Abstract
Pheromones are important for courtship and mate choice in many insects, but we know relatively little of their role in butterflies. The butterfly Heliconius melpomene uses a complex blend of wing androconial compounds during courtship. Electroantennography in H. melpomene and its close relative H. cydno showed that responses to androconial extracts were not species-specific. Females of both species responded more strongly to the H. cydno extract, suggesting conservation of peripheral nervous system elements across the two species. Individual blend components provoked little to no response, with the exception of octadecanal, a major component of the H. melpomene blend. Supplementing octadecanal on the wings of octadecanal-rich H. melpomene males led to an increase in the time until mating, demonstrating the bioactivity of octadecanal in Heliconius. Using quantitative trait locus (QTL) mapping, we identified a single locus on chromosome 20 responsible for 41% of the parental species’ difference in octadecanal production. This QTL does not overlap with any of the major wing color or mate choice loci, nor does it overlap with known regions of elevated or reduced FST. A set of 16 candidate fatty acid biosynthesis genes lies underneath the QTL.
Introduction
Chemical communication is the oldest form of sensory communication, and is found across the tree of life from bacteria (Taga & Bassler 2003) to humans (Jacob et al 2002; Wysocki & Preti 2004). In Lepidoptera, work on pheromones has largely focused on female pheromones of nocturnal moths, where relatively simple alterations to pheromones can produce drastic differences in mate attraction. For example, different populations of the corn borer moth Ostrinia nubialis exhibit a simple cis to trans switch in the pheromone 11-tetradecenyl acetate (Kochansky et al 1975) that leads to partial reproductive isolation between them (Lassance et al 2010). Sex pheromones have been less well studied in day-flying butterflies, where visual signaling is often assumed to play a more dominant role in mate choice (Vane-Wright & Boppré 1993). However, male butterflies also emit close-range pheromone bouquets which may act in concert with other wing pattern and behavioral cues (Mérot et al 2015), and are important in mate choice (Darragh et al 2017).
Despite the potential importance of pheromones in butterflies, pheromones have so far been identified in only a handful of butterfly species (Meinwald 1969, Pliske & Eisner 1969, Grula et al 1980, Nishida et al 1996, Schulz & Nishida 1996, Andersson 2007, Nieberding et al 2008, Yildizhan et al 2009). Within butterflies, the genetic basis of pheromone production is known only in Bicyclus anynana (Lienard et al 2014). As in other diurnal butterflies, the chemical bouquets of Heliconius (Darragh et al 2017, Mann et al 2018), are often complex, both in identity and quantity of compounds; however, just a few individual compounds may be biologically active pheromones. Identifying which components of these complex chemical bouquets are responsible for pheromonal communication is a considerable challenge. This is particularly true as even minor compounds can have major effects (McCormick et al 2014, Chen et al 2018). Determining pheromone bioactivity requires screening compounds via physiological activity followed by behavioral verification, and in many cases pheromone bouquet composition is known but the bioactive components remain unidentified.
To better characterize male butterfly pheromone production and the role it plays in mating and species recognition, we first comprehensively analyzed pheromone bouquets of the closely related Heliconius melpomene and H. cydno butterflies to determine the most important bioactive compounds. We further mapped a genetic locus likely responsible for the differential biosynthesis of the main fatty-acid compound between the two species. The two species are closely related yet strongly reproductively isolated (Jiggins 2017). Over the past decade there has been considerable research into genomic architecture of differences in wing pattern and mate preference (Jiggins 2017; Merrill et al 2019). Surprisingly, both wing color and mating preferences between these species have a relatively simple genetic basis with a large proportion of the difference between parental forms being controlled by a small handful of loci of large effect (Jiggins 2017, Merrill et al 2019). This has important implications for speciation, as theory predicts that large effect loci should facilitate speciation in the face of gene flow (Via 2012). Similarly, clustering of genetic loci that contribute to speciation can also make speciation easier (Merrill et al 2010, Smadja & Butlin 2011), and there is evidence for tight linkage of a gene controlling wing pattern and a QTL underlying divergent preference behaviors (Merrill et al 2019). Here, we identify an electrophysiologically active compound, octadecanal, explore its behavioral role, and investigate the genetic basis of its production in H. melpomene using QTL mapping.
Results
Androconial chemical bouquets of Heliconius melpomene and H. cydno
In order to identify candidate pheromone compounds, we first investigated the distribution of chemical compounds on the wings of H. cydno for comparison with published data on H. melpomene. We found that the bouquet of H. cydno was simpler than that of H. melpomene, with 7 main compounds versus 21 in H. melpomene. Heliconius cydno also had a less abundant overall bouquet, with an individual total of 1787 ± 776 ng vs. 3174 ± 1040 ng in H. melpomene (Table S1). Most of the main compounds (defined as those occurring in at least 90% of individuals) in H. cydno were linear alkanes (4 of 7 compounds), while H. melpomene has a more diverse array of different compound classes. Of the five major compounds (> 100ng per individual) in H. melpomene, only syringaldehyde was found in similar amounts in H. cydno; the major compounds octadecanal, 1-octadecanol, (Z)-11-icosenal, and (Z)-11-icosenol were absent or found in very low amounts in H. cydno. Comparison with previously published data for other Heliconius species, all of which lack octadecanal in the large amounts seen in H. melpomene, demonstrates that this high level of octadecanal is an evolutionarily derived state in H. melpomene (Mann et al 2017, Darragh et al 2019b). When focusing on the hindwing androconia of H. cydno, only two compounds [syringaldehyde (24.7% of the hindwing androconial bouquet) and (Z)-11-icosenol (1.7%)] were specific to this region. For details, please see SI text, Figures S1-S3, and Tables S1-S2.
Electroantennographic responses to con- and heterospecific pheromone stimulus sets
We next investigated the electroantennographic (EAG) response of both species to natural con- and heterospecific pheromone bouquets extracted from adult male butterflies. Females of both H. melpomene and H. cydno responded more strongly (i.e. showed a larger voltage displacement from the antennal baseline) to both natural wing extracts (Mnat and Cnat, respectively) as compared with the control solvent (dichloromethane plus 1 ng/μL 2-tetradecyl acetate, hereafter “DCM+IS”) (Figure 1, Figure S4; see Table S3 for statistical details). Males of both species also responded to both wing extracts. Females of both species showed stronger responses to H. cydno wing extracts than to H. melpomene wing extracts. Responses differed in males, with H. melpomene males responding equally to both species’ wing extracts, while males of H. cydno responded more strongly to wing extracts of H. melpomene.
We then explored antennal responses to synthetic compound blends. These were based on the most abundant compounds from each species (see Methods). Responses to the synthetic H. melpomene wing blend were mixed in H. melpomene females. In the first stimulus set (comparison of hetero- and conspecific stimuli) the synthetic blend (Msyn) yielded stronger responses than the natural H. melpomene wing extracts (Figure 1, Table S3) but in the second stimulus set (individual pheromone components) the synthetic blend yielded weaker responses than the natural H. melpomene wing extracts (Figure 2, Table S3). A similar pattern was seen in male H. melpomene. Both sexes of H. cydno produced only a modest response to the synthetic H. cydno wing blend (Csyn) when compared with the hexane solvent. In all cases this response was lower than to natural H. cydno wing extract.
Electroantennographic responses to individual pheromone components from both species
Finally, we explored the responses to individual compounds to identify specific biologically active pheromone components. Only octadecanal differed significantly from the controls in all four species-sex combinations (Figure 2, Table 1, Table S3). Both sexes of H. cydno also showed altered responses to the shared compound (Z)-11-icosenol, with an increased response in H. cydno males and a decreased response in H. cydno females. Both H. cydno females and H. melpomene males had decreased responses to (Z)-11-icosenal. Heliconius cydno females also showed a decreased response to (Z)-13-docosenal. Male H. cydno had increased responses to 1-octadecanol and henicosane. No other compound-species-sex combinations were significantly different from hexane. The shared compound syringaldehyde was the only pheromone component to show no change in EAG amplitude relative to the solvents in any species-sex combination.
In females of both species, response to octadecanal was stronger than response to the conspecific synthetic mixture. By contrast, male H. melpomene responded equally to the conspecific synthetic mixture and octadecanal, while male H. cydno responded equally to the conspecific synthetic mixture and both its components syringaldehyde and henicosane, with no evidence for a synergistic mixture effect. Individual components (with the exception of octadecanal) were not equivalent to the synthetic blends for females of both H. melpomene and H. cydno, suggesting a synergistic mixture effect. Antennal responses to a given stimulus can change over time, and this may reflect biological processes of neuronal adaptation. Female H. melpomene adapted equally quickly to octadecanal and the natural and synthetic H. melpomene pheromones, while adaptation to other stimuli was equal to the control, further supporting octadecanal’s salience as the main pheromone in H. melpomene (see SI text, Figures S5-S6, and Table S4).
Behavioral response to octadecanal supplementation in H. melpomene
We next confirmed a behavioral response to the most physiologically active substance, octadecanal, one of the dominant compounds in H. melpomene. A total of 29 behavioral experiments were conducted, with mating observed in 18 (62%); one trial was excluded due to wing damage, leaving 17 successful matings. With our small sample size, we found no evidence that females showed a preference for either treatment, mating with the hexane male 11 times (65%) and with the octadecanal male the remaining 6 times (35%) (p = 0.332). However, mating latency (time from experiment onset to mating) was significantly longer for the octadecanal matings (average 88.5 minutes) than for the hexane matings (average 43.7 minutes; t = 2.7848, df = 8.2491, p = 0.023). There was no evidence that this mating latency was due to evaporation of the octadecanal treatment, as there was no detectable drop in octadecanal quantity in the hindwing androconia over time [comparison of 30 minute and 2 hour treatments (Figure S7)]. Furthermore, little octadecanal rubbed off onto the forewing overlap region. Interestingly, although about 5.5x as much octadecanal as normal should have been present on the wings of treated males, only about 2-2.5x was seen after 30 minutes (the time at which the female would be introduced to the two males), suggesting some of the added octadecanal was lost before the start of the behavioral experiments, perhaps due to oxidization or pheromone hydrolysis, as has been shown in some moths (Ferkovich et al 1982).
Genetic basis of octadecanal production in H. melpomene
Analysis of octadecanal production by F1 males showed that the H. cydno octadecanal phenotype (little to no octadecanal) is dominant over the octadecanal-rich H. melpomene phenotype, and octadecanal production segregates in backcrosses to H. melpomene (Figure S10). Using the variance within the backcross individuals, we calculated a Castle-Wright estimator of 0.81 loci, suggesting a potentially monogenic basis for octadecanal production in H. melpomene. Quantitative trait locus (QTL) mapping revealed a single significant peak on chromosome 20 (Figure 3a, Figure S11). The chromosome 20 peak remained significant when kinship was taken into account and explained 41.31% of the difference between the two parent species. Bayesian confidence intervals for the peak on chromosome 20 were identical with and without kinship, spanning a range of 46.9-56.37cM with the peak at 47.66cM (with or without kinship), corresponding to a physical range of 3.4Mb along chromosome 20. To ensure that our findings were replicable across individual families, we also constructed effect plots at the kinship peak for each family separately, and all showed the same directionality (Figure 3b). The peak on chromosome 20 does not overlap with any of the major wing color loci (Jiggins 2017; van Belleghem et al 2017), nor does it overlap with mate choice QTL (Merrill et al 2019). The confidence interval region contains 160 genes, all of which represent potential candidates for octadecanal production.
We next evaluated the evidence in support of the 160 genes in this interval to identify top candidates for octadecanal production. In total, 14 were putative fatty-acyl CoA reductases (FARs), which catalyze the conversion of fatty acids to alcohols via a bound aldehyde intermediate (Table S5). Octadecanal is most likely produced via this pathway (Figure 3c), either as a direct product of a FAR-catalyzed conversion of 18-carbon stearic acid (by releasing the bound intermediate directly) or as a product of a further dehydrogenation of the alcohol intermediate (octadecanol) to the aldehyde product. Two candidate alcohol dehydrogenases, which might catalyze this reaction, were also contained within the region, yielding a total of 16 candidates. To ascertain whether octadecanol might serve as the precursor to octadecanal in H. melpomene, we also searched for QTL underlying octadecanol production (Castle-Wright estimator of 0.71 loci), and found a very similar pattern to octadecanal, with a single QTL peak on chromosome 20 (Figure 3d) explaining 25.36% of the difference between the two parent species. This peak broadly overlapped the octadecanal peak, with a much broader confidence interval from 10.91-56.37cm (12.9Mb) and a peak at 51.82cM regardless of whether kinship was taken into account (Figure S11). The 14 FARs in the region are highly clustered, with a set of eight found within a 133kb region. Sequence comparison between the H. melpomene and H. cydno alleles showed that nearly all of these genes harbor nonsynonymous SNPs between the two species (Table S5, Supplementary Data 1). One gene showed no coding SNPs between the two species; nine had between three and 24 nonsynonymous SNPs; and four had more substantial changes, including missing exons and frameshifts. The final two genes (one alcohol dehydrogenase and one FAR) could not be found in the H. cydno genome, and may instead represent annotation or assembly errors in H. melpomene or, alternately, deletions in H. cydno. Nearly all of the intact genes displayed purifying selection (ω between 0.001 and 0.2459), with only the remaining alcohol dehydrogenase (ω = 1.2478) under positive selection (Table S5). Taken together, these results suggest that either of the alcohol dehydrogenase candidates may underlie the production of bioactive octadecanal from octadecanol in H. melpomene, although functional experiments are required to confirm this hypothesis. Alternately, as QTL for octadecanal and its likely precursor octadecanol overlap, a single FAR may be responsible for producing both volatiles.
Discussion
Previous work has shown that male Heliconius butterflies use pheromones during courtship, and the presence of these pheromones is necessary for successful mating to take place (Darragh et al 2017). Here we demonstrate that the two closely related species with strong mating differences, H. melpomene and H. cydno, show strong differences in chemical bouquets with a number of major compounds including octadecanal, 1-octadecanol, (Z)-11-icosenal, and (Z)-11-icosenol found in high concentrations in H. melpomene but largely absent in H. cydno, consistent with prior results (Mann et al 2017, Darragh et al 2019b). Somewhat surprisingly, despite these strongly divergent pheromone signals, there was little difference in antennal response to both overall wing chemical bouquets and various chemical components between the species. Indeed, only octadecanal elicited a significant response in H. melpomene. Somewhat surprisingly, octadecanal was also physiologically active in H. cydno, where it is largely absent from the male wing bouquet.
These data stand in marked contrast to similar studies in other insect species and suggest that the peripheral nervous system architecture of H. melpomene and H. cydno has not diverged alongside their wing pheromone components. By contrast, the moth Ostrinia nubialis, whose E- and Z-strains diverged approximately 75,000-150,000 years ago (Malausa et al 2007), have opposite topologies in the antennal lobe and antennal sensillae (Kárpáti et al 2008, Koutroumpa et al 2014), and similar divergence in peripheral nervous system architecture has been seen in Rhagoletis pomonella (Frey & Bush 1990, Tait et al 2016) and Drosophila mojavensis (Date et al 2013, Crowley-Gall et al 2016) despite much shorter divergence times than between H. melpomene and H. cydno. Our results in Heliconius parallel what is seen in Colias butterflies, where C. eurytheme and C. philodice show very similar female electrophysiological responses to the con- and heterospecific pheromone compounds, despite a behavioral effect of treating males with heterospecific pheromones (Grula et al 1980). However, Colias use multiple signals when choosing between mates (Papke et al 2007), so rapid divergence in peripheral nervous system elements may not be necessary for full reproductive isolation. A similar pattern most likely occurs in Heliconius, with EAG responses very similar between species, and probable differences in the antennal lobe and higher brain regions (e.g. the mushroom body or lateral protocerebrum) driving behavioral differences (see e.g. Montgomery and Merrill 2017). Female Heliconius butterflies likely integrate multiple signals (including pheromones, male courtship flights, and visual cues) in these higher brain regions when making the decision to mate.
We have identified octadecanal as the major pheromone component in H. melpomene and showed that responses to it are conserved across both species despite its general absence in H. cydno. In the Lepidoptera, the activity of octadecanal as a pheromone has been tested in eight species across a variety of families (Tatsuki et al 1983; Cork et al 1988; Tumlinson et al 1989; Ho et al 1996; McElfresh et al 2000; Yildizhan et al 2009; El-Sayed et al 2011; Pires et al 2015; Chen et al 2018). Only in Cerconota anonella, and now in Heliconius melpomene, has some electrophysiological and behavioral activity been seen (Pires et al 2015). The closely related hexadecanal is a major pheromone component in the butterfly Bicyclus anynana (Nieberding 2008), and differs from octadecanal only in its origin from palmitic rather than stearic acid and carbon number, so octadecanal’s role as a pheromone in Heliconius is perhaps not entirely unexpected from a biochemical perspective. In contrast, we have failed to discover the physiologically active pheromone(s) in H. cydno. This is not entirely unexpected, as we focused on the most abundant compounds, but minor components have been shown to be important in other systems (McCormick et al 2014, Chen et al 2018).
Intriguingly, despite the strong EAG response, there was a marked negative behavioral response to increased octadecanal. A plausible explanation for the negative behavioral response to octadecanal supplementation is that disruption of the normal mixture ratios of H. melpomene may inhibit the female response, as seen in the butterfly Pieris napi, where synergistic processing of two volatile components in the male bouquet is necessary for acceptance behavior (Larsdotter-Mellström et al 2016). Octadecanal may also experience a dose-response curve with an aversive response to higher concentrations and an attractive response at lower ones. Potential mixture or dosage effects suggest that female H. melpomene may use octadecanal quantity or relative abundance to assess male quality or choose between courting males. The increased mating latency with octadecanal-treated males may reflect females undergoing a period of adjustment, either in the peripheral or central nervous system, to the higher dose of octadecanal; this would be consistent with our results showing long-term adaptation to octadecanal. We remain uncertain of what effect, if any, octadecanal would have on the behavior of H. cydno, where it is largely absent from the male pheromone bouquet. It may be used to avoid courtship with H. melpomene, supporting other divergent signals in maintaining reproductive isolation.
Given the strong physiological and behavioral response to octadecanal, and its possible role in reproductive isolation between H. melpomene and H. cydno, we sought to understand its genetic underpinnings. Fatty-acid derived compounds comprise the largest category of Lepidoptera sex pheromones (Ando & Yamakawa 2011), and are produced from fatty acyl-CoA precursors via the action of several enzymes. Since these pheromones are secondary metabolites derived from primary metabolic pathways, their production is likely to be relatively labile in evolutionary terms, allowing simple genetic changes to drive the wide diversity of lepidopteran sex pheromones. Even though we have a broad knowledge of pheromone diversity in Lepidoptera, our understanding of the genetics of pheromone biosynthesis is relatively weak. To date, only a handful of genes have been identified, mostly desaturases and fatty acyl-CoA reductases (Yew & Chung 2015).
Using a QTL mapping approach, we have shown that the production of octadecanal has a relatively simple genetic basis, with a region on chromosome 20 corresponding to production of both octadecanal and its likely precursor octadecanol in Heliconius. This locus therefore likely represents a region under divergent selection between H. melpomene and H. cydno that is unlinked to previously identified species differences in color and mate choice (Jiggins 2017, Merrill et al 2019). Patterns of FST between the species are highly heterogeneous and were not especially informative in further delimiting the locus (data from Martin et al 2013). Due to our small mapping population, the confidence intervals for these QTL therefore remain large: the octadecanal QTL spans 3.4Mb and contains 160 genes. Of these, we identified 16 likely candidate genes based on known biosynthetic pathways in moths and the butterfly Bicyclus anynana (Lienard et al 2014): 14 fatty acyl-CoA reductases and two alcohol dehydrogenases. Fatty acyl-CoA reductases have previously been identified in H. melpomene by (Lienard et al 2014), who noted lineage-specific duplications within H. melpomene on two scaffolds corresponding to H. melpomene chromosomes 19 and 20. All but one of the candidate FARs found on chromosome 20 were identified by Lienard et al, but all fall outside their clade of pheromone gland FARs. The identified Bicyclus FAR that produces hexadecanol does not also produce the major pheromone hexadecanal, implying the presence of an additional as yet undescribed alcohol dehydrogenase in Bicyclus, and the biochemical similarity between hexadecanal and octadecanal suggests Heliconius may also use an alcohol dehydrogenase to produce octadecanal. By contrast, the overlapping octadecanol and octadecanal QTL on chromosome 20 in Heliconius suggest the presence of a bifunctional FAR that produces both the alcohol and aldehyde together. Further studies, including functional assays and location of wing pheromone biosynthesis, will be required to tease apart our potential candidates.
Our studies of the electrophysiological and behavioral responses of Heliconius butterflies and the genetic basis of pheromone production add to the growing body of literature suggesting that pheromonal communication in Lepidoptera is not limited to nocturnal moths but can be found in day-flying butterflies that also use striking visual signals. Heliconius butterflies can detect con- and heterospecific wing compound bouquets, and a major component, octadecanal, is physiologically and behaviorally active in H. melpomene and its genetic basis appears relatively simple, consistent with other pheromone shifts found in insects (Symonds & Elgar 2007). Along with their striking wing color patterns, male Heliconius use chemistry to affect female mate choice, combining courtship behaviors, and chemistry in a dance to elicit female mating responses (Klein & de Araújo 2010, Mérot et al 2015). Despite our human bias towards visual signals, we are now beginning to understand how such visually striking butterflies communicate using chemistry.
Methods
Data and analysis code are deposited in Dryad under accession XXX. Sequencing reads leading to linkage map construction are deposited in the European Nucleotide Archive (ENA) under accession YYY.
Butterflies
Stocks from central Panama of Heliconius melpomene rosina and H. cydno chioneus (hereafter H. melpomene and H. cydno) were used for all experiments (Darragh et al 2017). Butterflies were reared in insectaries at the Smithsonian Tropical Research Institute, Gamboa, Panama under ambient temperature and light conditions. Eggs were collected from these breeding stocks and the resulting larvae fed on Passiflora platyloba var. williamsi, P. biflora, P. menispermifolia, and P. vitifolia until pupation. Data from (Darragh et al 2019a) indicate that larval diet does not affect the major compounds found in H. melpomene, suggesting that this dietary variation is unlikely to affect results. Newly-eclosed adult butterflies were separated by sex to ensure virgin status and supplied with flowers from Psiguria warscewiczii, Psiguria triphylla, Gurania eriantha, Psychotria poeppigiana, Stachytarpheta mutabilis, and Lantana sp. (most likely L. camara) as pollen sources, as well as a ∼20% sucrose solution as a nectar source. All experiments used virgin butterflies. For assessment of H. cydno wing bouquets, male butterflies were between 10-12 days post eclosion and had fed exclusively on Passiflora platyloba var. williamsi. For electrophysiology, female butterflies were between 1 and 20 days post eclosion, and males between 10-20 days post eclosion to ensure male sexual maturity (Darragh et al 2017). For behavior, female butterflies were used the day after eclosion and males were between 10 and 59 days post eclosion. Natural wing extracts of both species were extracted from males 10-12 days post eclosion as described in Darragh et al. 2017 using dichloromethane plus 1 ng/μL 2-tetradecyl acetate (hereafter “DCM+IS”) and concentrated approximately 10x prior to use under room air. All samples were stored at −20°C before use.
Identification and quantification of androconial compounds
To identify species specific compounds among our two species, the chemical composition of the H. cydno androconial bouquet was investigated in samples from 26 adult male H. cydno and compared with 31 adult male H. melpomene, the latter including samples previously analyzed in (Darragh et al 2017, 2019a), all collected as above. Samples were assessed using gas chromatography-mass spectrometry (GC-MS) with an Agilent 7890B gas chromatograph coupled with an Agilent 5977 mass spectrometer with electron ionization (Agilent Technologies, California, USA). The GC utilized an Agilent HP-5MS capillary column (30m length, 0.25mm inner diameter), helium carrier gas at 1.2 mL/minute, and an Agilent ALS 7693 autosampler. Injection was splitless with an inlet temperature of 250°C. The temperature ramp was isothermal at 50°C for 5 minutes, then increased at 5°C/minute to 320°C and was then isothermal for 5 minutes. Samples were identified using a custom MS library and quantified by comparison with the internal standard.
In line with (Darragh et al 2017), wings from eight male H. cydno were dissected into four regions: hindwing androconia, forewing overlap region, hindwing rest-of-wing, and forewing rest-of-wing, and all extracted identically after dissection. Wing region area was quantified by photographing the wings before dissection and measuring the total pixel area of each wing region in the GNU Image Manipulation Program v.2.8.20 (GIMP Development Team), with the pixel-mm conversion via measurement of a ruler in the photograph. Quantified compounds in each wing region for each individual were scaled by the area of the relevant wing region in that individual.
Chemicals
Syringaldehyde (3,5-dimethoxy-4-hydroxybenzaldehyde), 1-octadecanol, and henicosane were obtained commercially (Sigma-Aldrich). The aldehydes octadecanal, (Z)-11-icosenal, and (Z)-13-docosenal were obtained from the respective alcohols 1-octadecanol, (Z)-11-icosen-1-ol, and (Z)-13-docosen-1-ol by oxidation with iodoxybenzoic acid (IBX) in ethyl acetate according to More and Finney (2002). The required alcohols (Z)-11-icosen-1-ol and (Z)-13-docosen-1-ol were in turn obtained by lithium aluminum hydride reduction from commercially available (Z)-11-icosenoic acid and methyl (Z)-13-docosenoate (Larodan) according to Cha and Brown (1993). The seven target compounds (see SI for structures and reaction scheme) were chosen due to their quantitative dominance in the chemical profiles of H. melpomene and H. cydno. The solvent for all synthesized compounds was hexane, with the exception of the polar syringaldehyde, which was diluted in a 1:10 mixture of dichloromethane and hexane.
Synthetic blends of H. melpomene and H. cydno male wing bouquets were prepared from these synthesized compounds. Synthetic H. melpomene contained 23.22 ng/μL syringaldehyde, 23.33 ng/μL octadecanal, 6.92 ng/μL 1-octadecanol, 4.67 ng/μL (Z)-11-icosenal, 20.33 ng/μL (Z)-11-icosenol, and 4.83 ng/μL (Z)-13-docosenal. Synthetic H. cydno contained 47 ng/μL syringaldehyde and 93.33 ng/μL henicosane. Floral direct extractions from Lantana sp. (most likely L. camara) growing wild in Gamboa, Panama were used as a positive control. Single umbels were removed from plants at dawn and placed in a scintillation vial to which 400uL of DCM+IS was added. After 1 hour, the DCM+IS was removed to a glass vial and kept at −20°C before use.
Electroantennography
Electrophysiological preparations were assembled as follows: antennae were cut from the head of a virgin butterfly using fine scissors and the final 6.5 segments (approximately 1.5mm) cut off with a scalpel. Both antennae were then placed in parallel across an antenna fork (Syntech, Buchenbach, Germany) and held in place using electrode gel (Spectra 360 Electrode Gel, Parker Laboratories Inc., Fairfield, NJ, USA). The antenna fork was mounted on a Syntech EAG CombiProbe with 10x internal gain, and signal from this was routed through a Syntech IDAC4. EAG waveforms recorded using Syntech GcEad/2014 software. Stimulus pulses were delivered using a Syntech CS-55 Stimulus Controller with foot pedal trigger. Both continuous unhumified room air and stimulus pulses were delivered to the preparation at 1.5 liters/min. The stimulus pulses were delivered in triplets of 0.5 seconds each, separated by 5 seconds, with triplets initiated every 30 seconds. Stimulus delivery used odor cartridges assembled from Pasteur pipettes with a strip of filter paper plugged with cotton when not in use; each stimulus cartridge was ‘charged’ with 10uL of stimulus solution for each experiment. Each antennal preparation was used only once.
Two sets of stimuli were delivered: a species comparison set and a synthetic compound set. Both sets used air (nothing added to filter paper), hexane, and DCM+IS as negative controls and Lantana extract as a positive control. The species comparison set included male wing extracts from H. melpomene and H. cydno (“Mnat” and “Cnat” respectively) and synthetic blends representing the two species (“Msyn” and Csyn”). The synthetic compound set included air, hexane, DCM+IS, the conspecific male wing extract, the conspecific synthetic blend, and the seven synthetic compounds. Presentation order was randomized before each experiment. Species comparison experiments consisted of sixteen pulses each of the seven stimuli, interspersed with five pulses of Lantana extract at the start, between every three stimuli, and at the end. Synthetic compounds were similar, with eleven pulses of each of the twelve stimuli, interspersed with four pulses of Lantana at the start, between every four stimuli, and at the end. For analysis, the first triplet of each stimulus set was removed, leaving fifteen and ten pulses respectively. At least ten female and five male butterflies of each species were used with each experiment.
Onset time and amplitude of EAG responses were marked using GcEad/2014. To control for antennal signal degradation over time, a time-dependent correction factor was calculated using linear interpolation between each Lantana set and this applied to the absolute value of the EAG response amplitude. These corrected amplitudes were then scaled to the amplitude of the initial Lantana set to partially control for differences between preparations.
Analysis of antennal adaptation
Short-term adaptation (STA), as defined in (Zufall & Leinders-Zufall 2000), is adaptation occurring only over a very brief window from the initial stimulus that then resolves quickly with no further stimulation (e.g. the 10 second interval given for salamanders). By contrast, long-term or long-lasting adaptation (LTA) are defined in the same source as persisting over an extended period of time, up to several minutes in vertebrates and insects (Stengl 2010). Within our electrophysiological data set, there is the potential to measure both types of adaptation. STA, if present, should be evident within an individual triplet, as the stimuli within a triplet are separated by 5 seconds (and thus STA is likely to persist within a triplet), whereas LTA should be evident across an individual stimulus set, as these lasted 5.5-8 minutes with maximum intervals of 30 seconds between stimuli, insufficient for LTA to be abolished if we assume similar mechanisms as vertebrates. We assessed antennal responses for LTA by pooling all triplets within a stimulus-species-sex combination. We pooled these data within stimulus set types (species comparison and synthetic compound), treating butterfly preparation identity as a random effect. For each stimulus, the change in antennal response was assessed over the time since initial presentation of the stimulus. STA was assessed by looking for significant changes in the residuals from this analysis between members of a triplet.
Behavior
To test the potential role of octadecanal in H. melpomene female mate choice, behavioral experiments were conducted in insectaries at STRI, Gamboa, Panama, between April and July 2018. One day old virgin females were presented with both an octadecanal treated and a control H. melpomene male for two hours. Males were at least ten days old and were selected based on similarity of size, wing-wear and age, with treatment being allocated randomly by coin flip. Either 25 µl octadecanal (140 ng/µl octadecanal in hexane, thus adding 3500ng to the existing average 773.4ng for approximately a 5.5x dose) or 25 µl pure hexane (both evaporated down to a smaller volume under room air) was applied to the hindwing androconial region of each male, and males were then allowed 30 minutes to settle before beginning the two hour experiment period. Experiments began at or close to 9am, with observations being made every 15 minutes or until mating occurred.
To test the persistence of the octadecanal treatment on the wings of live butterflies, a separate set of H. melpomene males was treated as above with either hexane or octadecanal. Separate males were sampled at 30 minutes post treatment and two hours post treatment by extraction of the forewing overlap region (Darragh 2017) and the hindwing androconia in DCM+IS as above, with two males per treatment-time combination. Octadecanal was then measured using GC-MS as above and quantified by peak area comparison with the 2-tetradecyl acetate internal standard.
Quantitative trait locus mapping for octadecanal production
To map the genetic basis for octadecanal production in H. melpomene, we took advantage of the fact that H. cydno produces little to no octadecanal. Bidirectional F1 crosses between the two species revealed that the H. cydno phenotype (low to no octadecanal) is dominant over the high octadecanal production found in H. melpomene, so we constructed backcross families by crossing F1 males to female H. melpomene from our existing stocks. A total of ten families (nine with a female H. melpomene grandparent and one with a female H. cydno grandparent) were constructed, with each offspring representing a single recombination event from the F1 father. Butterflies were reared and wing extracts collected and analyzed from male offspring as described above, except that all larvae were reared on Passiflora platyloba var. williamsi. Bodies of male offspring were collected into dimethyl sulfoxide (DMSO) for later library preparation. The Castle-Wright estimators for octadecanal and octadecanol production were calculated using the phenotypic variance of the backcross individuals as the estimated segregation variance (Jones 2001).
Qiagen DNeasy kits (Qiagen) were used for DNA extraction. Individuals were genotyped either by RAD-sequencing as previously described (Davey et al., 2017, Merrill et al., 2019), or low-coverage whole genome sequencing using nextera-based libraries. For the nextera-based libraries a secondary purification was performed using magnetic SpeedBeads™ (Sigma) dissolved in 0.44mM PEG8000, 2.5M NaCl, 1mM Tris-Cl pH=8, and 0.1mM EDTA pH=8.0. High-throughput sequencing libraries were generated using a method based on the Nextera DNA Library Prep (Illumina, Inc.) with purified Tn5 transposase (Picelli et al 2014). Sample barcoding was performed using PCR extension with an i7-index primer (N701–N783) and the N501 i5-index primer. Libraries were purified and size selected using the same beads as above. Pooled libraries were sequenced by HiSeq 3000 (Illumina) by BGI (China).
Linkage mapping was conducted using standard Lep-MAP3(LM3) pipeline (Rastas, 2017). First, individual fastq files were mapped to the melpomene reference genome using BWA MEM (Li, 2013) and then sorted bams were created using SAMtools (Li & Durbin, 2011). The input genotype likelihoods were constructed by SAMtools mpileup and pileupParser2+pileup2posterior from LM3. The pedigree of individuals was validated and corrected using IBD (identity-by-descent) module and the sex of individuals was validated and corrected according to the coverage on the Z chromosome and autosomes using SAMtools depth. Then, ParentCall2 (parameter ZLimit=2) and Filtering2 (dataTolerance=0.001) modules were called on the input data and a random subset of 25% of markers (to speed up analysis) was used for the subsequent steps.
Initial linkage groups (chrX.map) and marker orders (orderX.txt) were constructed based on melpomene genome for each of 21 chromosomes. SeparateChromosomes2 was run on each of these groups with the default parameters (lodLimit=10) except for map=chrX.map (for X=1..21). Finally, OrderMarkers2 was run on each chromosome in the constructed order, with parameter scale=0.05, recombination2=0, evaluateOrder=orderX.txt and map=result_from_SeparateChromsomes2.chrX.txt. Another evaluation was done with data in the grandparental phase (additional parameter grandparentPhase=1). The phased data of these orders were matched using phasematch script (LM3) and obtaining all markers from the first evaluation in the grandparental phase. This obtained result was used as the final map.
Map construction resulted in the retention of 447,818 SNP markers across 89 individuals with phenotype data. To facilitate computation, markers were thinned evenly by a factor of ten, resulting in 44,782 markers with no missing data. Octadecanal production was log-transformed to obtain normality, then regressed against marker position using the R/qtl2 R library (Broman et al 2018). Significance thresholds were obtained by permutation testing in R/qtl2 with 1000 permutations, and QTL confidence intervals obtained using the bayes_int command. To account for the family structure present in our QTL mapping populations, we additionally included a kinship matrix calculated by R/qtl2 using the LOCO (leave one chromosome out) method in the marker regression and recalculated significance thresholds and confidence intervals with the kinship term included. Percent variance explained was calculated as the difference in phenotype means of individuals of each genotype divided by the difference in the parental phenotype. Since the genetic linkage map was based on whole genome data, we were able to obtain physical positions of QTL confidence interval endpoints. The physical positions of the kinship-included confidence interval were used to query Lepbase (Challis et al 2016) for potential candidate genes from the H. melpomene genome. To identify putative functions for each potential candidate, protein sequences from Lepbase were searched against the nr (non-redundant) protein database using BLASTp (Altschul et al 1990). For each candidate with a promising functional annotation, exons were pulled out of the H. cydno genome (Pessoa Pinharanda 2017) and aligned to the H. melpomene genes using BLASTn with each H. melpomene exon to search for SNPs between the two species. Selection was tested at the sequence level using codeml (Yang 2007) in pairwise mode with the F3X4 codon frequency model.
Statistical analysis
Differences in individual compounds between H. melpomene and H. cydno were assessed using a Welch’s t-test. Wing region differences in H. cydno were assessed for each individual compound found in at least four of eight samples of at least one wing region using a linear mixed model with wing region as a fixed effect and butterfly identity as a random effect using the package nlme v.3.1.137 (Pinheiro et al 2018). Statistical tests for these models were assessed with the Anova function in the package car v.3.0.0 (Fox & Weisberg 2011), and comparisons between wing regions were performed using the emmeans package v.1.3.1 (Lenth 2018).
For electroantennography, species comparison sets and synthetic compound sets were analyzed separately. Corrected and scaled EAG responses (for each experiment within sexes and species) were compared between stimuli with a linear mixed model for H. melpomene with stimulus as a fixed effect and butterfly preparation identity as a random effect using nlme as above. EAG responses for H. cydno were better represented as an inverse gamma distribution, so were assessed in the same fashion with a generalized linear mixed model (glmmPQL) using the package MASS v.7.3.50 (Venables & Ripley 2002). Statistical tests for these models were assessed with the Anova and emmeans (version 1.1.3) functions as above.
Long-term adaptation was assessed using robust linear mixed models with the package robustlmm v.2.2.1 (Koller 2016), with corrected but unscaled amplitude as the response variable, time since initial presentation of the stimulus as a fixed variable, and butterfly preparation as a random variable. Responses were pooled across samples within a species-sex combination and considered for each stimulus separately. As robustlmm does not provide p-values, confidence intervals were used to assess significance and difference between sample-species combinations. Short-term adaptation was assessed using the residuals from the same regression. Differences in the residuals between triplets within a triplet set were tested using a one-sample t-test with a hypothesized mean value of zero (i.e. no difference between residuals), performed on the subtractive difference between the residuals of the third (last) and first triplets.
Female mate choice was assessed using a binomial test, and treatment differences in time until mating were assessed with a two-sided t-test assuming unequal variances. Octadecanal persistence was not assessed statistically due to the small sample size. All statistical tests were performed in R v.3.5.0, v.3.5.1, or v.3.5.2 (R Core Team 2013).
Author contributions
KJRPB designed, performed, and analyzed electrophysiology experiments, performed QTL mapping and downstream analysis, and wrote the manuscript. KD designed and performed QTL mapping crosses, designed behavioral experiments, and prepared sequencing libraries. KD, SFG, DAA, KJRPB, and JM reared butterflies for all experiments. JM performed and analyzed behavioral experiments. IAW prepared sequencing libraries. YFC and MK provided Tn5 transposase for sequencing. SS provided synthetic compounds used in electrophysiology and behavior experiments as well as mass spectrometry libraries. PR analyzed sequencing data and constructed linkage maps. RMM helped design the QTL experiment. CDJ and WOM contributed funding, resources, and helped design the overall study. All authors contributed to the manuscript editing process.
Tables
SI Table 1: Comparison of Heliconius cydno and H. melpomene wing androconia bouquet. Amounts given are the mean in nanograms across 31 (H. melpomene) and 26 (H. cydno) samples, ± SD. Compounds are only included if found in at least 1/3 of samples for at least one species at levels of at least 1ng. #: compound found in at least 90% of samples of that species. Bold: species differ significantly. dof: Welch’s two sample t-test degrees of freedom.
SI Table 2: Wing region specificity of Heliconius cydno androconial compounds. Amounts given are the mean across eight samples ± SD. Numbers in parentheses after compound amounts indicate the number of samples the compound was found in. Bold: wing regions differ significantly. NA: post-hoc test not conducted as original linear model not significant.
SI Table 3: Details of statistical comparisons of stimuli from electrophysiological experiments. P1-P7: as in the main text and Figure 4.
SI Table 4: Short-term adaptation to different stimuli in H. melpomene and H. cydno males and females. t1, t2, and t3: first, second, and third members of a stimulus triplet. dof: degrees of freedom. P1-P7: as in the main text and Figure 4.
SI Table 5: Potential candidate genes for octadecanal production underlying the QTL peak on chromosome 20 in Heliconius melpomene.
Figures
SI Figure 1: Comparison of the major compounds in Heliconius melpomene and H. cydno. Compounds shown are those contributing at least 1% of the total bouquet amount in either species. Numbers under each bar indicate how many samples (out of 31 for H. melpomene and 26 for H. cydno) the compound was found in. Significantly different compounds: * p < 0.05; ** p < 0.01; *** p < 0.001. n.s., not significant.
SI Figure 2: Absolute and relative abundance of different compound classes in H. melpomene and H. cydno. n.s., not significant; * p < 0.05; ** p < 0.01; *** p < 0.001. The two unknown categories were not tested as the compound types are not known.
SI Figure 3: The seven compounds found in at least 0.1ng/mm2 of wing tissue in at least one wing region in Heliconius cydno. a: Compound abundance per square millimeter of tissue. b: Compound abundance without tissue area correction. Numbers under each bar indicate how many samples (out of eight) the compound was found in; letters above bars indicate significant differences between regions. n.s., not significant. A: hindwing androconia; O: forewing overlap region; H: hindwing excluding androconia; F: forewing excluding overlap region.
SI Figure 4: Structures and names of major components of the androconia of H. melpomene and H. cydno used in electrophysiological experiments.
SI Figure 5: Synthesis of target compounds used in electrophysiological experiments. IBX: iodosobenzoic acid; LiAlH: lithium aluminum hydride.
SI Figure 6: Heliconius melpomene responds to electrophysiological stimuli. Top to bottom: dichloromethane plus 2-tetradecyl acetate (internal standard) (negative control), Lantana extract (positive control), natural H. melpomene male wing extract, depiction of stimulus pulse timing. Data from a single virgin female.
SI Figure 7: Long-term adaptation to natural and synthetic stimuli in Heliconius butterflies. The 95% confidence intervals of the robust LMM slope are shown; a negative slope means that responses to that stimulus drop over time. P1-P7: see Figure 2.
SI Figure 8: Strength of long-term adaptation correlates with amplitude of EAG response in a sex-specific fashion. In females, a stronger response to a given stimulus correlates with a stronger degree of LTA both overall and for the synthetic compound set. In males the same is seen overall and for the natural extract set in H. cydno, but not in H. melpomene.
SI Figure 9: Octadecanal persistence in treated males over time. Bars show individual males, with two males per treatment-time point combination. Hindwing androconia and forewing overlap region males are shown in the same positions.
SI Figure 10: Octadecanal in H. melpomene, H. cydno, two F1 families (one in each crossing direction), and the ten backcross to H. melpomene families used in QTL mapping.
SI Figure 11: Chromosome 20 QTL map for production of octadecanal and octadecanol in H. melpomene. Shaded regions indicate the Bayesian confidence intervals with kinship structure taken into account and black line indicates the peak of the QTL.
Acknowledgments
The authors wish to thank Bill Wcislo and Callum Kingwell for advice on electroantennography. We also thank Ian Warren, the insectary team in Panama including Oscar Paneso and Chi-Yun Kuo, as well as the administrative support of Oris Acevedo and Melissa Cano. We also thank S. Dilek for technical assistance. Permits for research and collection of butterfly stocks were provided by the government of Panama. KJRPB, KD, JM, IAW, and CDJ were funded by the European Research Council (FP7-IDEAS-ERC 339873); KD was additionally funded by a Natural Research Council Doctoral Training Partnership and a Smithsonian Tropical Research Institute Short Term Fellowship; YFC and MK are supported by the European Research Council Grant No. 639096 “HybridMiX” and by the Max Planck Society; WOM is funded by the Smithsonian Tropical Research Institute; and SS by the Deutsche Forschungsgemeinschaft (grant DFG Schu984/13-1).