hadrianus majusculus and Ha. corsoni are then recovered as sister taxa in a basal branch after F. cassouleti. This clade is supported by a well-developed gular protrusion caused by a constriction in the gularo-humeral sulcus (character 119:1), and a wavy humero-pectoral sulcus medially (character 125:2).
Testudinidae is a clade supported by three synapomorphies: the first pleural not in contact with the nuchal (character 79:2), position of the costal rib head on or very near the neural/costal suture (character 94:2), and cervical scute wider than long (character 101:1). Pelorochelon soriana + 'Testudo' eocaenica + Manouria emys + Ma. impressa was recovered as the most-basal clade of Testudinidae, forming a monophyletic group. This clade is supported by a hexagonal third neural with short sides facing posteriorly (character 87:1), anterior lobe medially notched (character 118:1), and gular and entoplastron in contact with the anterior margin (character 122:1).
Then, Stylemys inusitata and St. nebrascensis branched as sister taxa. This clade is supported by three synapomorphies: second neural hexagonal with short sides facing anteriorly (character 86:0); third neural hexagonal with short sides facing anteriorly (character 87:0), lateral portion of the humero-pectoral sulcus perpendicular to the axial plane with anteriorly convex lateral parts (character 126:2) and straight or slightly rounded femoro-anal sulcus developed mainly perpendicular to the axial plane (character 136:1).
The clade Namibchersus namaquensis + Hesperotestudo osborniana + He. crassiscutata + He. bermudae + He. williamsi is weakly supported by one synapomorphy: a short major trochanter of the humerus not extending beyond the humeral head (character 152:0). However, this character is only coded for N. namaquensis and He. bermudae.
Hesperotestudo orthopygia branched out paraphyletically after the clade Namibchersus namaquensis + all other Hesperotestudo turtles. The position of He. orthopygia as sister taxa of Gopherus + Testudininae is supported by seven characters: contact between the parietal and the quadrate (character 14:1), lingual ridge on the triturating surface (character 31:1), sixth marginal scute with the third pleural scute not in contact (character 109:0), anterior lobe medially notched (118:1), well-developed gular protrusion (119:1), humerals equal to or shorter than gulars (124:0), and humeropectoral sulcus coinciding medially with the posterior contact of the entoplastron-hyoplastron (127:1). This clade is very well supported and suggests that He. orthopygia is not a Hesperotestudo and should be referred to as a distinct genus.
Gopherus + Testudinine are supported by three synapomorphies: symphysis of the maxilla/premaxilla shorter than the height of fossa nasalis, but with the development of a dorsal projection (character 18:1), absence of a circular or elliptical pit on the ventral side of the premaxillae (character 27:1) and serrated dentary (character 73:1).
After running 20,000,000 generations of Bayesian inference analysis, the average standard deviation of split frequencies was 0.013843 and it reached an Effective Sample Size (ESS) equal to 1,543 (see supplemental Fig. S3). However, the general topology of the BI consensus tree (Figure 7) is not fully resolved and it differs notably from the consensus tree obtained with the IWMP. aff. 'Stylemys' gisellae sp. nov. is recovered in a large polytomy along with several other taxa.
The Geochelona clade is recovered with a similar topology to the IWMP consensus tree, except for the exclusion of Stigmochelys pardalis and Floridemys nana that are retrieved from the large polytomy. This clade is supported by a low posterior probability (53). Three major clades have been recovered in Geochelona. The first clade comprises Kinixys erosa and K. homeana, and it is supported by a high posterior probability (100). The second clade includes Centrochelys sulcata, 'Chelonoidis' gringorum, Geochelone elegans, and Megalochelys atlas, and it is supported by a low posterior probability (59). The last clade is formed by Chelonoidis carbonaria, Ch. denticulate, Ch. chilensis and Ch. nigra, and it is supported with low posterior probability (77).

Phylogenetic framework
We analysed the phylogenetic relationships of aff. 'Stylemys' gisellae sp. nov. in a global Pan-Testudinidae context using a total evidence approach (morphological + molecular data). The resultant tree supports the inclusion of the new species in Testudinidae. The main difference between the published tree of Vlachos and Rabi (2018) and ours is that we used a larger number of taxa within Pan-Testudinidae. We included a larger sample of North American taxa derived from the original data matrix of Vlachos (2018). The topology of the resulting trees also varies according to the type of phylogenetic analysis used. Our discussion is based on the consensus tree resulting from the Implied Weighted Maximum Parsimony analysis carried out in this work, based on the evidence that implied weights outperform other methods for calculating trees and give better results (see Goloboff et al. 2018).
Fontainechelon cassouleti, from the early Eocene (Ypresian) of Saint Papoul, Aude, France, has been considered the basal most member of Pan-Testudinidae (Claude and Tong 2004;Pérez-García and Vlachos 2014;Vlachos and Rabi 2018;Pérez-García et al. 2020), although Hadrianus majusculus, from the early Eocene of New Mexico, USA, has been referred as the oldest taxon in the group (Joyce et al. 2013;Vlachos and Rabi 2018). The phylogenetic position of Fo. cassouleti and Ha. majusculus, with respect to each other, has been previously discussed. Vlachos (2018) considered that Ha. majusculus is derived more than Fo. cassouleti, because it has narrower and longer gular scutes, and 'wavy' humero-pectoral sulcus with medial part convex anteriorly. Conversely, Lichtig et al. (2021) disagree with that assessment because they consider that Hadrianus is the basal most testudinid, because of the absence of derived traits present in Fontainechelon, as are the lack of costal wedging and the overlap of the pectoral scutes onto the entoplastron. Our phylogenetic results agree with this last hypothesis as Fontainechelon resulted lower than Hadrianus on the stem since the combination of characters are in overall considered more primitive: third neural octagonal (homoplasic in Testudo oughlamensis), fourth neural rectangular (shared with 'Testudo' eocaenica and Te. oughlamensis), and medial contact of the seventh and/or eighth costal bones (shared with Hadrianus corsoni and 'Testudo' antiqua).
Hadrianus corsoni and Ha. majusculus branch sequentially and confirm their position within Pan-Testudinidae.
The position of the Manouria emys + Ma. impressa grouped with P. soriana and 'T'. eocaenica at the base of Testudinidae is contradictory. The phylogenetic molecular hypothesis suggests that Manouria, together with Gopherus, derives from a basal position within the extant Testudinidae (Guillon et al. 2012;Pereira et al. 2017). The basal position of Manouria spp. recovered in our IWMP analysis, implies that taxa traditionally recovered outside crown Testudinidae are included in this clade. In Vlachos and Rabi (2018), Manouria is retrieved in the basal-most position than other Oligocene and Miocene taxa (e.g., Cheirogaster maurini, Gigantochersina ammon, Namibchersus namaquensis, Paleotestudo canetotiana, and Stylemys nebrascensis), which is consistent with the results of our analysis. Crumly (1984) mentions that Manouria shares primitive characters as are a split supracaudal; gular and femoral scutes with equal midseam lengths; a broad cervical scute; the first two neurals hexagonal in shape with short articulation directed posteriorly; and the last five neurals hexagonal in shape with short articulation directed anteriorly. This can explain its phylogenetic position in our analysis. The inclusion of Manouria within a large data set where most of the taxa are represented by extinct turtles could influence the result. When Manouria spp. is excluded from the IWMP analysis, the position of almost all taxa remains at the same position, except for 'Testudo' costaricensis, which changes to Testudona. The exclusion of Manouria also changes the position of the Testudinidae node, including only Gopherus + Testudininae (see supplemental Fig. S2).
In all three analyses here presented, Gopherus is recovered as monophyletic, consistent with previous morphological and molecular phylogenetic hypotheses (e.g., Guillon et al. 2012;Pereira et al. 2017;Vlachos 2018;Vlachos and Rabi 2018). Stylemys has been considered a monophyletic taxon in previous phylogenetic hypotheses (e.g., Vlachos 2018). Nevertheless, in our IWMP ( Figure 5), only Stylemys inusitata and St. nebrascensis are recovered as sister taxa, while St. capax is recovered forming a clade with Ergilemys insolitus, Cheirogaster maurini, and Taraschelon gigas. This arrangement is the same as that obtained in the BI consensus tree (Figure 6), raising the possibility that St. capax should be referred to a new genus; it differs from St. inusitata and St. nebrascensis mainly in the shape of the second and third neural, the coincidence of the costo-peripheral suture with the pleuromarginal sulcus, the shape of the femoro-anal sulcus, and the width of the vertebrals in relation to the pleurals.
Hesperotestudo seems to be monophyletic, except He. orthopygia that branched alone outside the Hesperotestudo-clade. In our consensus tree resulting from the BI (Figure 7), He. orthopygia and He. osborniana are recovered in a large polytomy, outside the clade formed by He. bermude, He. crassiscutata and He. williamsi. Therefore, it could be considered that He. orthopygia should be referred to different new genera. Contrarily, in Vlachos (2018), He. orthopygia is considered a valid member of Hesperotestudo. Both genera, Stylemys and Hesperotestudo, are recovered as basal Geochelona in Vlachos and Rabi (2018), but in our IWMP strict consensus tree they are recovered as basal Testudinidae.
The Neogene European genus Titanochelon is recovered as a monophyletic clade, and it is the sister taxon of Impregnochelys pachytestis + Malacochersus tornieri. This arrangement differs from the previous topologies (Vlachos and Rabi 2018;Pérez-García et al. 2020). Similarly, 'Chelonoidis' gringorum (Simpson 1942), from the early Miocene of the Chubut Valley, Patagonia, Argentina, is resulted out of crown Chelonoidis in all three analyses here performed, consistent with the results of Vlachos and Rabi (2018). Although 'Ch'. gringorum has been considered part of the crown Chelonoidis (de la Fuente 1994; de la Fuente et al. 2018), our results suggest that 'Ch'. gringorum belong to a different new genus.
Testudo is a problematic taxon. Similar to previous results (Vlachos and Tsoukala 2016;Vlachos and Rabi 2018), in our IWMP, Testudo is recovered as polyphyletic. Here, the extant species Testudo graeca, Te. marginata,Te. kleinmanni,Te. horsfieldii and Te. hermanni and the extinct Te. marmorum are grouped together. In recent works, the species Te. horsfieldii and Te. hermanni have been placed in the genera Agrionemys and Chersine, respectively (e.g., Parham et al. 2006;Vlachos and Rabi 2018;Pérez-García et al. 2022 Rhodin et al. 2021). Based on our resulting phylogenetic arrangement in both IWMP and SMP consensus trees (Figures 6 and S1), we support the inclusion of these two species within Testudo, as suggested by molecular phylogenetic analysis (Fritz and Bininda-Emonds 2007;Guillon et al. 2012).
On the other hand, the arrangement of Te. shensiensis, Te. lunellensis, and Te. oughlamensis, together with Mesochersus orangeus resulted in a different position with respect to previous works (e.g., Vlachos and Rabi 2018). Also, 'Te'. eocaenica and 'Te'. costaricensis are recovered as basal into Testudinidae, while 'Te'. brontops, and 'Te'. kaiseni branch together into Testudona, despite it having been proposed that they belong to a different genera than Testudo (Vlachos and Rabi 2018).

Palaeobiogeography
Testudinidae is the most diverse and widely distributed crown-clade of turtles, with 67 extant species (Rhodin et al. 2021). Although the biogeographic history of the clade Pan-Testudinidae is complicated, the fossil record suggests that it possibly originated in Europe or North America, during the earliest Eocene. Based on our phylogenetic resolution, Fo. cassouleti is the known basal-most member of Pan-Testudinidae. Under this premise, a European origin of Pan-Testudinidae is here supported. During the middle Eocene Pan-Testudinidae spread into North America, with Hadrianus entering the scene. Later, in the late Eocene, Pan-Testudinidae reached Africa, and Asia, diversifying into new taxa included in Testudinidae . The presence of related taxa in North America (Stylemys nebrascensis), Asia (Ergilemys insolitus), and Europe (Cheirogaster maurini and Taraschelon gigas) during the late Eocene-early Oligocene suggest a second radiation, possibly from North America. The occurrence of aff. 'Stylemys' gisellae sp. nov. in the south of Mexico suggests that this group reached more southern areas during the early Oligocene. The dispersion of the basal members of Testudinidae coincides with the cooling and rapid expansion of Antarctic continental ice-sheets, in a period called Oi-1 Glaciation during the Eocene-Oligocene transition (Zachos et al. 2001) that possibly favoured the formation of intercontinental land bridges in tropical zones as a consequence of the drop in sea level.
The Miocene was the 'golden age' for land tortoises, when the 'Tortoise Miocene Explosion' occurred. In the early-middle Miocene, the expansion of Testudinidae in North America and Africa is notorious. In North America, Testudinidae is mainly represented by Hesperotestudo spp. and Gopherus edae (Hay 1908;Auffenberg 1964Auffenberg , 1974Vlachos 2018), while in Africa they are represented by Impregnochelys pachytestis, Mesochersus orangeus and Namibchersus namaquensis (Meylan and Auffenberg 1986;Lapparent de Broin 2003). Testudinidae reached South America in the late Oligocene or the early Miocene. Although the means of arrival of testudinids to South America is not clear, it has been proposed that they are derived from Africa by transoceanic drift (de la Fuente et al. 2014). The molecular evidence strongly supports an Afroasiatic origin for the South American taxa (Le et al. 2006;Guillon et al. 2012;Pereira et al. 2017). Between the mid-Miocene and the Pliocene, Titanochelon spread across Europe, becoming the best-represented large testudinid in this continent (Pérez-García and Vlachos 2014;Pérez-García et al. 2017). The higher taxonomic diversity of Testudinidae during the Miocene was possibly favoured by the Miocene Climatic Optimum, a warm climate period that was followed by a significant cooling (Zachos et al. 2001).
During the Pliocene and Pleistocene, members of the clades Testudona and Geochelona expanded across Europe, Asia and Africa. In the Pleistocene Geochelona became extinct in North America, but in South America and Africa they became the dominant group. The diversity of Testudinidae decreased considerably worldwide, as a consequence of the drastic climatic change that occurred in the Late Pleistocene.

Conclusions
The combination of morphological characters, size, geographical and temporal occurrences as well as the results obtained by multiple phylogenetic approaches here presented indicate that the specimen UMPE-443 represent a new taxon within Testudinidae, increasing the geographical distribution and diversity of this clade in North America (Figure 8). Although the combination of morphological characters present in the new taxon suggests that it could be a different genus from those previously described for North America, the lack of characters in the plastron, due to its poor conservation, does not allow this to be confirmed. Therefore, the new taxon is here referred to as aff. 'Stylemys' gisellae sp. nov. The new taxon represents the first Palaeogene tortoise described for Mexico, as well as the oldest Testudinidae known in the country.
Of the three performed analyses, the worst resolution was obtained using Bayesian inference because this type of analysis is more sensitive to data consistency. The best phylogenetic resolution was obtained by the Implied Weighted Maximum Parsimony analysis, with results similar to those obtained with the Standard Maximum Parsimony analysis. The difference between the topology of the three phylogenetic analyses shows that the position of several taxa within Testudinidae is sensitive to the phylogenetic method performed, probably due to the high homoplasy and polymorphism found in this clade, and to the great number of missing data in most extinct taxa. Future work, including even more taxa and betterpreserved specimens, may help to better resolve the phylogeny of Pan-Testudinidae.