The phylogenetic analyses were performed using the program TNT (Goloboff et al., 2008a, 2008b). For the parsimony analysis all characters were treated as unordered and equally weighted. The transformation cost between transitions and transversions was considered equal and gaps were considered as a fifth state to preserve its phylogenetic information (Giribet and Wheeler, 1999). A total evidence analysis was conducted combining the morphological and molecular information, as well as several partitioned analyses to analyze the phylogenetic signal of each dataset (e.g. molecular vs. morphological, extant vs. extinct taxa, comparisons of the five genes). Given the presence of highly divergent taxa, denoted by the presence of long branches, several molecular analyses were run excluding these terminal taxa (e.g. Trionychia and Pleurodira). This procedure, known as long branch extraction (LBE; Siddall and Whiting, 1999; Pol and Siddall, 2001), aims to test the influence of long branches on the resultant topologies (e.g. long branch attraction).
The heuristic tree search consisted of a thousand replicates of Wagner trees (using random addition sequences) followed by Tree Bisection Reconnection (TBR) branch swapping (saving 10 trees per replication). All the most parsimonious trees found in the replicates were subject to a final round of TBR.
Branch support was calculated using Jackknife resampling (Lanyon, 1985) as incorporated in TNT (see Goloboff et al., 2003). One thousand jackknife replicates were performed conducting a heuristic tree search consisting of 10 replicates of Wagner trees (with random addition sequences) followed by TBR (saving 10 trees per replicate). The results shown are difference in frequencies GC (for Group present/Contradicted) developed by Goloboff et al. (2003). The difference in frequencies GC was chosen because it is calculated as the difference between the frequency in which a given group is retrieved in the jackknife replicates and the most frequent contradictory group (Goloboff et al., 2003). Absolute frequencies (the usual method of counting frequencies in jackknife or bootstrap analysis) do not distinguish between a group with a frequency of 0.6 that is never contradicted, and a group with a frequency of 0.6 that is contradicted with a frequency of 0.4. GC frequencies distinguish these two cases, giving lower support values to the second type of groups (Goloboff et al., 2003).