‘Mainland-island’ population structure of a terrestrial salamander in a forest-bocage landscape with little evidence for in situ ecological speciation

Adaptation to different ecological environments can, through divergent selection, generate phenotypic and genetic differences between populations, and eventually give rise to new species. The fire salamander (Salamandra salamandra) has been proposed to represent an early stage of ecological speciation, driven by differential habitat adaptation through the deposition and development of larvae in streams versus ponds in the Kottenforst near Bonn (Germany). We set out to test this hypothesis of ecological speciation in an area different from the one where it was raised and we took the opportunity to explore for drivers of genetic differentiation at a landscape scale. A survey over 640 localities demonstrated the species’ presence in ponds and streams across forests, hilly terrain and areas with hedgerows (‘bocage’). Genetic variation at 14 microsatellite loci across 41 localities in and around two small deciduous forests showed that salamander effective population sizes were higher in forests than in the bocage, with panmixia in the forests (Fst < 0.010) versus genetic drift or founder effects in several of the small and more or less isolated bocage populations (Fst > 0.025). The system fits the ‘mainland-island’ metapopulation model rather than indicating adaptive genetic divergence in pond versus stream larval habitats. A reanalysis of the Kottenforst data indicated that microsatellite genetic variation fitted a geographical rather than an environmental axis, with a sharp transition from a western pond-breeding to an eastern, more frequently stream-breeding group of populations. A parallel changeover in mitochondrial DNA exists but remains to be well documented. The data support the existence of a hybrid zone following secondary contact of differentiated lineages, more so than speciation in situ.


Results
forest and bocage populations in the west of france. Larvae of S. salamandra were present in 251 localities (39%) and absent in 389 localities (61%). Localities are listed in a .kml file for use with e.g. Google Earth (Supplementary Information I). In logistic regression analysis the presence of the fire salamander is positively associated with forestation (P < 0.001), altitude (P < 0.001) and hedgerows (P < 0.05). At any amphibian breeding site, the probability for the occurrence of the fire salamander (P s ) is estimated by the equation (1/ (1 + exp(−0.0303*percent_forest_cover-0.00562*altitude-0.0299*percent_hedgerow_cover + 1.769))). The fit of the model is expressed by the 'area under the curve' statistic AUC = 0.721 ± 0.020, indicating that we achieved a good description of favourable versus unfavourable fire salamander terrestrial habitats. When habitat parameter values are standardized the formula is (1/(1 + exp(a*forest_cover + b*altitude + c*hedgerow_cover + 0.0210))), indicating the relative contributions to the model in the order forest (a = −0.910), altitude (b = −0.288) and hedgerows (c = −0.275). Extrapolating the model over the entire department suggests that habitats for the fire salamander outside forests are widespread in the hilly bocage landscape in the northeastern and northwestern corners of the department Mayenne. Low habitat suitability is found in the flat and deforested southern part of Mayenne, in particular the southwest (Fig. 1). The area studied for population genetic characteristics shows low habitat suitability in between and south of the forests Forêt de Bourgon and Bois de Hermet, intermediate habitat suitability surrounding the forests in the other cardinal directions and high habitat suitability further east of Bois de Hermet.
Localities studied for population genetic differentiation included 25 ponds and 16 streams and were 23 times from inside and 18 times from outside the forest. For the subdivision pond inside -outside and streams inside -outside the forests see Table 1. A high frequency of inferred null alleles (11.7%) and a substantial amount of missing data (5.0%) were observed for the microsatellite locus C2. After exclusion of this marker and the subsequent removal of seven individuals that had data missing for more than one locus (N = 734 remaining) the frequency of null alleles was estimated at 1.6% and 0.11% of data was missing. The genotypic data are presented in Supplementary Information II. The analyses with Colony software suggested that among individuals sampled from the same locality many were full siblings. With just one representative per family group population sample sizes went down by 357 under monogamy (48.6%) and by 85 under polygamy (11.6%). This approach also revealed a high potential for false positives since inferred siblings were frequent among larvae from different localities (N = 315 under monogamy, 42.9% and N = 108 under polygamy, 14.7%). Analytical results for the three data sets are summarized in Table 2.
The number of alleles observed per locus ranged from four in locus SalE5 to 15 in locus Sal3. A single locus significantly deviated from Hardy-Weinberg expectations and there were four instances of pairs of loci showing linkage disequilibrium. The overall F st was 0.0244. The statistical power to be able to detect the mixture of genetically differentiated populations with genetic clustering methods is dependent on sample size, the number of loci studied (14), the number of alleles per locus (average 9.4)  No association was observed between the pairwise distance matrices for genetic differentiation (F st /(1-F st )) and the logarithm of geographical distance (non-parametric matrix correlation, ρ = −0.033, P > 0.05), indicating the absence of isolation by distance over the study area. The observed amount of genetic differentiation is substantial but does not have a bimodal distribution as is illustrated by the UPGMA-dendrogram on F st ( Fig. 2A). A tight cluster at F st < 0.010 is mostly composed of forest populations. Populations that join the dendrogram at higher F st -values are mostly from outside the forests, which reflects a non-random distribution (G-test for independence, G = 11.34, df = 1, P < 0.001). At F st > 0.025 just non-forest populations are added. No significant genetic differentiation was found for pond-versus stream inhabiting population (G = 1.35, P > 0.05). Analyses with principal components revealed a wide overlap for the forest and non-forest groups. However, the forest populations form a tight group whereas the non-forest populations are more widely scattered over the bivariate plot (Fig. 2B). Accordingly, average distance to their nadir in the PC-plots is smaller for forest than for non-forest populations (Student's t-test, t = 3.18, df = 21.1, P < 0.01). No significant difference was found for the pond-versus stream-breeding classification (t = 0.58, df = 39, P > 0.05). The analysis with Structure yielded flat distributions with undifferentiated Q-values (Table 2). It is now realized that the methods employed frequently identify K = 2 as the top level of hierarchical structure, even when more subpopulations are actually present 19 .
Estimates for effective population size N e (Ň e ) were significantly higher for forest than for non-forest populations (Ň e/forested = 32.0, Ň e/non-forested = 22.4; Mann-Whitney U-test, U = 301, P < 0.05) whereas they were not significantly different for pond and stream populations (Ň e/pond = 27.0, Ň e/stream = 29.0; U = 240.5, P > 0.05). Among just pond populations, those from the forest were larger than those from outside the forest (Ň e/forested = 32.7, Ň e/ non-forested = 20.8; Kruskal-Wallis test statistic 5.25, P < 0.05) ( Table 1; see also Supplementary Information III). No significant indications for population genetic bottleneck effects were observed.
Mitochondrial DNA sequences were 755-756 bp. The shorter sequence (Genbank accession number MK395359, N = 40) conforms to 'type 1b' and the longer one (Genbank accession number MK395358, N = 89) to 'type 1a' 11 . Sequences are equivalent to Genbank accession number KU249989 20 with an indel at position 599. The distribution of the two mtDNA types bears no correspondence to the forest/non-forest inhabiting or to the stream-versus pond-breeding habitat classification (see Supplementary Information IV). For a compilation of mtDNA data from across northern Europe see Supplementary Information V. (1 + exp(−0.0303*percent_forest_cover-0.00562*altitude-0.0299*percent_hedgerow_cover + 1.769))) and was visualized with ILWIS 3.6 software 58 , available at https://52north.org/software/software-projects/ilwis/. Habitat suitability increases from deep blue with a probability of occurrence of zero to deep red with a probability of occurrence at unity (see colour bar). Prime fire salamander habitats are found at higher altitudes and are forested (in black) or with a dense hedgerow cover. Populations genetically investigated are located in and around the largely deciduous forests Forêt de Bourgon (FB) and Bois de Hermet (BH) and listed in Table 1 21 . All the investigated populations are situated inside the forest. Fire salamander larvae were observed in a variety of aquatic habitats (Table 3), with streams less frequent in the western section of the forest (one stream locality out of 27, 4%) than in the eastern section of the forest (six stream localities out of 20, 30%).
Analyses with Colony software suggested that many of the sampled individuals were full siblings. With only one representative per family group population sample sizes went down by 1628 under monogamy (63.5%) and by 1194 under polygamy (46.6%). Colony also revealed a high potential for false positives since inferred siblings  Table 1. Populations of the fire salamander in Mayenne, France with samples subjected to genetic analyses, with locality number, geographical coordinates, sample size with and without all but one of the inferred siblings excluded, classification of the habitat in which the larval salamanders were found (pond versus stream and forest versus bocage) and estimates of the effective population size Ň e . The microsatellite genetic profiles are summarized as the loadings on the first and second principal component axis. # All but one of the full siblings per inferred family group taken out under a monogamous/polygamous mating system.
were frequent among larvae from different localities (N = 624 under monogamy, 24.3% and N = 210 under polygamy, 8.2%). Results for genetic (dis)equilibria, genetic bottleneck effects and isolation by distance were markedly different for full and the sibling-excluded data sets. In the former we found 72 cases of significant deviation from Hardy-Weinberg expectations and 941 pairwise locus combinations with significant linkage disequilibrium ( Table 2). Numbers were by two orders of magnitude lower in the reduced data sets, suggesting that the significant signals for genetic equilibria are largely to be attributed to the sampling of family groups. In the sibling reduced data set we identified no significant signal for population genetic bottlenecks whereas there were seven in the full data set. Finally, the signal for isolation by distance was not significant in the full data set (non-parametric matrix correlation, ρ = 0.070, P > 0.05) and significant in the reduced data sets (ρ = 0.191, P < 0.001 under monogamy and ρ = 0.119, P < 0.05 under polygamy). Estimates for effective population size varied widely and were not significantly different for forest sections (Ň e/west = 31.2, range 4-104; Ň e/east = 55.6, range 2-272; Mann-Whitney U-test, U = 249, P > 0.05). Average effective population sizes for the Kottenforst were not significantly different from those in Mayenne (Mann-Whitney U-test, U = 1037, P > 0.05), but showed a wider range in the Kottenforst than in Mayenne (5 < Ň e < 56, see also Supplementary Information III).
The overall F st for the studied fire salamander populations was 0.0360, which value widely surpassed the threshold for 10% analytical power of 0.0016. The UPGMA-dendrogram of pairwise F st -values reveals two clusters that are differentiated at F st = 0.04 (Fig. 3A). One cluster is made up of one eastern plus 14 western populations and the other cluster is made up of two western and 14 eastern populations, which reflects a non-random distribution (G-test for independence, G = 23.54, df = 1, P < 0.001). This signal is stronger than the parallel separation in pond-and stream-breeding populations (G = 4.60, P < 0.05). Populations joining the dendrogram at higher F st levels are from either forest section. This set of populations is characterized by particularly low Ň e (average Ň e = 7.6 versus Ň e = 60.1 for the remainder; Mann-Whitney U-test, U = 439.5, P < 0.001), suggesting that a founder effect or genetic drift underlies the genetic differentiation of small populations. Analyses with principal components supports the population allocation of two spatial groups, in which the western and eastern group show no overlap along the first PC-axis if indeed small populations are excluded (Fig. 3B). The support for separation of a western versus an eastern group is several orders of magnitude stronger than that for the (more or less parallel) separation of pond-and stream-breeding salamander populations (Student's t-test, t = 8.58, df = 29.0, P < 0.000001 versus t = 2.81, df = 45, P = 0.007401). In line with these results, the plot of the PC1-scores over a  Table 2. Summary of results for full and sibling reduced data sets in the study of population genetic variation in the fire salamander in Mayenne, France and the Kottenforst, Germany. Note that the genetic differentiation of pond-versus stream-breeding populations is subordinate to a forest/non-forest differentiation in France and to a longitudinal differentiation in Germany. www.nature.com/scientificreports www.nature.com/scientificreports/ west to east axis describes a genetic transition in a sigmoid curve of the type reminiscent to those analyzed in the classical hybrid zone literature e.g. 22 . The cline that best fitted the data has a central position at km 365.3 of the longitudinal axis of the Universal Transverse Mercator coordinate system grid (UTM) and a width of 3.95 km (Fig. 4A). Because the PC-analysis extracted only a small proportion of the total variance in the data (2.1% along the first axis) we repeated the cline fitting procedure with the 'proportion of pond-versus stream-breeding genotypes' or 'assignment probabilities' of Hendrix et al. 21 under an inferred optimal number of two genetic clusters (K = 2), derived with Structure software 23,24 . The cline description that best accommodates these values has a central position at km 365.1 and a width of 1.11 km (Fig. 4B). The full model descriptions are presented in Supplementary Information VI.

Discussion
Adult fire salamanders are terrestrial amphibians that deposit larvae in ponds or streams, in where these develop until metamorphosis. The species figures as an example of local ecological and genetic differentiation and adaptation, possibly representing the first step in the speciation process 12,25 . We set out to replicate the observation of genetic differentiation of stream-and pond-breeding populations and we sought the opportunity to explore for additional drivers of genetic differentiation at a landscape scale. However, the microsatellite data we obtained for the fire salamander in Mayenne did not yield a signal for genetic differentiation of pond-and stream-breeding populations. Instead we found genetic differences for populations from inside and outside forests.    www.nature.com/scientificreports www.nature.com/scientificreports/ deciduous forest at 200-400 m a.s.l. 26,27 . In addition, we document the regular presence of fire salamanders outside forests, in particular in hilly areas where the network of hedges is dense. Such a pastural landscape is known as 'bocage' . Our data suggest that forests are the primary fire salamander habitat. Firstly, forestation represents the most prominent contribution to the habitat model. Secondly, the estimates for effective population sizes are larger for forest than for bocage populations. Thirdly, forest populations are genetically most similar suggesting ongoing www.nature.com/scientificreports www.nature.com/scientificreports/ gene flow, presumably covering historical times. In contrast, the bocage populations are frequently genetically differentiated from one another (F st > 0.025) indicating that they are more or less isolated and that genetic drift and founder effect operate more effectively in these small populations. In metapopulation terminology, the fire salamander system in Mayenne fits the 'mainland-island' model 28 in which forests are the mainland and the bocage represents an archipelago of islands. In Spain, genetic differentiation was observed to increase from F st < 0.05 among most continental S. salamandra populations to F st~0 .10 for peninsular populations separated for at least 2000 years, to F st~0 .19 for insular populations separated for 6-13 Kybp (thousands of years before present) (geological data 29,30 , genetic data 31,32 ). This system, literally fitting the mainland-island model and at a comparable spatial scale to Mayenne, confirms the propensity for genetic drift in small and isolated fire salamander populations.
The two forests in Mayenne may be interconnected by dispersal through the dense but patchy and declining bocage. While population connectivity operating through the network of hedgerows is likely, the study area does not offer a good setting for testing this hypothesis. First, we noted no consistent spatial genetic signal and it will be difficult to assess if this can be attributed to the counteracting forces of isolation by distance and connectivity by habitat. Second, to be able to disentangle these effects requires an explicit spatial configuration. For example, to test for fire salamander dispersal along hedgerows, the hypothesized corridor is to deviate from a straight line. If not, genetic differentiation will, by default, be associated to geographical, not ecological distance.
We found no strong evidence for population genetic bottlenecking, suggesting that fire salamander populations may be stable over time. Another line of evidence for the long-term persistence of fire salamander populations are several occurrences in the northwest of Mayenne ( Supplementary Information I) 33 . This area is currently devoid of forests (Fig. 2B) and also 18 th century maps by the Cassini family (accessible at https://www.geoportail. gouv.fr/donnees/carte-de-cassini) show no forest at the localities where fire salamanders were actually observed, suggesting that extant populations persisted in the bocage over the centuries and providing ample opportunity for genetic drift to operate. While we could not detect significant departures from panmixia in forests, the bocage populations are more or less isolated. Yet the bocage localities may serve as 'islands' or 'stepping stones' . Even a loose network may help to preserve population connectivity at a wider spatial scale and eventually promote the genetic exchange between forest populations, such as those of Forêt de Bourgon and Bois de Hermet (Fig. 2B). However, the pastural landscape in western France rapidly deteriorates from the perspective of wildlife, mostly through field size enlargements and agricultural reform, with the concomitant loss of small landscape elements such as hedgerows, spinneys and ponds 18,[33][34][35][36] . A further deterioration of the bocage is likely to isolate the forest populations from one another. To illustrate this point, in the flat, southern part of Mayenne the bocage has largely disappeared since the Second World War 37 and the fire salamander is locally rare with some scattered occurrences 33 (Supplementary Information I).
fire salamander population structure in the Kottenforst. The German Kottenforst supports two fire salamander population groups that are differentiated at F st = 0.030 and that are proposed to represent pond-versus stream-breeders. However, the ecological data in support of this interpretation are meagre, simply because stream-breeding populations are a minority in the Kottenforst system. Moreover, focusing on large and presumably healthy populations that are locally more frequently found in streams than in ponds (Table 3) strengthens the eco-geographical distinction more so than it sharpens the genetic differentiation.
The west to east distinction is to some degree obscured by a suite of 16 populations that are more deeply genetically differentiated (0.030 < F st < 0.153; Fig. 3). These latter populations are characterized by significantly smaller effective population sizes than average and 11 of them have Ň e ≤ 10. The small population sizes suggest that genetic drift might be prominent, but given the more or less unhampered dispersal of adult fire salamanders across the forest, populations are unlikely to be isolated and a genetic founder effect is a more plausible explanation.
With an average minimum distance of neighbouring populations of ca. 400 m the grid of localities studied in the Kottenforst is tighter than in Mayenne. Yet, unlike Mayenne, the pond-and stream-breeding populations appear to be spatially, behaviourally and genetically separated. This separation takes effect along a roughly longitudinal axis. The by approximation sigmoid shaped transition is 1.11-3.95 km wide, with inflection points that are 730-1440 m apart (Fig. 4). A cline this narrow cannot result from neutral processes exclusively. In the absence of selection, the width (w) of the cline can be predicted from a diffusion model as a function of dispersal distance (d) and the length of time since contact (t), as w = 2.51d √(t) 38 . Generation time is reported as six years for both sexes 39 and also as three years for males and four years for females 40 . At an average dispersal of 200 m per generation, which is at the lower limit because it just covers the regular home range, cline widths would exceed the measured width in a couple of generations and at higher dispersal, such as distances of up to 2000 m 24 , the collapse of the cline would be near-instantaneous. However, the Kottenforst genetic transition is sharp, suggesting that intrinsic selection operates against the two lineages' mixed offspring. The center of the transition is further characterized by a paucity of material studied, presumably caused by lack of aquatic habitat suitable for the deposition of larvae. This observation fits classical theory where a hybrid zone may be 'captured' where an area of low population density acts as a barrier to further movement 41,42 . If dispersal is high, selection against hybrids is to be strong. Evidence for genetic incompatibility of the lineages may have been overlooked by Caspers et al. 13 . These authors were surprised to find that offspring numbers in the Kottenforst increased with the degree of genetic relatedness between females and their sires. Puzzling as this observation might be for within-lineage data, in the context of selection against hybrids it makes direct sense.
The postglacial range expansion of S. salamandra has been linked to the spread of the beech, Fagus sylvatica L 11,12,43 . The contact zone, cq. intra-specific hybrid zone in the Kottenforst may have originated ca. 8 Kybp, by colonization from southern European glacial refuge areas, possibly involving the south of France and the northern Balkans, but more northerly locations cannot be excluded 44,45 . Regardless of the location of the glacial refugia, the hybrid zone in the Kottenforst appears to have been kept in check over hundreds or thousands of years, in which selection against hybrids counteracted dispersal into the zone. As an alternative explanation to in situ ecological speciation, we suggest that the Kottenforst is an area of secondary contact of a pond-breeding western lineage and a stream-breeding eastern lineage. Pond-breeding in the Kottenforst is considered a recent adaptation 25,46 , but we consider this unlikely because the condition is widespread in western Europe 26,28,35,47,48 .
An argument against the secondary contact scenario may be that the Kottenforst fire salamander populations are more similar to one another than to other, nearby populations for mitochondrial 11 and nuclear genetic markers 49 . The fire salamander in northern Europe is characterized by two mitochondrial DNA lineages with 'type 1' in the west and 'type 2' in the east. According to Weitere et al. 11 the pond-and stream-breeding populations in the Kottenforst are all descendants of the western mitochondrial DNA lineage. However, a sharp and complete transition from type 1 to type 2 is found not far away (at 70-80 km north, in between the localities Felderbachtal and Bochum; Supplementary Information V). We propose that the mtDNA cline is displaced relative to the cline from the nuclear genetic markers, a phenomenon regularly observed, especially for uniparentally transmitted markers [50][51][52] . Unfortunately, a wide sampling gap (>350 km) precludes measuring position, width and shape of the mtDNA transition east of the Kottenforst. With microsatellite data Steinfartz et al. 49 also found that the Kottenforst populations are the most closely related in the wider area. This conclusion, however, depends on the position of the 'root' of the graph (which is not provided) and the supporting evidence is not unambiguous given the graph's short internal branches. Future research might reveal a continuation of the fire salamander contact zone outside the Kottenforst and possibly show that the western and eastern lineages represent the currently described subspecies S. s. terrestris Lacépède, 1788 and S. s. salamandra. A wider phylogeographic survey is required to solve these issues. concluding remarks. The fire salamander is a species prone to population genetic differentiation.
Substantial genetic drift has been reported for small and isolated populations across its range 31,53-56 . We analyzed (2020) 10:1700 | https://doi.org/10.1038/s41598-020-58551-0 www.nature.com/scientificreports www.nature.com/scientificreports/ microsatellite genetic data in two metapopulations of the fire salamander in the northwest of Europe, with contrasting results. In France we documented small, isolated, persisting and genetically differentiated populations in the bocage versus panmixia in two small deciduous forests. This population structure fits the classical mainland-island model. A mainland-island population structure may have been paramount to the fire salamander prior to the recent intensification of agriculture, the field size enlargements that came with it and the widespread conversion from pasture for cattle-breeding to arable for growing crops. We did not find a bimodal distribution coinciding with stream-and pond-reproduction as detected in the Kottenforst 12,25,57 . Accordingly, the fire salamander system in France offers no support for habitat driven genetic differentiation other than through drift and founder effects operating in more or less isolated bocage populations. In Germany we found small and genetically differentiated populations scattered over the Kottenforst, in addition to larger populations occupying ponds in the western section and ponds and streams in the eastern section of the forest. The finding is somewhat puzzling, given the unrestricted gene flow that we documented for the French forest populations and the rampant migration documented for the Kottenforst per se, with large home ranges and frequent long-distance dispersal 24 . An ad hoc explanation, supported by the high frequencies of full-siblings and low effective population sizes, is that these localities represent a particularly intensive sampling scheme, with larvae included from unusual aquatic habitats such as puddles and wheel ruts ( Table 3). These small and frequently impermanent water bodies may represent sub-optimal conditions for larval growth, survival and metamorphosis and be transient satellites to the core fire salamander populations. Finally, to explain the significant spatial component in the remaining genetic variation, we propose that secondary contact between a western and an eastern fire salamander lineage better explains the available data than an ecological sympatric speciation scenario.

Materials and Methods
We recorded the presence or absence of fire salamander larvae in 640 amphibian pond and stream breeding sites across the department Mayenne. Ponds represent focal points of amphibian presence and harbour more or less isolated populations (or 'demes') that together constitute a metapopulation. The deme structure in streams is less obvious, but clearly not all sections of a stream are equally suitable for reproduction, for example in sections with and without predatory fish, or sections falling dry. Yet, for consistency, we consider the amphibian sites that we study to represent local populations. Locality data on altitude (m above sea level, a.s.l.), forestation and hedgerows (percent cover) were extracted from digital IGN maps (Institute Géographique Nationale) with a 25 m spatial resolution, provided by department officials. The area considered around each pond or stream sampling locality had a radius of 200 m. This scale should represent a typical fire salamander home range because 200 m is the median distance travelled by adults in a capture-recapture study in the Kottenforst 24 . The biological and environmental data were analyzed by logistic regression with a weight parameter, so that in analysis the number of fire salamander presences equaled the number of absences. Habitat models were visualized with ILWIS 3.6 58 . We did not observe a disease-born population decline that was reported ca. 500 km to the northwest, in The Netherlands 10 .
The area selected for population genetic research is situated around two small, largely deciduous forests and positioned at the transition from flat and deforested in the south of Mayenne, to hilly with a bocage landscape in the north (Fig. 1). Larvae were captured with dip nets in 41 ponds and streams which were located inside as well as outside the forests. The larvae were released on the spot directly after tail tip tissue sampling. Altogether 741 salamander larvae were genotyped for 15 polymorphic microsatellite loci. The loci Sal3, Sal29, SalE11, SalE5, SalE6, SalE7 and SalE8 were studied following Steinfartz et al. 59 and the loci B11, C2, C3, E11, G6, G9, IA6, IIA6 were studied following Hendrix et al. 60 . Locus SalE5 is a dinucleotide microsatellite locus and the others are composed of tetramer motifs. After a quality check (see Results) data for the locus C2 were excluded so that 14 markers remained. A total of 129 individuals from 14 localities was sequenced for the mitochondrial D-loop (control region, 756 bp) as described in Steinfartz et al. 61 . Nuclear genetic data from the German Kottenforst involved 2563 larval fire salamanders studied at 17 polymorphic microsatellite loci and were retrieved from Hendrix et al. 21 . The 47 Kottenforst sampling localities were classified in two spatial groups ('west' with localities K01-K27 and 'east' with localities K28-K47) that represent opposite sections of a sharp genetic transition (see Results).
Computer programs employed for the genetic data were as follows. FreeNA 62 to analyze for the presence of null-alleles. FreeNA was run with 1000 replicates using the EM algorithm. GenePop version 4.2 63 to estimate F st -values and analyze for Hardy-Weinberg and linkage (dis)equilibria. GenePop was run with dememorization number 1000, 100 batches and 1000 iterations per batch. Linkage disequilibrium was determined using the log likelihood ratio statistic. Hardy-Weinberg equilibrium and linkage disequilibrium results were interpreted under the Benjamini-Hochberg correction for multiple comparisons. Statistical power for the detection of genetically differentiated yet sympatric populations was estimated following Jorde et al. 64 . The genetic diversity among populations was summarized by clustering with the unweighted pair group method with arithmetic mean with Primer-e software (UPGMA 65 ) and by principal component analysis (PC) with Adegenet version 2.0.0 following the manuals 66,67 . We also carried out a Bayesian assignment analysis with Structure 68 , to which we adopted the program settings for the similar analyses by Hendrix et al. 21 . The parameter K (the number of genetic groups suggested by the data) was evaluated under the 'Evanno-criterion' that is implemented in StructureHarvester 69 . We used Colony version 2.0.6.2 70 to analyze family groups and to estimate effective population size (Ň e ) from the frequency of siblings. Settings for Colony were both sexes either monogamous or polygamous (see below), no allelic dropout or typing errors, diploid, codominant markers, no inbreeding, scaled full sibship, unknown population allele frequency, ten 'medium long' or 'very long' runs under full-likelihood with 'medium' precision and a 'strong/optimal' sibship prior for effective population size. Results for Ň e reported are under α is zero, assuming that deviations from Hardy-Weinberg equilibrium are negligible. We used Bottleneck 71 to test for reductions in effective population size by considering that alleles are generally lost faster than heterozygosity and thus, populations that have experienced a reduction in effective population size are expected to have excess heterozygosity relative to that expected under mutation-drift equilibrium. Samples smaller than for ten individuals were ignored.
Statistical relevance of the results were determined with the Wilcoxon signed-rank test. We applied the two phase model with 93.6% single step mutations, variance 30 and 10,000 replications. The proportion of single-step mutations was determined with Misat version 1.0 72 with dimer code 2, tetramer code 4, gridsize 40, moments estimated under the one-step model, 100,000 runs through the Markov chain, with continuously updating the value of θ 0 . We used HZAR 73 to formalize the geographical cline suggested by the first PC-axis for the Kottenforst. The chain length was 100,000 with a burn-in of 10%, randomized seeds and model selection based on AICc scores. For details on the cline fitting procedure and model selection see e.g. 74 . Mitochondrial DNA sequences were aligned with ClustalW version 2 75 under default settings. Matrix correlations were done with Primer-e 65 under 10,000 permutations. Other statistical analyses were with SPSS v. 20 76 .
Reproduction in the fire salamander may be monogamous or polygamous. The available data indicate that both systems occur in nature about equally frequent 49 and we explored both possibilities. Including siblings in analyses of population diversity and structure can introduce a variety of biases ( 77 and references therein). Clutch sizes in the fire salamander are ca. 30 27 and the more or less frequent sampling of full-and half-siblings in our study cannot be excluded. However, excluding all but one members of a family group is not exempt from problems neither and will cause other biases 78,79 . Considering the moderate number of markers and alleles in either of the two data sets we tested for the presence of full-siblings in the material at two levels. Firstly, we reconstructed family groups per population. The numbers of inferred full-siblings were substantial. Secondly, we compared all individuals across localities. Although female fire salamanders may deposit larvae in more than one water body 13 this phenomenon will be rare in our sampling of disparate localities with average minimum distances of ca. 1.1 km in Mayenne and 400 m in the Kottenforst. Consequently, inferred full-siblings from different localities possibly qualify as false positives. It is altogether not straightforward to determine the balance between redundant information from siblings included versus independent information from false positives excluded. We here present results from the full data set not to hamper comparability with published results 24 , with the parallel results for the siblings excluded data set in Table 2. The mitochondrial DNA data considered include 80

Data availability
The genotypic data for fire salamanders from Mayenne, France are presented in Supplementary Information II. The data for the Kottenforst, Germany are accessible at https://doi.org/10.5061/dryad.h0r6q.