Background

There is increasing evidence of the role of chromosomal inversion polymorphisms in adaptation and speciation [1, 2]. Several lines of evidence attest this namely: (1) the repeated occurrence of latitudinal clinal variation across multiple invaded continents, particularly in Drosophila species ([3,4,5,6]; (2) consistent associations between inversions and several relevant phenotypic traits [7,8,9,10,11,12,13]; and (3) inversion frequency changes beyond genetic drift expectations across replicated populations adapting to imposed laboratory conditions [14,15,16].

Several hypotheses have been advanced to explain the adaptive character of inversions (see [1]), with the most widely accepted hypotheses relying on recombination reduction in inversion heterokaryotypes and consequent protection of genes with positive effects on fitness [17, 18]. The coadaptation hypothesis [17] postulates that specific favourable combinations of alleles interacting epistatically within a particular inversion arise as an adaptive response to a given environment. The local adaptation hypothesis [18] states that the spread of a new inversion results from advantageous locally adapted alleles captured within the inverted segment, without epistasis being necessarily involved. Inversion frequencies changes will result from a balance between selection and migration. A general expectation in both these scenarios is that there is genetic differentiation between inversion types, as a result of local adaptation shaping the genetic content of different inversions. Coalescent models suggest that this differentiation is likely higher near breakpoints – where recombination is greatly reduced - and near locally adapted alleles within the inversion [19]. In addition, under the coadaptation hypothesis there is a clear expectation of genetic differentiation within inversions sampled from different geographical populations, as a result of population-specific epistatic interactions evolving in response to different environmental conditions [20]. However, this pattern might not be exclusive of the co-adaptation model as some within-inversion differentiation might occur depending on the balance between migration and selection on locally adapted alleles (with or without epistasis involved) as postulated by the local adaptation.

Empirical studies on the molecular variation associated with inversions have frequently detected high levels of genetic differentiation between chromosomal arrangements ([21,22,23,24,25,26,27] but see [28]). This pattern is consistent with the aforementioned hypotheses and is expected as a result of reduced recombination in inversion heterokaryotypes leading to distinct genetic pools. Studies in the O chromosome of D. subobscura found an extensive genetic differentiation between arrangements occurring regardless of the distance to the nearest breakpoint, with selection likely implicated [29, 30]. On the other hand, several studies have reported a general lack of measurable genetic differentiation within arrangements sampled from different populations ([20, 31,32,33], contrary to the expectations of the coadaptation hypothesis. An important exception is the genetic clinal variation within the inversion In(3R)Payne of Drosophila melanogaster, both in North American and Australian localities, implicating a role of this inversion in the response to climatic selection [6, 31, 34]. In addition, there is some evidence for epistasis - as expected under the coadaptation model - based on patterns of linkage disequilibrium (LD) with regions within arrangements showing high levels of LD and differentiation interspersed with other regions of low LD [20, 31] as well as LD between arrangements and genes located outside them [30].

One of the most emblematic examples of the adaptive role of chromosomal polymorphisms comes from Drosophila subobscura, a species presenting repeatable clinal patterns for inversion frequencies across Europe, the ancestral area, and North and South America [3]. Thermal adaptation or associated factors might be implicated in such patterns. Evidence for this hypothesis comes from the association between temporal frequency changes of D. subobscura chromosomal arrangements and increased temperature due to climate warming at a worldwide scale [5]. Also, seasonal inversion frequency changes in D. subobscura have been reported, even shortly responding to heat waves [35] as well as associations between inversions and several phenotypic traits, including thermal preference and tolerance [36,37,38] and body size [10, 39]. Importantly, differences in gene expression patterns were also observed in D. subobscura from a colonizing South American population evolving in the laboratory at different thermal regimes (13, 18 and 22 °C); with an enrichment for candidate genes located inside arrangements [40]. This suggests that different inversions may contribute to the maintenance of genes with differential expression and potential impact on fitness. Indeed, this was the case in D. pseudoobscura with multiple genes presenting expression differences among arrangements [41]. Consequently, it is relevant to analyse overall patterns of molecular variation in these candidate genes in the vicinity/within the D. subobscura inversions as well as address their variation across the species geographic gradient to search for targets of climatic selection.

Pegueroles et al. [33] analysed 6 of these genes previously associated with thermal adaptation located in the O chromosome of D. subobscura and found no differentiation for the same arrangement between Barcelona and Greece indicating high levels of gene flow and the absence of geographic varying selection at least in those two southern regions. In a study in D. subobscura involving several localities across western Europe, Simões et al. [32] using microsatellite loci also detected low within-arrangement differentiation in several inversions of two autosomes (J and U) and the sex chromosome (A) confirming an extensive gene flow in this species (see also [42, 43]). Nevertheless, significant clinal patterns at few specific microsatellite loci within arrangements emerged (one in the distal region of A2), suggesting adaptation at neighbouring loci.

The aim of the present study is to examine the genetic variation associated with two arrangements of the sexual chromosome of Drosophila subobscura across a geographic gradient. Specifically, we intend to (1) characterize genetic variation and differentiation between arrangements, addressing the impact of inverted regions at the genetic level; (2) analyse patterns of within-arrangement differentiation, searching for evidence of different epistatic combinations in different populations; (3) identify latitudinal patterns of genetic variation, which might indicate targets of climatic selection at a geographical scale. In this study we targeted AST and A2, the two most frequent inversions of the A chromosome in our sampled localities. These two arrangements present clinal variation of frequencies in Europe, as well as in the two recently colonized American hemispheres, with A2 increasing frequency towards warmer locations and AST showing the corresponding opposite pattern [32, 44]. We analysed DNA sequence variation in four gene regions (PhKgamma, Ubc-E2H, Hfw and Nipsnap). These genes are good candidates to search for evidences of climatic latitudinal selection associated with the Drosophila subobscura inversion clines as they (1) have been previously implicated in thermal adaptation ([40], see also above) and (2) are located relatively close to inversion breakpoints, with reduced recombination in these regions possibly protecting genetic variants and showing the footprint of selection.

Methods

Fly samples and chromosomal inversion scoring

Wild Drosophila subobscura samples were collected from four European locations (Fig. 1). Samples from Málaga (Mal, 36° 43’N, Spain) and Rasquera (Ras, 40° 57’N, Spain) were obtained in October 2008 with a total of 169 and 152 individuals, respectively. Samples from Dijon (Dij, 47° 18’N, France) and Groningen (Gro, 57° 13’N, Netherlands) were obtained in August/September 2009 with a total of 344 and 326 individuals, respectively [45]). These locations vary greatly in environmental conditions, for instance the two most extreme sampled locations have presented clearly contrasting range of yearly average temperatures (7–31 °C in Málaga vs. 0–23 °C in Groningen) and also average precipitation values (44 mm in Málaga vs. 75 mm in Groningen) between 2000 and 2012. We performed the collections in the late summer/early fall to capture the second peak of abundance for the species in the different locations allowing a more reliable comparisons between warmer and colder populations [46].

Fig. 1
figure 1

Sampling sites of D. subobscura and frequency, as coloured pies for each locality, of the three sex chromosome arrangements detected (data from [25]). AST – dark grey, A2 – light grey; A1 – white

Full size image

Our study complies with National and International guidelines, either involving field collections or experimentation in animals. We use Drosophila subobscura, which is not a threatened species nor requires legal ethical consent for experimentation.

This study focused on the two most frequent arrangements present on the acrocentric A (sexual) chromosome (AST/A2), and the only two arrangements present in all analysed locations (Fig. 1). AST is considered an ancestral chromosome arrangement and A2 a derived medium-sized inversion (~ 7.1 Mb) with an estimated age of 160.000 years [23]. The breakpoints of A2 (Fig. 2) are 8C/D-12C/D [47]. In order to score the chromosomal arrangements of the A chromosome, wild-caught males or F1 male descendants from wild isofemale lines were individually crossed with virgin females of the chcu strain, homokaryotypic for the AST arrangement. Polytene chromosomes from salivary glands of one female larva descendant per cross were stained with 2% orcein in 60% acetic/lactic acid (50:50) (see details in [32]). This protocol allowed identifying the chromosomal arrangement of each parental male from the chromosomal configuration of their female larva progeny.

Fig. 2
figure 2

Cytological location of the genes analysed. The grey box represents the region inverted in the A2 arrangement. Black bars indicate the cytological location of each gene in each chromosomal arrangement

Full size image

Genomic regions, DNA extraction and sequencing

DNA was individually extracted from wild males (or male descendants of wild females) whose chromosomal arrangements (A2 or AST) had been previously scored as described above. By amplifying and directly sequencing males we avoided the problems associated with sequencing of diploid data. The karyotype of these males was inferred from their descendants (see above) making it possible to obtain both genetic and karyotypic data for the same individual. Genomic DNA extraction was performed following the protocol described in [48]. The four genomic regions analysed were localized again by in situ hybridization with digoxigenin following standard protocols [49], as some genes in Laayouni et al. [40] have proved to be mislocated [33]. All probes hybridized in the same band previously reported [40] with the exception of Nipsnap that was localized nearby but outside the inversion (see Fig. 2 and Additional file 1). Thus, PhKgamma (cytological location 8C; CG1830) and Nipsnap (13B; CG9212) are located outside the A2 inversion, and Ubc-E2H (9A; CG2257) and Hfw (12A; CG3095) within that inversion (Fig. 2). Following the AST arrangement, being the ancestral one, we will refer to PhKgamma and Ubc-E2H as the proximal loci, the region being closer to the centromere, and Hfw and Nipsnap as the distal loci. Ubc-E2H (Ubiquitin conjugating enzyme E2H) is involved in lateral inhibition and protein polyubiquitination, PhKgamma (Phosphorylase kynase enzyme) is involved in embryonic morphogenesis and glycogen metabolism and Hfw (Halfway) is involved in intracellular signal transduction and pupariation and no involvement in biological processes is yet ascribed to the Nipsnap gene although a putative role in vesicular transport has been proposed (the function of each gene is according to Flybase). Primers used for amplification and sequencing reactions are listed in Table 1. The fragments amplified correspond to partial sequences, not comprising the totality of each gene (see Table 1 for details). Distance to the nearest breakpoint (Mb) was estimated by counting the number of cytological bands in the map of Kunze-Mühl and Müller [50] assuming that all bands contain the same DNA content (82 Kb) as in [51]. Distances were around 0.082 Mb for PhKGamma and around 0.574 Mb for the other 3 genes (Fig. 2).

Table 1 Chromosomal location, length and primers used for each of the 4 genes studied
Full size table

DNA amplification and sequencing was carried out as in [45] with an annealing temperature of 5 s at 54 °C during the amplification reaction. For each individual two sequences were obtained, from forward and reverse primers (see primer sequences in Table 1). Both sequences were assembled using SeqMan II from the Lasergene program (DNASTAR) to get one final sequence per individual. All sequences were then aligned with Clustal W, in BioEdit v.7.2 [52]. The sequence length of the total alignment for each of the four analysed genomic regions is given in Table 1 with the 5’UTR, exonic and intronic regions assigned by comparison with the orthologous sequences of Drosophila pseudoobscura retrieved from Flybase and visual inspection of the resulting expected proteins. The number of males analysed per locality and inversion are presented for each gene fragment in Table 2. Our sampling strategy targeted a comparable number of males to be analysed for each population and arrangement, and as such the inversion frequencies in our samples do not match natural inversion frequencies as seen in Fig. 1.

Table 2 Nucleotide variation for each gene and arrangement across all localities
Full size table

Data analysis and statistical methods

Several standard parameters for each locality and arrangement were estimated using DnaSP v5 [53]: the number of haplotypes (h), number of polymorphic sites (S), the nucleotide diversity (Ï€), nucleotide diversity in synonymous sites and non-coding positions (Ï€sil), heterozygosity in silent sites (Ï´sil) and divergence per silent site (Ksil) using D. pseudoobscura as outgroup [54], with the Jukes-Cantor correction. Comparisons of the diversity values across localities and arrangements were performed with the Wilcoxon matched pairs test using Statistica v10. Genetic differentiation between or within arrangements was assessed using FST [55] and Dxy [56] as well as significant levels for Snn [57] were obtained with 10,000 replicates. False Discovery rate - FDR - correction [58] was applied to account for multiple comparisons in the differentiation analyses. Analyses were performed excluding gaps. For a visual representation of the genetic differences between arrangements and populations in the PhKgamma and Ubc-E2H gene regions, pairwise FST values were used to compute a Principal Coordinate Analysis (pcoa in R package ape version 3.5 (http://ape-package.ird.fr/). A locus-by-locus analysis of Molecular variance (AMOVA) was performed on Arlequin v3.5.2 [59] to test for genetic differentiation (FCT) between arrangements at each specific polymorphic site. A hierarchical structure was considered with two groups corresponding to the two arrangements (A2 vs AST), with the different localities and individuals sampled within these groups.

Gene conversion tracts (GCTs) were detected using the algorithm defined by [60]. Linkage disequilibrium (LD) between pairs of informative sites was assessed and the statistical significance analysed with a chi-square test after Bonferroni correction. These analyses were performed for each arrangement and gene region, as well as for the concatenated dataset. The overall linkage disequilibrium statistic ZnS was estimated for informative sites. LD was also estimated between polymorphic informative sites through r2 estimates. Deviations from neutrality were tested using Tajima’s D [61] for each chromosomal arrangement and locality. Population size changes were also determined using R2 [62]. Statistical significance was obtained by coalescent simulations (5000 replicates) considering free recombination, as it seems most suitable when analysing D. subobscura genes whether located inside or outside chromosomal arrangements (see [33]). All neutrality tests were performed after removing recombinant individuals (with indication of GCTs). The DnaSP v5 program [53] was used to perform the GCTs and linkage disequilibrium analyses as well as neutrality tests.

Results

Nucleotide variability and genetic differentiation

For the two arrangements (AST and A2), we analysed the four genomic regions in the two most geographically distant localities (Málaga and Groningen), and two gene regions for the two central localities (Rasquera and Dijon) (Table 2). The two gene fragments being analysed in all localities were PhKgamma and Ubc-E2H as they showed genetic differentiation between arrangements (see Figs. 3 and 4 and Additional file 2). The genetic variation in those regions is likely to be affected by inversions to a greater extent and thus could possibly present clinal variation.

Fig. 3
figure 3

Principal Coordinate analysis based on genetic differentiation (FST) among arrangements and localities for the PhKgamma gene. AST groups are represented in blue and A2 in orange; different localities are indicated by different symbols

Full size image
Fig. 4
figure 4

Principal Coordinate analysis based on genetic differentiation (FST) among arrangements and localities for the Ubc-E2H gene. AST groups are represented in blue and A2 in orange; different localities are indicated by different symbols

Full size image

Diversity values were in general quite similar between localities for the same gene and chromosomal arrangement (see Additional file 3). Higher nucleotide diversity was found in AST than A2 arrangements at all gene regions (Table 2 and Additional file 3), with significant differences when including information from all regions and localities for all diversity measures (Wilcoxon matched pairs test; Z = 3.06, P = 0.002 for π and πsil; Z = 2.93, P = 0.003 for S; Z = 2.98, P = 0.003 for Ɵsil). No differences in the divergence per silent site values (Ksil) relative to D. pseudoobscura were found between the two arrangements (Wilcoxon matched pairs test; Z = 0.24, P = 0.814). Higher nucleotide diversity was found for Hfw and Nipsnap (Table 2 and Additional file 3), which might possibly indicate a higher substitution rate in these two more distally located regions.

High significant genetic differentiation between the AST and A2 arrangements was detected within each locality and also between localities for Ubc-E2H and PhKgamma (see Figs. 3 and 4 and Additional file 2) and also the concatenated dataset of all genes (see Additional file 4). On the other hand, Hfw and Nipsnap did not show significant differentiation between arrangements whether from the same or different localities (see Additional file 4). In general, no significant differentiation was observed when analysing groups of individuals carrying the same arrangement across localities with the exception of the comparison of AST individuals from Málaga and Rasquera in PhKgamma (see Additional file 2 and Additional file 4).

As low levels of within-population variation – as is the case in PhKgamma and Ubc-E2H – can inflate FST estimates [63, 64] we have used Dxy, which is not affected by such low variation. The results for each gene region were qualitatively similar to those reported above for FST, with consistently higher between-arrangement than within-arrangement differentiation for PhKgamma and Ubc-E2H, although the magnitude of such differences was lower than that obtained with FST estimates (Additional file 2). However, both statistics are highly correlated as shown by a Mantel test for PhKgamma (r2 = 0.754, P = 0.003) and Ubc-E2H (r = 0.741, P = 0.020).

Analysis of genetic differentiation, done separately in coding (exon) and non-coding (non-exon) regions for the four genes concatenated data set for Málaga and Groningen (the two geographically distant localities), showed that the significant differentiation between groups of individuals carrying different arrangements was due solely to variation in non-coding regions (see Additional file 5). Interestingly, these analyses indicated low but significant differentiation for AST individuals from different locations in non-coding regions, but not for A2 individuals. Considering each gene region, we found high differentiation between arrangements in upstream regions for Ubc-E2H and both upstream and intronic regions for PhKgamma while no relevant differentiation was observed for the other two genes in any region analysed (Fig. 5, see also Additional file 5). Curiously, in the intronic region of PhKgamma significant genetic differentiation between AST individuals from different localities was observed (see Additional file 5). A locus-by-locus analysis of Molecular variance (AMOVA) was performed to test the genetic differentiation (FCT) between arrangements at each specific polymorphic site using a hierarchical structure (see Methods). Again, differentiation between arrangements was observed exclusively in non-coding regions (see Additional file 6).

Fig. 5
figure 5

Genetic diversity (π) within each arrangement and Dxy between arrangements along all gene regions. Dxy – black line; π AST – dashed line; π A2 – dotted line. 5’UTR, intronic and exonic regions are discriminated (see also Table 1). The four gene regions are ordered following the AST arrangement (see Fig. 2)

Full size image

On average FST values between arrangements found in the same locality or in different localities showed similar values in both PhKgamma and Ubc-E2H and the concatenated dataset (see Additional file 2 and Additional file 4; see also Figs. 3 and 4). In these regions, differentiation between arrangements was similar across localities with the exception of Rasquera for PhKgamma (see Additional file 7). However, this exception was due to the recombinant individuals from Rasquera (four individuals presenting gene conversion tracks, see below).

Pairwise genetic distances, based on the concatenated dataset of PhKgamma and Ubc-E2H, were correlated with their geographic distance with Mantel tests. No significant association between genetic and geographic distances was found for individuals carrying A2 (r2 = 0.004, P = 0.462), or AST (r2 = 0.017, P = 0.284). Similar results are also obtained for the two genes separately (see Figs. 3 and 4). In agreement with this, no particular haplotype in any arrangement showed consistent indications of directional changes in frequency with latitude.

Gene conversion tracts (GCT) were detected only in PhKgamma and in four individuals, all of them from Rasquera. When removing individuals presenting GCTs in the PhKgamma gene, four fixed differences were detected between arrangements (position 205 in the 5’UTR, and positions 358–360 in the intron). For Ubc-E2H, one fixed difference was found between the two arrangements (position 60 in the 5’UTR). No fixed differences were found in the other two genes.

Linkage disequilibrium

In general, levels of linkage disequilibrium (LD) were low in all gene regions and arrangements analysed (see Additional file 8). Using the concatenated dataset of all four genes and considering both arrangements, LD between pairs of polymorphic sites (R2) showed only 35 significant comparisons out of a total of 2415 (P < 0.001, chi-square with Bonferroni correction). Importantly, the majority of the significant LD comparisons were observed between sites from the same gene region, indicating short-range LD (see also Additional file 9). In fact, only 12 of the 35 significant comparisons corresponded to LD between distinct gene regions - 6 of them between PhKgamma and Ubc-E2H (namely between positions 358-360 bp from PhKgamma and position 60 bp from Ubc-E2H, see below). In the analysis including only the individuals carrying the A2 inversion just 3 out of a total of 496 comparisons indicated significant LD after Bonferroni correction (P < 0.001, chi-square with Bonferroni correction). However, in these three cases, the significant LD comparisons were always between sites within the same gene region (see Additional file 9). No significant comparisons were found between polymorphic sites for AST individuals. Importantly, these results show that within-arrangement LD was not observed across populations. However, fixed associations (complete LD) were found between arrangements and sites in PhKgamma (positions 205, 358–360 of the total alignment with D. pseudoobscura; P < 0.001, chi-square with Bonferroni correction) and also in Ubc-E2H (position 60; P < 0.001, chi-square with Bonferroni correction). Interestingly, all positions in PhKgamma are also in complete LD among themselves (P < 0.001, chi-square with Bonferroni correction). All these sites are located in non-coding/intronic regions. The fixed position in the Ubc-E2H gene is located 547 bp upstream from the first exon. The fixed positions at 358–360 bp in the PhKgamma gene are located in an intronic region just 26 bp downstream of the first exon. Comparison against D. pseudoobscura orthologous gene indicates more similarity to AST sequences. The fixed position between arrangements at 205 bp is within the 5’UTR region 36 bp upstream of the start coding site of PhKgamma, and possibly within the core promoter region, despite the absence of characteristic sequence motifs such as TATA box and the TFIIB recognition element [65].

Neutrality tests

All Tajima’s D values were negative for each gene, arrangement and locality, and also for all locations combined (see Additional file 10). Tajima’s D values were generally more negative in the A2 arrangement. R2 values also indicated stronger deviations from neutrality in the A2 arrangement (see Additional file 10). The general scenario points to an excess of low frequency variants, consistent with a population expansion scenario, particularly in the A2 arrangement.

Discussion

AST and A2 are respectively a cold and a warm-adapted arrangement: while AST shows increased frequencies towards higher, colder latitudes, the opposite pattern is found for A2 [32, 44], with both gene flow and local adaptation likely implicated [7]. In this study we analysed patterns of genetic variation associated with these arrangements and searched for patterns of geographical selection at the genetic level in candidate genes for thermal adaptation [40], located within or near these inverted regions.

Our study shows that these arrangements shape the patterns of genetic variation likely by reducing recombination between certain regions within chromosomes. In fact, the two proximally located regions in AST (PhKgamma and Ubc-E2H) presented significant differentiation between arrangements particularly in non-coding regions, regardless of the arrangement’s geographical origin. Several molecular studies in Drosophila have documented significant genetic differentiation between chromosomes carrying different arrangements, particularly within inverted regions [22, 23, 25, 29, 33, 66] and also specifically near inversion breakpoints (e.g. [14, 21, 23]). The differentiation between arrangements that we observe could result from locally adapted alleles protected within the inversions which would be in agreement with the expectations of both the coadaptation and local adaptation hypotheses [17, 18]. Furthermore, the fact that these differences are observed in 5’UTR and intronic regions might indicate that regulatory regions could be mostly the targets of selection (see [67]). In addition, the low differentiation between arrangements at coding regions - and the low variability observed - might be an effect of purifying selection acting in these regions. However, interpreting these patterns as being due to local adaptation is not straightforward, as neutral processes might also be involved. Non-adaptive associations between the inversion and neutral genetic variants generated during the origin and/or initial spread of the inversion could lead to longstanding differentiation and linkage disequilibrium [68], particularly near breakpoints where recombination is expectedly very low [19, 69]. This signal is more likely to be maintained when population size is high and the inversion is young – see [19]. In our case, despite the high effective size of the species [70] the age of A2 - around 160.000 years [23], makes this expectation less likely. Moreover, the fact that microsatellite loci located in the same (proximal) regions as PhKgamma and Ubc-E2H did not show similar patterns [32] - namely the observed depletion in genetic variation and increase in differentiation between arrangements - further suggests that selection might be involved in the regions analysed here.

In the specific context of the AST/A2 complex, Nóbrega et al. [23] also found that the extent of genetic exchange between arrangements was limited, and not sufficient to homogenize nucleotide variation in the centre of A2. In our study, we show that the reduction of recombination is not homogeneous in regions around the vicinity of the breakpoints of the AST/A2 complex, with significant genetic differentiation and linkage disequilibrium between inversions occurring only in the proximal region of AST. Interestingly, linkage disequilibrium occurred in one region (PhKgamma) located outside the inversion, although near the breakpoint. Other studies have suggested that suppressed recombination can extent to 2.5 Mb or more outside inversion breakpoints [25, 71]. Functional analyses might elucidate if the non-coding variation regions found in PhKgamma and Ubc-E2H is in fact adaptive.

We did not find evidence for geographical (i.e. climatic) selection within arrangements, as very low within-arrangement genetic differentiation was detected across localities. It is possible that the differences in gene expression described previously for these genomic regions between thermal regimes [40] were controlled by other genes/sequences besides the upstream and coding regions analysed here. Importantly, the differential expression [40] was observed at the population level -between populations evolving under cold (13 °C) vs warm (22 °C) regimes - and not addressed at the karyotype level. It is possible that the gene expression in a particular karyotype does not present clinal variation, with differential gene expression occurring between distinct karyotypes. Most interestingly, the expression of PhKgamma and Nipsnap have shown to be positively associated to latitude in semi natural populations of D. subobscura from a similar geographic gradient, but not when the same populations were maintained in a common garden [72]. Thus, although evidence is not conclusive, these gene regions could be involved in local adaptation to temperature.

Other studies in D. subobscura found very limited evidence for geographical molecular variation within inversions. Pegueroles et al. [33] also analysed candidate genes for thermal adaptation located in the O chromosome of D. subobscura and found no significant differentiation within each arrangement for individuals sampled in Barcelona (Spain) and Mount Parnes (Greece). Evidence for similar patterns of generally low within-arrangement differentiation was also described in other D. subobscura studies [73, 74] as well as studies in D. melanogaster ([31], but see [75]) and D. pseudoobscura [20]. However, recent evidence in D. melanogaster indicates several instances of genetic clinal variation in regions linked with inversions particularly associated with the In(3R)Payne [6, 34], perhaps due to the higher resolution power of these later studies in terms of the genomic scan performed and number of sampling sites studied.

Within-arrangement differentiation across localities would be expected if different epistatic interactions evolved (and eventually got fixed) in different localities due to adaptation to the specific local conditions as proposed by the Dobzhansky’s coadaptation model [17]. The lack of within-arrangement differentiation of the selected genes across populations suggests that population-specific epistatic interactions are not occurring in the candidate genes for thermal adaptation that we sampled. It is of course possible that other loci within the inversion might present such varying epistatic interactions in the different localities. Some studies have reported evidence of epistasis based on patterns of low LD between neighbouring sites and high LD between distant sites within the inversion [20, 31]. Recently, indications of epistatic selection have also been obtained from observations of regions with significant genetic differentiation outside arrangements presenting strong linkage disequilibrium with the arrangements. In Atpα gene, another candidate gene for thermal adaptation, Pegueroles et al. [30] found strong genetic differentiation associated to the analysed arrangements of the O chromosome despite this gene being located outside them. Higher resolution genomic scans of adaptive variation as well as functional analyses are needed to conclusively assess the presence of relevant epistatic interactions and if these differ across populations.

Conclusions

Our genetic analysis of candidate loci for thermal adaptation [40] across geographically distinct D. subobscura localities of the European cline found very low within-arrangement differentiation across localities as well as no clinal patterns of genetic variation. This indicates that gene flow across the European continent has homogenized the genetic content of these potentially selected regions at a wide geographical scale. Given the strong effect of gene flow in this species, it is possible that considerable within-arrangement differentiation at a geographical scale is only maintained at relatively few loci subjected to stronger selective pressures in the different environments. In this scenario, one can argue that restricted gene flow cannot explain the observed inversion frequency clines and that local adaptation is maintaining the clinal variation in this species, counteracting the homogenizing effects of the high gene flow. We also found an important effect of the AST/A2 complex in shaping genetic variation and preventing relevant gene exchange in different genomic regions, although its impact was not uniform across the different regions analysed. More generally, our study adds to a body of evidence showing that inversions act as barriers to recombination even outside inversions and can potentially have a role in the spread of favourable genic combinations and consequent adaptation to local habitats.