Phylogeography and molecular rates of subterranean aquatic Stenasellid Isopods with a peri-Tyrrhenian distribution

Levels and patterns of COI sequence variation

Different individuals of the same population have identical mtDNA sequences, thus we used a single sequence per population for our analyses. Percentages of variable sites range from 2.1% for the second codon positions to 85.7% for the third codon positions with an average value of 34.8% on all positions. The majority of parsimony informative sites are in third codon positions (45%), and in total 17.4% of changes were parsimony informative. As seen generally in mitochondrial DNA genes, this COI fragment is A + T rich, especially in third codon positions (65.2% all positions; 78.1% third codon positions). This pattern of nucleotide composition and sequence variation is similar to those reported in 33 other crustacean species analysed for the same COI fragment (Wetzer 2001).

In Table 1 the Tamura & Nei distances (D_T&N) are reported together with the absolute number of substitutions (Tv + Ti) and the number of Tv changes between each species pair. Sequence divergence between the S. racovitzai and S. virei is remarkably high (0.720 ± 0.030 Tv + Ti; 0.170 ± 0.007 Tv only), and close to the levels found between these two species and P. coxalis (0.838 ± 0.028 Tv + Ti and 0.185 ± 0.047 Tv only). This is partly at odds with the close phyletic relationship between S. virei and S. racovitzai suggested by Magniez (1974) on the basis of the morphology of the exopod of the IV pleopod. The average sequence divergence within the S. racovitzai complex is also remarkably high for comparisons between populations assigned to the same nominal species (0.293 ± 0.061 Tv + Ti; 0.093 ± 0.017 Tv only). Indeed, they are of the same order of magnitude of the values obtained for interspecific comparisons among several Isopods species (Wetzer 2001). Granted that levels of sequence divergence should not be used as a direct indication of taxonomic status, the high divergence values found in this study raise the possibility of a taxonomic revision of the complex. A complete survey of all the species included in the genus, coupled with the analysis of a larger DNA data set, and the integration of the genetic and morphological data needs to be accomplished to properly address this aspect.

Phylogenetic analyses and comparison with allozymes

The unweighted MP search produced a single most parsimonious tree 226 steps long (Fig. 2a). We obtained the same topology in all the MP analyses, regardless of the weighting scheme adopted. ML searches and NJ clustering produced trees with identical topologies. All three phylogenetic reconstructions have high bootstrap support values ranging from 72 to 100 over 1000 replicates. Similarly, Bremer support values for the MP tree are high (Fig. 2). Templeton's (1983) and Shimodaira & Hasegawa's (1999) tests indicate that the topology of the MP and ML trees (Fig. 2a) are significantly better than all other possible topologies (results available from the first author). Thus, the phylogenetic relationships obtained using this COI fragment are remarkably robust. The S. racovitzai complex is clearly a monophyletic group. This is in agreement with morphological analyses, which group these disjunct populations into the same phyletic lineage (Magniez 1974; Messana et al. 1995). Within S. racovitzai the two Sardinian populations are not each other's closest relatives. QUI, the population from SE Sardinia, is the sister taxon to a lineage which includes the NW Sardinian population (PTO) and the Corsican (COR) and Tuscan (UCC) populations.

Allozyme data are available for the same organisms, although not for exactly the same populations. Figure 2b (from Ketmaier et al. 1999) shows a dendrogram representing the genetic relationships among four populations of the S. racovitzai complex, and two other species within the same genus, S. virei (with two subspecies virei and hussoni) and S. assorgiai Argano, 1968. The tree in Fig. 2c summarizes the genetic relationships between all the known S. racovitzai populations, but does not include a outgroup (Ketmaier et al. 2000). Unfortunately, the three different data sets are not completely comparable because they do not share the same set of populations. However, it is interesting to note that within the S. racovitzai species-group the branching pattern recovered by the two data sets follows the same geographical scheme. The Corsican and Tuscan populations are always sister taxa (D_Nei’78 = 0.807 ± 0.139, Corsican vs. Tuscan populations; D_Nei’78 = 1.113 ± 0.134 Corsican and Tuscan vs. all Sardinian populations). The two Sardinian populations are not each other's closest relatives and the NW Sardinian population is the sister taxon of the Corsican–Tuscan clade (D_Nei’78 = 1.015 ± 0.055, NW Sardinia from Tuscany and Corsica; D_Nei’78 = 1.277 ± 0.024 SE Sardinia from the Corsican and Tuscan populations).

Molecular rates

Linearity of accumulation of substitutions over time was assessed by using two types of rate tests and and by taking advantage of the availability of absolute time estimates for two of the cladogenetic events in this group, the split of Corsica–Sardinia from the Pyrenees (29 Myr ago) and the separation of Sardinia from Corsica (20–9 Myr ago). A total of 96 Tajima (1993) tests was carried out, which included all the possible three taxa comparisons in the data set on all substitutions and on Tv only (in all positions and in third codon positions). We could not reject the null hypothesis of constant substitution rates at the P < 0.05 level (data available from first author). The ML ratio test produced similar results: the log likelihood from the ML analysis without enforcing a molecular clock was not significantly different from the log likelihood of the ML tree with a molecular clock enforced (χ² = 5.270; d.f. = 5; 0.250 < P < 0.500).

We also checked the results of the Tajima and ML ratio tests against absolute times. We assume that the split of the Corsica–Sardinia plate from the Pyrenees is responsible for the separation of the S. racovitzai and S. virei lineages, and that the split between Corsica and Sardinia led to the isolation of the S. racovitzai populations on the two islands. If our mtDNA sequences were accumulating substitutions linearly over time we would expect to obtain similar sequence divergence estimates between Sardinian and Pyrenean and between Corsican and Pyrenean species. Indeed, levels of sequence divergence between the Pyrenean (PEY) and the two Sardinian populations (PTO/QUI) are quite close to each other and remarkably similar to the value between PEY and the Corsican population (D_T&N= 0.692 and 0.765 for PEY vs. PTO and QUI; D_T&N= 0.735 for PEY vs. COR, Table 1). It is important to emphasize that the time estimates obtained in this study are based on a biogeographical event and thus they must be considered as minimal estimates of divergence, because they do not have an upper and a lower bound as dating based on fossils might have. Thus, we cannot exclude the possibility that genetic divergence occurred some time before the geological events we are assuming were the causal factors behind the cladogenetic events. However, the fact that several taxa of organisms with limited dispersal abilities share a similar distribution points strongly to a causal relation between geological events and cladogenetic patterns. A similar logic has been used to justify biogeographical based datings in a diverse array of marine species across the Panama Isthmus (Knowlton et al. 1993; Schubart et al. 1998; Wares 2001).

Table 2 lists COI rates for the fragment studied here using 29 Myr as the dating of the divergence of the Pyrenean population (PYR) from the Corsican (COR) and Sardinian (PTO and QUI) populations and compares them with rates for the same gene fragment in other Crustacea. Interestingly, the rates on all substitutions for Stenasellus are remarkably similar to the rates in a variety of marine Crustacea and in two lineages of cave beetles inhabiting the same regions as the S. racovitzai–S. virei group. This similarity of rates holds among taxonomically (Cirripeda vs. Decapoda, Crustacea vs. Insects) and ecologically (marine vs. interstitial, aquatic vs. terrestrial) disparate groups. Rates based on a subset of substitutional changes vary by a factor of 2.7–5 and do not seem be associated with either taxonomy or ecological requirements.

Another way to take advantage of the absolute time estimates and to check for linearity of sequence evolution in our data set is to compare geological and molecular ratios. If substitutions accumulate in a linear fashion, then the ratio of the geological estimates for the separation of the land masses (dating of the split between Corsica–Sardinia from the Pyrenees/dating of the split between Corsica and Sardinia) should be similar to the ratio of the molecular distances (average distance between S. virei and all the S. racovitzai populations/average distances between the Corsican and Sardinian S. racovitzai populations). The ratios range from 1.5 to 3.2 (29/15–29/9) for the geological divergences and from 1.4 to 2.9 for the molecular distances, a result consistent with a linear accumulation of changes over time in this COI fragment. If the Pyrenean lineage of S. racovitzai (PEY) split from the Corsica–Sardinia–Tuscany clade (QUI–PTO–COR–UCC) 29 Myr ago, we can date the differentiation events between the Corsican and Sardinian taxa and see if they fall within the range of the geological estimates. Depending on the type of rate used (Table 2), time estimates for the split between the Sardinian (PTO) and Corsican (COR) populations range from 12 to 20 Myr ago. This result suggests that regardless of the type of rate used the differentiation between the Sardinian and Corsican/Tuscan clade occurred within the range of the geological estimates for the separation of the islands and that it predates the complete separation of the two islands (9 Myr ago). A similar time was obtained for the cladogenetic events leading to the separation of newts with the same distribution (Caccone et al. 1997). This result is in line with the consideration that geological datings are minimal estimates of genetic divergence and that separation of these taxa is likely to have started before the completion of the geological event.

We also used the mtDNA rates to provide times of divergences for the other cladogenetic events in the Stenasellus phylogeny that have been associated with geological events but have uncertain dating: the split between between Proasellus and Stenasellus, the separation of the two Sardinian taxa (QUI and PTO) and the split between the Corsican and Tuscan populations (COR and UCC).

Using the COI rates in Table 2 the divergence between the two genera occurred at least 33–29 Myr ago. We do not have the proper taxon sampling scheme to address this question, which is peripheral to the scope of this study. However, it is interesting to note that the Miocene time was an intense orogenetic period in Europe (Steininger & Rögl 1984), which might have played a role in the separation of the two genera. This implies that the separation of the Pyrenean and Sardinian–Corsican Stenasellus might have happened almost at the same time as the divergence of the whole genus from Proasellus.

Within Stenasellus the two Sardinian populations separated 17–22 Myr ago, according to the rates in Table 2. This timing is in agreement with the Miocene sea introgressions which formed a geographical barrier that split Sardinia into two distinct calcareous blocks (Fig. 1c; Cherchi & Montadert 1982; Boccaletti et al. 1990). All Stenasellus populations have always been found to be associated with limestone areas. These areas are surrounded by granite rocks, which isolate the different water systems in the area. Between the northern and southern Sardinian populations there is a deep and old break in the limestone deposits (Fig. 1b), which makes it impossible for this type of organisms to disperse actively.

Our estimates for the split between the Corsican (COR) and Tuscan (UCC) populations are more recent than those for the Sardinian populations, ranging from 2 to 0.4 Myr ago, between the Donau and the Mindel glaciations during the Quaternary period (La Greca 1998). This result supports the idea that the presence of Stenasellus in coastal Tuscany is not due to an old vicariance event but is rather the result of active dispersal. Enlarged fluvial basins (a condition ideal for dispersal of interstitial organisms) during interglacial periods could have facilitated dispersal through a continuous hydrographic system on the land bridge that connected these regions during the Quaternary (Giusti 1976; Lipparini 1976; Cherchi & Montadert 1982; Burgassi et al. 1983; La Greca 1990).

Phylogeographical scenario and biogeographical considerations

The phylogenetic relationships presented in this study are in remarkable agreement with the known geological history of the area. The Corsican and Sardinian populations are related more closely to each other than to the Pyrenean species, as expected on the basis of the history of the land masses they occupy. The deep divergence of the two populations from SE and NW Sardinia is also in good agreement with the history of sea regressions, that in Miocene–Pliocene times inundated the rift that crosses Sardinia from North to South producing a powerful geographical barrier for freshwater organisms with poor dispersal capacities (Fig. 1b). A comparable pattern of genetic divergence has also been found in other organisms from Sardinia with similar distributions and limited dispersal abilities, such as freshwater populations of the Proasellus coxalis species group (Ketmaier et al. 2001) and cave beetles (Caccone & Sbordoni 2001). The close phyletic relationships between the populations from NW Sardinia, Corsica and Tuscany also reflect the geology and the well-known connections between Sardinian, Corsican and coastal Tuscan faunas (Lanza 1984). A similar situation occurs in some other stygobiont taxa, such as the harpacticoid Parapseudoleptomesochra minoricae (Chappuis & Rouch, 1960) and the amphipod Bogidiella chappuisi Ruffo, 1952, which share the same disjunct distribution, and similar ecological requirements to Stenasellus. This parallel distribution pattern (Sardinian and Tuscan islands and coasts) has led to the recognition of a Sardinian stygofaunal province that also includes the insular and coastal areas of Tuscany (Pesce 1985).

In conclusion, we suggest that a combination of vicariance and dispersal events need to be invoked to explain the distribution of these taxa and their patterns of genetic differentiation. Vicariance can explain the separation of the Pyrenean populations from the Corsican and Sardinian ones, and can also be the main process responsible for the separation of the SE and NW Sardinian populations. However, we need to invoke active dispersal to explain the low levels of genetic divergence between the Corsican and Tuscan populations. However, is it reasonable to hypothesize active dispersal for organisms with very strict ecological requirements and very limited dispersal abilities? According to Magniez (1981), Stenasellids are adapted to live in all types of subterranean waters. Although they can be found in free water, they live primarily in interstitial waters. In such environments they can dig ramified galleries and, if continuous, migrate and extend their settlement. During the Pleistocene and later on in the Quaternary the lowering of sea levels determined water captures between rivers flowing into the Tyrrhenian Sea (Bartolini & Pranzini 1981). This created a favourable environment for dispersal of different phreatobic organisms, including a number of stygobionts such as copepods, ostracods, amphipods, and syncarids, which share the same disjunct distribution as Stenasellus, and have comparable ecological requirements. However, this still does not explain why Stenasellids arriving at the Tuscan Archipelago did not disperse any further and why Stenasellids are known for only two locations in Tuscany (see Fig. 1c). A possibility we cannot discount is that Stenasellids might have been more widespread at some time in the past and became extinct when the fossil islands were incorporated into the mainland because of the higher competitive pressures they had to face, moving from a continental habitat from an insular one (Lanza 1984). This could be the reason why their range is so small even in the fossil island, where these isopods are still present. At the same time, it should also be considered that these organisms are not common and easy to collect and it could be that further surveys in Tuscany could surprise us with new localities of these elusive organisms. However, it needs to be noted that in the past 10 years both the Sardinian and Corsican coast have been surveyed intensively for interstitial fauna (Messana & Argano, pers. com.) and no new locations for these organisms have been reported.