A chi-square test of homogeneity of base frequencies across taxa was carried out using PAUP 4.0, b. 10 (Swofford, 2001). We executed the likelihood mapping method (Strimmer and von Haeseler, 1997) using TREE-PUZZLE (Schmidt et al., 2002) to test the a priori phylogenetic signal in the mtDNA portions studied. Phylogenetic congruence among 12S and 16S data partitions was also performed using the partition-homogeneity test implemented in PAUP* (Swofford, 2001).
Testing of the evolutionary model that best fits our data was conducted with MODELTEST 3.04 (Posada and Crandall, 1998), based on a likelihood ratio test. Different models of nucleotide substitutions were fitted to each data set and for the combined data set. For 12S, the TRN model was the best one selected (Tamura and Nei, 1993), while for 16S the GTR model was selected (Tavaré, 1986). Both models were corrected for rate heterogeneity among sites with a Gamma (G) distribution (Yang, 1993). For the 12S and 16S data sets, the GTR+G model was selected. We carried out a phylogenetic reconstruction by Maximum Parsimony (MP) (Kluge and Farris, 1969) and Neighbour-Joining (NJ) (Saitou and Nei, 1987) analyses using PAUP and by the Bayesian method using MrBayes 3.0B4 (Huelsenbeck and Ronquist, 2001). MP and NJ analyses were performed separately for each set of DNA sequences and in a combined analysis (total evidence approach) for the taxa sequenced for both genes. Neighbour-Joining trees were constructed with distances computed with the best-fit model obtained with MODELTEST. Parsimony analysis was carried out using the heuristic search algorithm, using 100 random-taxon-replicates for all analyses. The analysis was performed with ACCTRAN optimization and tree bisection TBR branch swapping, considering all characters as unordered and equally weighted, and gaps treated as fifth state. A strict consensus tree was calculated when there was more than one tree. Branch supports were assessed by 1000 non-parametric bootstrap replicates. Non-parametric bootstrapping with heuristic searches of 2000 replicates for MP and NJ was used to assess confidences of branches in MP and NJ. A Bayesian analysis was also performed with MrBayes 3.0B4, with clade support assessed by posterior probability. Four Markov chains, one heated and three cold, were allowed to run for two million generations using random starting trees, trees were sampled every 100 generations, with a burnin amounted to 20%.
Regarding the choice of outgroups for the Mediterranean stygobitic cirolanids, no marine cirolanid isopods were available to us. However, the direct marine ancestors of stygobitic cirolanids are unknown at present. The marine cirolanids described for the Mediterranean sea are 12 species belonging to three genera (Eurydice, Cirolana, Natatolana, Bruce, 1986; Keable, 2006). No samples of these taxa were collected during our research activities and we used as outgroup a marine cirolanid taxon present in the GenBank Database: Cirolana rugicauda Heller, 1861.