Phylogeny, biogeography and diversification patterns of side-necked turtles (Testudines: Pleurodira)

Pleurodires or side-necked turtles are today restricted to freshwater environments of South America, Africa–Madagascar and Australia, but in the past they were distributed much more broadly, being found also on Eurasia, India and North America, and marine environments. Two hypotheses were proposed to explain this distribution; in the first, vicariance would have shaped the current geographical distribution and, in the second, extinctions constrained a previously widespread distribution. Here, we aim to reconstruct pleurodiran biogeographic history and diversification patterns based on a new phylogenetic hypothesis recovered from the analysis of the largest morphological dataset yet compiled for the lineage, testing which biogeographical process prevailed during its evolutionary history. The resulting topology generally agrees with previous hypotheses of the group and shows that most diversification shifts were related to the exploration of new niches, e.g. littoral or marine radiations. In addition, as other turtles, pleurodires do not seem to have been much affected by either the Cretaceous–Palaeogene or the Eocene–Oligocene mass extinctions. The biogeographic analyses highlight the predominance of both anagenetic and cladogenetic dispersal events and support the importance of transoceanic dispersals as a more common driver of area changes than previously thought, agreeing with previous studies with other non-turtle lineages.

As in all previous analyses (e.g. Gaffney et al., 2006Gaffney et al., , 2011Cadena, 2015), no taxon is recovered in the stem-lineage of Pelomedusidae, resulting in a large gap from the Barremian to the first records of the extant taxa (Fig. 2). Sokatra antitra and Atolchelys lepida were recovered as successive sister-taxa to Podocnemidoidea (Bothremydidae + Podocnemidoidae). The position of Atolchelys lepida is still controversial: in the original description (Romano et al., 2014) this taxon is retrieved as sister to all bothremydids and in a later analysis (Cadena, 2015) it is inside a clade also including So. antitra and euraxemyidids.
The interrelations of Bothremydidae are problematic in two points. The first one is related to the Kurmademydini clade (Gaffney et al., 2006(Gaffney et al., , 2009Rabi et al., 2012;Joyce et al., 2016), which is not supported by our analyses (Fig. 1, S1). A paraphyletic "Kurmademydini" was already retrieved by previous analyses (Romano et al., 2014;Cadena, 2015), but in these Kinkonychelys rogersi and Kurmademys kallamedensis are grouped in a clade, whereas our result support the former as closer to the other bothremydids. The second point of disagreement is related to the position of Foxemydina which is alternatively closer to Bothremydini (Gaffney et al., 2006;Cadena et al., 2012a;Rabi et al., 2012) or sister to the clade including Taphrosphyni + Bothremydini (Cadena, 2015;Joyce et al., 2016). Our results agree with the former hypothesis with a good support from morphological characters ( Fig. S1-S6).

The European erymnochelyin Papoulemys laurenti was recently reallocated inside
Neochelys based on a redescription that identified common features between those taxa (Pérez-García & Lapparent de Broin 2015). Unlike the Erymnochelys-group, this hypothesis was supported by one phylogenetic analysis (Cadena, 2015). Here, however, Papoulemys laurenti is retrieved closer to Kenyemys williamsi and 'Neochelys' fajumensis than to the included Neochelys taxa, namely N. arenarum and N. franzeni. This is supported only by wider than long cranial margin of intergular (ch. 242), while Neochelys is diagnosed by slightly elongated heart-like shaped interparietal scale (ch. 27) and their position closer to Erymnochelys and Turkanemys than to Papoulemys laurenti is supported by a foramen palatinum posterius restricted to the palatine (ch. 71) and neural series reaching costal plates 6 (ch. 175). Based on our analyses, to reallocate Papoulemys laurenti into the same genus as Neochelys arenarum and N. franzeni would require to do the same with Turkanemys pattersoni and Erymnochelys madagascariensis. Hence, we choose to maintain the status of Papoulemys as a valid genus, with Papoulemys laurenti as its type and only included species.

Detailed description of the results
Although South American and Australasian chelids were forced to form clades (Fig. S7), the extinct chelids maintain the same position as in the original tree (Fig. S8), with Chelus colombianus as sister taxon to C. fimbriatus, and the other fossils grouped in a clade sister to Hydromedusa spp.
Araripemydidae, Euraxemydidae, Bothremydidae and non-Podocnemididae Podocnemidoidae taxa maintain the same positions, but Sokatra antitra is in a politomy including Pan-Pelomedusidae and Pan-Podocnemididae (Fig. S8). Erymnochelys madagascariensis is forced closer to Podocnemis spp. than to Peltocephalus dumerilianus (Fig. S7), but the latter is still recovered as sister to Stereogenyini, and Caninemys tridentata and Cerrejonemys wayuunaki form successive sister taxa to Podocnemis spp as well (Fig. S8). Similarly, the taxa closer to E. madagascariensis (e.g. Turkanemys pattersoni, Neochelys franzeni and Kenyemys williamsi) were again recovered in the same clade (Fig. S8). On the other hand, Carbonemys cofrinii, Dacquemys paleomorpha, Stupendemys geographicus, and UCMP 42008, recovered in a clade inside Erymnochelyinae in the unconstrained analysis, were not grouped together and appear in a politomy with the other podocnemidid clades (Fig. S8).   The biogeographic analysis requires fully dichotomic topologies, hence, the polytomies of the original tree had to be manually resolved by deliberately choosing an arrangement. However, some considerations were taken in order to minimize the effect of arbitrariness related to this procedure.
As advocated by previous studies (e.g., Upchurch et al., 2015), one alternative is to define clades in a way that the arrangement minimizes biogeographic changes (e.g. in a polytomy including two taxa in one area and a third in another, we define the first two as sister-taxa in exclusion of the latter). This was possible to be done for one case of the original tree ((Carbonemys cofrinii +     Euraxemydidae. DIVALIKE-M2 favors an African distribution in all three topologies (Fig. 2), whereas DEC-M2 favors South American ancestors (Appendix S2).
As expected, the probabilities for the alternative areas in the Pleurodira node are relatively low, but an African/Australian distribution is best supported by DIVALIKE-M2, and African/Australian/South American distribution by DEC-M2. For the Pan-Chelidae node an Australian distribution is favored and one dispersal event to South America occurred during the Early Cretaceous. From this area the ancestors of Chelodina colliei dispersed back to Australia before the Aptian (Fig. 2).
Removing the marine taxa (Appendices S5, S6) from the analyses did not alter the results described above, except for the Bothremydidae node, which is estimated as distributed in Madagascar during the Early Cretaceous. On the other hand, the analysis of the 'molecular constrained tree' showed noteworthy differences (Appendices S7, S8  (Fig. 3). From these areas, Stereogenyini dispersed to Africa during the beginning of the Paleocene and then to the other areas during the Eocene and Oligocene.