We have sampled 14 A. franciscana populations distributed in North and South America and obtained sequence information from a total of four mitochondrial and nuclear loci. The 16S rRNA sequences ranged in length from 455 to 458 bp, with a total of 19 parsimony-informative sites. Base frequencies were homogeneous across taxa (χ² = 13.22, df = 60, p = 0.99). Similarly, inspection of plots between transitions/transversions and corrected distances (TrN + G, α = 0.20) revealed no signs of substitution saturation (data not shown). The g₁ value (g₁ = −0.659, p < 0.05) indicated that the tree-length distribution was significantly more left-skewed than expected from random data, thus suggesting the presence of phylogenetic signal. Pairwise corrected distances for the ingroup ranged from zero to 3.7%. Distances to outgroup (A. salina) ranged from 45.3 to 62.8%.

Similar patterns were also obtained for the COI data set (582 bp, 31 parsimony-informative sites). Pairwise ingroup distances (HKY + G, α = 0.27) ranged from zero to 4.7%, while ingroup-to-outgroup divergence values ranged from 39.4 to 44.1%.

A similar picture regarding data composition (base frequencies) and quality (substitution saturation, presence of phylogenetic signal) was obtained for both nuclear loci. For ITS1 (ingroup sequence length 1165–1182 bp), base frequency homogeneity could not be rejected (χ² = 6.82, df = 60, p = 0.99), while the tree-length distribution was significantly left-skewed (g₁ = −0.520, p < 0.05). Intraspecific and interspecific (to A. salina) distance estimates (HKY + G, α = 0.37) ranged from zero to 4.7% and from 23.5 to 28.1%, respectively.

In a previous study using the second intron of p26 as a discriminant marker in A. franciscana, Maniatsi et al. [29] identified a distinct pattern of allelic polymorphism between populations from San Francisco Bay (SFB) and Great Salt Lake (GSL). To ensure adequate resolution of the allelic diversity within our data set we preliminary screened several individuals (25–40) from each population. Three different p26 alleles were recovered as gauged by visualization of PCR-amplified fragments under UV light: the most common p26 allele corresponds to a fragment of ~1500 bp, the second allele corresponds to a fragment of ~2000 bp while the third bears a length of ~1400 bp. Populations GSL, CEJ, and CON were polymorphic for these three alleles and sequencing was performed in selected individuals (Table 1). Ingroup diversity estimates for p26 (HKY + G, α = 0.58) varied from a minimum of 0.1% to a maximum of 15.2%. Highest values were observed for the haplotypes corresponding to ~1500 bp (0.4–14.8%) while for the ~2000 and ~1400 bp alleles genetic diversities were 0.1–7.5% and 0.2–3.4%, respectively.

The pairwise 16S and COI genetic distance matrices were significantly correlated (Mantel test: Z = 0.88, r = 0.99, p < 0.001, 1000 randomizations). Three well-differentiated clades with high nodal support were recovered with the mitochondrial DNA data. The same topology was also recovered when considering third codon positions for the COI gene (data not shown). The maximum parsimony analysis of the concatenated genes (16S + COI) recovered 124 most parsimonious trees (tree length = 230). The consensus phylogeny is shown in Figure 1. The first clade consists of the North American populations SFB, GSL, the Brazilian feral populations MAC, GAL, ABG, and the northernmost Chilean population of IQU. In the second clade, the Chilean populations of CON and LLA are clustered with all three Argentinean representatives (LTU, SET, CHI). The third clade consists of the remaining Chilean populations (LVI, CEJ, CHA). Analysis of population structure on the basis of the mitochondrial DNA topology revealed a strong pattern of population differentiation. The K_ST statistic was 0.74 (p < 0.001, 10,000 randomizations) meaning that sequence differences segregate according to clade origin. In addition, there was no evidence for an isolation-by-distance pattern as judged by the lack of correlation between genetic and geographic distances, excluding the three Brazilian feral populations (Mantel test: Z = 130.43, r = 0.21, p = 0.13).

For the combined nuclear data, significant correlation was also evident between the ITS1 and p26 genetic divergence matrices (Mantel test: Z = 2.85, r = 0.98, p < 0.05, 1000 randomizations). The inferred phylogeny using maximum parsimony (38 equally parsimonious trees, tree length = 1224) is shown in Figure 2. Four groups are recovered corresponding to the different lengths of the p26 alleles (1400, 1500, and 2000 bp) with the 1500 bp group splitting further in two clusters. The most divergent group is the one bearing the 2000 bp haplotypes: it shares a more distant common ancestor with the others groups (see Figure 2), with distance values of 8.2% (2000 versus 1500) and 7.2% (2000 versus 1400), while 1500 versus 1400 equals 4.8%. Interestingly, the 2000 bp group retains a similar phylogenetic structure with that of the entire mitochondrial DNA tree (contains representatives from the three mitochondrial DNA clades). The remaining three groups (1400, 1500-A, 1500-B) are characterized by various degrees of cohesion and concordance to the recovered mitochondrial clades. Of particular importance are the LVI and IQU denoted haplotypes whose closest relatives reside in different mitochondrial and nuclear clades. It should be emphasized that allele lengths (1400, 1500, 2000) are used only as tags in the nuclear phylogeny; the resulting groups are based solely on sequence differences/similarities since gaps were considered as missing data.

The inferred mitochondrial genealogy is a typical example of pronounced phylogenetic gaps between allopatric lineages. The presence of three reciprocally monophyletic mitochondrial phylogroups (clades 1–3, see Figure 1) with statistically significant support is reminiscent of phylogeographic discontinuities modulated by long-term extrinsic barriers to genetic exchange or low gender-specific dispersal and gene flow (see [1]). Between-clade distances for 16S and COI range from 2.2 to 4.2%, with a proportion of interclade differentiation of 81.3%.

Our results are in agreement with previous estimates based on allozymes reporting considerable differentiation (mean genetic distance of 12.6%, [37]) and substructuring (mean F_ST = 0.24, [38]) between geographic populations of the A. franciscana group. For South American A. franciscana populations, mean F_ST values range from 0.38 (allozymes, [39]) to 0.91 (16S RFLPs, [40]). Marked genetic distances have also been reported between Mexican and North American A. franciscana [41]. In a study of the intraspecific phylogeography of the Mediterranean A. salina [42], COI distances between higher-level clades (as determined by nested clade phylogeographic analysis) ranged from 1.1 to 2.7%. The authors found a pattern of extensive regional endemism, suggestive of an early Pleistocene split of Mediterranean lineages and subsequent population isolation in a number of glacial refugia. Analogous interpretations have also been advanced to explain the presence of localized distinct phylogroups in the rotifer Brachionus plicatilis (mean COI divergence 2.8%) from the Iberian Peninsula [43] and the freshwater cladoceran Daphnia magna (COI divergences 0.73–1.74%) from Europe [13].

Our mitochondrial DNA data point to a similar pattern of regional endemism and phylogeographic structure at the microscale, typically seen in continental zooplankters [43]. The substantial amount of genetic differentiation between phylogroups, the support of monophyly of clades, and the absence of co-distributed lineages reflect accumulation of mutational differences following population separation. In other words, the lineage sorting process has been completed and the mitochondrial genealogy conveys a strong signal of historical fragmentation. Alternatively, gene flow between proximate populations may have been severely restricted in contemporary time, a possibility which can be effectively evaluated with additional and more extensive sampling.

Informative insights to the observed regional patterns of lineage endemism may be provided by consideration of biogeographic data pertinent to South America and, in particular, to the recent geological history of the Andes mountain range. Recent data indicate that the final phase of the central Andean uplift took place between 6.0 and 2.7 mya [44] affecting the diversification of taxa on either side of the mountain range [45]. Divergence rates for snapping shrimp COI sequences have been calculated at 1.4% per million years [46] and useful approximations are allowed when sequence divergences are in the order of <8%. Implementation of the above COI rate yields an estimated time of divergence between the most distantly related mitochondrial clades (clades 2 and 3, 4.2%) at 3.0 mya which is just before the last uplift of the Andes, in the late Pliocene. This is a strong indication that the Andean orogeny had a major effect on lineage splitting between clades 2 and 3 and extant populations have probably descended from gene pools which persisted through the recent periods of repeated glaciations. Thus, our results imply a longer demographic history than that expected by populations evolving during postglacial times. This is a signature of potential isolated refugial areas on both slopes of the Andes [45]. A rather similar pattern of spatial structuring of distinct genetic lineages has also been observed for A. persimilis populations on either side of the Andes [11]. Although such interpretations may involve a large margin of error, they provide an evolutionary timescale of lineage separations on which, given additional data, a more statistical phylogeographic framework can be devised (see [47,48]).

The most salient feature of the inferred phylogeny is the incomplete lineage sorting of nuclear genes. Under neutrality, the expected time for lineage sorting to reciprocal monophyly is fourfold longer for nuclear lineages than for mitochondrial lineages [2]. Thus, due to a fourfold larger effective population size, the coalescent process proceeds more slowly for alleles at nuclear genes than for mitochondrial haplotypes. The nuclear tree quite clearly captures a snapshot of the above process for the ITS1 and p26 allelic lineages. In addition, due to the stochastic nature of genetic drift, its effects, simply by chance, may have not been registered strongly in the investigated nuclear loci. The dynamics of the specific sequences may also be involved by virtue of their being influenced by different forces, as ITS1 is a spacer region and p26 an intron region.

The incomplete sorting of ancestral polymorphisms is most apparent for the 1500 bp group of p26 alleles, and particularly for 1500-B (see Figure 2), which shows higher sequence diversity (0.4–14.8%) compared with the other groups of allelic lineages (1400, 2000) and contains representatives of all the respective mitochondrial clades. This paraphyletic pattern, as well as the subtopologies within the other groups (1400, 1500-A), may have likely arisen as a result of male-mediated dispersal. This mechanism erodes geographic structure in nuclear genes, yet leaves the mitochondrial DNA phylogeny unaffected (see [49]). It is also worth noting that the 1500-B allelic lineage is represented by populations with the highest proximity, a fact relating to the previous dispersal scenario. Moreover, allozyme analyses of South American A. franciscana have identified the populations of LVI and IQU as the most divergent ones, bearing a mixture of both private and shared polymorphisms [39]. An obvious critical requirement for the dispersal mechanism is that transported cysts developing into females do not reproduce with resident males. Although there is currently no evidence for intraspecific directionality in reproduction, clues may be obtained from the interspecific level where such a mechanism seems to work [12]. Together, these findings support the view that gene flow has been operating to such a degree as to prevent genetic differentiation by genetic drift or, alternatively, selection has favoured the maintenance of certain alleles in South American sites. It is interesting to see that populations CON and CEJ (together with GSL from North America) display a pattern of polymorphism for the p26 allelic lineages. In particular, all three p26 alleles are found in the CON population (1400, 1500, 2000) whereas for CEJ and GSL two p26 alleles were detected (1500, 2000) with the 1500 alleles grouped in different lineages for each population (see Figure 2). Although additional data are needed to explain the distribution of this polymorphism, it could be tentatively ascribed to the fact that the three populations (members of distinct mitochondrial clades, see Figure 1) represent the most ancestral stocks (possibly refugial sites) from which lineage colonization and subsequent random loss occurred.

Irrespective of the specific etiology for the observed nuclear genealogy, the overall phylogeographic evidence in the investigated A. franciscana populations supports a genealogical phase at the interval between gene splitting and population fragmentation. Put differently, this interval is where speciation occurs. The mitochondrial DNA evidence upholds a specific demographic independence between sets of A. franciscana populations. This is registered in the presence of reciprocally monophyletic clades, being, by definition, genetically distinct and free of exchange. Depending on the evolutionary timescale, these population sets may qualify as evolutionary significant units [50], thus justifying to a large extent arguments for incipient speciation in A. franciscana. Our data come to reinforce previous evidence on significant population differentiation in A. franciscana. A distinct correlation between chromocentre frequency and latitude, certain types of aneuploidy as registered in fluctuating chromosome numbers, extreme examples of ecological isolation, and an association between habitat type and heterozygosity in A. franciscana populations [23] are all conducive to lineage diversification and suggest the presence of both geographical and ecological forms. However, before solid statements on the possible superspecies status of A. franciscana are made, more thorough phylogeographic nuclear sequence assays, accounting for the temporal and spatial aspects of population genealogical structure, are needed in order to unite the observed genetic attributes of the species with its postulated past biogeography.

We analyzed 14 A. franciscana populations from North (USA) and South America (Brazil, Chile, Argentina) (see Figure 3, Table 2). The three Brazilian feral populations originate from material inoculated more than 30 years ago in the respective sites [28]. We used published ITS1 and 16S rRNA sequences from Baxevanis et al. [11] and Kappas et al. [12] and also obtained novel sequences of these markers following the protocols by Kappas et al. [12] (for GenBank accession numbers see Table 1). Additionally, we amplified a 601 bp fragment of the mitochondrial cytochrome oxidase subunit I (COI) gene region and the entire second intron of the nuclear p26 gene. Whenever possible, we used the same individuals to those registered in GenBank for ITS1 and 16S rRNA.

For COI, a set of newly designed primers were employed (available upon request) based on deposited COI sequences of Anostraca. PCR conditions were: 4 min at 94 °C, 40 cycles of 50 sec at 94 °C, 50 sec at 50 °C, 50 sec at 72 °C and a final extension of 5 min at 72 °C. Total reaction volumes of 25 μL consisted of 2.5 μL template DNA, 5 μL 5 × PCR buffer, 2.5 mM MgCl₂, 1 mM dNTPs, 0.001 mM of each primer and 1.25 U of Taq DNA polymerase (Expand High Fidelity^PLUS PCR System, Roche).

For p26, PCR conditions are detailed in Maniatsi et al. [29]. In all cases, bands were purified prior to direct sequencing (complementary strands). All amplifications were performed on a PTC-100^® Peltier thermal cycler (MJ Research). Sequencing reactions were electrophoresed on a PRISM 3730xl DNA analyzer (Applied Biosystems). The newly derived COI and p26 gene sequences were deposited in GenBank (see Table 1). Sequences were aligned using ClustalX 1.8 [30]. In all cases (16S, COI, ITS1, p26), genealogies were rooted using A. salina as outgroup (accession numbers: 16S: FJ007839, COI: GU248381, ITS1: DQ201303, p26: GU248407).

Newly derived sequences were subjected to BLAST searches for confirmation of homology. For all data sets we conducted tests for homogeneity of base frequencies, substitution saturation, and presence of phylogenetic signal following the methods in Kappas et al. [12]. Values of genetic diversity within and distance between populations were computed in MEGA version 4 [31]. Corrected distances were used based on the best-fit substitution model as determined by Modeltest 3.7 [32].

Phylogenetic reconstruction was implemented in PAUP* 4.0b10 [33] under maximum parsimony. Phylogenetic inference was based on concatenated sequence data (mitochondrial: 16S + COI; nuclear: ITS1 + p26). Prior to data combination, conflicting phylogenetic signal was evaluated through Mantel tests on the corresponding genetic distance matrices. For each joint data set, trees were generated using heuristic searches with TBR (tree-bisection-reconnection) branch swapping and 500 random taxon additions. Nodal support was assessed by 1000 bootstrap replicates. Gaps were treated as missing data. Phylogenetic analysis was also conducted considering third codon positions only for COI.

For mitochondrial DNA, we evaluated isolation-by-distance (IBD) patterns by correlating geographic and genetic distances (corrected) between A. franciscana populations. We used the IBDWS version 3.14 (Isolation by Distance Web Service, [34]) to perform Mantel tests (both raw and log-transformed matrices, 30,000 randomizations) and Reduced Major Axis (RMA) regression analysis. We also tested for the presence of phylogeographic structure using DnaSP version 5 [35]. We used the nucleotide sequence-based K_ST statistic as given in Hudson et al. [36] in order to control for the presence of unique haplotypes in different groupings. Group assignment was based on the inferred topology during phylogenetic reconstruction. Statistical significance was assessed using a permutation test (10,000 randomizations).