High-quality reference genome assembly for Heliconius sara magdalena

We assembled the genome of Heliconius sara magdalena using two laboratory-reared females from a stock population originating from Panama (S1 Table). Using the first individual (BioSample SAMEA8947140), we obtained 24 Gbp PacBio CLR data with a 50x coverage and an N50 subread length of 16.5 kb and 122 Gbp of 10X linked-read Illumina data with a 294x coverage. From the second individual (BioSample SAMEA8947139), we obtained 137 Gbp Hi-C Illumina data with a 111x coverage. The final assembled genome consists of 348.8 Mbp in 384 scaffolds. The contig N50 was 8.2 Mbp and that of scaffold N50 was 17.8 Mbp. Our genome showed the highest contiguity (S2 Table) and BUSCO statistics (S3 Table) of Heliconius charithonia, Heliconius erato and H. melpomene genomes. The BUSCO completeness, using the Lepidoptera gene set, achieved 98.2% single-copy BUSCOs and fewer duplicated, fragmented, and missing BUSCOs than in the genomes of H. erato [40], H. melpomene [41] and H. charithonia [42] (S3 Table). We assigned the largest 22 scaffolds to 20 autosomes (one scaffold for each chromosome, except for chromosome 11 which is composed of three scaffolds) and one scaffold to the Z chromosome based on synteny with the Heliconius melpomene genome. H. sara chromosomes are collinear with this genome, as well as with H. erato and H. charithonia (S1 Fig). For more information on the genome see https://tolqc.cog.sanger.ac.uk/durbin/jiggins/Heliconius_sara/ and https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_917862395.2/.

The W chromosome did not assemble well, as is commonly seen in lepidopteran genomes [30]. We thus used whole genome resequencing data from 114 individuals collected in this study (see below “Whole-genome resequencing dataset section”) to assign scaffolds to the W chromosome based on sequencing depth differences between males and females. Among the 360 scaffolds not assigned to a specific chromosome, 32 exhibited a higher mean depth in females than in males in Heliconius sara. This pattern suggests that these 32 scaffolds likely constitute a part of the W chromosome (S2 Fig). Interestingly, the reads from the sapho subclade species (H. antiochus, H. sapho, H. hewitsoni, H. eleuchia, and H. congener) did not align to these 32 scaffolds, suggesting that W chromosome of these species is either too divergent from the H. sara W chromosome (S2 Fig) or not present. These putative W scaffolds in H. sara correspond to a single homolog in H. charithonia (S1C Fig).

Whole-genome resequencing dataset

A total of 114 individuals were successfully whole-genome resequenced. Our dataset exhibits high taxonomic completeness covering all 7 species within the sara/sapho subclade (H. sara, H. leucadia, H. antiochus, H. sapho, H. hewitsoni, H. eleuchia, and H. congener) and 19 out of the 28 described subspecies [43] (S1 Table). The average mapping percentage to the H. sara genome was 95.56% (range: 77.38%–99.14%) (S1 Table). We observed a strong phylogenetic signal in the mapping proportion and, consequently, in the proportion of missing data per individual (S1 Table and S3 Fig). The mapping proportion was 97.7%, 97.6%, and 96.4% for H. sara specimens, its sister species H. leucadia, and the sapho subclade, respectively. One H. congener and one H. antiochus individual exhibited a particularly high proportion of missing data (19.5% and 39.5%, respectively) and low mean depth of coverage (7.4X and 9.1X, respectively) (S1 Table and S3 Fig), and were thus excluded from further analyses.

Phylogenetic analysis reveals two main subclades and uncovered incongruence across the genome

We reconstructed a Maximum Likelihood (ML) phylogenetic tree using all 112 individuals and 183,282,470 concatenated sites. This phylogeny separated individuals into two main subclades, consistent with the PCA analyses (S4 Fig): (i) sara and (ii) sapho (Fig 1). The sara subclade is composed of two species, namely H. sara and H. leucadia, where H. sara is subdivided into an Andean group (H. s. magdalena, H. s. sprucei, and H. s. elektra) and an Amazonian group (H. s. sara). The sapho subclade was split into two well-resolved lineages (H. antiochus and a subclade composed of two monophyletic groups: H. eleuchia/H. congener and H. sapho/H. hewitsoni). H. antiochus appeared as a monophyletic group split into an Andean group (H. a. aranea and H. a. araneides), and an Amazonian group (H. a. antiochus and H. a. salvini) (Fig 1). H. antiochus nested into the sapho subclade, whereas H. hewitsoni was found to be sister to H. sapho, thus resolving the previously undetermined position of these species [38].

To complement the concatenated phylogeny, we also reconstructed a species tree with Astral using males only. The species phylogeny closely mirrored the genome wide phylogeny (S5A Fig). Although H. hewitsoni and H. sapho were recovered as sister species, their branch lengths (measured in coalescence units) (S5A Fig) were short, indicating high levels of discordance. This was also the case for the subspecies of H. eleuchia, where H. e. eleusinus and H. e. primularis sometimes group with H. congener (S5A Fig). These phylogenetic discordances were also evident in the DensiTree analysis visualising 271 phylogenies together (S5A–S5B Fig).

We found strong phylogenetic incongruence across chromosomes. The whole-genome topology was recovered on only eight chromosomes (S6–S27 Figs), while nine chromosomes showed H. congener appearing as sister either to H. e. eleuchia or to a clade composed of H. e. eleusinus + H. e. primularis (S6–S27 Figs). Similarly, H. hewitsoni was not sister to H. sapho on eight chromosomes (S6–S27 Figs). Interestingly, we observed sex-specific clustering on three chromosomes (S27B Fig). On chromosome 4 (Chr4), all species in the sapho subclade (H. antiochus, H. sapho, H. hewitsoni, H. eleuchia, and H. congener) showed females and males forming separate clades within each species, whereas the males of H. congener and H. eleuchia formed a shared clade and their females formed a shared clade (S27B and S9 Figs). Sex-specific clades were also observed on chromosome 9 (Chr9) in the sapho-hewitsoni subclade (S27B and S14 Figs) and on chromosome 14 (Chr14) in the congener-eleuchia subclade (S27B and S19 Figs).

Haplotype-based phylogenetic analysis on chromosomes 4, 9 and 14

The grouping by sex we observed in the phylogenetic trees of Chr4, Chr9 and Chr14 suggests possible fusions between these autosomes and either the Z or W chromosome, or possibly both (females of Heliconius are ZW and males are ZZ) [43]. As females of Lepidoptera lack crossing over and their meiosis is achiasmatic, they do not recombine [46]. This means that if the W chromosome is involved in the fusion (Fig 2A), the Sex-A fusion would be restricted to females and the fused chromosome would tend to accumulate mutations and/or structural variants leading to divergence from its unfused homologue. The unfused Chr4 would become a neo-Z2 chromosome in all species of the sapho subclade, Chr9 would become a neo-Z3 in the sapho-hewitsoni subclade, and Chr14 would become a neo-Z3 in the congener-eleuchia subclade. Alternatively, if the Z chromosome is involved in the fusion (Fig 2B), females would initially still have the unfused homologue (neo-W) that would start to accumulate mutations and/or structural variants, leading to the divergence from the Z-fused homologue. If the sex-autosome fusion was with the Z in the sapho subclade, Chr4 would become neo-W2 in all five species, Chr9 would become neo-W3 in the sapho-hewitsoni subclade and Chr14 would become neo-W3 in the congener-eleuchia subclade (Fig 2B). Lastly, it is possible that the autosomes fused both with the Z and the W, or that some autosomes fused with the Z and others with the W, leading to neo-Ws and neo-Zs, and reducing the number of chromosomes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Scenarios of Sex-A fusions involving either the W or Z chromosomes.

Scenario of (A) WA fusion and (B) ZA fusion for Chr4, Chr9 and Chr14 differentiating the expected pattern by sex: (i) Female-specific haplotype clustering, (ii) low female F_ST, (iii) high female heterozygosity and (iv) low female mean depth. In both fusion scenarios, we would also expect that the unfused chromosomes 4/9/14 would become neo-Zs in the case of W-A fusions or neo-Ws in the case of Z-A fusions. For more details see main text. Yellow: Z chromosome. Pink: W chromosome. Green: autosome. Dark-grey semicircles: fusions. Red dots: mutations and/or structural variants. Lastly, it is possible that both the W and Z chromosomes may have fused with these autosomes, resulting in a neo-W and neo-Z sex chromosome constitution.

https://doi.org/10.1371/journal.pgen.1011318.g002

Under either fusion to the Z or the W, we predict i) genealogies where each female has one haplotype that forms part of the same clade as the male haplotypes, while the other haplotype forms a female-specific clade (hereafter called female-specific haplotype clustering), ii) low genetic differentiation (F_ST) on Chr4, Chr9, and Chr14, due to higher divergence between males and females within populations and lower variation between populations if they share the same sex-autosome fusions (see “Patterns of genetic differentiation”), iii) high proportion of sites where all females are heterozygous due to the presence of two different haplotypes (see “Sex-specific differences in heterozygosity and mean depth”), and iv) low sequencing depth in females due to poor mapping of the female-specific haplotypes that have accumulated mutations and structural variants (see “Sex-specific differences in heterozygosity and mean depth”) (Fig 2).

Consistent with our hypotheses, we identified 218,839 SNPs on Chr4 where all males within the sapho subclade were homozygous and most females were heterozygous (up to one female per species was homozygous). To study the phylogenetic relationships among male and female haplotypes at these sites, we phased our dataset and inferred marginal phylogenies from ancestral recombination graphs constructed using Relate [47]. We subsampled the dataset to every 1000th SNP with high female heterozygosity (214 SNPs). At 31% of these sites, we recovered the expected marginal phylogeny with a female-specific haplotype clade (Fig 3A). The sites displaying this pattern were distributed across the entire chromosome and were not concentrated in a specific region (Fig 3B). Another set of SNPs (53%) exhibited marginal phylogenies where female haplotypes from at least two species within the sapho subclade clustered as expected for a shared Sex-A fusion (S28 Fig). The remaining sites (16%) exhibited a mixed signal similar to the unphased phylogenetic tree obtained for the entire chromosome (S27B Fig). This is likely due to phasing errors where the fused and unfused homologues of the females are mixed. On Chr9, we found 66,344 sites where the species in the sapho-hewitsoni subclade exhibited the expected pattern given a Sex-A fusion. In this case, we found that the 23% of the 66 sampled sites displayed the expected marginal phylogenies (S29A Fig). The remaining 77% of the SNPs showed a mixed signal as in Chr4 (S27B Fig). Finally, on Chr14 of the species in the congener-eleuchia subclade we identified 57,953 sites with the described pattern. Among the 57 subsampled sites, 44% showed the expected genealogical clustering (S30A Fig) while the remaining 56% displayed a mixed signal (S27B Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Marginal tree at a focal SNP inferred from an ancestral recombination graph of Chr4.

Genealogy of the region around SNP 1,900,983 on Chr4 showing a pattern fully consistent with a shared Sex-A fusion ancestral to all five species of the sapho subclade; a total of 66 subsampled SNPs across this chromosome showed the same pattern. Each vertical line represents an individual haplotype, and the haplotypes of all individuals are shown differentiating those of females (orange) from those of males (blue). Note that one allele of the females in the sapho subclade clustered with the alleles of males, while the other female allele formed a separate group (highlighted in orange). (B) Position of each of the 66 focal SNPs on Chr4 that showed genealogies consistent with a Sex-A fusion; they were not clustered in a specific region but rather distributed along the entire chromosome. The position of the SNP 1,900,983 used for the genealogy in A, is indicated by an arrow.

https://doi.org/10.1371/journal.pgen.1011318.g003

Patterns of genomic differentiation

Genomic differentiation (F_ST) was higher between the putative sister species H. sapho and H. hewitsoni than between the sister species H. congener and H. eleuchia, and also higher than the comparison between Andean and Amazonian subspecies in H. antiochus (average F_ST = 0.33, 0.26, and 0.07 respectively) (S31 Fig). H. sara vs. H. leucadia were the least differentiated pair (average F_ST = 0.05) (S31 Fig). We also observed elevated F_ST values on the Z chromosome compared to autosomes in all but one comparison (H. eleuchia vs. H. congener) (S31 Fig). In line with predictions from sex-autosome fusions in the sapho subclade, Chr4 showed lower than average F_ST values in this subclade, but not in H. sara vs. H. leucadia (Figs 4A and S31). This pattern is expected if the females have two highly divergent non-recombining haplotypes (one fused and one unfused), leading to high within-population nucleotide diversity (π) (S32B Fig), even if there is some divergence between populations (D_XY) (S32A Fig). Chr9 showed lower F_ST in sapho-hewitsoni subclade (Figs 4A and S31), and Chr14 in the congener-eleuchia subclade (Figs 4A and S31). The same pattern was observed when we compared F_ST between subspecies (S33–S35 Figs). In line with expectations (Fig 2), the observed pattern of lower F_ST on Chr4, Chr9, and Chr14 was exclusive to females and absent in males across all comparisons mentioned above within all five species of the sapho subclade (Fig 4B). The observed differences in F_ST between males and females were statistically significant for these three chromosomes (Wilcoxon test p < 0.01) (S36 Fig). The F_ST values of these chromosomes within females were also significantly lower compared to those of the other chromosomes (Wilcoxon test p < 0.01) (S36 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Genome-wide F_ST in the sara/sapho clade.

(A) F_ST between pairs of species. Each point represents a 50Kb window, whereby the top 5% windows are shown in black. The numbers below correspond to H. sara chromosomes. Subspecies comparisons are shown in S32–S34 Figs. (B) F_ST between pairs of species by sex. Each circle represents a chromosome, and chromosomes with evidence of Sex-A fusions are colour coded (* indicates outlier chromosomes, p<0.01). The observed reduction in F_ST in females is due to higher genetic diversity within species due to the divergence between the sex chromosome-fused and unfused haplotypes in females.

https://doi.org/10.1371/journal.pgen.1011318.g004

Sex-specific differences in heterozygosity and mean depth

Consistent with our hypotheses (Fig 2), Chr4, Chr9, and Chr14 showed striking sex-specific differences in the proportion of heterozygous sites and mean depth only in species of the sapho subclade, supporting three fusions of these chromosomes with the Z or W chromosomes or both (Fig 5). The strongest difference in the proportion of heterozygous sites was observed on Chr4 where females of H. eleuchia, H. congener, H. sapho, H. hewitsoni and H. antiochus showed a higher proportion of heterozygous sites than males and the other autosomal chromosomes in females (Fig 5B). Females of congener-eleuchia subclade also showed a high proportion of heterozygous sites on Chr14 and in sapho-hewitsoni subclade on Chr9 (Fig 5B). Differences between males and females were significant on these three chromosomes for all species of the sapho subclade (Wilcoxon test, p <0.01) (S37 Fig), except for H. hewitsoni where differences could not be tested due to low sample size. The proportion of heterozygous sites of females was also significantly higher for Chr4 than for the other chromosomes in H. eleuchia, H. sapho, H. antiochus and H. congener (Wilcoxon test, p <0.01) (S37 Fig). The same was true for Chr14 in congener-eleuchia subclade, and Chr9 in H. sapho (Wilcoxon test, p <0.01) (S37 Fig). A high proportion of heterozygous sites was not observed in the females of the species H. sara and H. leucadia on any chromosome (Figs 5 and S37). Chromosome Z exhibited lower proportion of heterozygous sites than autosomal chromosomes in females across all species (Figs 5 and S37).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Genome-wide topology and patterns of heterozygosity and sequencing depth across the genome.

(A) Genome-wide topology with grey squares highlighting nodes with putative Sex-A fusions. (B) The proportion of heterozygous sites vs. mean sequencing depth per chromosome, by sex and in each species. Each circle represents a chromosome, and chromosomes with evidence of Sex-A fusions are colour coded. The high heterozygosity in females is due to the presence of the fusion in only one of the haplotypes, which becomes divergent from its counterpart. The low mean depth is because the haplotype limited to females diverged enough to be difficult to map onto the reference genome of H. sara.

https://doi.org/10.1371/journal.pgen.1011318.g005

Females of H. eleuchia, H. sapho, H. hewitsoni, and H. antiochus also showed a reduced mean depth on Chr4, whereas the mean depth on that chromosome in males was normal (Fig 5B). However, these differences were only significant in H. sapho (Wilcoxon test, p ≤ 0.01) (S38D Fig). This pattern was not true for Chr14 in congener-eleuchia subclade, nor Chr9 for sapho-hewitsoni subclade (Fig 5). The mean depth of Chr4 was also lower than that of all other autosomes in females of H. sapho (Wilcoxon test, p ≤ 0.01) (S37D Fig). However, this was not true for Chr14 and Chr9 (S38 Fig).

The sliding window analyses on Chr4, Chr9, and Chr14 revealed that the excess heterozygosity in females is present in most windows along the entire chromosomes (i.e., it is not concentrated in a specific region on the chromosomes; Wilcoxon test, p <0.01) (Figs 6 and S39–S40). Also, the mean depth values were lower for females than males in most windows on Chr4 for all sapho subclade species except H. hewitsoni, Chr9 for H. sapho and Chr14 for congener-eleuchia subclade (Wilcoxon test, p <0.01) (Figs 6 and S39–S41). As expected, some peaks of high sequencing depth, likely due to expansions of repeats or duplications were also visible in both females and males (Fig 6). Interestingly, one female of H. congener has a region of eight windows on Chr4 with very high sequencing depth (Fig 6), indicating that repeat expansions may still be ongoing. There were also few windows on Chr4, Chr9, and Chr14 where females exhibited both lower mean sequencing depth than males and (almost) no heterozygous sites (highlighted with a grey rectangle) (Fig 6). In these regions, the female-specific haplotype likely diverged too much from the reference genome to map well and thus the heterozygosity is low in females because only one haplotype is represented in the data. H. sara and H. leucadia were the only species in the clade that did not show sex-specific patterns in heterozygosity and mean depth in Chr4, Chr9 and Chr14 (Figs 6 and S39–S41). Interestingly, the heterozygosity in females of H. congener and H. sapho dropped at the end of Chr14 and Chr9, respectively, to values similar to those of the males (S39 Fig), indicating that the last part of these chromosomes may not be fused to a sex chromosome in this species. However, the mean depth pattern did not change in these regions (S39 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Patterns of heterozygosity and depth across Chr4.

Proportion of heterozygous (Het) sites and mean sequencing depth in 50 kb sliding windows for each species. Males are shown in blue and females in orange. Grey rectangles highlight regions where females show very low sequencing depth and no heterozygous sites, suggesting that the female-limited haplotype diverged enough to become unmappable to the H. sara reference genome. The mean depth figures were trimmed to a maximum value of 20 to better visualize the differences between sexes.

https://doi.org/10.1371/journal.pgen.1011318.g006

Evidence for three W-autosome fusions

In order to elucidate if the W or Z chromosome or both are involved in the sex-autosome fusions, we produced Illumina Hi-C data for a female H. congener (0.71 Gbp) and a male H. sapho (0.73 Gbp). We mapped the Hi-C data of these two sapho subclade individuals and the H. sara Hi-C data produced for the reference genome to our H. sara genome and to a previously published H. charithonia genome which has the W chromosome assembled [42]. In line with phylogenetic distances, the proportion of reads mapping to the H. sara reference genome were much higher (96%, 93% and 92%) than to the H. charithonia reference genome (69%, 70%, 71% for H. sara, H. congener and H. sapho, respectively). The mean mapping quality for all three species against both genomes exceeded a Phred quality score of 35.

We did not observe an excess of Hi-C contacts either between autosomes or between autosomes and sex chromosomes in the H. sara female (Fig 7A), as expected if the sex-autosome fusions were only present in the five species of the sapho subclade. The Hi-C signal of the H. sapho male and the H. congener female showed that their genomes are split into 56 and 33 chromosomes, respectively (Fig 7B and 7C), consistent with findings by Brown et al [37] and suggesting a high number of chromosome fissions. However, no fusion was observed in the H. sapho male suggesting that the Chr4 and Chr9 fusions are likely not with the Z, but with the W (Fig 7B). In contrast, the female H. congener showed an excess of contacts between Chr4 and Chr14, in line with a fusion (Fig 7C). There was no excess of Hi-C contacts between Chr4 and Chr14 with the Z chromosome (Fig 7C and 7D), indicating that the Chr4 and Chr14 are likely fused with the W instead. Even though there is some signal of Hi-C contact between these chromosomes and the W, the low mapping rates of Illumina reads of the sapho subclade to the W of the Heliconius sara genome (S2 Fig) likely explains the absence of a stronger signal. The same patterns were also observed when we mapped Hi-C data to the H. charithonia genome (S42 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Hi-C contact density map confirmed a high number of chromosome fissions and W-A fusions in the sapho subclade.

Hi-C contact heatmaps of the H. sara reference genome of (A) the H. sara female, (B) the H. sapho male, and (C) the H. congener female. Note that the chromosomes of the H. sara reference genome are labelled and their boundaries are denoted with vertical and horizontal black lines. The Hi-C signal of the H. sapho male and the H. congener female showed that their genomes are split into 56 and 33 chromosomes, respectively (B-C), consistent with the cytological chromosome counts of Brown et al [37]. (D) Zoom in (C) showing Hi-C contacts on Chr4, Chr14, ChrW, and ChrZ. (E) Hi-C reads assigned to one of the two haplotypes in the H. congener female showing the W-A fusion. (F) Hi-C reads assigned to the other haplotype in the H. congener female showing no evidence of a fusion.

https://doi.org/10.1371/journal.pgen.1011318.g007

As the Hi-C signal of chromosome fusions in the H. congener female represents a mix of signals from the fused and unfused haplotypes (Fig 7C and 7D), we phased the Hi-C data across the chromosomes of interest (Chr4, Chr14, W, Z) and used Chomper [48] to split the Hi-C data into two subsets of read pairs representing the two haplotypes (47.5% and 52.2% of the read pairs in each subset). The excess of Hi-C contacts between Chr4 and Chr14 was completely absent in one haplotype (Fig 7E) and very strong in the other (Fig 7F). The Hi-C contacts between Chr4 and Chr14 were unevenly distributed (Fig 7E) suggesting not only a fusion between these chromosomes but also the presence of other chromosomal rearrangements such as inversions or translocations.

Genome assembly of Heliconius sara magdalena

We used two laboratory-reared females from a stock population from Panama to generate a reference genome for H. sara magdalena (BioSamples SAMEA8947140 and SAMEA8947139; S1 Table). We assembled the genome by combining PacBio, 10X data and Hi-C data, all generated by the Tree of Life Programme at the Wellcome Sanger Institute (https://www.sanger.ac.uk/programme/tree-of-life/). The BioSample SAMEA8947140 was used to generate the PacBio continuous long reads (CLR). Libraries were sequenced on four Single Molecule Real-Time (SMRT) cells using the PacBio Sequel II system. The linked-reads from 10X Genomics Chromium technology were generated with the same sample and sequenced in four lanes on the Illumina HiSeq X Ten platform. The second BioSample SAMEA8947139 was used to produce Dovetail Hi-C data and sequenced on a HiSeq X Ten platform.

An initial contig assembly was generated from the PacBio (CLR) data using wtdbg2 v2.2 [58]. The PacBio data was then used to polish the contigs using Arrow (https://github.com/PacificBiosciences/GenomicConsensus). We then retained haplotig identification with the Purge Haplotigs pipeline [59]. The 10X data were mapped to this assembly using Longranger v2.2 (10X Genomics) and variant calling was performed using freebayes v1.1.0-3-g961e5f3 [60]. Next, this first assembly was polished using BCFtools consensus v1.9 [61] by applying homozygous non-reference calls as edits. The 10X linked-reads were then used to scaffold contigs using Scaff10X v2.3 (https://github.com/wtsi-hpag/Scaff10X). A round of manual curation was performed on these polished scaffolds using gEVAL [62]. Lastly, Dovetail Genomics Hi-C data was used to scaffold the assembly further using SALSA v2.2 [63], followed by another round of manual curation with gEVAL [62]. The chromosome-scale scaffolds were named by synteny to the Heliconius melpomene melpomene assembly Hmel2.5 in LepBase. We assessed the genome contiguity with gnx-tools (https://github.com/mh11/gnx-tools/blob/master/README) and genome completeness with BUSCO v5.1.2 [64] using the Lepidoptera gene set. To obtain synteny plots between H. sara vs. H. melpomene, H. erato, and H. charithonia, we first performed pairwise alignments between these genomes using minimap2 v. 2.24 [65]. Subsequently, we plotted the minimap2 results using custom scripts from (https://github.com/simonhmartin/asynt).

We used whole genome resequencing data from 114 individuals obtained in this study (see sample collection for genome resequencing section) to identify the W chromosome within the genome of Heliconius sara. We calculated the mean depth across the scaffolds that were not yet assigned to a chromosome. We first generated a file containing the mean depth per site averaged across all individuals of the same sex and species using the—site-mean-depth option of vcftools v.0.1.14 [66]. Then, we used the R package windowscanr v. 0.1 (https://github.com/tavareshugo/WindowScanR) to calculate the average of the mean depth per species, per sex, and per 500 bp windows. Scaffolds where we observed a higher mean depth in females compared to males were assigned to the W chromosome.

Sample collection for genome resequencing

We collected 114 Heliconius individuals from 7 species and 18 subspecies in the sara/sapho clade across their distribution range: 48 H. sara, 2 H. leucadia, 21 H. antiochus, 13 H. sapho, 3 H. hewitsoni, 17 H. eleuchia and 10 H. congener (S1 Table). The body of each individual was preserved in NaCl-saturated DMSO solution and stored at -80°C; wings were kept for phenotype reference.

Whole-genome resequencing and genotype calling

Genomic DNA was extracted from thoracic tissue using a DNeasy Blood and Tissue Kit (Qiagen). Library preparation and whole-genome Illumina resequencing (PE reads) was carried out on Illumina’s HiSeq X system by Novogene (Beijing, China), with 30X coverage per individual. We also downloaded two samples of H. charithonia (SRR4032025 –SRR4032026) from SRA (https://www.ncbi.nlm.nih.gov/sra) to include them as outgroups in phylogenetic analyses. Our H. sara genome (HelSar1) was used as a reference to map the reads of each individual using BWA mem v0.7.12 [67] with default parameters. We then used samtools v1.12 to sort and index the alignment files [68]. PCR-duplicate reads were identified and removed using Picard tools v2.9.2 [69], and variant calling was conducted with HaplotypeCaller (GATK, v3.7.0) in BP-resolution mode [70]. Then, samples were jointly genotyped using GATK’s GenotypeGVCFs (68). We used vcftools v0.1.14 [66] and the final VCF to calculate: (i) mean depth per individual and site, (ii) quality per site, (iii) the proportion of missing data per individual and (iv) the proportion of missing data per site, and (v) mapping percentage per individual. Based on these results, we kept sites with quality value (—minQ) ≥ 30 and less than 5% missing data. We also excluded sites with a sequencing depth below 5 and mean depth per individual more than 1.5 times the mean to exclude paralogous regions. For this, we used the custom script removeTooHighDepthSites.sh from (https://github.com/joanam/VictoriaRegionSuperflock/BashPipelines). We additionally removed sites with excess heterozygosity across all individuals using the vcftools option—hardy and a p-value cut-off of <1e-5 to remove reads from paralogous regions that are collapsed in the reference genome.

Analysis of population structure within the sara/sapho clade

We performed a principal component analysis (PCA) to study the genetic structure of populations. We filtered out monomorphic or multiallelic sites, and sites with minor allele frequency (MAF) smaller than 0.1 with vcftools [66]. To reduce the linkage disequilibrium effect, we used the python script ldPruning.sh from (https://github.com/joanam/scripts), which removes sites with r²>0.2 in windows of 50 Kbp sliding by 10 Kbp. This resulted in a vcf file with 3,685,916 high-quality SNPs sites. We conducted the PCA using Plink v2.0 with default parameters [71,72].

Phylogenetic relationships among sara/sapho clade species

We generated a whole-genome Maximum Likelihood (ML) tree using a vcf containing all sites as input in RAXML v8.2.9 [73], with the GTRGAMMA model and 100 bootstrap replicates. We applied the same procedure to obtain ML trees for each chromosome to study the phylogenetic incongruence across the genome. We also inferred the species tree using the coalescent-based method ASTRAL [74]. For this, we used vfctools v0.1.14 [66] to: (i) extract two males per subspecies so Sex-A fusions present only in females do not alter the species tree, and (ii) extract 2 kbp loci spaced at least 10 kbp apart to ensure no linkage disequilibrium between them [75]. Then, samtools [68] was used to generate 271 multilocus blocks, each resulting from concatenating 100 loci. Each block was converted into PHYLIP format using our custom script vcf2phylip.py from (https://github.com/joanam/scripts/blob/master/vcf2phylip.py) and used to estimate a ML tree in IQ-tree selecting the best model with ModelFinder and assessing node support with 1000 ultrafast bootstraps [76,77]. The resulting 271 topologies were used as input in ASTRAL. We also investigated these 271 topologies with DensiTree [78] to visualize discordance.

Haplotype-based phylogenetic analysis on chromosomes 4, 9 and 14

To further investigate sex clustering in the phylogenies of autosomes 4, 9 and 14, we phased them to infer their haplotypes. Haplotype phasing was done by combining two methods: WhatsHap, which is a haplotype assembly technique [79], and SHAPEIT4, a statistical phasing method [80]. To implement WhatsHap, we used the BAM file of each individual as well as the reference genome to group nearby genetic variants into fully resolved haplotype blocks or phase sets [79]. Then, we used the WhatsHap output file to run SHAPEIT4, which further phases haplotypes based on population-level information using default parameters [80]. Despite combining these different phasing approaches, our dataset still included phasing errors leading to haplotype switching in the females between the W-linked chromosomes and the neo-Z chromosomes. We were thus not able to generate phylogenies of the complete chromosomes and instead used Relate v.1.1.2, a method that uses short-range phasing information to infer phylogenies.

With Relate v.1.1.2 [47] we inferred ancestral recombination graphs (ARG) along chromosomes 4, 9, and 14. This software generates genealogies by first identifying the relative order of coalescence events at each genomic position using a nonsymmetric distance matrix and considering a mutation rate and a recombination map. This matrix is constructed based on posterior probabilities from a hidden Markov model (HMM), which considers ancestral and derived SNP status to enhance accuracy. Subsequently, a rooted binary tree is constructed from this distance matrix [47]. Initially, the phased vcf file was transformed into haplotype format using RelateFileFormats with the “—mode ConvertFromVcf" flag. We performed this analysis using an effective population size of 1x10⁷ individuals, as estimated for H. erato [81], and a mutation rate of 2.9x10^-9 per site per generation from H. melpomene [82]. As Relate further requires a genetic map, we created a uniform recombination rate map using the average recombination rate of 6 cM/Mb calculated in H. erato [83], and our custom script (https://github.com/joanam/scripts/blob/master/createuniformrecmap.r). Finally, the ancestral allele was assigned as the one more common in the outgroup H. charithonia.

Because any possible Sex-A fusion would produce heterozygous females and homozygous males, we identified sites with this pattern (see hypotheses in Fig 2). We first calculated the number of heterozygous and homozygous individuals per species and site using the—hardy option in vcftools v0.1.14 (46). Then, we used custom scripts to find sites where each species in the sapho subclade met the following criteria: (i) no males were heterozygous, (ii) at least one female was heterozygous, and (iii) not more than one female was homozygous (allowing for one female with one allele not called). Finally, we selected one SNP for every 1000 SNPs from these filtered sites and visualized their genealogies using the script TreeView.sh (58). This subsampling approach enabled us to examine genealogies from various sites evenly distributed along the chromosomes.

Patterns of genetic differentiation

We calculated F_ST by pairs of sister species along chromosomes in non-overlapping 50 Kbp windows. Because H. antiochus did not have a sister species, we calculated these statistics between Andean and Amazonian subspecies. Windows that contained less than 2,500 high-quality genotyped variable sites were rejected. We used a dataset including SNPs and monomorphic sites and the popgenWindows.py script from (https://github.com/simonhmartin/genomics_general). We also calculated F_ST per sex following the same methodology.

Patterns of heterozygosity and mean depth by chromosome

To study chromosomes with F_ST patterns different from the genome average, we used the options—het and—depth of vcftools v.0.1.14 [66] to calculate heterozygosity and mean depth per chromosome for each individual of each species. We also calculated these statistics in 50 Kbp non-overlapping sliding windows along the ‘outlier’ chromosomes identified. On these, we calculated π specifying each individual as its own population, so π became a measure of proportion of heterozygous sites. This was done with the Python script popgenWindows.py from (https://github.com/simonhmartin/genomics_general). We then averaged these values across all individuals of the same sex and species. For sequencing depth, we first generated a file containing the mean depth per site averaged across all individuals of the same sex and species using the—site-mean-depth option of vcftools v. 0.1.14 [66]. We then used the R package windowscanr v.0.1 from (https://github.com/tavareshugo/WindowScanR) to calculate the mean of the mean depth per species, per sex, and per window. The few individuals of H. hewitsoni and H. leucadia were not included in the sliding windows analysis. Statistical tests were applied to assess significant differences in heterozygosity and mean depth between sexes and between chromosomes. As the data were not normally distributed, we performed a Wilcoxon signed rank test to compare the sexes. To assess differences between chromosomes, we applied a Kruskal-Wallis test and a post hoc test (pairwise Wilcoxon test for Kruskal-Wallis).

Identification of fusions with Hi-C data

To investigate whether autosomes 4, 9, and 14 are fused with the W or Z chromosomes, we constructed Hi-C libraries from the thorax of one female of Heliconius congener (BioSamples SAMEA112329098; S1 Table) and a male of Heliconius sapho (BioSamples SAMEA112696452, S1 Table) using the Arima2 kit (Arima Genomics, Inc). These libraries were then sequenced on an Illumina NovaSeq S4 platform with 150 bp paired-end reads. We used BWA mem2 v 2.2.1 [84] to map the reads against the genomes of Heliconius sara (this study) and Heliconius charithonia [42] using default parameters. We also mapped the Hi-C data of H. sara to both references to confirm the absence of the fusion in the sara subclade. Next, we removed PCR duplicates, eliminated poorly aligned reads, and filtered out reads with a mapping quality <10 from the resulting BAM files using samtools v1.12 [68]. Additionally, we generated contact maps with pretextview and pretextSnapshot (https://github.com/wtsi-hpag/PretextView and https://github.com/wtsi-hpag/PretextSnapshot). All these steps were done using a custom Perl pipeline developed by Shane McCarthy at the Wellcome Sanger Institute.

To further investigate the excess of Hi-C contacts between autosomes 4, 14 and W in the female of H. congener, we generated haplotype-specific Hi-C maps for these chromosomes. First, we created a version of the H. sara reference genome in which chromosomes 4, 14, and W were concatenated together to allow for phasing across chromosomes. Next, we mapped the Hi-C reads of H. congener to this modified reference following the mapping pipeline by Arima Genomics, Inc. (https://github.com/ArimaGenomics/mapping_pipeline). Then we called heterozygous variants using freebayes v1.3.2-dirty [60]. These variants were then normalized with bcftools v1.8 [61], decomposed with vcfallelicprimitives [85] and filtered for coverage (>21 and <141 reads) with vcftools v. 0.1.14 [66]. Next, the remaining SNPs were phased using HAPCUT2 v1.3.3, using both the bam and the vcf files as input [86]. We used the Python script chomper.py from to separate haplotype aligned Hi-C reads [48]. Finally, these haplotype-specific sets of Hi-C reads were realigned to the original H. sara assembly using a custom Perl pipeline developed by Shane McCarthy at the Wellcome Sanger Institute. Hi-C contact map was generated using pretextview and pretextSnapshot (https://github.com/wtsi-hpag/PretextView and https://github.com/wtsi-hpag/PretextSnapshot).