Two reproductively isolated cytotypes and a swarm of highly inbred, disconnected populations: a glimpse into Salicornia’s evolutionary history and challenging taxonomy

Authors


A. Vanderpoorten, Institute of Botany, University of Liège, B22 Sart Tilman, B-4000 Liège, Belgium. Tel.: +32 4366 3842; fax: +32 4366 2925; e-mail: a.vanderpoorten@ulg.ac.be

Abstract

The main factor of differentiation at six nuclear microsatellite and seven cpDNA loci in Salicornia from the Mediterranean and Atlantic coasts of France is cytotypic identity, suggesting the presence of a strong reproductive barrier among sympatric cytotypes. Within cytotypes, a substantial proportion of the differentiation between species is due to confounded phylogeographic signal. Conspecific individuals tend to be significantly more related than individuals from different species at the population scale, but mean kinship coefficients among pairs of conspecific and nonconspecific individuals from different populations are not significantly different, suggesting the absence of reproductive isolation among species of the same cytotype. The observed association between morphology and genetic variation within populations would thus result from the selfing mating system (Fis = 0.70) generating substantial linkage within the genome, linkage that would quickly disappear among unrelated individuals from different populations. Salicornia species thus function as a network of inbred populations, strongly challenging taxonomic concepts.

Introduction

Polyploidization is an important evolutionary force. Estimates suggest that 70% of all angiosperms have experienced one or more episodes of polyploidization (Masterson, 1994), and polyploids represent the bulk of the species diversity in many major lineages of land plants including pteridophytes (95%, Soltis & Soltis, 1999) and mosses (79%, Wyatt et al., 1988). As opposed to the traditional view of polyploids as genetically depauperate species because of the sudden reproductive isolation after a single polyploidization event, an increasing body of literature suggests that individual polyploid species evolved multiple times (see Albach, 2007, for review), often after the fusion of nonreduced gametes from different species. The greater diversity resulting from increased heterozygosity found within individual genomes theoretically provides selective advantages in changing environments, explaining for example the striking increase in polyploids with latitude, associated intensity of climatic fluctuations, and frequent recolonization events (Abbott & Brochmann, 2003; Brochmann et al., 2004). At the population level, by contrast, recently formed polyploids might, owing to founding effects, harbour less genetic diversity if polyploid formation is a rare and/or recent event (see Soltis & Soltis, 1999, for review). The recurrent series of range contractions and expansions during the Pleistocene glacials and interglacials would further reinforce this effect (Hewitt, 2004). An ancient, possibly polyphyletic origin of polyploids, as well as subsequent introgression, may, however, account for the high genetic diversity that has increasingly been reported among polyploid populations (Ramsey & Schemske, 2002; Luttikhuizen et al., 2007).

Salicornia (Amaranthaceae), a sub-cosmopolitan genus of annual species, is a model of choice for exploring the origin and evolution of polyploid formation and intercytotypic barriers. Two chromosome series, one diploid with 2n = 18 and the other tetraploid with 2n = 36, are present in the genus (Dalby, 1962; Shepherd & Yan, 2003), and the distinction between cytotypes is fully supported by amplified fragment length polymorphism (AFLP) and DNA sequences (Le Goff, 1999; Kadereit et al., 2007; Murakeözy et al., 2007). Tetraploids, whose seedling radicles grow more rapidly than those of diploids, typically occur in areas exposed to tidal disturbance, whereas diploids colonize more stable, upshore habitats (Davy et al., 2001). Thus, perhaps more than in other groups of plants that are often characterized by cytotypic vicariance along major ecological gradients (see Duckert-Henriod & Favarger, 1987; and Vamosi et al., 2007; for review), diploid and tetraploid Salicornia species are commonly found in sympatry along tidal gradients (Davy et al., 2001), if not completely intermixed in complex vegetation mosaics (Lorenzoni et al., 1993; van Hulzen et al., 2006), making the genus a keystone for the phytosociological classification of European salt marshes (Géhu, 1992; Géhu & Bioret, 1992).

Owing to their strict occurrence in extremely constraining salty environments, Salicornia species display a strikingly reduced, convergent morphology offering a limited number of characters for taxonomy. The flowers are uniformously tepaloid and the leaves reduced to scarious rims. The level of variation of ITS, ETS and trnL sequence data is extremely low within cytotypes (Kadereit et al., 2007; Murakeözy et al., 2007), which results in poorly resolved relationships among species. None of the traditionally defined species, even those whose distinction is straightwforward, such as S. pusilla, which is the only species among diploids with solitary flowers, were, however, resolved as monophyletic (Kadereit et al., 2007; Murakeözy et al., 2007). The genus has, consequently, long puzzled taxonomists, and systematic treatments range from recognition of an array of species and their putative hybrids (e.g. Lahondère, 2004; Stace, 2010) to severe reductions to synonymy, keeping one or a few species within each cytotype (e.g. Valdés & Castroviejo, 1990; Piirainen, 2001).

In the present study, we use a combination of PCR-RFLP markers across a range of cpDNA loci as well as specific nuclear microsatellites to define the hierarchy of factors accounting for Salicornia’s genetic variation patterns along the Mediterranean and Atlantic coasts of France and address the following questions at different nested taxonomic and spatial scales: (i) What is the origin of the tetraploids and what are the genetic relationships between diploid and tetraploid lineages? (ii) Are species reproductively isolated? (iii) What are the dispersal ranges of seeds and pollen and does dispersal rates exceed mutation rates, or is a strong phylogeographic signal present in the data?

Materials and methods

Taxon sampling

In total, 566 specimens were collected from 13 French Mediterranean (including Corsica) and 30 localities along the Atlantic coast (sensu lato, including the English Channel and the North sea) (see Table S1, Fig. 1). Diploids and tetraploids differ in a suite of nonequivocal morphological features (e.g. Dalby, 1962; Lahondère, 2004), and a recent study combining flow cytometry and thorough morphometric analyses confirmed that morphological variation conforms to cytotypic differentiation (Kaligaric et al., 2008). The degree of size difference between the median and the lateral flowers proved, in particular, to be most useful to distinguish individuals attributed to either the diploid or tetraploid lineage based upon their morphology. The reliability of the morphological criterion was confirmed by a trial on a subset of 30 freshly collected specimens whose ploidy level was determined by flow cytometry using the protocol described in Kaligaric et al. (2008). Based on the morphological criterion, the 566 specimens included 194 tetraploids (comprising collectively S. fragilis, S. emericii and S. dolichostachya) and 372 diploids (comprising collectively S. ramosissima, S. obscura, S. patula and S. pusilla). The application of the morphological criterion was, however, complicated in the case of S. obscura, a species that differs from all other diploids by flowers of sub-equal size. Therefore, the cytotype of 98 such specimens was scored as ambiguous, and those specimens were not included in the analyses explicitly partitioning the genetic variation among cytotypes.

Figure 1.

 Localization of the 43 populations of Salicornia sampled along the Mediterranean and Atlantic coasts of France.

The taxonomic system of Lahondère (2004), which closely matches the most recent treatment of Stace (2010), was used here for species identification. We followed, however, the species names accepted by Kadereit et al. (2007), and hence, considered S. brachystachya to be synonym to S. ramosissima. Our sampling thus included specimens of S. pusilla, S. ramosissima, S. patula and S. obscura, which are diploid, and S. emericii, S. dolichostachya and S. fragilis, which are tetraploid. Four species are restricted to the Atlantic coasts of Europe: S. pusilla and S. obscura have been reported from France to The Netherlands and the British Isles; and S. dolichostachya and S. fragilis from the Iberian Peninsula to Denmark and the British Isles. One species, S. patula, is restricted to the Mediterranean, where it has been reported from the Iberian Peninsula to the Adriatic. Finally, S. emericii and S. ramosissima occur along both the Atlantic and Mediterranean coasts of Europe. Each specimen was placed in individual Eppendorfs and kept at −80 °C pending DNA extraction.

Molecular protocols

Samples were cooled in liquid nitrogen and immediately ground using a Retsch MM 301 mill, followed by DNA isolation using the CTAB protocol of Doyle & Doyle (1987) without any RNase treatment. Each specimen was genotyped at six nuclear microsatellite loci as described in Vanderpoorten et al. (2010a), and at seven cpDNA loci using PCR-RFLP (Table 1).

Table 1.   Chloroplast DNA amplicons analysed in a PCR-RFLP analysis of diploid and tetraploid Salicornia, with primer information, PCR conditions and restriction enzymes.
AmpliconReference for primersTa (°C)Extension time (min)Restriction enzyme(s)Number of polymorphic bands (alleles)
  1. Ta, annealing temperature.

trnH-trnKDemesure et al. (1995)602Hsp92II1 (9)
trnK1-trnK2Demesure et al. (1995)583HinfI + MboI3 (2, 2, 5)
trnK2-trnQDumolin-Lapègue et al. (1997)543HinfI + RsaI4 (5, 2, 2, 2)
trnC-trnDDemesure et al. (1995)583HinfI + RsaI4 (2, 6, 3, 2)
trnT-psbCDumolin-Lapègue et al. (1997)524HinfI + MboI5 (2, 3, 3, 3, 2)
trnS-trnfMDemesure et al. (1995)562HinfI + MboI2 (3, 2)
trnV-rbcLDumolin-Lapègue et al. (1997)563HinfI + MboI4 (4, 2, 2, 6)

Amplification of cpDNA regions was performed in 20 μL containing 1× enzyme buffer, 3.5 mm MgCl2, 200 μg mL−1 BSA, 200 μm dNTPs, 0.2 μm of each primer and 0.5 U Taq polymerase (Roche, Penzberg, Germany) or DreamTaq polymerase (Fermentas, St Leon-Rot, Germany). PCR cycling conditions were as follows, with annealing temperature and elongation time depending on the amplified region (Table 1): 94 °C for 3 min, 35 cycles of 30 s at 93 °C, 30 s at 52–62 °C, 2–4 min at 72 °C and a final extension of 7 min at 72 °C. Amplicons were then digested with one, or a cocktail of two restriction endonucleases (Table 1). Restriction products were separated on 9% polyacrylamide gels (29 : 1 acrylamide : bisacrylamide) for trnK1-trnK2, trnK2-trnQ, trnC-trnD, trnT-psbC, trnS-trnfM, and trnV-rbcL, and 1.8% agarose gels for trnH-trnK. Gels were stained with ethidium bromide and photographed under UV light. For each polymorphic band in the restriction profile, size variants were numbered with decreasing size, whereas restriction-site polymorphisms were coded as 0 for the highest molecular weight band when it contained the restriction site or, if one of the two bands observed when the restriction site was present showed size variants, this band was coded as 0 in specimens in which the restriction site was absent.

For simple sequence repeats (SSRs), the genotypes at each locus were directly scored for diploid species from the electropherograms using GeneMapper 4.1 (Applied Biosystems, Foster, CA, USA). For tetraploids, the dosage of the different alleles within a genome was unknown. Microsatellite DNA Allele Counting using Peak Ratios was introduced to quantify allelic configuration in polyploids (Esselink et al., 2004). This technique relies on the assumption that during PCRs, abundant alleles within a locus should amplify more often than less abundant ones. The relative peak areas found in peak diagrams are therefore thought to be correlated with the relative number of copies of that allele within the genome. However, PCR selection caused by differential primer affinity, allele size and PCR drift resulting from events during early cycles of PCR causes a bias in simultaneously amplified products (Wagner et al., 1994). Peak area ratios in heterozygous diploid individuals thus nearly always differ from 1, requiring a correction for differences in amplification success of alleles. In hybrids, however, unequal amplification efficiency at microsatellite loci between species may occur, so that individuals that are hybrids between these two groups may not possess the same allele dosage (Vergilino et al., 2009). In allotetraploids such as Salicornia (see below), wherein differences in amplification patterns between primer pairs are most likely due to incomplete complementarity of the designed primers to the target regions in the heterologous chromosomes (Catalán et al., 2006; Palop-Esteban et al., 2007), the definition of clear correction factors proves most difficult (Scheepens et al., 2007; Gonzalez-Perez et al., 2009; Helsen et al., 2009). As a consequence, each microsatellite allele was scored for each specimen using a binary coding similar to that employed with dominant markers, i.e. each allele was scored as present/absent. This way of coding genotypes can cause some redundancy (one locus can occur several times in the data matrix), but it solves the problem of assessing allele dosage and reduces the risk of mixing genotypes from homeologous loci in tetraploids.

Data analysis

Global patterns of genetic variation

We first explored the SSR data matrix without making any a priori hypothesis of structure in the data. The mating system of Salicornia is characterized by high levels of selfing (Dalby, 1962; Ferguson, 1964; Jefferies & Gottlieb, 1982), hampering the use of techniques such as those implemented by Structure (Pritchard et al., 2000) that rely on Hardy–Weinberg equilibrium to perform groupings. Instruct, an extension of Structure that performs groupings from expected genotype frequencies based on inbreeding or selfing rates (Gao et al., 2007), and which is thus potentially well suited to strong selfers like Salicornia, does not allow for both diploids and tetraploids to be analysed simultaneously. We therefore submitted the matrix of presence–absence of alleles to a principal components analysis (PCA) performed on the correlation matrix and implemented the Markov chains Monte Carlo (MCMC) of Instruct within each of the diploid and tetraploid cytotypes. The admixture model of Instruct was run for K = 2–20 groups, and the deviance information criterion was employed to determine the optimal value of K. For each K, three MCMCs were run for 500 000 iterations each and 100 000 steps of burnin. Convergence of the three runs was assessed by means of the Gelman–Rubin statistics.

For the chloroplast data, the minimum spanning haplotype network was computed using Arlequin 3.1. Support for clades was assessed by a cladistic analysis of the matrix of presence/absence of bands through an adaptation of Jukes-Cantor’s substitution model for binary characters (Lewis, 2001). Restriction-site data typically suffer from a severe sampling bias because constant characters are never recorded, leading to an overestimation of the transition rates and hence, the necessity to apply a correction to the model (Felsenstein, 1992). This correction was implemented by using the ‘variable’ coding option of MrBayes 3.1 (Ronquist & Huelsenbeck, 2003). The model was implemented in a Bayesian framework. Four Metropolis-coupled Markov chain Monte Carlo of four chains each were run for 10 000 000 generations with MrBayes 3.1. Trees and model parameters were sampled every 10 000 generations. The number of generations needed to reach stationarity and chain convergence was estimated by visual inspection of the plot of the log-likelihood score at each sampling point. The trees of the ‘burn-in’ for each run were excluded from the tree set, and the remaining trees from each run were combined to form the full sample of trees assumed to be representative of the posterior probability distribution.

Cytotypic partitioning of genetic diversity and variation

The MrBayes analysis was used to explicitly test whether the cpDNA patterns were compatible with the expectations of a monophyletic origin of the tetraploids. The analysis described above was re-run under the constraint that only trees, which are compatible with a monophyletic origin of the diploid and tetraploid cytotypes, were sampled. Bayes factors, measured by twice the difference of the log marginal likelihoods of the two competing models, were used to assess the significance of the difference of the log-likelihoods returned by the constrained and unconstrained analyses. Threshold values of 2, 5 and 10 were taken as positive, strong and very strong evidence for selecting a model over another, respectively (Raftery, 1996).

We then measured the level of genetic diversity and differentiation within and among cytotypes. Allelic richness (expected number of alleles when resampling a constant number of individuals, here three individuals) was computed using the method of El Mousadik & Petit (1996) for samples of unequal size. A t-test was then used to compare the average allelic richness per population in diploids vs. tetraploids. Fst and Nst were employed as a measure of genetic differentiation for the SSR and cpDNA data matrix, respectively. Nst is a measure of genetic differentiation among populations analogous to Fst but taking into account the phylogenetic relationships between alleles (Pons & Petit, 1996). The latter was derived from a matrix of mean character differences among haplotypes as implemented by paup 4.0b10 (Swofford, 2003). Significance of Fst and Nst was tested by constructing the distribution of the null hypothesis by means of 999 random permutations of individuals among populations as implemented by Spagedi 1.3 (Hardy & Vekemans, 2002).

Finally, a discriminant analysis was employed to find the allelic combinations that best allow for the determination of the cytotypes. Discriminant functions were constructed for each of the SSR and cpDNA data sets. To avoid that the frequencies of diploids and tetraploids in the data directly influence the analysis, equal a priori probabilities to belong to one of the two cytotypes were given to each specimen. The discriminant functions were then used to determine the probability that the specimens characterized by sub-equal flowers and attributed to S. obscura, belong to the diploid or tetraploid cytotype. The robustness of the discriminant functions was tested by a cross-validation procedure, wherein the data were divided into a training set and a test set. For that purpose, the 5th, 10th, 15th, etc. lines of the matrix were removed from the training set and included in the test set. The analyses were re-performed on the training set, and the probability to belong to the diploid or tetraploid cytotype was determined for each specimen of the test set based upon the discriminant functions independently derived from the training set.

Geographic and taxonomic partitioning of genetic variation

Genetic variation was partitioned among species and geographic regions (Mediterranean and Atlantic) using Fst and Nst for the SSRs and cpDNA data, respectively. The existence of a phylogeographic signal in the chloroplast data was tested by assessing the significance of the observed difference between Nst and Fst values by means of 999 random permutations of the mean character difference matrix among haplotypes. Indeed, when Nst is significantly larger than Fst, it means that mutations creating new haplotypes occur at a higher rate than the gene flow of haplotypes among subdivisions, generating a phylogeographic pattern. Fis, whose significance was determined by means of 999 permutations of alleles among individuals from the same population, was computed using Spagedi 1.3.

Finally, we investigated patterns of genetic differentiation at the scale of individuals within geographic regions, both within and among species, along gradients of geographic distance. For that purpose, we computed pairwise similarity coefficients between conspecific individuals on the one hand, and between individuals belonging to two different species on the other. For SSR data, we estimated pairwise kinship coefficients between individuals, Fij, using J. Nason’s estimator (Loiselle et al., 1995). For the cpDNA data, Fij was also computed as well as a Fij analogue for ordered alleles, called Nij, taking the phylogenetic relationship among haplotypes into account. More specifically, the estimated parameters Fij and Nij are defined as inline image and inline image, where hij is the probability that two gene copies from individuals i and j carry different alleles (or haplotypes), νij is the phylogenetic distance between the haplotypes carried by individuals i and j (mean character differences among haplotypes), while inline image and inline image are the averages over all pairs of individuals in the sample of hij and νij, respectively. Both Fij and Nij were computed from global allele frequencies within each geographic region. To test for isolation by distance, the significance of the slope of the regression of Fij or Nij on the logarithm of spatial distance between individuals, ln(dij), was tested by means of 999 random permutations of population locations (Mantel test). The mean Fij or Nij values were also computed over i, j pairs separated by predefined geographic distance intervals, d, giving F(d) and N(d). Threshold distance separating intervals were 0, 10, 50, 100, 250, 500 and 1000 km, the first interval corresponding to pairs of individuals from the same population. For cpDNA data, the difference between N(d) and F(d) was tested by means of 999 random permutations of the genetic distance matrix to test the presence of a phylogeographic signal at different spatial scales. All computations were performed using Spagedi 1.3.

Results

Nuclear SSR and cpDNA data

The number of nuclear microsatellite alleles at each locus is given in Table 2. Diploids and tetraploids substantially differ by their degree of heterozygosity at all loci but S10, at which the vast majority of specimens were homozygous for both cytotypes (Table 2).

Table 2.   Number of alleles and proportion (%) of homozygous and heterozygous genotypes at six nuclear microsatellite loci in diploid and tetraploid Salicornia sampled along the Atlantic and Mediterranean coasts of France.
 2n4n
Nb of allelesHomozygous/heterozygousNb of allelesHomozygous/heterozygous
S2398/2410/90
S5498/2335/65
S7582/18546/54
S8491/9551/49
S10588/12588/12
S19593/767/93

The PCR-RFLP analysis of the cpDNA revealed a total of 23 polymorphic bands, allowing for the distinction of 37 haplotypes. Fifteen haplotypes were restricted to diploids, which exhibited a strong biogeographic differentiation because 10 were sampled from the Atlantic (haplotypes F, G, H, I, J, K, L, Q, R, AH) and four from the Mediterranean (D, T, U, V). Sixteen other haplotypes were restricted to tetraploids, with 11 sampled from the Atlantic (E, W, Z, AB, AD, AE, AF, AG, AI, AJ, AL) and four from the Mediterreanean (A, B, C, AC). The six remaining haplotypes were shared by diploids and tetraploids, with five restricted to the Atlantic (M, N, O, P, AK) and one to the Mediterranean (X).

Global genetic structure

The PCA performed on the presence/absence of alleles at the six nuclear microsatellite loci shows that the main differentiation among specimens is among cytotypes (Fig. 2). Along PCA1, which accounts for 18.1% of the total variance, the correlation coefficient between the specimen scores and their cytotypic identity is 0.87 (P < 0.001). PCA2, which accounts for another 7.1% of the total variance, discriminates diploid Mediterranean and Atlantic specimens. PCA1 is positively correlated with allele S5-106 (0.72), which is opposed to alleles S2-124 (−0.85), S5-102 (−0.77), S5-104 (−0.76), S8-128 (−0.71), S19-183 (−0.83). Several alleles are thus diagnostic for cytotypic identity, but none are strictly restricted to a cytotype. For example, allele S2-124 and S19-183 have a frequency of 2 and 1% in diploids, whereas they reach a frequency of 90 and 88% in tetraploids, respectively. Along PCA2, S8-124 (0.52) and S19-175 (0.49) are opposed to S19-177 (−0.46), S10-172 (−0.58) and S8-126 (−0.66).

Figure 2.

 Principal component analysis of diploid and tetraploid Salicornia sampled along the Atlantic and Mediterranean coasts of France and genotyped at six nuclear microsatellite loci.

For the Instruct analyses, the Gelman–Rubin statistics was systematically < 1.10 for each value of K, indicating good chain convergence. In both the diploid and tetraploid datasets, the log-likelihoods increased with the value of K, and the model with the highest value of K set here, i.e. K = 20, was systematically identified as the best-fit model based upon the deviance information criterion.

The network representation of the relationships among the cpDNA haplotypes of the sampled Salicornia is presented in Fig. 3. Five groups of haplotypes, hereafter referred to as groups I–V, can be distinguished. Groups I, IV and V, which have a posterior probability (p.p.) of 0.83, 1.00 and 1.00 in the Bayesian analysis, respectively, include nearly exclusively tetraploids. Group V almost only contains Atlantic tetraploid haplotypes, but AK includes 6% of diploid specimens. Group I mostly groups both Mediterranean and Atlantic tetraploid haplotypes, although some Mediterranean diploids (9% of haplotype X and 58% of haplotype AA) are also included. Groups II and III, which have a p.p. of 0.97, only contain diploids. None of the species were defined by monophyletic chloroplast lineages, and most haplotypes are shared among several species. By contrast, most haplotypes tend to be unique to either the Mediterranean or Atlantic regions, and a strong geographic structure is evident within each of the diploid and tetraploid lineages. In diploids, all the Mediterranean haplotypes are clustered within group II, whereas all the Atlantic haplotypes cluster together within group III.

Figure 3.

 cpDNA haplotypic network from the analysis of PCR-RFLP markers in diploid and tetraploid Salicornia sampled along the Atlantic and Mediterranean coasts of France. Bars along the branches represent single mutational steps. Roman numbers identify clades of haplotypes discussed in the text. Each circle represents a haplotype. Circle size is proportional to the number of specimens included in the haplotype, and colour patterns indicate the species identification and geographic origin of those specimens. Numbers within clades are the posterior probabilities, as assessed by a Bayesian inference implementing a transition model among states for the PCR-RFLP profiles (see text for details).

Genetic diversity and differentiation between diploid and tetraploid cytotypes

The 50% majority-rule consensus from the trees sampled from the posterior probability distribution generated by the MrBayes analysis of the cpDNA dataset included a large polytomy at the basis of the tree. However, constraining the diploid and tetraploid cytotypes to monophyly led to a significant difference in marginal log-likelihood (−1482 and −1542 in the unconstrained and constrained analyses, respectively).

Fst values among cytotypes are all significant at the 0.001 level and are only marginally higher than intra-cytotypic comparisons among biogeographic regions for both the nuclear and cpDNA markers (Table 3).

Table 3.   Genetic differentiation between geographic regions (Atlantic and Mediterranean) and cytotypes (diploid and tetraploid) in Salicornia: pairwise Fst values for the SSRs and Nst for the cpDNA markers are indicated above and below the diagonal, respectively. All Fst and Nst values are significant at the 0.001 level.
 2n Mediterranean2n Atlantic4n Mediterranean4n Atlantic
2n Mediterranean0.260.430.49
2n Atlantic0.890.310.41
4n Mediterranean0.630.700.19
4n Atlantic0.520.630.38

The cytotypic identity of all the specimens was correctly recovered by the discriminant analyses employing the cpDNA data at 96.4% (97% for the diploids and 95.5% for the tetraploids). After cross-validation, the correct classification rate remained at 96%. With the SSR data matrix, the correct classification rate was 99.4% (99.7% for the diploids and 99% for the tetraploids), and these values only slightly dropped after cross-validation (overall correct classification rate of 96.5%, with 98.9% for the diploids and 92% for the tetraploids). The discriminant functions derived from the SSR and cpDNA data were consistent in identifying the same 35% of the specimens attributed to S. obscura as being tetraploids.

Allelic richness for diploid and tetraploid populations is documented in Table S1. Average allelic richness did not significantly differ between diploids and tetraploids, neither in the chloroplast (1.38 vs. 1.34) nor in the nucleus (1.89 vs. 1.78).

Taxonomic and geographic partitioning of genetic variation

Taxonomic partitioning (Table 4) accounted substantially less than geographic partitioning (Table 5) for the observed patterns of genetic variation. The global Fst and Nst resulting from taxonomic partitioning (i.e. among species) of the SSR and cpDNA data, respectively, were 0.010 (P < 0.001) and 0.69 (P < 0.001). By comparison, the Fst and Nst resulting from the geographic partitioning of the SSR and cpDNA data between the Atlantic and Mediterranean reached 0.27 (P < 0.001) and 0.89 (P < 0.001), respectively. The latter was significantly higher than the cpDNA Fst (0.41), indicating the presence of a significant phylogeographic signal.

Table 4.   Pairwise comparisons of Fst (SSR, above the diagonal) and Nst (cpDNA, below the diagonal) among Salicornia patula, S. ramosissima, S. obscura and S. pusilla sampled along the Atlantic and Mediterranean coasts of France. All of the presented Fst and Nst values are significantly different from 0 at the 0.001 probability level.
 S. patulaS. ramosissimaS. obscuraS. pusilla
S. patula0.170.250.30
S. ramosissima0.670.030.05
S. obscura0.880.170.02
S. pusilla0.940.180.25
Table 5.   Geographic (Atlantic vs. Mediterranean) and taxonomic (species assignation) partitioning of genetic variation in diploid Salicornia from the Mediterranean and Atlantic coasts of France. Values above and below the diagonal are Fst derived from the analysis of nuclear SSR data and Nst derived from the analysis of PCR-RFLP patterns at seven cpDNA loci, respectively. All Fst and Nst are significant at the 0.001 level except for the Fst between S. patula and S. ramosissima from the Mediterranean (P < 0.05).
 S. patula, MediterraneanS. ramosissima, AtlanticS. obscura, AtlanticS. pusilla, AtlanticS. ramosissima, Mediterranean
S. patula, Mediterranean0.280.250.300.03
S. ramosissima, Atlantic0.910.030.050.30
S. obscura, Atlantic0.880.110.020.28
S. pusilla, Atlantic0.940.050.250.33
S. ramosissima, Mediterranean< 0.010.880.880.92

Pairwise species differentiation levels substantially droped when performed within the same geographic region (Table 5). For example, while SSR Fst and cpDNA Nst between S. ramosissima and S. patula were 0.17 and 0.67, respectively (Table 4), these values both droped to 0.03 within the Mediterranean. By comparison, SSR Fst and cpDNA Nst within S. ramosissima reached respectively 0.30 and 0.52 between Mediterranean and Atlantic populations.

Similar results were obtained with the tetraploids. Geographic partitioning of the genetic variation included in the SSR and cpDNA data resulted in Fst and Nst values between the Mediterranean and Atlantic coasts of 0.18 (P < 0.001) and 0.37 (P < 0.001), respectively. By comparison, taxonomic partitioning among the three tetraploid species resulted in a global Fst for SSR and Nst for the cpDNA data set of 0.08 (P < 0.001) and 0.20 (P < 0.001). Pairwise comparisons among species and geographic regions (Table 6) similarly showed that the highest values of genetic differentiation were reached at the inter-regional level.

Table 6.   Geographic (Atlantic vs. Mediterranean) and taxonomic (species assignation) partitioning of genetic variation for tetraploid Salicornia from the Mediterranean and Atlantic coasts of France. Values above and below the diagonal are Fst derived from the analysis of nuclear SSR data and Nst derived from the analysis of PCR-RFLP patterns at seven cpDNA loci, respectively. All Fst and Nst are significant at the 0.001 level.
 S. fragilisS. dolichostachyaS. emericii (Mediterranean)S. emericii (Atlantic)
S. fragilis0.050.210.05
S. dolichostachya0.090.240.04
S. emericii (Mediterranean)0.560.60 0.11
S. emericii (Atlantic)< 0.01< 0.010.56

F and N statistics and Mantel tests among individuals within geographic regions

The variation of the average kinship coefficients within and among species along a gradient of geographic distance was very similar in diploids (Figs 4 and 5) and tetraploids (see Figs S1 and S2). Globally, mean SSR Fij among conspecific individuals were only significant at the local scale. Average Fij among pairs of individuals from different species were lower, but significant at that scale. Beyond the population scale, none of the Fij comparisons were significant. For the cpDNA data, all mean Nij values within and among species reached 1 at the local scale and were significantly different from 0, but not significantly higher than the respective mean Fij values. There was a progressive decrease of Nij values with geographic distance; hence, the slope of regression between the Nij and geographic distance was significantly different from 0.

Figure 4.

 Mean Fij resulting from the comparisons of individual SSR genotypes (a) and mean Nij resulting from the comparisons of individual cpDNA haplotypes (b) within and among the diploid Salicornia patula and S. ramosissima from the Mediterranean coasts of France depending on the geographic distance separating them. The P-value of the slope of the regression between the Fij and Nij values and geographic distance are given in the upper right corners.

Figure 5.

 Mean Fij resulting from the comparisons of individual SSR genotypes (a) and mean Nij resulting from the comparisons of individual cpDNA haplotypes (b) within and among the diploid Salicornia ramosissima, S. obscura and S. pusilla from the Atlantic coasts of France depending on the geographic distance separating them. The P-value of the slope of the regression between the Fij and Nij values and geographic distance are given in the upper right corners.

Discussion

Differentiation and reproductive isolation between diploids and tetraploids

The clear and almost complete segregation of diploids and tetraploids based on the PCA of the nuclear SSRs and the cpDNA network suggests the presence of a strong reproductive barrier among cytotypes in Salicornia, in agreement with several observations of complete reproductive barriers among sympatric cytotypes (Hardy et al., 2000; Husband & Sabara, 2003; Kloda et al., 2008). SSR and cpDNA alleles typical for tetraploids were, however, occasionally found in diploids Allele sharing between diploids and tetraploids at SSRs loci may be because of homoplastic mutations or null alleles. The sharing of SSR alleles or cpDNA haplotypes by diploid and tetraploid lineages may also be explained by the retention of ancestral polymorphisms from their diploid ancestors within the tetraploid lineages. Such an interpretation is fully compatible with the very recent origin of the genus, dated to 1.8–1.4 myrs (Kadereit et al., 2006). However, the sharing of alleles among sympatric individuals from different cytotypes, along with the occurrence of specimens with a typical diploid morphology and nuclear SSR patterns but with a cpDNA haplotype characteristic for tetraploids, is rather indicative of instances of inter-cytotypic gene flow. This hypothesis is favoured by Kaligaric et al. (2008) to explain the sharing of cpDNA haplotypes between Mediterranean diploids and tetraploids and in fact, although there are few examples of the process in the wild, the existence of gene flow among cytotypes has long been recognized (see Chapman & Abbott, 2010, for review).

The high correlation between cytotypes and genotypes observed here confirms previous assessments on the reliability of flower morphology to distinguish among diploid and tetraploid Salicornia (e.g. Lahondère, 2004; Kaligaric et al., 2008). Those cpDNA and SSR markers may therefore prove useful for the determination of the ploidy level of herbarium material, juvenile specimens or specimens with an ambiguous flower morphology. This is especially the case of S. obscura, whose lateral and median flowers are subequal in size, rendering the confusion with both diploid and tetraploid species possible (see Lambinon & Vanderpoorten, 2009, for review). In fact, the molecular analysis of the specimens attributed to S. obscura in the present study revealed that about one-third of them were tetraploids. This suggests that S. obscura has served as a convenient taxonomic repository for morphologically ill-characterized diploid or tetraploid specimens.

Origin and evolution of the tetraploids

The tetraploids exhibit high levels of heterozygosity. Heterozygosity is fixed in the vast majority of populations at loci S2 and S19, with > 95% of heterozygous profiles. More than 50% of specimens are heterozygous at loci S5, S7 and S8. These patterns are consistent with previous isozyme analyses, wherein diploids were strictly homozygous and tetraploids showed either a homozygous or a fixed heterozygous profile (Wolff & Jefferies, 1987). Although the chances of producing an autopolyploid from two unreduced gametes fusing are greatly increased in strong selfers like Salicornia (Shepherd & Yan, 2003), the most parsimonious explanation is that the tetraploids are of allopolyploid origin. Indeed, although the lack of fixed heterozygosity may strengthen the evidence for autopolyploidy (Soltis & Rieseberg, 1986), fixed heterozygosity at many loci in a polyploid is commonly used as evidence for allopolyploidy (Arft & Ranker, 1998; Såstad et al., 2001; Nyberg Berglund et al., 2006). In allopolyploids, indeed, parental genomes may be different enough for chromosome pairing to occur only between chromosomes that originate from the same parental genome. Alleles of each parental genome segregate as if they were from a diploid with disomic inheritance. If the parental genomes are homozygous for different alleles, all gametes will be heteroallelic and all offspring will be heterozygous, i.e. heterozygosity will be fixed. The increased heterozygosity resulting from hybridization in tetraploid Salicornia may, ultimately, result in the formation of new gene combinations and generation of new forms of enzymes, and be critical for the successful establishment in unstable environments with recurrent flooding periods where they typically occur.

At the population level, given the strong reproductive barrier among cytotypes, and provided that the diploids did not suffer more substantially than the tetraploids in the course of the last glaciations, the comparable levels of diversity observed in tetraploids and diploids either suggest an ancient and/or multiple origin of polyploids. Indeed, polyploids are expected to harbour less genetic diversity if polyploid formation is a rare and/or recent event. This is because, although allotetraploids potentially accumulate genetic variation at a faster rate than diploids, newly formed polyploids start out with limited genetic diversity because of founding effects. It therefore takes a considerable amount of time to reach equilibrium between mutation and drift, and ultimately higher levels of genetic diversity (Luttikhuizen et al., 2007). An ancient origin of allotetraploid Salicornia would be consistent with the idea that fast mutation rates at the microsatellite loci would have regenerated the loss of diversity following speciation. The only marginally higher differentiation among than within cytotypes with the SSRs, and the actually higher differentiation within diploids than between diploids and tetraploids in the chloroplast, contrast with the ancient origin hypothesis. The latter is further weakened by the recent origin of the genus (Kadereit et al., 2006) and the lack of phylogenetic resolution within cytotypes, which has been interpreted in terms of a recent and rapid expansion (Kadereit et al., 2007; Murakeözy et al., 2007).

Thus, an alternative explanation for the observed patterns of diversity in tetraploids is that the latter evolved recurrently from diploids. Although a monophyletic origin of European tetraploids was resolved from the phylogenetic analysis of nrDNA ETS sequences (Kadereit et al., 2007), which is consistent with the clear differentiation among cytotypes observed here using nuclear SSRs, a monophyletic origin of the cpDNA lineages observed among tetraploids was statistically rejected. This suggests that allopolyploidization has happened several times from a common gene pool, adding to the mounting evidence for a recurrent origin of polyploids (see Albach, 2007, for review).

Geographic and taxonomic partitioning of genetic variation within cytotypes

Although none of the species, as circumscribed by the most complex taxonomic treatments of, e.g. Lahondère (2004) and Stace (2010), are defined by monophyletic chloroplast lineage, partitioning of genetic variation among species is weak, but significant. The strong underlying phylogeographic structure, however, largely contributes to this differentiation. Levels of differentiation between populations of different species within the same region are indeed very low when compared to those among conspecific populations from different regions. A significant phylogeographic signal is in fact present in the cpDNA data between the Mediterranean and the Atlantic, suggesting that these two regions were colonized anciently and accumulated mutations at a faster rate than migration events. These findings are consistent with the eastern/western pattern of differentiation found by Kadereit et al. (2007).

At the local scale, the results further indicate that taxonomy cannot be retrieved from analyses of genetic relationships. In fact, mean cpDNA Nij kinship coefficients are equal to 1 within and among species. Similarly, although conspecific individuals tend to be significantly more related than individuals from different species at the population scale (see below), kinship coefficients derived from SSR variation among individuals from different species are not significantly lower than those within species as soon as individuals from different populations are considered. The poor relation between taxonomy and genetic variation documented here is consistent with previous analyses using AFLPs (Le Goff, 1999) and DNA sequences (Murakeözy et al., 2007; Kadereit et al., 2007), which failed to resolve monophyletic species groups. This is, however, counter-intuitive given the apparently clear circumscription of at least some species, like S. pusilla, which is readily recognized by its solitary flowers.

In Salicornia, phenotypic plasticity has been suggested as the main factor accounting for the incongruence observed between traditional species concepts and patterns of genetic differentiation (Murakeözy et al., 2007). Owing to their extremely reduced morphology, Salicornia species are defined based on global branching architecture and shape; colour; and size of the lateral vs. central flowers, i.e. a suite of quantitative characters that are indeed arguably more prone to plasticity than complex flower characters found in other groups. The hypothesis that plasticity accounts for the incongruence between species concepts and patterns of genetic differentiation is, however, at odds with the fact that mean Fij among pairs of conspecific individuals in the SSR data tend to be significantly higher than those between pairs of individuals from different species at the population scale, thereby suggesting that the morphological differentiation has a genetical basis. Transplantation experiments indeed revealed that individuals tend to retain their specific morphology (Kadereit et al., 2007).

Another interpretation of the incongruence between taxonomy and genetic variation in Salicornia is the lack of inter-specific reproductive barriers. Salicornia species have, however, traditionally been considered as strong if not complete selfers (Dalby, 1962; Ferguson, 1964). A 100% inbreeding rate was, for instance, reported based upon the genetic identity between 38 maternal plants and 2112 F1 progenies (Noble et al., 1992). In diploid species indeed, the anthers usually dehisce before they are exserted and ripe dehiscing anthers may be seen in contact with presumably receptive stigmas, their pollen spilling onto the stigmatic papillae. Cleistogamy has, in addition, been reported in many instances (see Kadereit et al., 2007, for review). Pistillate flowers can, however, be observed in S. ramosisima (Kadereit et al., 2007), suggesting that outbreeding by wind pollination cannot be completely ruled out. In addition, heterozygosity was recurrently observed in diploids, and the mean level of inbreeding observed within diploid populations (Fis = 0.70) indicates an outcrossing rate of 18% if only selfing accounts for the heterozygote deficit. Such an interpretation is consistent with the existence of fertile inter-specific hybrids such as S. × marshallii between S. pusilla and S. ramosissima, which is readily distinguished by a combination of inflorescences with a single flower, as in S. pusilla, two flowers, and three flowers, as in S. ramosissima. The hybrid nature of such specimens with an intermediate morphology was confirmed by use of the present SSR markers because they display heterozygous genotypes at loci with different homozygous genotypes within sympatric S. pusilla and S. ramosissima. The absence of interspecifric reproductive barriers is also suggested by the partial incongruence reported among nuclear and cpDNA sequences (Murakeözy et al., 2007; Kaligaric et al., 2008). Salicornia is comparable, in this respect, to Quercus, wherein species are not well differentiated genetically owing to the sharing of common cpDNA haplotypes among sympatric species, suggesting appreciable localized cytoplasmic gene flow and high levels of cpDNA fixation within populations (Whittemore & Schaal, 1991), whilst haplotypes sampled among allopatric conspecific specimens are highly divergent (Petit et al., 1993; Manos et al., 1999).

A third and not mutually exclusive interpretation is that the observed morphological differences are not the result of divergent genomes, but are based on single or few point mutations, or even to changes in the mechanisms of gene regulation affecting when and where a gene is expressed. In the beach mouse, for example, a single amino acid mutation contributes to adaptive colour pattern (Hoekstra et al., 2006). In other taxa with reduced morphologies like mosses, Hedenäs & Eldenäs (2008) and Sotiaux et al. (2009) similarly evoked the possibility that a single or a few genes may be responsible for dramatic morphological modifications, whereas the remaining of the genome had no time to sort out. This interpretation is in line with Kadereit’s et al. (2007) hypothesis of recurrent evolution of a species such as S. pusilla from S. ramosissima. The observed association between morphology and genetic variation within populations would thus result from the selfing mating system generating substantial linkage within the genome, linkage that would quickly disappear among unrelated individuals from different populations.

Kinship coefficients derived from both the nuclear and cpDNA markers significantly decreased as soon as individuals from different populations were considered, indicating an extremely low mobility of both seeds and pollen. In fact, Salicornia seeds lack specialized devices for dispersal. Fifty per cent of the seeds can be found within a distance of 10 cm from the mother plant, and most are trapped in sediments, algae or marsh vegetation (see Kadereit et al., 2007, for review). Whilst rare events of long-distance dispersal, probably aided by zoochory, might explain some trans-oceanic phylogenetic patterns (Kadereit et al., 2007), this suggests that routine dispersal at the regional scale is extremely limited in the genus.

The existence of strongly inbred, disconnected populations might explain why, although a significant geographic partitioning of genetic variation is evidenced by the F statistics described above, the log-likelihood values of the models implemented by Instruct steadily increased with the value of K. Under this hypothesis, distinctive traits among species of a same cytotype would represent genetic variation within a single gene pool displaying discontinuous variation because of the coexistence of inbred lineages.

Whatever the evolutionary mechanisms behind it, our results thus strongly suggest that the observed range of morphological variation in Salicornia is unparalleled by genetic differentiation. The results presented here do not support the most complex taxonomic treatments of Lahondère (2004) and Stace (2010) but rather fit with the broad species aggregate concept of, e.g. Valdés & Castroviejo (1990) and Piirainen (2001). Given the strong geographic signal in the data, one possibility would be to recognize, within the study area, one Mediterranean and one Atlantic species within each of the diploid and tetraploid lineages. Such species would be ‘cryptic’ in the sense that they would not be characterized morphologically. ‘Cryptic’ species have increasingly been recognized in other organisms with reduced morphologies like bryophytes (see Vanderpoorten et al., 2010b, for review). Recognition of such lineages solely characterized by their geographic range would, however, dramatically increase the number of ‘species’ in taxa with a substantial geographic structure across their distribution range. Definite taxonomic conclusions will, therefore, be presented elsewhere based upon a comparative analysis of all source of information available, including morphology, cytology and various genetic markers collected for a representative numbers of all Salicornia species across their entire distribution range.

Acknowledgments

This study was made possible thanks to grant 2. 4504.05 of the Belgian Funds for Scientific Research (F.R.S. FNRS). The authors are very thankful to C. Lahondère, G. Paradis and B. Toussaint, who contributed to the fieldwork and checked our identifications, and to two reviewers for their constructive comments on the manuscript. We also acknowledge the Cornell Computational Biology Service Unit (CBSU).

Ancillary