We sampled nine populations that were separated by distances between a minimum of 42 km between VAL and CUN and a maximum of 920 km between VFE and CER (Table 1, Figure 1). In the 2009 and 2010 seasons, a total of 153 of M. arion adults were captured, and a middle left leg was removed from each individual. After this procedure, the butterflies were immediately released, and the collected legs were stored in 99.9% ethanol. For every population, 5 to 25 samples were obtained.

a.s.l. – above sea level. LHP – larval host plant.

Sampling sites were divided into three categories: woodland clearings on upland localities (350 - 600 m above see level) where M. a. ligurica exploits Origanum vulgare as LHP (CUN, LOA, BDR); mountain pastures in the Alps (1400 - 2200 m a. s. l.) where M. a. obscura is dependent on Thymus spp. (VAL, VFE, CDF); and clearings or pastures in the Apennines (1000 - 1200 m a. s. l.) where M. a. arion uses Thymus spp. (AUR, CER, CET) (Table 1, Figure 1).

We used DNA isolated from the legs or thoraxes of specimens previously collected for microsatellite analyses 35 36 . A total of 110 samples were analysed; all were from the main areas of M. arion’s distribution in the country and the sampling included eleven Polish populations. The distances between the Polish populations ranged from 8 km at a minimum (KLU and SRO) to a maximum of 460 km (GUG and SRO). The localities can be divided into two types: dry grasslands (often clearings in pine forests) on sandy flat areas where Thymus serpyllum was used as LHP and xerothermic grasslands on southerly exposed slopes with T. pulegioides (Table 1, Figure 1).

Genomic DNA was extracted using a Genomic Mini Kit (A&A Biotechnology, Gdańsk, Poland). The Ron and Hobbes primers were used for the COI mtDNA fragment amplification, and ef51.9 and efrcM4 primers were used for the EF-1α gene 37 . PCR was performed on a GeneAmp PCR System 9700 (Applied Biosystems, Foster City, CA, USA) using a 5 μL reaction volume containing multiplex PCR Master Mix (1x) (Qiagen, Hilden, Germany), 0.2 μm of each primer, approximately 10-20 ng of genomic DNA and RNase-free water. Each PCR consisted of an initial activation step at 95°C for 15 min; 40 - 45 cycles of denaturation at 94°C for 30 s, annealing at 57°C for 90 s and extension at 72°C for 60 s; and a final extension at 60°C for 30 min. Direct sequencing was performed using the BigDye™ Terminator Cycle Sequencing Ready Reaction Kit 3.1 (Applied Biosystems). Both primers were used for sequencing both strands. The sequencing products were run on an ABI PRISM 3130 capillary automated sequencer (Applied Biosystems).

In addition, PCR to amplify the 16S rDNA to test for the presence of Wolbachia in M. arion was performed with the same amplification protocols using the W-Specf and W-Specr primers 38 . We screened all sampled populations, choosing 2 to 3 randomly selected individuals from each.

DNA sequences were aligned in BioEdit v 7.0.4 39 and revised manually. Haplotype reconstruction of the nuclear EF-1α gene from genotype data was conducted using the algorithms provided in PHASE as implemented in DnaSP 5.0 40 . The number of haplotypes (N _h ), haplotype diversity (h), nucleotide diversity (π), and the number of segregating sites were calculated using ARLEQUIN 3.11 41 . Relationships among the COI and the EF-1α haplotypes were represented as a haplotype network obtained by the statistical parsimony method using the TCS software 42 . Population divergence estimates were obtained as F _ST values in ARLEQUIN 3.11 41 . The significance of pairwise F _ST values for the COI and the EF-1α genes was ascertained by 1000 permutations. The isolation by distance (IBD) pattern was tested by comparing the genetic differentiation between populations, as measured by pairwise F _ST /(1- F _ST ), to the logarithm of geographical distance using IBD software 43 . Analysis of molecular variance (AMOVA) 44 was performed using ARLEQUIN 3.11 41 (with 10000 permutations) to assess structuring within the data, and sampling sites were grouped as a single population (using Φ _ST ). To explore patterns of genetic divergence in more detail, we applied the spatial AMOVA procedure using SAMOVA ver. 1.0 45 . This allowed us to identify the groupings of sampling sites that maximised the F _CT value based upon 10000 simulated annealing steps.

To test the phylogeographic structuring of haplotype distributions, we compared the average G _ST values (based on haplotype frequencies) to N _ST (based on haplotype frequencies and distance between haplotypes), as described by Pons and Petit 46 , using DnaSP 5.10.01 39 .

Mitochondrial DNA sequences (COI) were obtained for the 804-bp fragment in 225 individuals. We found twelve (eight singletons) haplotypes (GenBank accession nos. KC316050 - KC316061), which were defined by thirteen variable sites (eleven transitions and two transversions). The haplotypes differed by one to five substitutions. Possible relationships between the COI mtDNA haplotypes were estimated as a network, as shown in Figure 1. The network exhibited a star-shaped topology with H1 being the ancestral haplotype. This haplotype was the most common and most widespread, found in 18 out of 20 populations studied and only absent in two Italian sites (CER and CET). Generally, eleven COI haplotypes were present in Italy, ranging from one (CUN) to four (AUR and VFE) in any given population. However, all but one Polish population of M. arion were fixed for the same haplotype (H1). Moreover, H1 was the only haplotype shared between the two countries (Table 2 and Additional file 1: Table S1).

N – sample size, h – haplotype and gene diversity, number of haplotypes (N _h ) is given in parentheses, π – nucleotide diversity. COI – cytochrome oxidase subunit I (mtDNA), EF-1α – elongation factor (nuclear gene). For full site names and other details see Table 1.

For the pooled samples, the average COI mtDNA nucleotide (π) and haplotype (h) diversity values were estimated at 0.12% and 0.480, respectively. In the Italian populations, the average π and h were 0.17% and 0.651, respectively. For M. arion in Poland, the corresponding values were much smaller (π = 0.003%, h = 0.024) because the polymorphism was found only in the SRO population that had only two haplotypes (one being a singleton) (Table 2 and Additional file 1: Table S1). For the populations studied, π ranged from 0.00% to 0.38 and h ranged from 0.000 to 0.733 (Table 2).

All samples, regardless of the population origin, were positive for Wolbachia by PCR analysis. Our 16S rDNA sequences (334-bp, GenBank accession no KC337261 - KC337262) differed by a single substitution from the Wolbachia sequence found in the chrysomelid beetle Diabrotica virgifera 47 and therefore can be assigned to supergroup A with the highest probability.

Nuclear DNA sequences (EF-1α) were obtained by comparing a 460-bp fragment from 263 individuals. Altogether, 26 segregating sites (consisting of 29 substitutions, 20 transitions and 9 transversions) comprised the 30 haplotypes (17 singletons) (GenBank accession nos. KC316062 - KC316091) in the combined samples. Of these 30 haplotypes, the Italian populations contained 18 (4.56 haplotypes per population on average) and the Polish contained 17 (average per population 3.73) Table 2 and Additional file 1: Table S2). The possible relationships between the EF-1α haplotypes were estimated using a network, shown in Figure 2. The network exhibited a star-shaped topology, with H1 and H5 being the two ancestral haplotypes. The EF-1α haplotypes differed from one to six substitutions.

Only five out of the 30 EF-1α haplotypes (17%) were shared between the two countries. For all populations studied, the haplotype numbers ranged from two in the Mt. Cervati (CER), Sowlany (SOW) and Piaski (PIA) populations, up to seven in the Colle delle Finestre (CDF) populations and eight in the Hutki Kanki (HUT) population (Table 2 and Additional file 1: Table S2). The average π for the pooled sample was 0.21% (range: 0.02% - 0.47) and haplotype diversity was 0.632 (range: 0.094 – 0.743, Table 2). For Italy, the average π and h were 0.22% and 0.542, respectively. For M. arion in Poland, the corresponding values were similar (0.19% and 0.609, respectively).

Based upon the COI genetic data, there was a statistically significant and high degree of genetic differentiation among the 20 M. arion populations studied (F _ST  = 0.436, P < 0.001). The pairwise F _ST values between the populations were highly variable and ranged from 0.000 to 0.955 (Figure 3, Additional file 1: Table S3). In Italy, we observed very high and statistically significant population differentiation (average F _ST  = 0.432, P < 0.001; range: 0.000 – 0.955). However, in Poland, the average genetic differentiation at the COI gene among all the populations was very low and not statistically significant (F _ST  = 0.005, P > 0.05; range: 0.000 - 0.054).

The average genetic differentiation among all 20 populations of M. arion for the EF-1α gene was also high (F _ST  = 0.179) and statistically significant (P < 0.001). The pairwise F _ST values ranged from 0.000 to 0.678. The average pairwise F _ST value between Italian populations was high and statistically significant (F _ST  = 0.176, P < 0.001; range: 0.000 – 0.499). Moderate and significant, genetic differentiation was found among the Polish populations (average F _ST  = 0.059; P < 0.001; range: 0.000 – 0.392) (Figure 3, Additional file 1: Table S3). No pattern of isolation by distance (IBD) was detected among Italian populations of M. arion, with respect to either the COI gene (r ² = 0.02, P > 0.05) or for the EF-1α gene (r ² = 0.03, P > 0.05) nor among Polish populations with respect to the EF-1α gene (r ² = 0.01, P > 0.05). The IBD analysis was not performed on the Polish samples for the COI gene because the majority of pairwise F _ST values were zero.

The geographical structuring among the M. arion haplotypes was highly supported by AMOVA results, for which all sampling sites were treated as a single group (Φ _ST  = 0.438, P < 0.001 for the mtDNA COI gene; Φ _ST  = 0.195, P < 0.001 for the EF-1α gene). When the Italian and Polish samples were treated as two separate groups, the AMOVA was not significant for the COI gene (Φ _CT  = 0.073, ns). However, significant structuring between Italy and Poland was found for the EF-1α gene (Φ _CT  = 0.122, P < 0.01). In Italy, no significant structuring was observed with respect to the LHP for either marker (mtDNA: Φ _CT  = 0.089, ns; EF-1α: Φ _CT  = 0.054, ns). The same result was obtained when the Italian populations were regrouped as lowland (up to 600 m) and highland (over 600 m).

SAMOVA was used separately on the Italian and Polish populations to identify which subdivision was most likely to explain the genetic structure observed in M. arion for the COI and EF-1α genes. For the COI gene in Italy, the data were best explained by assuming four groups of populations. The first group consisted of the AUR and CET samples; group two included populations from the BDR, LOA, CUN, VAL and VFE; and groups three and four were represented by the CER and the CDF populations, respectively. The percentage of variation among these four groups was high at 48.43% (P < 0.001); among the populations within the groups, the variation was 3.26% (P < 0.001), and within the populations, the variation was 48.31% (P < 0.001). The maximum percentage of variation for the EF-1α gene in Italy (28.34%, P < 0.05) occurred under the assumption of two population groups (CUN and VAL compared to all other Italian samples), while the variation among populations within groups was 5.37% (P < 0.001) and within populations was 66.29% (P < 0.001). The analysis of the Polish samples revealed no grouping of populations, and the EF-1α gene data were best explained by assuming a single population group (Φ _ST  = 0.059, P < 0.001). No such analysis was performed for the mtDNA on Polish samples because of the almost complete lack of polymorphism.

For the EF-1α gene, the average N _ST value (0.156) was significantly (P < 0.001) higher than the average G _ST value (0.076) among 20 populations studied, indicating a phylogeographic structuring of M. arion haplotype distributions. The same pattern was also found when the Italian samples (N _ST  = 0.143, G _ST  = 0.050; P < 0.001) and the Polish samples (N _ST  = 0.056, G _ST  = 0.026; P < 0.001) were analysed separately. Interestingly, both the N _ST and G _ST values for the EF-1α gene were significantly higher among Italian samples than among Polish ones (P < 0.001). No phylogeographically significant structuring was found for the mtDNA in Italy (N _ST  = 0.319, G _ST  = 0.262; ns).

Perfectly matching rear edge theory 7 , this study revealed great genetic differentiation among M. arion Italian populations, both at the COI mtDNA and the nuclear EF-1α genes. This is in contrast to the almost total absence of mtDNA polymorphism and the low to moderate genetic differentiation at the nuclear gene found for M. arion populations inhabiting Poland, a recently deglaciated area. Moreover, the rear edge populations (Italy) possessed as many as 11 out of the 12 mtDNA haplotypes recorded in this study; thereby, emphasising their relevance in the context of biodiversity conservation and, most importantly, from the evolutionary perspective.

The mtDNA diversity in Italy may be thus an example of 'southern richness’ as opposed to ‘northern purity' as defined by Hewitt 1 3 48 , e.g. high polymorphism and divergence in the south and low variation and lack of divergence in the north. “Southern richness” was also visible for the nuclear gene EF-1α as the average number of alleles per population and the genetic differentiation among them was significantly larger than among Polish populations.

Thus, our data do not support the alternative scenario that suggests lower genetic diversity in populations at the edge of a species’ distribution compared to that in the centre 26 27 . The high genetic diversity found in Southern Europe can be attributed to prolonged demographic stability in the populations occurring in these areas 1 and by the high geographic structuring of populations evolving in allopatric conditions 49 50 . In Italy, prolonged demographic stability was assumed to create a common history for many taxa and, recently, allopatric differentiation (refugia-within-refugia) was described in Bombina pachypus 51 . The identification of four distinct genetic groups for mtDNA as well as similar N _ST and G _ST values emphasize the role of the genetic drift as the main force responsible for the differentiation among Italian populations. In Italy, we found more structuring and greater genetic differentiation for the COI gene than for the EF-1α gene. Mitochondrial markers are characterised by much faster lineage sorting rates and more frequent haplotype extinctions than the bi-parental nuclear markers. Consequently, as is widely recognised for many Lepidoptera e.g., 52 53 , differentiation in mtDNA among populations is usually much stronger than in nuclear DNA (e.g., EF-1α).

Although our data on mitochondrial genetic material suggested a significant population genetic structure, they did not support the subdivision of the populations into macrogeographic groups (Poland versus Italy). The most common mtDNA haplotype in our samples (H1) is indeed the most widespread throughout the species’ range because it was recorded also from Catalonia, Germany, Denmark, Sweden, Estonia, Finland, the Czech Republic, Slovakia, Romania, Hungary, Ukraine, Russia and Kazakhstan. Moreover, the second most common haplotype (H5), which we found in five Italian sites (CER, AUR, VAL, CDF, VFE), is known to occur in Catalonia, the Czech Republic and Sweden 28 . As a result, our mtDNA data do not fully support the colonisation pattern hypothesized by Ugelvig 28 . The author suggests the existence of three different genetic groups in the Western Palaearctic originating from different southern refugia in the Iberian Peninsula, the Balkans and Asia and thus three postglacial colonisation routes. The Alps are known to have acted as an initial barrier to the possible expansion of Italian genetic pools towards Northern Europe for many species 3 9 , but our data do not allow us to exclude Italy as a possible source area for Central and Northern European M. arion populations. In Poland, we almost exclusively found a single haplotype, which was also the most frequent in northern Italy (H1). Of course, a “European” origin of north Italian populations cannot be ruled out either. This would mean that the dispersion from the East has reached northern Italy. Our results have indeed shown that the ancestral haplotype (H1) is dominant in northern Italy, but is rare (AUR) or absent (CET, CER) in central and southern Italy. Ugelvig 28 suggests that, in the case of M. arion, Italy was colonised from Iberia. It should be noted, however, that only two samples from Italy were analysed, while our extensive study showed that Italy was rather colonised from two different routes.

We detected three unique and closely related mtDNA haplotypes (H3, H4, H6) in southern Italy. A similar pattern was found in Italy for the Euscorpio carpathicus complex, where Salomone et al. 54 concluded that the presence of ancestral haplotypes in the northern part of the country meant that other refugia, outside of the Apennine, may have existed for these taxa. Possible sources are the Balkans, which northern Italy was connected to by the Adriatic bridge during the quaternary cold periods 2 , and the Iberian Peninsula perhaps via Col di Tenda, as was the case for the western lineage of Melanargia galathea 55 . However, both directions (i.e., also from Italy towards Iberia) are possible, as it was suggested for Polyommatus coridon 5 . This conclusion is supported by our data showing the presence of one haplotype (H5) in Catalonia, in the Alpes Maritimes (VAL is 20 km away from Col di Tenda) and all along the Italian latitudinal gradient.

On the other hand, the presence of unique haplotypes in southern Italy concurs with the hypothesis that, contrary to common belief, most relic rear edge populations may not have been the source of major postglacial re-colonisations, thereby preserving their high genetic distinctiveness 7 . Rear edges, as ‘relic hotspots’, may be considered as regions of special conservation concern, as they often coincide with centres of high biodiversity and endemism. To fully resolve M. arion phylogeographical pattern further studies should be based on a much higher number of specimens, especially coming from other South European areas as well as central populations.

M. arion showed moderate to high levels of diversity in the nuclear gene EF-1α. Altogether, 30 haplotypes were found, 18 occurring in Italy and 17 in Poland. Interestingly, a study carried out on M. alcon in a closely overlapping area found only six haplotypes among the 179 individuals sampled 32 . Our results from the EF-1α gene support the allozyme and microsatellite data 56 57 by showing that M. arion exhibits a rather high nuclear genetic variability when compared to other members of the genus Maculinea 33 35 58 . It has been suggested that the specialised parasitic life style of Maculinea butterflies makes them prone to genetic drift and therefore to a loss in genetic variation 58 . However, M. arion is considered a less advanced social parasite because it preys on ant larvae and shows a lower level of host ant specificity. In this case, it is possible that the effective population size of M. arion is higher than in other members of the genus, especially M. alcon and M. ‘rebeli’. In this regard, it is interesting to observe that recent genetic data obtained on Danish and Swedish populations suggest that M. arion is more dispersive and less sensitive to genetic erosion than was previously assumed 59 60 .

The AMOVA analysis of nuclear genes, through EF-1α, showed significant genetic differentiation between the Italian and Polish populations. This may indicate a different population origin occurred in both countries and some indications of this can be found among 24 sequences available in Genbank. A total of twelve EF-1α haplotypes in addition to our samples (30 haplotypes) were deposited by Als et al. 58 . In the pooled sample, the most common haplotype (H1) was shared by all populations in Poland and Italy and has been found in Slovakia, Sweden and Spain. The H5 haplotype was the most common haplotype throughout Poland and network suggests its expansion here. It was also present in six out of nine Italian populations, although never as a dominant haplotype, and has been identified in Slovakia and Sweden as well. This suggests that Poland was not colonised exclusively from northern Italy. Poland and Italy shared five out of the 30 EF-1α haplotypes and one of them (H8) was also found in Slovakia 58 . The presence of two common haplotypes in Poland may indicate that this part of Europe was colonised from two different routes (from the east and west). Finally, the H15 haplotype, which we identified only in northern Italy (CDF), was also found in Spain. The latter finding (and the presence of H1 in both countries that seems to be ancestral for the majority of Italian haplotypes; see network), may corroborate the existence of relationships between the Iberian and the Italian peninsula, as indicated by mtDNA and postulated by Ugelvig 28 .

The low to moderate population differentiation observed in Poland for the EF-1α gene concurs with results from a previous study involving microsatellite markers that was conducted on the same DNA material; in which, rather low levels of differentiation for the species were determined, although some populations were highly isolated and the genetic differentiation was indicated to be mainly shaped by habitat fragmentation 36 . Interestingly one population (HUT), where suitable biotope for M. arion was extensive, yielded the highest level of polymorphism in both EF-1α and microsatellite markers 36 .

We did not find any evidence for IBD for any population grouping or for molecular marker. The lack of any correlation between geographic and genetic distances can be explained by historical demographic processes or cycles of colonisation and extinction. A similar pattern was identified for some Apennine populations of Bombina pachypus by Canestrelli et al. 51 .

Our genetic data do not support any subspecies divisions in M. arion. Among the Italian samples there were representatives of both M. a. ligurica (LOA, CUN, BDR) and M. a. obscura (VFE, CDF, VAL), while the remaining Italian populations (CET, AUR, CER), as well as all Polish populations belonged to M. a. arion. The analysis of the bar code sequence showed that two of the three populations belonging to M. arion ligurica were characterised by fixed or almost fixed haplotype H1, which was also fixed in 10 of 11 Polish populations and almost fixed in the remaining population. No species or subspecies-level differentiation was detected also as far as EF-1α gene is concerned. Hence, the results of our study are in agreement with the allozyme analysis of Hungarian populations indicating no differences between Origanum and Thymus dependent populations 57 . This suggests that, despite the high ecological and morphological variability which led to the description of several subspecies, M. arion is a species with an absence of deep historic population isolation. Sielezniew & Dziekańska 61 hypothetise that clinal adaptation is a more likely explanation for the observed wing pattern variation in this butterfly than incipient speciation.

It is worth noting that in the case of two other Maculinea taxa (M. alcon and M. ‘rebeli’), molecular analysis revealed no differences supporting division into separate species 33 62 and possible selective sweeps cannot be ruled out as well 33 . A lack of host-associated genetic differentiation in the barcoding gene COI is reported by Craft et al. 63 from New Guinea, where 7 out of 28 analysed Lepidoptera species exhibited the same haplotypes despite the use of multiple LHPs. A similar pattern was observed by Hulcr et al. 64 in the moth Homona mermerodes.

We can speculate that use of different LHPs by a locally monophagous butterfly species, such as M. arion, has not been a mean to promote speciation, but can be ascribed to local adaptations in the species’ phenology that have evolved recently. For instance, when one of the glacial refugia was localised at the southern base of the Alps, after deglaciation, some populations could expand their range to higher altitudes tracking hosts re-colonisation, while others were able to survive in the lowlands by shifting to a novel host plant 11 . If this scenario is true, the M. a. ligurica populations may represent a relic of M. arion populations that survived at the base of the Alps and were a source for the re-colonisation of the Alps.

All the populations in this study were infected by Wolbachia, and screening of all our samples suggested 100% prevalence of a single strain of the bacterium (supergroup A). The most common effect of a Wolbachia infection is cytoplasmic incompatibility (CI), which results in infected males that are unable to successfully reproduce with either uninfected females or females infected by a different strain. As a consequence, the distribution of mtDNA variation in Wolbachia infected populations may not conform to the expectations of the neutral theory 34 65 , even though endosymbiotic infections may not explain the divergent haplotype groups within some butterfly populations 66 . The star-like network of haplotypes found in our study is typical for a rapid expansion model 67 , but could also be shaped by selective sweeps of Wolbachia, as hypothesised for Polygonia c-album 68 .

The congeneric Maculinea alcon (including the xerothermophilous ecotype ‘M. rebeli’) is also reported to have Wolbachia infection 33 . However, a different strain (supergroup B) was detected in samples of this species collected across Poland and Lithuania. Sielezniew et al. 33 hypothesise that a selective sweep could have reduced mtDNA diversity in the past, as only a single COI mtDNA haplotype was found in that area. Interestingly, our study indicates great genetic differentiation for mtDNA among studied populations in Italy, all infected by the same strain of the bacterium. This could mean that Wolbachia is not responsible for the observed pattern of genetic differentiation among the studied M. arion populations. It is probable that the Wolbachia infection in M. arion populations preceded both the expansion from glacial refugia and ecological specialisation.