We sequenced a mitochondrial region of 1572 bp representing 799 bp of CoI, 62 bp of the tRNA^LEU and 711 bp of CoII from 11 individuals. Including sequences from GenBank, the alignment had 1572 nucleotide sites examined from 69 individuals, of which 155 (10%) were variable. We obtained 584 bp of the Z-linked Triose phosphate isomerase exon 3 (31 bp), intron 3 (430 bp) and exon 4 (123 bp), for 11 alleles from 11 individuals. Including sequences from GenBank the total alignment had 25 alleles from 22 individuals. Identity was confirmed by comparison with reference sequences for H. c. chioneus (AF413788) and H. m. rosina (AF413790).

The four autosomal regions were portions of nuclear loci initially chosen for their potential role in the development of butterfly wing patterns 30. Three of these, Distal-less, invected and scalloped, are transcription factors 313233, while the fourth, white, is member of the ommochrome biosynthesis pathway that generates the yellow, orange and red pigments in Heliconius 34. However, linkage mapping has shown that none of these genes are linked to the switch genes that control pattern divergence in Heliconius 3536, so there is currently no evidence that they are involved in pattern evolution. We included 558 bp of Distal-less, corresponding to exon 4/5 (175 bp), intron 5 (333 bp) and exon 6 (50 bp) of Drosophila (NM166689- intron 4 is absent in Heliconius) for 32 alleles from 20 individuals; 439 bp of invected exon 2 (67 bp), intron 2 (265 bp) and exon 3 (107 bp) for 42 alleles from 23 individuals. We obtained 499 bp of white exon 4 (49 bp), intron 4 (397 bp) and exon 5 (53 bp) for 25 alleles from 17 individuals; 483 bp of the scalloped gene from exon 7 (13 bp), intron 7 (334 bp) and exon 8 (136 bp) for 22 alleles from 12 individuals.

None of the three species formed a monophyletic group at all of the genes sampled. In the phylogeny derived from mitochondrial DNA, individuals of the three species fell into three well-supported monophyletic clades (Figure 1): 1) an eastern melpomene clade; 2) the cydno clade including all the H. cydno and H. heurippa sampled and seven H. m. melpomene from the eastern Andean foothills; 3) a western melpomene clade. Within the cydno clade, the H. heurippa haplotypes form a monophyletic group with five fixed differences from H. cydno (0.92% net divergence). Genetic diversity in H. heurippa was the lowest of the three species (Table 1). Tajima's D was not significantly different from zero (Table 1), suggesting that a small but constant population size rather than a recent bottleneck is more likely to explain the lack of variation in H. heurippa. Interestingly Tajima's D estimates were significantly negative for two of the three populations of H. melpomene (H. m. rosina and H. m. mocoa) included here for comparison (Table 1), possibly reflecting a recent bottleneck or mtDNA selective sweep.

Among the five nuclear loci, the sex-linked locus Tpi was the only marker that clearly separated all three species. H. c. cordula and H. m. melopomene alleles formed distinct clades separated by five fixed differences and one shared polymorphism, with 1.3% net divergence (Figure 2A). H. heurippa alleles also formed a distinct cluster (Figure 2A) separated by five fixed differences from H. m. melpomene (1.3% net divergence) and by six from H. c. cordula (1.4% net divergence). In concordance with the network groups, F_ST values showed the species as three distinct populations, with H. heurippa showing greater differentiation from H. c. cordula (F_ST = 0.791) and H. m. melpomene (F_ST = 0.719) than that observed between H. c. cordula and H. m. melopomene (F_ST = 0.498 P < 0.05; Table 3; Figure 2A).

Three of the autosomal loci, Dll, inv and w, show a striking pattern in which H. c. cordula and H. m. melpomene are clearly differentiated, but H. heurippa shares variation with both species. At Dll, H. c. cordula and H. m. melpomene (Figure 2B) are separated by 10 fixed differences (3% net divergence) and share three polymorphisms. Similarly at inv, H. c. cordula and H. m. melpomene have ten fixed differences (4.6% net divergence) and two shared polymorphisms (Figure 2C). At w, one allele of H. m. melpomene was shared with H. c. cordula (M6-1; Figure 2D), such that there was only one fixed difference (2% net divergence) and eight shared polymorphisms. Nonetheless, estimates of F_ST between H. c. cordula and H. m. melpomene were high for all three loci (Dll, 0.621, inv, 0.593 and w, 0.327). In contrast, H. heurippa did not have any fixed differences with either H. c. cordula or H. m. melpomene at any of these three loci. At Dll, H. heurippa was more similar to H. m. melpomene (net divergence 0.41%) than H. c. cordula (net divergence 2.28%). At the other two loci, H. heurippa was more similar to H. cydno, with inv showing 0.56% and 2.76% divergence with H. c. cordula and H. m. melpomene respectively, while w showed 0.024% and 1% for the same two comparisons. Estimates of F_ST supported these observations, with Dll showing H. heurippa not significantly differentiated from H. m. melpomene but strongly differentiated from H. c. cordula (Table 3), while inv and w showed the opposite pattern with H. heurippa more similar to H. c. cordula (Table 3).

In contrast with the patterns observed in the other five loci, the sd locus showed no fixed differences among any of the three species (Figure 2E). H. c. cordula and H. m. melpomene had seven shared polymorphisms and net divergence of 0.62%. H. heurippa shared 14 and ten polymorphisms with H. c. cordula and H. m. melpomene respectively, representing net divergence of 0.10% and 0.5%. Genetic diversity was generally high in all three species (Table 2). Tajima's D was significantly negative in H. c. cordula (Table 2). F_ST values showed that H. heurippa was more similar to H. c. cordula than to H. m. melpomene (Table 3).

In order to select the 'basic' dataset required by the IM software, indel-free alignments were investigated to search for evidence of recombination for each species pair comparison 37. In the H. cydno – H. melpomene comparison, recombination was only detected at Dll with its maximum significant breakpoint found between sequences C6-2 and M5-2 (p = 0.007) between 16^th and the 17^th sites, as was indicated using the maximum chi-square method. A recombination free block of 206 bp was selected in the 5' to 3' direction from the site of probable recombination (Table 4).

In the H. cydno – H. heurippa comparison, inv had recombination (p = 0.015) with the most significant breakpoint found between sequences H3-1 and C8-2 between 52^nd and 53^rd sites. A non recombining block of 250 bp preceding the sites involved in the recombination was selected, in the 5' to 3' direction (Table 4). In this comparison, sd also showed significant recombination between sequences H1-1 and H4-1 with the most significant breakpoint at the 16^th and 17^th sites (p = 0.0001). For this locus, a 308 bp region (5' to 3') lacking recombination was included in further analysis (Table 4). In the H. melpomene – H. heurippa comparison, the only locus with recombination was inv in which, the exchange was between the sequences H7-2 and M7-1, with the most significant breakpoint between 12^th and 13^th sites (p = 0.005). A 125 bp 5' to 3' region after the sites involved in recombination was selected for the IM analysis (Table 4).

Isolation-Migration model 38 was used to infer the population history of H. melpomene melpomene and H. cydno cordula. The IM program uses a coalescent model to estimate effective population size, time since divergence, and ongoing migration parameters from multilocus sequence data sampled from two sister species. Here, H. cydno cordula was estimated to have a two-fold greater population size compared to H. melpomene melpomene (Table 5). This result is consistent with other studies involving different H. cydno and H. melpomene geographical races 3739. The ancestral population size was 1.23 × 10⁶ (0.76 × 10⁶-1.98 × 10⁶) individuals and the speciation event probably took place around 2 million years ago, similar to a previous estimate using a different dataset 37. However, because of the weak scalar estimation for this parameter, this estimated time is a very approximate value (Table 5).

Two loci (mtDNA and w) showed evidence of gene flow. In both cases it was asymmetric (Table 5). All remaining genes had estimates asymptotically near to zero in both directions (Table 5). The estimated migration rate from H. c. cordula to H. m. melpomene (m₁ = 1.32 × 10^-6 per generation per locus) was high at mtDNA, as was expected from the pattern of shared variation seen in the mtDNA tree. Gene flow estimated in the other direction was nearly zero (Table 5). The w locus had a 0.61 × 10^-6 per generation per locus migration rate in the same direction, a result that is consistent with the presence of the M6-1 H. melpomene haplotype in the H. cydno allele cluster.

The IM model was then used to compare H. heurippa with each of the parental taxa. The estimated H. heurippa population size was similar (0.27 – 0.30 × 10⁶) in each comparison and smaller than those estimated for the other species (Table 5). Unlike the comparison between H. m. melpomene and H. c. cordula, both comparisons involving H. heurippa failed to yield precise estimates of either the ancestral population size or the divergence time. The likelihood surfaces obtained for these parameters were either flat or rising over the parameter range investigated.

More importantly here, was the incidence of gene flow observed at several loci between H. heurippa and the other species, the only exceptions being mtDNA and Tpi, where the migration rate was not significantly different to zero in both directions (Table 5). At all other loci migration parameters were significantly greater than zero in at least one direction, with evidence for gene flow both from parental species into H. heurippa and vice-versa (Table 5).

Heliconius heurippa was initially identified as a putative hybrid species based on its intermediate colour pattern, which shows a striking similarity to phenotypes produced after just a few generations of hybridization between H. c. cordula and H. m. melpomene 26. The sequence data presented here provides independent evidence that hybridization has played an important, and ongoing, role in the evolution of the H. heurippa genome. All four autosomal loci showed a pattern in which H. heurippa shares similar alleles with both H. m. melpomene and H. c. cordula. At three of these loci (inv, w and sd), H. heurippa was most closely related to H. c. cordula, while at the fourth (Dll) it was closer to H. m. melpomene. Nonetheless, our analysis suggests that the pattern of shared alleles between H. heurippa and its relatives is mainly due to introgression in the fairly recent past. The results are consistent with historical gene exchange playing an important role in the origin of H. heurippa, but do not provide evidence for a 'mosaic' hybrid genome as has been demonstrated in other examples of hybrid speciation.

The comparison of H. c. cordula and H. m. melpomene under the IM model indicates ongoing introgressive hybridization at two loci (Table 5). In particular, there are very closely related mtDNA haplotypes shared between the species (gene flow from H. c. cordula to H. m. melpomene, m₁ = 1.32 × 10^-6 per generation per locus; Figure 1) and asymmetric moderate gene flow in w (m₁ = 0.61 × 10^-6 per generation per locus, gene flow from H. c. cordula to H. m. melpomene and 0 in the other direction). These species are known to hybridize in the wild and previously, shared alleles have been observed at another autosomal locus, Mpi for which a symmetrical migration rate of 1.54 ×10^-6 per generation per locus was estimated in Panamanian populations 3740. Furthermore, substantial shared DNA sequence variation at 16 loci was observed in H. cydno and H. melpomene from Costa Rica 39.

Notably, shared mtDNA variation between H. cydno and H. melpomene in eastern Colombia (Figure 1) suggests recent introgression. The shared mtDNA haplotypes may have been retained as an ancient polymorphism since the speciation of H. melpomene and H. cydno, but this seems unlikely given the evolutionary distance between the species (1.5–2% divergence between the two mtDNA clades). Female hybrid sterility following Haldane's rule 41 would be expected to prevent introgression of mtDNA, although limited fertility of female hybrids has been documented and would provide a route for infrequent introgression 41.

Pairwise IM analyses of H. heurippa and either one of the other taxa, suggest a small species population size (Table 5). The absence of consistent negative Tajima's D across loci (Table 2) indicates a historically small constant size instead of a short bottleneck. No reliable estimates for divergence time and ancestral population size were obtained in either comparison. Two loci mtDNA and Tpi did not show gene flow for any pair evaluated that involves H. heurippa. The remaining loci showed recent gene interchange among H. heurippa and the two other species. There is no hybrid sterility between H. cydno and H. heurippa, and insectary mate choice experiments produce frequent matings in one direction (between H. heurippa males and H. cydno females). It seems likely that there is a hybrid zone between these species somewhere in the Andes between Villavicencio (Colombia) and San Cristobal (Venezuela; Linares Pers. Obs.), which would explain the observed levels of gene flow.

In the H. melpomene – H. heurippa species pair, inv (64) and w (348.2) in particular showed high rates of gene flow from H. melpomene to H. heurippa. These two species are broadly sympatric, which might facilitate hybridization. Nonetheless, although hybrid females are sterile in one direction of cross, female offspring of backcrossed males segregate for sterility which should allow some gene flow 24.

In a broad sense, our Isolation-Migration analysis suggests that different parts of the H. heurippa genome are subject to very different levels of gene flow with the two sister taxa. We also carried out linkage disequilibrium tests for introgression 42, which should be sensitive to very recent introgression events, but these were not significant for any of the loci, suggesting that gene flow is sufficiently ancient for linkage disequilibrium to have been lost. There are a number of possible explanations for such interlocus variation. One is that some of the loci sequenced for this study are either themselves subject to natural selection or else are linked to such regions. Selection does seem plausible as we know that there is a strong correlation between the Tpi locus and hybrid sterility in interracial H. melpomene, H. cydno x H. melpomene and H. melpomene x H. heurippa crosses 244143. If Tpi were associated with genes causing hybrid sterility, this might explain the clear lack of gene flow among the three species at this locus. This is also consistent with the fact that a disproportionate number of species differences, including morphology, physiology, oviposition preference, mate selection and pheromones, map to the Z chromosome in other Lepidoptera species 44.

At the other extreme, sd shows far more allelic mixing between species than the other loci studied. A similar pattern in the cydno-melpomene group races from Panama and Costa Rica has been observed at other loci 3739. This has led to suggestion that balancing selection could be maintaining diversity and perhaps promoting introgressive hybridization 373940. It is possible that this process is occurring at the sd haplotypes obtained here. There is currently no evidence for the role of sd, or any of the other loci studied here playing a role in the evolution of colour pattern, so it is unclear what the selection pressures on this locus might be.

In addition, analysis of protein coding sequence did not provide any direct evidence for positive or balancing selection on any of the loci, although the power of this analysis was limited by the fact that most of our sequence data is for non-coding regions (analysis using CODEML, data not shown, 45). The genealogical pattern and IM analysis do suggest that there is both ancestral variation and recent gene flow at the sd locus (Table 5).

Overall, the species relationships are consistent with the H. heurippa genome containing a greater contribution from H. cydno than H. melpomene. The H. heurippa mtDNA haplotypes fall within an H. cydno clade, and at three of the five nuclear loci F_ST values show H. heurippa closer to H. cydno (Table 3). This is consistent with what is known about the three species. H. heurippa is a geographic replacement of H. cydno that flies in similar habitats. Where H. cydno flies sympatrically with H. melpomene, the former is associated more with closed canopy forest and tends to be found at higher altitudes 46, both characteristics also observed in H. heurippa populations in eastern Colombia (Linares and Jiggins Pers. Obs.). Furthermore, laboratory reconstruction of the H. heurippa colour pattern from H. c. cordula and H. m. melpomene involves backcrossing F₁ male hybrids to H. cydno, with a correspondingly greater contribution from the H. cydno genome. Finally, patterns of hybrid sterility show that H. heurippa is more compatible with H. cydno 2426. Nonetheless, the Tpi genealogy contrasts with this general pattern, in t