Population structure is dominated by passive dispersal
Like in most flightless freshwater taxa, dispersal of R. balthica between unconnected habitat patches depends on passive dispersal mechanisms . In particular lentic habitats are ephemeral on an intermediate time-scale, thus selecting on populations with good dispersal capacities . In R. balthica, this passive dispersal mechanism is presumably transportation by water fowl . The minimum spanning tree (Figure 3) adds credibility to this assumption, as it clusters the respectively most similar populations mainly along the major bird migration route of the East Atlantic flyway in Southwest-Northeast direction. The suggested connection pattern of the minimum spanning tree beard a striking resemblance to the inferred initial postglacial recolonisation dispersal pattern, where also bird migration routes were implicated . This suggests that it either presents the remnant of this saltatory postglacial colonisation process or that recurrent dispersal follows the same routes. The connection lines of the respectively most similar populations appeared to be distance independent (Figure 2). This was also reflected in the spatial distribution of the inferred genotype clusters (Figure 2), where similar genotypes could be found hundreds of kilometres apart and/or in close proximity. Both findings are substantiated by the complete lack of correlation between population differentiation and geographic distance (Figure 4). Thus, distance independent passive transport seemed to be the primary process for gene-flow and/or colonisation of empty habitats along the Southwest-Northeast axis from virtually any part of the environmental gradient to any other. As a consequence of this unpredictable long range dispersal, colonisers originating from one part of the range must cope with very different environmental conditions upon arrival, arguing for a high phenotypic plasticity leading to the observed broad ecological tolerance.
Despite the possibility for virtually unrestricted long range dispersal, only few populations were found to show admixture; most sites harboured primarily individuals that clearly belonged to the same inferred genotype cluster (Figure 2). This matched the observation that despite the large overall number of alleles per locus (15.6) and haplotypes (132), at single sites, only a very limited number of haplotypes was found (3.2 +/- 1.3 alleles per locus and 4.2 +/- 2.1 haplotypes, respectively). Such a pattern is compatible with a scenario of site colonisation by one or few individuals, followed by a rapid increase of the population size, supported by the inferred mixed mating system in R. balthica. The widespread presence of selfing supplements thus the finding of a preferentially outcrossing system in a local flood-plain system of the Rhône river by Evanno et al. .
Indeed, the mating system, in particular the ability to reproduce uniparentally has long been considered to influence colonisation success . Selfing, like any form of uniparental reproduction, has the automatic advantage of increased gene-transmission to the next generation (no cost of sex), thought to be balanced by the costs of inbreeding depression . Selfing can evolve as reproductive assurance strategy in the absence of mating partners, because it is always better to self-fertilise offspring whose fitness may suffer from inbreeding than to leave no offspring at all . Predominant selfing as mating system should therefore evolve mostly in cases where mating partners are rare or absent , which is in particular the case for the first colonisers of a previously empty habitat. Even Darwin  suggested that selfing or monoecious plants should expand their ranges more easily because already a single individual can found a reproducing population. Indeed, the average proportion of selfing was slightly increased in the recent expansion area (Figure 5C). Such populations, made up of selfing and/or inbred individuals, would be relatively inert against the effect of subsequent gene-flow, as the establishment probability of immigrating alleles in a demographically large population is low . Another, not mutually exclusive explanation for the observed pattern would be short population persistence times, not allowing to accumulate genetic variation by gene-flow or mutation over time. Other studies on freshwater snails have shown that high population turn-over and large size fluctuations are indeed typical for this taxon in general [39–42] and for R. balthica in particular . The bottleneck analysis with the non-selfing populations indicates that the population dynamics of the species is indeed high and not restricted to certain parts of the species range. Nine out of 34 populations (26%) tested showed signs of a population bottleneck within the last few generations (see Additional file 2 Fig. A4).
The observed pattern could also point to a low incidence of successful dispersal events, resulting in low gene-flow rates. This is, however, difficult to evaluate, because direct estimates of passive dispersal rates are not available for freshwater snails.
Current climate change left its mark in the distribution of genetic variability
The influence of the various predictors on all measures of genetic variability was remarkably similar in terms of direction of deviation from the overall mean (Figure 5). This confirmed that both nuclear and mitochondrial markers were subject to similar demographic forces, as might be expected in simultaneous hermaphroditic animals where e.g. sex biased dispersal or sex ratio bias are by definition impossible. The effects on the number of rarefied microsatellite alleles per locus A and expected heterozygosity H
were so similar (correlation coefficient r = 0.85) that we will discuss them together hereafter (Figure 5). Even though selfing proved to be a substantial issue in R. balthica, the mating system population differences had a surprisingly low effect on the distribution of genetic variability, as shown by the low correlation between the degree of selfing and genetic variability measures H
and A (r = 0.30, p = 0.006 and r = 0.22, p = 0.045, respectively). This means that high selfing rates are not predominantly responsible for the loss of genetic variability. A low correlation further allowed investigating whether the factors considered influenced the mating system. The predominant factor in all models with substantial support was the recent, climate driven range expansion (exp), which lowered the level of variability for all genetic markers considerably (Figure 5). This is not surprising, as an ongoing or recent expansion represents a non-equilibrium situation caused by repeated bottlenecks and founder events both of which decrease genetic variability [44, 45]. In R. balthica, this effect might be enhanced by the possibility of self-fertilisation , which facilitates the colonisation of newly emerging habitats by one or few individuals [47–49]. The factor exp was part of the best model to explain the distribution of selfing, showing that this trait may have played a role in the swift colonisation of newly emerging habitat in the course of a climate change. 'However, given enough time, one may predict that the effects of this non-recurrent, historic event at the current range limit will be transient and eventually assume a level of genetic variation either by immigration or mutation comparable to the remaining distribution area. In the past, this has obviously been the case for the expansion from the Pleistocene refugia into the Holocene expansion areas, where nowadays no appreciable difference in genetic variability was detectable (Figure 5).
Biotic interactions had a positive effect on the intra-population variability of both nuclear and mitochondrial markers (Figure 5A, B, Table. 2), however in the GLM analysis only on A and H
. According to Eckert et al. , such an increase may be explained by introgression from neighbouring, closely related species through inter-specific hybridisation. However, close inspection of the alleles and mitochondrial haplotypes found at the sampling sites in question revealed, with the exception of one private allele and one private haplotype in one population, respectively, solely alleles and haplotypes that also occurred in other R. balthica populations throughout the species' range. Moreover, the allelic size range of the microsatellite loci in the potentially hybridising undescribed Radix species is known  and none of these alleles were found in the present data set. Also the mitochondrial haplotypes found at these sites fit very well in the haplotype variability of R. balthica . Inter-specific hybridisation with neighbouring taxa is thus an unlikely explanation for the pattern reported here.
However, secondary contact of two more R. balthica lineages, e.g. from different refugial populations, could be the reason for the increase of genetic variability in these areas, as has been shown for other snail species [50, 51]. Several sites throughout the range show signs of increased nuclear admixture, in particular in Southern Sweden and around the LGM refugia (Figure 2, 5). Since most sites grouped in the variables bio are situated around the refugial area and overlap in these more than average variable populations with the predictor bar (Figure 1), an increased variability of nuclear and mitochondrial markers predicted by these variables may indeed be due to few admixed, secondary contact sites and not due to the biological process tested for.
Population size, as rather crudely estimated from the size of the water body, had no detectable effect on the distribution of genetic variability (Figure 5, Table 2). This may have two major reasons: first, population densities of more than 50 individuals per m2 were observed and thus population sizes of several thousand individuals even in small water bodies can be reached (personal observation M. Pfenninger). Thus, the effect of drift in small populations may be difficult to estimate from habitat size alone, but depend rather on the mating system or the founding history. Second, freshwater snail populations are often subject to high population turnover or size fluctuations [21, 52] which lead to a discrepancy between the demographic and the effective population size and thus, potential loss of genetic variability. The high proportion of bottleneck populations detected, argue in that direction.
The size of a water body, however, did have an effect on the selfing rate (Figure 5C, Table 2). Surprisingly, larger habitats were associated with more selfing. This is perhaps due to a dilution effect in larger habitats, which makes selfing as a reproductive assurance strategy more often necessary, because potential mates are less often encountered.
Loss of variability by extinction-recolonisation dynamics was also substantiated by some of the models incorporating environmental marginality (lim) that received substantial support in the data (Table 1). Sites facing more extreme environmental variation exhibited a slightly decreased level of genetic variability at nuclear markers (Figure 5). This is probably a result of extreme climatic events, like e.g. droughts too severe for the snails, flash-floods or too cold winters, in these areas. Such events are expected to decrease genetic variation by decimating or extinguishing local populations and have been shown for R. balthica on a local scale .
Geographic marginality per se contributed little to the distribution of genetic variability in R. balthica (Table 1). In nuclear marker loci, populations close to the inferred range limits even tended to harbour slightly more genetic variation than the total average (Figure 5). Contrary to the majority of empirical studies reviewed by Eckert et al. , the distribution of genetic variability in R. balthica does not follow the predictions for the genetic extension of the ACH. However, contrary to all previously discussed factors, the reliability of this inference depends crucially on the quality of the inference and sampling of the range and its margins. Apart from the multitude of possible definitions for a species range , its practical determination is inherently difficult, because it depends as well on the presence of unequivocally identified populations of the focal species in certain areas as on their absence in others. While the former often enough presents a practical problem due to unrecognised cryptic species, varying observation density and -quality , it is virtually impossible to prove the absence of most species from an area. A species range and in particular its margin is therefore rather an effort-dependent estimate than a fact.
In the case of Radix, unequivocal species determination is possible only with molecular methods and in particular R. balthica can be easily mistaken for other species . Therefore, range estimates of R. balthica based on morphology or even anatomy are prone to error and were not considered here. Our estimate of the R. balthica range represents therefore the best currently available estimate. However, given the postglacial expansion history as inferred by phylogeography , it cannot be excluded that the species also occurs in Norway, Ireland and Scotland. On the other hand, the absence of R. balthica and the confirmed presence of other MOTUs in the sites sampled in the South-West, South, South-East and East argues for a good coverage of the range limits in this area (see Additional file 1). For the South-East, the absence of R. balthica from the Balkans is confirmed by another recent study . In Sweden, no Radix snails were found further North than the populations reported here during our surveys. In total, we are confident that our sample represents i) the larger part of the present species range and ii) that with the possible exception of the North-West, also the range margins were adequately sampled.
However, the ACH does not predict precisely, how variation should decrease towards range margins . By testing the distance to the closest range margin, we assumed that the decline is steady and linear from the core range. If the decline is actually steep and starts only close to the margins, we would have missed it with our sample strategy, because we have probably missed the respectively most marginal populations. On the other hand, a range margin effect requires distance-dependent dispersal , which we have shown to be absent in this species.
The factors evaluated here had also an impact on the variability in the mating system. The common quality of the factors identified to trigger changes in mating system towards more self-fertilisation seemed to be increased population turn-over (Table 1). Actually, self-fertilisation should be advantageous in any metapopulation system with high population turn-over rates . However, even the best model (marg + exp + lim + size) explained not even half of the variance in selfing, indicating that probably additional, untested factors significantly shaped the mating system.