Butterfly genome reveals promiscuous exchange of mimicry adaptations among species.

The evolutionary importance of hybridization and introgression has long been debated. Hybrids are usually rare and unfit, but even infrequent hybridization can aid adaptation by transferring beneficial traits between species. Here we use genomic tools to investigate introgression in Heliconius, a rapidly radiating genus of neotropical butterflies widely used in studies of ecology, behaviour, mimicry and speciation. We sequenced the genome of Heliconius melpomene and compared it with other taxa to investigate chromosomal evolution in Lepidoptera and gene flow among multiple Heliconius species and races. Among 12,669 predicted genes, biologically important expansions of families of chemosensory and Hox genes are particularly noteworthy. Chromosomal organization has remained broadly conserved since the Cretaceous period, when butterflies split from the Bombyx (silkmoth) lineage. Using genomic resequencing, we show hybrid exchange of genes between three co-mimics, Heliconius melpomene, Heliconius timareta and Heliconius elevatus, especially at two genomic regions that control mimicry pattern. We infer that closely related Heliconius species exchange protective colour-pattern genes promiscuously, implying that hybridization has an important role in adaptive radiation.

The butterfly genus Heliconius (Nymphalidae: Heliconiinae) is associated with a suite of derived life-history and ecological traits, including pollen-feeding, extended life-span, augmented ultraviolet colour vision, ‘trap-lining’ foraging behavior, gregarious roosting and complex mating behaviours, and provides outstanding opportunities for genomic studies of adaptive radiation and speciation[4, 6]. The genus is best known for the hundreds of different colour pattern races seen among its 43 species, with repeated examples of both convergent evolution among distantly related species and divergent evolution between closely related taxa[3]. Geographic mosaics of multiple colour pattern races, such as in Heliconius melpomene (Fig. 1), converge to similar mosaics in other species, and this led to the hypothesis of mimicry[2]. Heliconius are unpalatable and Müllerian mimicry of warning colour patterns enables species to share the cost of educating predators[3]. Divergence in wing pattern is also associated with speciation and adaptive radiation due to a dual role in mimicry and mate selection[3, 5]. A particularly recent radiation is the melpomene-silvaniform clade, where mimetic patterns often appear polyphyletic (Fig. 1a). Most species in this clade occasionally hybridise in the wild with other clade members[7]. Gene genealogies at a small number of loci indicate introgression between species[8], and one non-mimetic species, H. heurippa, has a hybrid origin[9]. Adaptive introgression of mimicry loci is therefore a plausible explanation for parallel evolution of multiple mimetic patterns in the melpomene-silvaniform clade.

Figure 1

Distribution, mimicry and phylogenetic relationships of sequenced taxa

a Phylogenetic relationship of sequenced species and subspecies in the ‘melpomene- silvaniform clade’ of Heliconius. H. elevatus falls in the ‘silvaniform’ clade, but its colour pattern mimics melpomene-timareta clade taxa. b Geographic distribution of ‘postman’ and ‘rayed’ H. melpomene races studied here (blue, yellow and purple), and the entire distribution of H. melpomene (grey). The H. timareta races investigated have limited distributions indicated by arrows (red) and mimic sympatric races of H. melpomene. H. elevatus and the other silvaniform species are distributed widely across the Amazon basin (Supplementary Information 22).

A Heliconius melpomene melpomene stock from Darién, Panama (Fig. 1) was inbred via five generations of sib mating. A single male was sequenced to 38x coverage (after quality filtering) using combined 454 and Illumina technologies (Supplementary Information 1-8). The complete draft genome assembly of 269 Mb consists of 3,807 scaffolds with an N50 of 277 kb and contains 12,657 predicted protein-coding genes. RAD linkage mapping was used to assign and order 83% of the sequenced genome onto the 21 chromosomes (Supplementary Information 4). These data permit a considerably improved genome-wide chromosomal synteny comparison with the silkmoth Bombyx mori[10, 11]. Using 6,010 orthologues identified between H. melpomene and B. mori we found that 11 of 21 H. melpomene linkage groups show homology to single B. mori chromosomes and ten linkage groups have major contributions from two B. mori chromosomes (Fig. 2a and Supplementary Information 8), revealing several previously unidentified chromosomal fusions. These fusions on the Heliconius lineage most likely occurred after divergence from the sister genus Eueides[4], which has the lepidopteran modal karyotype of n=31[12]. Three chromosomal fusions are evident in Bombyx (Fig. 2a, B. mori chromosomes 11, 23 and 24), as required for evolution of the Bombyx n=28 karyotype from the ancestral n=31 karyotype. Heliconius and Bombyx lineages diverged in the Cretaceous >100 MYA[11], so the chromosomal structures of Lepidoptera genomes have remained highly conserved compared to those of flies or vertebrates[13, 14]. In contrast, small-scale rearrangements were frequent. In the comparison with Bombyx, we estimate 0.05-0.13 breaks/Mb/MY, and with the Monarch butterfly, Danaus plexippus, 0.04-0.29 breaks/Mb/MY. Although lower than previously suggested for Lepidoptera[15], these rates are comparable to Drosophila (Supplementary Information 8).

Figure 2

Comparative analysis of synteny and expansion of the chemosensory genes

a Maps of the 21 Heliconius chromosomes (above) and of the 28 Bombyx chromosomes (below in grey) based on positions of 6010 orthologue pairs demonstrate highly conserved synteny and a shared n=31 ancestor (Supplementary Information 8). Dotted lines within chromosomes indicate major chromosomal fusions. b Maximum likelihood tree showing expansion of the chemosensory protein genes (CSP) in two butterfly genomes.

The origin of butterflies was associated with a switch from nocturnal to diurnal behaviour, and a corresponding increase in visual communication[16]. Heliconius have increased visual complexity through expression of a duplicate UV opsin[6], in addition to the long wavelength, blue, and UV-sensitive opsins in Bombyx. We might therefore predict reduced complexity of olfactory genes, but in fact Heliconius and Danaus[17] genomes have more chemosensory proteins (CSPs) than any other insect genome: 33 and 34 CSPs respectively (Supplementary Information 9), versus 24 in Bombyx and 3- 4 in Drosophila[18]. Lineage-specific CSP expansions were evident in both Danaus and Heliconius (Fig. 2b). In contrast, all three lepidopteran genomes possess similar numbers of odorant binding proteins and olfactory receptors (Supplementary Information 9). Hox genes are involved in body plan development and show strong conservation across animals. We identified four additional Hox genes located between the canonical Hox genes pb and zen, orthologous to shx genes in B. mori (Supplementary Information 10)[19]. These Hox gene duplications in the butterflies and Bombyx share a common origin, and are independent of the two tandem duplications known in dipterans (zen2, bcd). Immunity-related gene families are similar across all three lepidopterans (Supplementary Information 11), contrasting with extensive duplications and losses within dipterans[20].

The Heliconius reference genome enabled rigorous tests for introgression among melpomene-silvaniform clade species. We used RAD resequencing to reconstruct a robust phylogenetic tree based on 84 individuals of H. melpomene and its relatives,sampling on average 12 Mb, or 4% of the genome (Fig 1a, Supplementary Information 12, 13, 18). We then tested for introgression between the sympatric co-mimetic postman races of H. melpomene aglaope and H. timareta ssp. nov. (Fig. 1) in Peru, employing ‘ABBA-BABA’ single nucleotide sites and Patterson’s D-statistics (Fig. 3a), originally developed to test for admixture between Neanderthals and modern humans[21, 22] (Supplementary Information 12). Genome-wide we found an excess of ABBA sites, giving a significantly positive Patterson’s D = 0.037 ± 0.003 (two tailed Z-test for D = 0, P = 1 × 10−40), indicating greater genome-wide introgression between the sympatric mimetic taxa H. m. amaryllis and H. timareta ssp. nov., than between H. m. aglaope and H. timareta ssp. nov., which do not overlap spatially (Fig. 1b). These D-statistics yield an estimate of 2-5% of the genome exchanged[21] between the two taxa (Supplementary Information 12). Eleven of the 21 chromosomes have significantly positive D-statistics (Fig. 3b,); interestingly, the strongest signals of introgressions were found on two chromosomes containing the known mimicry loci B/D and N/Yb (Fig. 3b, Supplementary Information 15).

Figure 3

Four-taxon ABBA-BABA test of introgression

a ABBA and BABA nucleotide sites employed in the test are derived (– – B –) in H. timareta, compared with the silvaniform outgroup (– – – A), but differ among H. melpomene amaryllis and H. melpomene aglaope (either ABBA or BABA). As this almost exclusively restricts attention to sites polymorphic in the ancestor of H. timareta and H. melpomene, equal numbers of ABBA and BABA sites[22] are expected under a null hypothesis of no introgression, as depicted in the two gene genealogies. b Distribution among chromosomes of Patterson’s D-statistic ± s.e., which measures excess of ABBA vs. BABA sites[22], here for the comparison H. m. aglaope; H. m. amaryllis; H. timareta ssp. nov.; silvaniform. Chromosomes containing the two colour pattern regions (B/D red; N/Yb yellow) have the two highest D-statistics; the combinatorial probability of this occurring by chance is 0.005. The excess of ABBA sites (0 < D < 1) indicates introgression between sympatric H. timareta and H. melpomene amaryllis.

Perhaps the best known case of Müllerian mimicry is the geographic mosaic of ~30 bold postman and rayed colour pattern races of H. melpomene (Fig. 1b, Supplementary Information 22), which mimic a near-identical colour pattern mosaic in H. erato (Fig. 1a), among other Heliconius. Mimicry variation is generally controlled by a few loci with major effects. Mimetic pattern differences between the postman H. melpomene amaryllis and rayed H. melpomene aglaope races studied here (Fig 1a) are controlled by the B/D (red pattern) and N/Yb (yellow pattern) loci[23, 24]. These loci are located on the same two chromosomes showing the strongest D- statistics in our RAD analysis (Fig. 3b). To test whether mimicry loci might be introgressed between co-mimetic H. timareta and H. melpomene (Fig. 1a)[7], we resequenced the colour pattern regions B/D (0.7 Mb) and N/Yb (1.2 Mb), and 1.8 Mb of unlinked regions across the genome, from both postman and ray-patterned H. melpomene and H. timareta from Peru and Colombia, and six silvaniform outgroup taxa (Fig. 1a, Supplementary Information 12). To test for introgression at the B/D mimicry locus we compared rayed H. m. aglaope and postman H. m. amaryllis as the ingroup with postman H. timareta ssp. nov. (as in Fig. 3a) and found large, significant peaks of shared fixed ABBA nucleotide sites combined with an almost complete lack of BABA sites (Fig. 4b). This provides evidence that blocks of shared sequence variation in the B/D region were exchanged between postman H. timareta and postman H. melpomene, in the genomic region known to determine red mimicry patterns between races of H. melpomene[23, 24] (Fig. 4a).

Figure 4

Evidence for adaptive introgression at the B/D mimicry locus

a Genetic divergence between H. melpomene races aglaope (rayed) and amaryllis (postman) across a hybrid zone in N.E. Peru. Divergence, F, is measured along the B/D region (Supplementary Information 14). F peaks in the region known to control red wing pattern elements between the genes kinesin and optix[23]. b, c Distribution of fixed ABBA and BABA sites (see Fig. 4a) along B/D for two comparisons. Excesses of ABBA in b and BABA in c are highly significant (two-tailed Z-tests for D = 0; D = 0.90 ± 0.13, P = 5 × 10−14 and D = –0.91 ± 0.10, P = 9 × 10−24 respectively), indicating introgression. d, e, f, Genealogical change along B/D investigated with maximum likelihood based on 50 kb windows. Three representative tree topologies are shown. Topology A, the species tree, is found within the white windows. In topologies B (dark green window) and C (light green windows) taxa group by colour pattern rather than species. Within striped windows H. melpomene and/or H. timareta are paraphyletic but the taxa do not group by colour pattern. Support is shown for nodes with > 50% bootstrap support (Supplementary Information 19).

For a reciprocal test, we used the same H. melpomene races as the ingroup to compare with rayed H. timareta florencia at the B/D region. In this case, correspondingly large and significant peaks of BABA nucleotide sites are accompanied by virtual absence of ABBA sites (Fig. 4c) indicating that variation at the same mimicry locus was also shared between rayed H. timareta and rayed H. melpomene. Equivalent results in the N/Yb colour pattern region, controlling yellow colour pattern differences, are in the expected directions for introgression and highly significant for the test using postman H. timareta ssp. nov. (P = 6 × 10−34), although not significant with rayed H. timareta florencia (P = 0.13, Supplementary Information 17). In contrast hardly any ABBA or BABA sites are present in either comparison across 1.8 Mb in 55 genomic scaffolds unlinked to the colour pattern regions (Supplementary Information 21). These concordant, but reciprocal patterns, where fixed ABBA and BABA substitutions occur almost exclusively within large genomic blocks at two different colour pattern loci (449 and 99 sites for B/D and N/Yb respectively, Figs. 4b,c and Supplementary Information 17) would be very hard to explain via convergent functional site evolution or under coalescent fluctuations. Instead, our results imply that derived colour pattern elements have introgressed recently between both rayed and postman forms of H. timareta and H. melpomene.

To test whether colour pattern loci might be shared more broadly across the clade, we used sliding-window phylogenetic analyses along the colour pattern regions. For regions flanking and unlinked to colour pattern loci, tree topologies are similar to the overriding signal recovered from the genome as a whole (Supplementary Information 18). Races of H. melpomene and H. timareta each form separate monophyletic sister groups and both are separated from the more distantly related silvaniform species (Fig. 4d). By contrast, within the region of peak ABBA/BABA differences, the topologies switch dramatically. Races of H. melpomene and H. timareta group according to wing pattern, while the species themselves become polyphyletic (Figs. 4e,f, Supplementary Information 19, 20). Remarkably, the rayed H. elevatus, a member of the silvaniform clade according to genome average relationships (Fig. 1a, Supplementary Information 18), groups with rayed races of unrelated H. melpomene and H. timareta in small sections within both B/D and N/Yb colour pattern loci (Fig. 4e, Supplementary Information 19, 20). These results are again most readily explained by introgression and fixation of mimicry genes.

We have developed a de novo reference genome sequence that will facilitate evolutionary and ecological studies in this key group of butterflies. We have demonstrated repeated exchange of small (~100 kb) adaptive genome regions among multiple species in an adaptive radiation. Our genome-scale analysis provides considerably greater power than previous tests of introgression [8, 25, 26]. As with H. heurippa[9], our evidence suggests that H. elevatus was formed during a hybrid speciation event. The main genomic signal from this rayed species places it closest to H. pardalinus butleri (Fig. 1a), but colour pattern genomic regions resemble those of rayed races of H. melpomene (Fig. 4e and Supplementary Information 18-20). Colour pattern is important in mating behaviour in Heliconius[5], and the transfer of mimetic pattern may have enabled the divergent sibling species H. elevatus to coexist with H. pardalinus across the Amazon. Although it was long suspected that introgression might be important in adaptive radiation[1], our results from the most diverse terrestrial biome on the planet suggest that adaptive introgression is more pervasive than previously realized.

Methods summary

A full description of methods can be found in the Supplementary Information.

Supplementary Information is linked to the online version of the paper at www.nature.com/nature.