Introduction

Most plant species in Europe have been extensively inventoried, described and documented. However, some of them show high intraspecific genetic, phenotypic and/or ecological variation leading to among-population differentiation that might ultimately result in speciation (Coyne and Orr, 2004; Gavin et al., 2014). We may wonder whether one single species might consist of several Evolutionary Significant Units (ESUs; Fraser and Bernatchez, 2001) evolving in relative isolation from each other, or even corresponding to cryptic species, as a result of recent or past evolutionary processes. Identifying these ESUs has implications in terms of sustainable plant diversity conservation. Depending on their distribution (wide or restricted), abundance (common or rare), habitat requirements (more or less specialised) and patterns of genetic diversity and structure, the conservation status of ESUs may differ, and in situ and ex situ conservation strategies should be evaluated for each ESU separately (Palsbøll et al., 2007). Besides, plant communities are facing strong and fast environmental changes owing to human activities (for example, climate change, habitat fragmentation), threatening their survival in the long term (Hautekèete et al., 2015). Recent studies of the effects of contemporary climate warming on plant species distribution and migration have emphasised the need for a better understanding of species response to past climatic changes (Petit et al., 2008).

The study of population genetic structure and phylogeographic patterns across a species range, using a combination of plastid and nuclear molecular markers, allows retracing past and present species spread and identifying past and recent evolutionary processes responsible for the current patterns of genetic variation (Avise, 2000; Gavin et al., 2014). Past climate events, especially the series of major ice ages that occurred during the Quaternary (2.4 million years ago (Mya) to the present), have driven the evolution of divergent genetic lineages by isolating populations in distant refugia, for instance, in the Iberian peninsula, Italy or the Balkans in Europe (Hewitt, 2000). Ultimately, these historical refugia may have promoted allopatric genetic divergence leading to different ESUs or to sibling species that may not be morphologically distinguishable, that is, cryptic species (Hewitt, 2000; Gavin et al., 2014). Subsequently, postglacial migration and species range expansion from these different refugia have allowed divergent lineages to come into secondary contact in suture zones (for example, Taberlet et al., 1998; Hewitt, 2011). Contemporary processes, such as gene flow, genetic drift and spatially heterogeneous selective pressures, in combination with species life-history traits, in particular the breeding system and seed and pollen dispersal capabilities, can also contribute to shape species’ distribution, genetic diversity and population differentiation (Hamrick and Godt, 1996; Duminil et al., 2007).

In this paper, we examine the genetic structure of the perennial herb Silene nutans in Western Europe. In Great Britain, northern France and Belgium, this species exhibits a disjunct distribution, is locally rare and vulnerable and populations show allozyme and/or morphological divergence (for example, Hepper, 1956; Fitter, 1978; Van Rossum et al., 1997, 2003). In Great Britain, two varieties, S. nutans var. salmoniana Hepper and var. smithiana Moss, have been differentiated based on reproductive traits (Hepper, 1951), but this distinction is no longer considered in recent taxonomy (for example, Stace, 2010). In Belgium, two parapatric edaphic ecotypes (calcicolous Ca and silicicolous Si) with contrasting morphological traits and substantial genetic divergence have been described. Isolating reproductive barriers have been identified between them, suggesting incipient speciation processes (De Bilde, 1973; Van Rossum et al., 1996, 1997). Our hypothesis is that populations of S. nutans in Western Europe may correspond to several ESUs or to cryptic biological species.

Using plastid single-nucleotide polymorphisms (SNPs) and nuclear microsatellite markers, we investigated the large-scale phylogeographic and population genetic structure patterns on a comprehensive sampling of populations of S. nutans from Western Europe. Using F-statistics, Bayesian clustering and spatial multivariate methods, we addressed the following questions:

  1. 1

    What are the range-wide levels of genetic diversity and genetic differentiation in S. nutans? Can the patterns be related to historical gene flow, mating system and/or seed and pollen dispersal abilities?

  2. 2

    If spatial genetic analyses reveal genetically differentiated lineages in S. nutans in Western Europe, what is the most plausible evolutionary process involved in the formation of these distinct lineages? We hypothesised that range shifts driven by the last Quaternary glaciations and leading-edge expansions associated with founding events may be the main processes explaining the large-scale patterns of genetic variability we observed.

  3. 3

    Is there evidence for suture zones among genetically divergent lineages, and if so, can we detect hybridisation events mediated by seed and/or pollen flow in the contact zones?

Our results will be discussed in the light of biogeographical scenarios, climate refugia and past historical migration events since the Last Glacial Maxima, leading to divergent ESUs and possible cryptic speciation.

Methods

Studied species and plant material

S. nutans L. (Caryophyllaceae) is a diploid, herbaceous, long-lived perennial rosette-forming plant species occurring in dry habitats, on rock outcrops, sand or shingle, such as xerothermophilous grasslands, open forests and forest edges. Its wide continental distribution range extends from western Europe to central Siberia and South Caucasus (Hepper, 1956; Fitter, 1978). Flowers are protandrous and insect-pollinated, mainly by moths (Jürgens et al., 1996). S. nutans is self-compatible and expected to exhibit a mixed-mating system. Inbreeding depression in selfed progeny has been reported (Dufaÿ et al., 2010).

Leaf material was sampled from a total of 1979 individuals from 111 populations located at the western border of S. nutans geographic range (Supplementary Table S1, Figure 1a). The sampling covered six countries, including 16 populations from the United Kingdom (populations 1–11 and 16 for S. nutans var. smithiana, and populations 12–15 for var. salmoniana; population assignment to varieties was based on fruit measurements; Van Rossum, unpublished data), 50 populations originating from France, 10 populations from Luxemburg, 13 populations from Belgium (for which 5 have been identified as belonging to the Ca ecotype labelled 77, 78, 80, 87 and 88, and 5 to the Si ecotype labelled 79, 81, 82, 86 and 89; population assignment to ecotypes was based on De Bilde, 1973 and Van Rossum et al., 1997), three populations from The Netherlands and 19 populations from Germany. Leaf samples were dried and conserved in silica gel until DNA extraction. Sample sizes ranged from 2 to 133, with a mean of 17.8 individuals per population (±14.8).

Figure 1
figure 1

Map of the studied populations of Silene nutans in Western Europe. (a) Geographical location and population numbers. (b) Geographical distribution and within-population proportion of plastid haplotypes based on six SNPs. (c) Map of mean population membership probabilities of belonging to the two modal clusters based on nuclear microsatellite loci. (d) Assignment probabilities of individual membership into K=2 and K=6 clusters for Western lineage. Each individual is represented by a thin line partitioned into K coloured segments displaying the individual’s estimated membership fractions in K clusters. (e) Neighbour-joining tree based on Cavalli-Sforza and Edwards’ distance (1967), using nuclear data; only bootstraps >50% are indicated. Coloured circles next to population codes referred to plastid haplotype found in populations. (f) Assignment probabilities of individual membership into K=2 clusters for Eastern lineage. Each individual is represented by a thin line partitioned into two coloured segments displaying the individual’s estimated membership fractions in two clusters. The inset box in panels (a), (b) and (c) zoom on a densely sampled area in greater detail.

Full size image

Molecular analyses

All sampled individuals were genotyped using both plastid and nuclear DNA markers. DNA was extracted from 15 to 20 mg of leaf tissue using Macherey-Nagel (Düren, Germany) NucleoSpin 96 Plant II Kits following the standard protocol outlined in the manufacturer’s handbook.

Multilocus nuclear DNA genotypes were characterised at 13 recently isolated microsatellite loci named B09, E08, G01, H07, D10, Sil16, Sil19, Sil24, Sil31, Sil35, Sil36, Sil37 and Sil42 and described in Godé et al. (2014). Amplification procedures, multiplexing and genotyping were carried out following the standard protocols described in Godé et al. (2014).

Plastid diversity was investigated using SNPs recently developed from plastid sequences (pDNA) of S. nutans representative samples (Lahiani et al., 2013). Plastid SNPs were defined based on six polymorphic pDNA nucleotides in the intergenic spacer sequences psbA-trnH, named Cp 42, and the matK gene fragment, named Cp 397, Cp 540, Cp 656, Cp 730 and Cp 804. The Kompetitive Allele Specific PCR (KASP) genotyping assays was used to detect the SNPs. This method is based on competitive allele-specific PCR and enables bi-allelic scoring of SNPs. The SNP-specific KASP assay mix (designed by LGC group, www.lgcgroup.com) and the universal KASP Master mix are added to DNA samples, and a thermal cycling reaction is then performed, followed by an end-point fluorescent read. Bi-allelic discrimination is achieved through the competitive binding of two allele-specific forward primers, each with a unique tail sequence that corresponds with two universal FRET (fluorescence resonant energy transfer) cassettes; one labelled with FAM dye and the other with HEX dye. The sequences flanking the SNPs that were used to define the primers are given in Supplementary Table S2.

Amplification reactions were carried out by PCR with 30 ng of DNA in 8.11 μl reaction volume containing 4 μl of universal KASP Master mix (2 × ) and 0.11 μl of the SNP-specific KASP Assay mix (composed of the two forward primers and the reverse primer). Cycling conditions for PCR amplification were 15 min at 94 °C, 10 cycles of 20 s at 94 °C and 60 s at 65 °C with a step-down of 0.8 °C per cycle, 26 cycles of 20 s at 94 °C and 60 s at 57 °C and finally 12 °C for 10 min. Fluorescence was detected using a LightCycler 480 (Roche Diagnostics, Basel, Switzerland). Individual haplotypes were defined as a combination of allelic states for all six SNPs.

Nuclear genetic variation within populations

Prior to further analyses, genotypic linkage disequilibrium among all pairwise locus combinations was checked with a log-likelihood ratio test, implemented in FSTAT version 2.9.3.2. (Goudet, 1995), using 10 000 randomisations of two-locus genotypes and Bonferroni correction. The following measures of genetic variation were calculated within and over all populations: allelic richness (Ar, El Mousadik and Petit, 1996) based on a minimum sample size of eight individuals (six populations with sample size below this threshold were not included, see Supplementary Table S1), observed heterozygosity (HO), and expected heterozygosity (HE) corrected for sample size. Some populations failed to be genotyped for the B09 locus (populations 59, 102 and 103) and the D10 locus (population 21); allelic richness was thus only estimated for the 11 remaining loci. Using GENEPOP version 4.3 (Rousset, 2008), deviations from Hardy–Weinberg equilibrium were also tested by estimating the intra-population fixation index (FIS) following Weir and Cockerham (1984). Significance of FIS estimates was tested using exact probability tests across loci and populations. Markov chain method provided unbiased estimates of the Fisher’s exact test probability using the following parameters: 100 000 dememorisations, 1000 batches, and 100000 iterations per batch. To obtain corrected FIS estimates (FIS_corr) accounting for null allele occurrence, we used the Bayesian individual inbreeding model described in Chybicki and Burczyk (2009) and implemented in the software INEST version 2.0 available at http://genetyka.ukw.edu.pl. Using SPAGeDi V1.5 (Hardy and Vekemans, 2002), indirect and independent estimates of selfing rates (s) were derived from the multilocus correlation structure of population samples, following a multilocus estimate of the standardised identity disequilibrium and described in David et al. (2007). To examine geographic trends in the distribution of large-scale genetic diversity, we tested the relationship between allelic richness, longitude and latitude using a linear multiple regression model. Spatially interpolated values of allelic richness were then generated for a better visualisation using the thin plate spline method implemented in R version 3.0.2 (R Core Team, 2014) and represented on a geographical heat map.

Lineage distinctiveness and spatial patterns of genetic structure

Three complementary approaches were used on nuclear microsatellite data to investigate the genetic affinities among populations. First, genetic distance-based population clustering was performed to infer the genetic relationships among S. nutans populations using an unrooted neighbour-joining tree. It was constructed using Cavalli-Sforza and Edwards (1967) genetic chord distance (DCE) based on allele frequencies. Bootstrapped confidence values on obtained branches were determined using random resampling replications over loci (1000 replicates) implemented in the POPULATIONS software 1.2.32.

Second, a non-spatially explicit Bayesian clustering was carried out to infer the number of distinct gene pools in the whole data set, using the model-based Bayesian algorithm implemented in the STRUCTURE software (Pritchard et al., 2000). The number of potential K clusters was assessed from 10 different runs of K ranging from 1 to 80. To ascertain adequate convergence of the Markov Chain Monte Carlo (MCMC) model, we allowed a burn-in of 100 000 iterations, followed by 2 × 106 MCMC replications without any prior geographic information on the putative affiliation of individuals. To identify the most likely number of K clusters, the ad hoc statistic Δ K was calculated as described in Evanno et al. (2005). CLUMPP version 1.1.2. (Jakobsson and Rosenberg, 2007) accounted for label switching and was subsequently used to find the optimal alignment of independent runs by averaging the top runs. The resulting most likely grouping of populations was plotted using DISTRUCT version 1.1. (Rosenberg, 2004). This method was applied (i) to the whole data set to identify potential lineages, (ii) to each identified lineage separately to investigate more subtle lineage-specific genetic structure, and (iii) to populations from areas where different haplotypes (potential distinct lineages) co-occurred (potential contact zones), to detect potential admixture events. This kind of model-based Bayesian algorithm infers subtle population structure by minimising linkage disequilibrium and departures from Hardy–Weinberg equilibrium within each inferred cluster (Pritchard et al., 2000). However, the assumption of panmixia does not necessarily hold in S. nutans, which can exhibit a mixed-mating system in natural populations (Dufaÿ et al., 2010).

Third, to avoid this potential bias, we performed a multivariate ordination analysis that does not require any genetic assumptions: a spatial principal component analysis (sPCA, reviewed in Jombart et al., 2009). We calculated independent synthetic variables that maximise the product of the genetic variance among populations and their spatial autocorrelation (based on Moran’s I). The computation of was Moran’s I carried out based on a Delaunay graph. Genetically distinguishable groups, clines in allele frequency and intermediate states can lead to what is called ‘global structure’ and are identified by positive components, that is, showing positive spatial autocorrelations (for example, DiLeo et al., 2010). sPCA and tests for statistical significance of global structure were performed using the R ADEGENET package (Jombart, 2008). The main results of sPCA can be depicted on maps of population scores, allowing a visual assessment of spatial genetic structuring (Jombart et al., 2008). Using the same strategy as that used for Bayesian clustering, we constructed maps for (i) the whole data set to investigate potential genetically distinct lineages, and (ii) each inferred lineage separately to identify specific evolutionary and phylogeographic processes.

Genetic differentiation among nuclear–plastid lineages

Between-population genetic differentiation was investigated for nuclear data (i) over all populations for each locus and at the multilocus level, (ii) within the identified nuclear–plastid Western and Eastern lineages and (iii) between populations (over all loci) using the ANOVA procedure by Weir and Cockerham (1984). Pairwise FST values and their significance (G test) were calculated using GENEPOP. The partitioning of nuclear genetic diversity within and among genetically distinct lineages inferred from the spatial genetic structure analyses was estimated following the procedure proposed by Yang (1998) and implemented in the R package HIERFSTAT (Goudet, 2005). The among-plastid haplotype component is given by FH/T, and its significance was tested by permuting individuals of each population across plastid haplotypes. The between-nuclear–plastid Western-Eastern lineage component is given by FL/T and its significance was tested by permuting individuals of each haplotype across nuclear–plastid lineages. Genetic differentiation among each pairwise comparison of plastid haplotypes was also performed. Variance in microsatellite allele size due to stepwise changes of repeat number can also account for the level of genetic differentiation and may give insights on the relative contribution of mutation versus migration rates in genetic divergence. Using SPAGeDi, we thus also calculated the microsatellite allele size-based estimates of genetic differentiation RST (Slatkin, 1995) and tested them for significance by allele-size permutations following Hardy et al. (2003). Finally, we estimated the proportion of specific alleles of each identified lineage for nuclear microsatellite loci.

Spatial genetic structure within nuclear–plastid lineages

To detect more subtle cryptic genetic structure, we reanalysed the nuclear data set by carrying out Bayesian clustering and sPCA separately for the Eastern and Western lineages. We also tested for isolation by distance by performing a spatial autocorrelation analysis on (i) the whole plastid data set and (ii) for the nuclear data within the two identified nuclear–plastid lineages. For plastid data, we used classic one-dimensional spatial correlograms based on Moran’s I. Random permutations of population locations were carried out to test the statistical significance of each Moran’s I. For nuclear data, one-dimensional Mantel correlograms were designed according to Oden and Sokal (1986). The normalised Mantel statistic rz (Smouse et al., 1986) was used as described in Oden and Sokal (1986) to estimate the relationships between pairwise genetic distances (DCE distance) and pairwise geographical distances among populations. All calculations were carried out using PASSAGE version 2 (Rosenberg and Anderson, 2011).

Results

Nuclear and plastid genetic variation within populations

Exact tests for linkage disequilibrium between microsatellite loci only showed 38 significant P-values out of 7690 comparisons (0.49%, 384 expected from type 1 error at α=0.05). When analysed over all populations, 11 out of the 13 loci showed significant deviation from Hardy–Weinberg expectations, with mean single-locus FIS values ranging from −0.007 to 0.284 (Table 1). The mean multilocus FIS value was 0.146. At the population level, FIS values ranged from −0.230 to 0.513, with 89 out of the 111 populations (80.1%) showing significantly positive multilocus FIS values (Supplementary Table S1). Corrected within-population multilocus FIS estimates (FIS_corr) accounting for null allele occurrence always yielded lower values with an arithmetic mean of 0.049. Using the classical relationship (s=2FIS/(1+FIS)) expected at genetic equilibrium for a mixed-mating system, estimated mean selfing rates were of 0.254 for FIS and of 0.093 for FIS_corr. Estimations of selfing rates based on standardised identity disequilibrium ranged from 0 to 0.379, with an arithmetic mean of 0.060 (Supplementary Table S1). It should be kept in mind that selfing rate estimations should be interpreted with caution. Indeed, populations with small sample sizes lead to large variance in s estimates and an upward bias is expected when propensity for selfing is low.

Table 1 Estimates of nuclear genetic variation for 13 microsatellite loci over 111 populations of Silene nutans from Western Europe: total number of alleles (An), the allelic richness (Ar), the observed heterozygosity (HO), expected heterozygosity (HE) and the mean fixation indices (FIS, FIT, FST, following Weir and Cockerham, 1984)
Full size table

Nuclear microsatellite loci displayed high levels of polymorphism, with the number of alleles ranging from 26 (Sil19) to 86 (Sil24), for an average number of 47.9 alleles over all populations (Table 1). Mean HO and HE ranged from 0.437 to 0.779 and from 0.512 to 0.854, respectively (Table 1). The map interpolating levels of allelic richness (Figure 2) and the multiple regression analysis (adjusted R2=0.422, F(2,102)=38.9, P<0.001, Variance Inflation Factor=1.003; Pearson's correlation coefficient between latitude and longitude r=0.054) indicated a significant decrease in allelic richness along a southern–northern (β=−0.150, P<0.001) and an eastern–western gradient (β=0.073, P<0.001).

Figure 2
figure 2

Map of the observed level of allelic richness (Ar) found in each population of Silene nutans (displayed as proportional circles). Spatially interpolated levels of allelic richness can also be visualised as colour plot gradient (from dark blue for low values to dark red for high values).

Full size image

For plastid haplotype genetic diversity, we found a total of six different haplotypes, of which four corresponded to 99.5% of the overall plastid diversity observed (Figure 1b, Supplementary Table S1): the TACTCT SNP combination (hereafter called the blue haplotype), TCCTCG SNP combination (orange haplotype), GCCTTG SNP combination (red haplotype), and TCTGCG SNP combination (yellow haplotype). All populations were monomorphic, except seven populations that showed intra-population plastid polymorphism, with the second haplotype occurring at very low frequency (Supplementary Table S1). The Belgian Ca ecotype and the British var. smithiana showed the blue haplotype, whereas the Si ecotype and the British var. salmoniana showed the orange haplotype (Supplementary Figures S1b and c, Supplementary Table S1).

Lineage distinctiveness and spatial patterns of genetic structure

The four main plastid haplotypes showed geographic structuring (Figure 1b, Supplementary Figure S2) and several geographical areas in which two haplotypes co-occurred (Figure 1b, Supplementary Figures S1b and d). Given the lack of intra-population polymorphism and the strong spatial structure observed, these plastid haplotypes were used as a diagnostic criterion for population lineage identity. Based on nuclear microsatellite data, and taking plastid identity into account, a clear phylogeographic pattern and genetically differentiated nuclear–plastid lineages emerged from (i) the neighbour-joining tree based on allele frequencies using DCE (Figure 1e); (ii) the non-spatially explicit Bayesian clustering method (Figures 1c–f); and (iii) the sPCA (Figure 3). Populations clearly clustered according to their plastid haplotype identity, with a Western lineage containing the red, orange and yellow plastid haplotypes, and an Eastern lineage characterised by the blue plastid haplotype. The highest likelihood found using the ΔK statistic corroborated the occurrence of the two main nuclear–plastid lineages (Supplementary Figure S3). The sPCA performed on the whole nuclear data set revealed significant global structure (P<0.001). The first two sPCA axes accounted for most of the spatial genetic structure, covering a high proportion of the spatial autocorrelation and most of the variance in the genetic data (I=0.631, var=0.851 and I=0.610, var=0.223, respectively). The first two eigenvalues were 0.537 and 0.136, while the remaining eigenvalues were <0.090. The first axis in the sPCA revealed a striking longitudinal cline, differentiating a Western and an Eastern nuclear–plastid lineage (Figure 3).

Figure 3
figure 3

Geographic map of the first global scores of an sPCA performed on the whole data set in Silene nutans (I=0.631, var=0.851, P<0.001). Each square represents a population. Populations that are more closely related in multivariate space share the same colour of squares. Sizes of squares indicate the magnitude of spatial positive autocorrelation.

Full size image

Genetic differentiation among nuclear–plastid lineages

The overall and pairwise genetic differentiation for nuclear data were high and highly significant, respectively, with single-locus FST estimates ranging from 0.177 to 0.369, and a mean multilocus FST value of 0.247 (all significant at P<0.001, see Table 1). Pairwise FST estimates between the 111 studied populations ranged from 0 to 0.608, with 52.8% of significant estimates (Supplementary Figure S4).

The overall genetic distinctiveness among plastid haplotypes was supported by a significant hierarchical nuclear genetic differentiation (FH/T=0.143, P<0.001). The highest levels of nuclear genetic differentiation among haplotypes were found between blue and orange haplotypes (FBO=0.157, P<0.001) and between blue and red haplotypes (FBR=0.146, P<0.001). The differentiation was weaker but still significant between orange and red haplotypes (FOR=0.078, P<0.010). We found no significant nuclear genetic differentiation between the yellow and any other haplotype, likely due to small sample size (the yellow haplotype occurred in only eight populations). Nuclear genetic differentiation between the Eastern and Western lineages was highly significant (FL/T=0.126, P<0.001). Mean genetic differentiation (as measured by multilocus FST estimates) within these two main lineages was 0.187 and 0.195 for Eastern and Western lineages, respectively (all at P<0.001). Mean multilocus allele size-based estimate of genetic differentiation RST were (i) 0.357 (P<0.01) among the 111 sampled populations, (ii) 0.173 (P>0.05) between the Western and Eastern lineages, (iii) 0.014 (P>0.05) between the two main genetic clusters within the Eastern lineage, and (iv) 0.274 (P<0.01) between the two main genetic clusters within the Western lineage. Over all loci, only 43% of the alleles were shared between the Eastern and Western nuclear–plastid lineages (Table 1).

Spatial genetic structure within nuclear–plastid lineages

When focussing on the Western lineage, the best mode suggested by the Bayesian clustering was K=2 (Supplementary Figure S3), mirroring the nuclear genetic divergence between the red and the yellow plastid haplotype populations and the orange haplotype populations (Figure 1d). Additional hierarchical structure was detected at K=6 (Supplementary Figure S3): red and yellow haplotype populations were distinct and the orange haplotype was partitioned into four further population clusters, corresponding to four geographical regions: southern Belgium (Si ecotype), northern France, western France together with southern England (var. salmoniana), and south-central France (Figure 1d, Supplementary Table S1). sPCA revealed a latitudinal (north–south) cline regardless of haplotype identity (Figure 4a). This global structure (P=0.001) showed a strong signal of positive spatial autocorrelation (I=0.694) and represented a high proportion of the genetic variability (var=0.471). The Mantel correlogram displayed a clear continuous decrease in genetic similarity with increasing geographic distance (Figure 4a).

Figure 4
figure 4

Spatial genetic structure within the two main Western and Eastern nuclear-plastid lineages in Silene nutans. One dimensional Mantel correlograms of (a) Western and (b) Eastern lineage. To reduce the random effects of sampling and ensure statistical supports for calculations, correlograms were designed with a minimum number of, respectively, 99 and 102 pairs of populations in each distance class. rz: normalised Mantel statistics. For each correlogram plot, the geographical map of the first global scores of sPCAs is also given. Each square represents a population. Populations that are more closely related in multivariate space share the same colour of squares. Sizes of squares indicate the magnitude of spatial positive autocorrelation.

Full size image

When analysing the Eastern nuclear–plastid lineage, a partition between (i) inland continental populations (Germany, Luxemburg, eastern France and the Ca ecotype from southern Belgium) and (ii) populations located near the Channel coast (northern France, Belgium and The Netherlands) and in the UK (var. smithiana) was underlined by both Bayesian clustering with the highest likelihood for K=2 (Figure 1f, Supplementary Figure S3) and the plot of scores from the first global component of the sPCA (Figure 4b). The global structure (P=0.034) showed a strong signal of positive spatial autocorrelation (I=0.784). The Mantel correlogram pictured a typical pattern of long-distance differentiation owing to a substantial geographic break in allele frequencies across the range of the Eastern lineage (Figure 4b).

Detection of hybridisation events in contact zones

Although most populations were fixed for a single plastid haplotype, we tested for the possibility of hybridisation events mediated by pollen in contact zones. We focussed on three areas where populations of different haplotypes co-occurred and ran further Bayesian assignments on these subsets of populations (see Supplementary Figure S1). Admixture was absent: groups identified by Bayesian clustering always referred clearly to haplotype identity with no signature of admixed nuclear genotypes or evidence of discordance between nuclear membership probability and plastid haplotype identity (Supplementary Figure S1).

Discussion

Broad patterns of nuclear and plastid genetic diversity

S. nutans has been described as a self-compatible, gynomonoecious–gynodioecious species with a censer mechanism of seed dispersal (Hepper, 1956). Hermaphroditic plants are able to self by geitonogamy, with an extreme variation in selfing rates among individuals (Dufaÿ et al., 2010). FIS values were positive and significant for several populations and most loci, suggesting propensity for selfing even when accounting for null allele occurrence. Assuming that populations are at genetic equilibrium, the classic relationship between the intra-population fixation index and the selfing rate (s=2FIS/(1+FIS)) gave a mean selfing rate of 25.4% and 9.3% for FIS and FIS_corr, respectively. However, positive significant FIS values may also result from biparental inbreeding owing to restricted pollen and/or seed dispersal. As a consequence, mean within-population selfing rate based on identity disequilibrium and less sensitive to biparental inbreeding (David et al., 2007) gave a much lower rate of 6%. Overall, biparental inbreeding due to mating between relatives, along with a low rate of self-fertilisation, is expected to amplify the potential for isolation by distance pattern developing kin-structure at a local scale (Heywood, 1991; Vekemans and Hardy, 2004).

Inbreeding, along with restricted gene flow, leads to high genetic differentiation among populations. The mean level of genetic differentiation within Western (FST=0.195) and Eastern (FST=0.187) lineages of S. nutans was concordant with a mixed-mating system and moderate levels of gene flow among populations (Duminil et al., 2009). The nearly complete fixation of plastid haplotypes and their strong geographical clustering may either confirm a spatially restricted seed dispersal and/or a lack of within-lineages plastid diversity with density-dependent processes or ecological factors that constraint lineages’ distribution (Waters et al., 2013). Moreover, a continuous decrease in genetic similarity with increasing geographical distance was shown, which suggested limited pollen/seed dispersal. Therefore, to conserve substantial levels of genetic diversity, in situ as well as ex situ, the preserved populations should reflect the nuclear–plastid differentiation patterns. Attention should particularly be paid to the southern populations, which have conserved the highest allelic richness.

Phylogeographic patterns in S. nutans

At a continental scale, leading-edge expansions following the Last Glacial Maximum may result in multiple phylogeographic patterns that are species specific as each taxon responded independently to Quaternary cold periods (Taberlet et al., 1998). Rapid postglacial recolonisation through repeated population bottlenecks of founding lineages, blocking establishment of latecomers, is expected to reduce within-lineage genetic diversity with increasing distance from its refugium (Waters et al., 2013).

In Western Europe, S. nutans showed a phylogeographic pattern with two main, strongly differentiated Western and Eastern nuclear–plastid lineages (FL/T=0.126). Patterns of long-distance differentiation and spatial trends in allelic richness suggested a Western lineage originating from the western Mediterranean Basin and an Eastern lineage, probably originating from Eastern Europe. The Eastern lineage exhibited a north-westward expansion, from eastern Germany to northern Wales with a genetic break found along the continental European coastline. The Western lineage harboured three plastid haplotypes, spatially structured, consistent with a northward expansion from distinct potential refugia. These hypotheses are supported by the asymmetrical patterns of private alleles between the two main nuclear–plastid lineages. Most private alleles were found in the Western lineage. Moreover, allele size-based measure of genetic differentiation yielded a highly significant RST estimate of 0.274 between the two main genetic clusters found in this Western lineage. Overall, this could suggest (i) a predominant role of stepwise-like mutation processes over migration and relative cluster isolation, and (ii) different glacial refugia composed of distinct genetic pools. We should keep in mind that our microsatellite loci may not follow a strict stepwise mutation process and that homoplasy is likely to occur, which may explain why no significant RST estimate was detected between Western and Eastern lineages.

Secondary contacts: from diverging lineages to cryptic species

Beyond shaping the distribution of present genetic diversity, Quaternary climate oscillations have been shown to favour speciation processes (Hewitt, 2000). Genetic drift and selection operating in different, isolated glacial refugia can lead to genetic divergence. Patterns of postglacial expansion from different refugia found in animals, plants, fungal taxa and described in S. nutans here produce suture zones in which lineages come into secondary contact (Taberlet et al., 1998; Hewitt, 2000, 2011). Many of these suture zones have been described across Europe, such as in the Alps, the Pyrenees or in eastern Central Europe along the eastern borders of Germany (Hewitt, 2000). Accordingly, our results in Western Europe identified three suture zones where the Eastern and Western nuclear–plastid lineages of S. nutans co-occur: in southern Belgium, eastern France, and surprisingly, in southern England. Although previous phylogeographic studies have reported that only one lineage colonised Great Britain (for example, oaks expanding from Spain or grasshoppers from the Balkans, reviewed in Hewitt, 2011), our study on S. nutans is, to our knowledge, the first that shows the existence of a contact zone in Great Britain shaped by postglacial expansion.

Our study shows that the two morphological varieties previously described in Hepper (1951) in Great Britain, S. nutans var. salmoniana and var. smithiana, belong to separate nuclear–plastid lineages, Western and Eastern, respectively. Morphological differences, for example, differences in fruit size, have been also found elsewhere, in particular between blue and orange/red haplotype populations, whereas the continental populations belonging to the yellow and blue haplotypes did not show such contrasting differences (Van Rossum, unpublished data). Moreover, the two reproductively isolated Belgian Ca and Si parapatric edaphic ecotypes also correspond to the Eastern and Western nuclear–plastid lineages, respectively. Bayesian clustering performed on the three contact zones identified in S. nutans did not reveal any signatures of admixture between haplotype lineages, suggesting that no hybridisation events occur in natural populations. Such result was somewhat unexpected given that range shifts and expansions of a colonising species into the range of a local species may lead to haplotype sharing owing to longer pollen than seed dispersal (for example, Petit and Excoffier, 2009). Altogether, this strict association between nuclear gene pools and plastid haplotype suggests cryptic speciation in S. nutans, at least for Eastern and Western nuclear–plastid lineages for which a strong reproductive barrier can be suspected. From an evolutionary point of view, suture zones resulting from endogenous selection are not stabilised geographically and their dynamics are thought to be maintained by a balance between dispersal and selection against hybrids (Barton, 1979). These tension zones may be associated with an environmental transition where different genotypic combinations allow differential adaptive values, resulting in a genetic–environment association and ecotypic differentiation (Barton and Hewitt, 1985; Hewitt, 1988; Bierne et al., 2011). Although in southern Belgium the genetic lineages are clearly associated with edaphic properties (for example, Van Rossum et al., 1997), to date, no evidence of such association between genetic variation and soil characteristics have been found elsewhere (Van Rossum et al., 2003; Van Rossum and Prentice, 2004). This inconsistency suggests that the divergence of the two main lineages precedes the specific adaptation to the nature of soil and indicates locally restricted association between edaphic adaptation and lineage identity.

Conclusion

Based on our findings on genetic structure in S. nutans, two strongly differentiated evolutionary lineages were identified with no hybridisation events detected using nuclear markers. Besides, no plastid haplotypes were shared between these two lineages. This suggests the occurrence of at least two cryptic species within S. nutans, with no introgression events affecting species integrity. According to Mayr’s biological species concept (Coyne and Orr, 2004), two main ESUs can therefore be defined in Western Europe, corresponding to Eastern and Western nuclear–plastid lineages. Moreover, within the Western lineage, three potential ESUs might also be considered, corresponding to the three sub-lineages. Given the differences in their distribution range, the conservation status of these distinct ESUs should be evaluated separately. In situ preservation of populations and genetic rescue implying ex situ conservation techniques should take the lineage identity into account. Applying such conservation approaches are particularly true in Great Britain, northern France and Belgium, where S. nutans is rare and where distinct lineages co-occur in close contact.

Data archiving

  1. 1

    DNA sequences: Genbank accession numbers KJ671551–KJ671562.

  2. 2

    Coordinates of sample locations and nuclear genetic data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.vq4j4.