Species specificity and intraspecific variation in the chemical profiles of Heliconius butterflies across a large geographic range

Abstract In many animals, mate choice is important for the maintenance of reproductive isolation between species. Traits important for mate choice and behavioral isolation are predicted to be under strong stabilizing selection within species; however, such traits can also exhibit variation at the population level driven by neutral and adaptive evolutionary processes. Here, we describe patterns of divergence among androconial and genital chemical profiles at inter‐ and intraspecific levels in mimetic Heliconius butterflies. Most variation in chemical bouquets was found between species, but there were also quantitative differences at the population level. We found a strong correlation between interspecific chemical and genetic divergence, but this correlation varied in intraspecific comparisons. We identified “indicator” compounds characteristic of particular species that included compounds already known to elicit a behavioral response, suggesting an approach for identification of candidate compounds for future behavioral studies in novel systems. Overall, the strong signal of species identity suggests a role for these compounds in species recognition, but with additional potentially neutral variation at the population level.

Chemical compounds, such as sex pheromones, mediate intraspecific communication in many systems (Wyatt, 2003(Wyatt, , 2014. The role of chemical signaling in behavioral isolation is also well established, especially among moth species (Löfstedt, 1993;Smadja & Butlin, 2008). Pheromone evolution requires changes in both the detection of pheromone by the receiver and the production of pheromone by the sender. Due to this coordination between detection and production, these pheromone blends are traditionally regarded as being under stabilizing selection toward a species stereotype (Löfstedt, 1993). Nonetheless, even when species-specific characteristics are present, chemical composition can exhibit intraspecific variation, with both qualitative and quantitative differences found across a species range (Carde & Allison, 2016).
The role of chemical signaling is likely to be especially important in comimics, where visual signals alone are not sufficient to identify conspecifics (Estrada & Jiggins, 2008;Giraldo, Salazar, Jiggins, Bermingham, & Linares, 2008;Mérot et al., 2013;Sánchez et al., 2015). In contrast, chemical compounds could be part of a multimodal aposematic warning signal Rothschild, 1961), with some tentative evidence that comimics exhibit similar chemical bouquets to aid recognition by predators .
Here, we describe the chemical profiles of seven species of Heliconius from over 250 individuals collected across the Neotropics.
We focus on the comimetic species H. melpomene and H. erato that are distributed widely across the Neotropics and analyzed both wing androconial and genital compounds of male butterflies. We hypothesize that compounds found consistently across the geographic range of a species are likely to be behaviorally active compounds, important for mate choice. We use H. melpomene as a test species due to the availability of behavioral and electrophysiological data to investigate this approach, by evaluating consistency in compound blends across different localities.
The extensive dataset analyzed here allows us to test evolutionary hypotheses, as well as identifying interesting candidate compounds for future behavioral studies. As well as interspecific variation, we also investigated intraspecific variation in chemical profiles of H. melpomene and H. erato. In both inter-and intraspecific datasets, we correlated chemical profile data with both geographic and genetic distances. Furthermore, to investigate if the chemical compounds are part of the aposematic comimicry signal, we sampled two different mimicry rings in western Ecuador and Panama. Between two and fifteen males were chemically analyzed per population ( Figure 1, Table A1 in Appendix 1), and one representative from each subspecies of H. erato and H. melpomene was used for whole-genome sequencing (Table A2 in Appendix 1). We follow the latest Heliconius taxonomy (Lamas & Jiggins, 2017).

| Extraction and chemical analysis of tissues
The androconial region of the wing, previously described as the gray-brown overlapping region of the hind wing (Darragh et al., 2017), as well as the genitalia, was dissected for analysis immediately after collection. For chemical extraction, the tissue was soaked in 200 μl dichloromethane containing 200 ng 2-tetradecyl acetate (internal standard) in 2-ml glass vials with PTFE-coated caps (Agilent) for one hour. The solvent was then transferred to new vials, maintained cool in the field, and stored at −20°C upon return. Androconial samples were evaporated to a reduced volume at room temperature prior to analysis. Extracts were analyzed by GC/MS using an Agilent model 5977 mass-selective detector connected to an Agilent GC model 7890B and equipped with an Agilent ALS 7693 autosampler. HP-5MS fused silica capillary columns (Agilent, 30 m × 0.25 mm, 0.25 µm) were used. Injection was performed in splitless mode (250°C injector temperature) with helium as the carrier gas (constant flow of 1.2 ml/min). The temperature program started at 50°C, was held for 5 min, and then rose at a rate of 5°C/min to 320°C, before being held at 320°C for 5 min.
Components were identified by comparison of mass spectra and gas chromatographic retention index with those of authentic reference samples and also by analysis of mass spectra. Components were quantified using 2-tetradecyl acetate as an internal standard.
Only compounds eluting earlier than hexacosane were analyzed in androconial samples and those earlier than nonacosane in genital samples (Darragh et al., 2017). We globally removed compounds that were not found in at least half of all individuals from a given population.

| DNA extraction and library preparation
We used a representative individual from each subspecies of H. erato and H. melpomene from across their range. Individuals were genotyped with medium-to high-coverage whole-genome sequencing.
We used two sequencing approaches. Genomic DNA of individuals whose ID starts with SR or KK (C. Kozak Table A2 in Appendix 1.
An interspecific genetic distance matrix was constructed using the function "cophenetic.phylo" from the ape package (Paradis & Schliep, 2018) with a previously published phylogeny (Kozak et al., 2015). Geographic distance matrices were created by inputting the coordinates of collection localities into the function "distm" in the geosphere package to calculate the Haversine great-circle distance between points (Hijmans, 2017).

| Inter-and intraspecific indicator compounds
We carried out indicator analysis using the indicspecies package (Cáceres & Legendre, 2009). Groupings are decided a priori (in this case, species or population), and compounds are determined which act as indicators of these groups. The best indicators are those which are only found in a single group (specificity), and all group members possess the compound (coverage); such a compound would have an indicator value of 1. The specificity of a compound is calculated based on the amount of compound found in each individual, while the coverage considers only presence or absence of the compound.
We used the function "indicators" to investigate both which single compounds and which combinations of compounds best predict group membership. We used the function "pruneindicators" to find the single compounds or combinations of compounds which had the highest indicator values.

| Variation in chemical profiles
Divergence in chemical profiles across species and populations was estimated with nonmetric multidimensional scaling (NMDS) ordination in three dimensions, based on a Bray-Curtis similarity matrix using absolute peak areas. We used the "metaMDS" function in the vegan package version 2.5-1 (Oksanen et al., 2017) and visualized the NMDS using the ade4 package (Dray & Dufour, 2007).
We assessed the relative importance of relevant factors in driving the variation in chemical profiles with multivariate statistical analyses. These factors included species identity, geographic region, and individual locality. We excluded subspecies as a factor because, in Heliconius, these are determined based on their, sometimes very subtle, difference in wing color pattern, with extensive gene flow across the genome between subspecies (Van Belleghem et al., 2017). It is therefore more biologically relevant to include locality in the model, to account for genetic drift between subspecies, and since locality and subspecies are highly correlated, we cannot include both. To compare overall variation in chemical composition between groups, we carried out PERMANOVA (permutational multivariate analysis of variance) testing based on a Bray-Curtis distance matrix, using the "adonis2" function in the vegan package with 1,000 permutations. We investigated each term in the model sequentially, starting with species identity, the main clustering factor found from visualization with NMDS, followed by geographic region (Panama vs. Western Andes vs. Eastern Andes vs. Amazon), and finally individual collecting localities. Model goodness of fit was evaluated by Akaike's information criterion (AIC). In general, we chose the model with the lowest AIC value; however, if two models were within two AIC of each other, we chose the simplest model as the best fit (Table A3 in Appendix 2). We followed these PERMANOVA tests with post hoc pairwise testing using the function "pairwise.perm.MANOVA" in the RVAideMemoire package, with Bonferroni correction, to identify which grouping factors were significantly different (Hervé, 2018).
We repeated the PERMANOVA within species, in H. erato and H. melpomene, to investigate fine-scale intraspecific geographic patterns. In the within-species analysis, we included geographic region (Panama vs. Western Andes vs. Eastern Andes vs. Amazon) and individual collecting localities as the two factors.
One issue with distance-based analyses such as PERMANOVA is that differences in dispersion between groups can be confounded with differences in location . To confirm these analyses and account for this issue, we implemented multivariate generalized linear models using the function "ManyGLM" from the mvabund package (Wang, Naumann, Wright, & Warton, 2012). We modeled the data using a negative binomial distribution, which we found to be appropriate through examination of residual plots. For interspecific analyses, we included species, region, and locality nested within region in the model. For intraspecific analyses, we included region and locality nested within region. The "ManyGLM" function fits models to each chemical compound, summing the test statistics to give a multivariate test statistic known as Sum-of-LR. This statistic can be tested for significance using resampling methods. We carried out backward elimination and compared the fit of models by using the "ANOVA.manyglm" function with a likelihood ratio test (Table A4 in Appendix 2). We can also determine which compounds are driving between-group differences by looking at the individual contribution of each compound to the Sum-of-LR, with p-values adjusted for multiple testing using the "adjust" option.

F I G U R E 1
Map indicating species collected from twelve localities across the Neotropics. See Table A1 in Appendix 1 for sample numbers. The phylogeny was previously published by Kozak et al. (2015) 2.6 | Phylogenetic and geographic distance Shared ancestry can explain part of the variation in a species' chemical profile. Using the interspecific genetic distance matrix calculated above, we tested for a correlation between phylogenetic distance and chemical profile divergence. We carried out partial Mantel tests, controlling for geographic distance, using the vegan package (Oksanen et al., 2017). To investigate the role of geographic distance in chemical profile divergence, we compared geographic and chemical distances matrices, controlling for genetic distance, with partial Mantel tests. To visualize the species phylogeny (Kozak et al., 2015), we used the "plot.

| Genomic and chemical distance within species
We calculated intraspecific genetic distances using genome sequences from 11 H. erato and 13 H. melpomene populations. We visualized genetic distances in two dimensions using MDS with the function "cmdscale." We tested for a correlation between intraspecific genetic distance and chemical profile divergence with partial Mantel tests, controlling for geographic distance, using the vegan package (Oksanen et al., 2017). Hybrids between populations of the same species were excluded from this analysis ( Table A2 in Appendix 1). We also used partial Mantel tests to investigate the role of geographic distance, while controlling for genetic distance.

| Comimics and similarity of chemical profiles
We used samples of two mimicry rings from two localities, Panama and western Ecuador. H. melpomene and H. erato form one mimicry ring, while H. cydno and H. sapho form another, with the addition of H. eleuchia in western Ecuador ( Figure 1). We visualized these samples but did not carry out statistical analyses due to the pseudoreplication caused by the similarity of individuals within a species. More species comparisons would be needed for further analysis.
All statistical analyses were performed with R version 3.5.1 (R Core Team, 2018). Figures were made using a palette of colors optimized for color blindness (Wong, 2011). We used ggplot2 for violin and boxplots (Wickham, 2009

| Chemical compounds in androconia and genitals
We sampled 252 androconia and 275 genitals across 42 populations of seven species and identified 349 compounds in the genitals and 157 in the androconia (Tables S1 and S2). Of the total number of androconial compounds, 38% are fatty acid derivatives, 20% aromatics, 10% terpenoids, 1% macrolides, <1% lactones, and 31% unknown or unidentified compounds. Of the genital compounds, 17% are fatty acid derivatives, 7% aromatics, 10% terpenoids, 1% lactones, 12% macrolides, and 44% unknown or unidentified compounds. The main difference is that there are more macrolides in the genitals than in androconia.  Nieberding et al., 2012). Using H. erato as an example, the androconial bouquet is 0.00002% and genital bouquet 0.0007% of total body weight (Montgomery, Merrill, & Ott, 2016). In general, a higher number of compounds and total amount of compounds are found in the genitals than in the androconial patches of Heliconius wings.

| Are there species-specific chemical compounds?
In order to identify candidate species recognition pheromones, we examined our data for species-specific compounds using indicator analysis. In most species that we examined, there were single androconial compounds that were strong indicators of species identity (Table 1) There were similarly species-specific genital compounds in all species except H. sapho and H. timareta, where a combination of two compounds was the best predictor (Table 2). Similar to the androconia, in H. melpomene, the best indicator compound for genitalia has known behavioral activity, in this case the anti-aphrodisiac, (E)β-ocimene (Schulz, Estrada, Yildizham, Boppré, & Gilbert, 2008 Note: A is a measure of species specificity of the compounds, B is a measure of species coverage, and sqrtIV is the indicator value which considers both A and B and ranges from 0 (compound not present in any individuals of that species) to 1 (compound only present in that species, and present in all individuals).

| What factors affect interspecific variation in chemical profiles?
Our sampling allowed us to investigate how variation in chemical composition is partitioned within and between species, and determine the extent to which chemistry is a species-diagnostic trait.
Visualization of the chemical profiles reveals that individuals mostly group by species for both androconial and genital chemical bouquets ( Figure 3). Species significantly differ in their androconial bouquet, with species identity accounting for 58% of the overall variation in chemical profiles (PERMANOVA, Species, F 6,251 = 72.16, p < .001). All pairwise comparisons of species are significantly different (Table A5 in Appendix 3 Finally, 6% of variation is explained by an interaction between species and region (PERMANOVA, Species*Region, F 6,274 = 6.52, p < .001).
For both androconial and genital chemical profiles, most variation is explained by species identity, rather than geographic location, as confirmed by ManyGLM (Tables A7 and A8 in Appendix 3). We also confirmed this by comparison of within and between species and locality Bray-Curtis distances ( Figure A1 and Figure A2, Appendix 4).

| Does phylogenetic distance explain chemical profile divergence?
Using whole-genome sequence data, we explored the degree to which variation between species can be explained by geographic and genetic distance among the samples. We carried out partial Mantel tests to investigate the correlation between two variables while controlling for a third variable. When controlling for geographic distance, genetic divergence is strongly correlated with both androconial and genital chemical divergence (partial Mantel test, androconia, r = .7871, p = .001; genitals, r = .6936, p = .001). When controlling for genetic distance, geographic distance is significantly but weakly correlated with androconial and genital chemical divergence (partial Mantel test, androconia, r = .072, p = .001; genitals, r = .046, p = .007).

| Do we find population-specific chemical compounds?
We used an indicator analysis to search for compounds unique to specific populations of H. erato and H. melpomene. Most intraspecific differences are due to quantitative rather than qualitative differences between populations, perhaps explaining why many population indicators were weak as they are also found in other regions at different amounts (Tables A9 and A10 in Appendix 3). The only exception is H. e. cyrbia (western Ecuador) that has many genital compounds unique to this region (Table A9 in Appendix 3).

| What factors affect intraspecific variation in chemical profiles of H. erato and H. melpomene?
We also wanted to determine the sources of variation within species These geographic differences in chemical profiles are not as strong in H. melpomene ( Figure 5). For H. melpomene androconial TA B L E 2 Genital compounds which are the best indicators of species identity. A, B, and sqrtIV as in

| Is there evidence for similarity between comimics in chemical profile?
We investigated the effect of mimicry ring on chemical profile using individuals collected in Panama and western Ecuador from two mimicry rings ( Figure 6). Consistent with our interspecific analyses, we find that species is the main determinant of androconial and genital bouquets. H. sapho and H. eleuchia group closely in the NMDS visualization; however, they are closely related and so it is unclear whether this similarity is due to comimicry or shared ancestry. Especially for the androconia, H. erato and H. melpomene seem to be more similar than we might expect given their phylogenetic distance.
All the results described above show a consistent pattern when unidentified compounds were not included in the analysis (Appendix 5). Interspecific analyses were also consistent when repeated without populations with a sample of fewer than five individuals (this removed seven populations from androconial analysis and five from genital analysis) (Appendix 5).

| D ISCUSS I ON
Heliconius butterflies represent a continental-scale adaptive radiation (Kozak et al., 2015). Speciation in this group is often associated with divergence in wing color pattern, and pattern variation plays an important role in speciation and mate preference (Jiggins, 2008;Jiggins et al., 2001;Merrill et al., 2011Merrill et al., , 2015Merrill et al., , 2019Sánchez et al., 2015). H. erato cyrbia which has compounds not found in other H. erato populations. Our results are also in agreement with the prediction of convergence between comimics, supporting an earlier hypothesis . Our work sets the stage for further research into the biology and function of chemical profiles, and their role in within-and between-species signaling.
It would be challenging to conduct behavioral experiments on large numbers of species and populations, and therefore, identifying the behaviorally active components in pheromone blends across a radiation is beyond the scope of a single study. Other studies have also attempted to predict male sex pheromones without behavioral data, by selecting based on multiple criteria such as male specificity and abundance (Bacquet et al., 2015). This stepwise selection of candidates focuses on within-species characteristics such as abundance, without considering the presence of the compound in other species.
We hypothesized that consistent species-specific compounds are likely to be biologically important. We present an alternative method Chemical profiles are predicted to be highly species-specific if they are involved in species recognition during mating. For instance, orchid bee chemical blends, presumably important for mating and species recognition, show high species specificity, as well as within-species variability, which can be partly explained by geography (Brand et al., 2019;Weber et al., 2016;Zimmermann, Roubik, & Eltz, 2006). We see similar patterns in Heliconius, with greater interspecific than intraspecific differences in chemical profiles. The magnitude of intraspecific differences is smaller in Heliconius, likely due to the fact that orchid bees collect their blends from the environment (Eltz, Whitten, Roubik, & Linsenmair, 1999 We found a correlation between chemical distance and genetic distance. This suggests that neutral evolutionary forces are important in the evolution of chemical bouquets. The correlation between genital chemical distance and genetic distance is a much stronger correlation than previously reported (Estrada, Schulz, Yildizhan, & Gilbert, 2011), possibly due to the quantitative nature of our data.
The strong signal of neutrality suggests that the majority of compounds in the bouquets are neutrally evolving. For example, in the genital bouquet of H. melpomene, one compound, (E)-β-ocimene, can act by itself as an anti-aphrodisiac, with other components of the bouquet thought to moderate its evaporation rate (Schulz et al., 2008). In the future, focusing on the evolutionary patterns of only compounds which exhibit behavioral or electrophysiological responses, rather than the entire bouquet, may disentangle the processes involved in the evolution of these profiles.
Heliconius erato and H. melpomene both exhibit extensive color pattern variation across their geographic range (Sheppard, Turner, Brown, Benson, & Singer, 1985) and these populations also differ in their androconial and genital bouquets. While traditionally predicted to be under stabilizing selection, intraspecific variation between populations in chemical profiles has been documented in other Lepidoptera (Carde & Allison, 2016). Chemical divergence in putative male sex pheromones between populations of Bicyclus anynana is reported to be as large as differences between Bicyclus species and is greater than predicted by genetic divergence (Bacquet et al., 2016). This is in contrast to what we find here, where interspecific differences are much greater than intraspecific ones.
Interestingly, Heliconius erato cyrbia produces many unique genital compounds and is also the most genetically divergent H. erato population in our study, suggesting that genetic drift is important for the evolution of chemical profiles in Heliconius. Across all H. erato populations, we find a correlation between chemical distance and genetic distance, which is weaker for androconial bouquets. In H. melpomene, genetic distance is also weakly correlated with genital chemical divergence. These correlations suggest that some of the geographic variation between populations could be neutral, with stochastic processes important for bouquet evolution in Heliconius.
In contrast, androconial chemical variation in H. melpomene is better explained by geographic distance. This might imply that other evolutionary forces are important for chemical profile evolution in H. melpomene.
One factor potentially involved in geographic variation is larval host plant use. Feeding on different host plants as a larvae affects the production of some minor components of both androconial and genital chemical bouquets . The major components, however, are unaffected by larval host plant, suggesting that any dietary precursors required for compound production are present in different Passiflora species . In Panama, H. cydno and H. melpomene both feed on P. menispermifolia (Merrill, Naisbit, Mallet, & Jiggins, 2013), and yet have different chemical profiles, highlighting that from the same precursors different species can produce different compounds. Furthermore, it is often unclear which is the major Passiflora host plant of any particular Heliconius population. The composition of Passiflora species varies geographically (Benson, 1978;Benson, Brown, & Gilbert, 1975), and both host preference and level of host specificity vary between populations of the same Heliconius species (Castro, Zagrobelny, Cardoso, & Bak, 2018). A greater understanding of the variation in larval diet of Heliconius across the Neotropics will help us understand how much geographic variation in chemical profile can be attributed to host plant use.
Heliconius butterflies are an excellent example of visual mimicry, with different species converging on the same warning color patterns (Merrill et al., 2015;Sheppard et al., 1985;Sherratt, 2008). It has been suggested that chemical compounds could also contribute to mimicry between species (Dettner & Liepert, 1994;Mann et al., 2017). In this study, we find patterns consistent with predictions of convergence between comimics. Individuals within particular comimicry groups, such as H. melpomene and H. erato, seem to converge on a more similar chemical profile. Most known examples of chemical mimicry come from systems of deception, for example, mimicry of ant alarm pheromones by rove beetles to avoid predation, rather than mimicry of aposematic warning signals (Dettner & Liepert, 1994;Stoeffler, Maier, Tolasch, & Steidle, 2007;Vereecken & McNeil, 2010). We suggest that in Heliconius different components of the bouquet could be important for chemical mimicry and species recognition, reducing conflict between these selection pressures.
Convergence of genital bouquets between comimics could be due to the anti-aphrodisiac function of these compounds (Gilbert, 1976;Schulz et al., 2008). Anti-aphrodisiac compounds are transferred from males to females during mating to deter future matings from other males. Convergence in wing pattern between comimics could result in harassment not only by conspecific but also heterospecific males (Estrada & Jiggins, 2008). The use of the same anti-aphrodisiac by comimics could combat interspecific attraction by deterring males of both species, as highlighted by the production of (E)-β-ocimene by H. erato and H. melpomene, as well as other Heliconius species (Estrada et al., 2011).
Compounds could also play a role in predator deterrence. Genital compounds were originally suggested to form part of the antipredation signal (Eltringham, 1925 1984;fireflies, Vencl et al., 2016), effective against avian predators (Guilford, Nicol, Rothschild, & Moore, 1987). Further investigation will be required to determine if odors of Heliconius butterflies act as antipredation signals.
Overall, our study reveals strong species differences in bouquets and the presence of species-specific compounds, as well as intraspecific variation. A pattern of species specificity alongside intraspecific variation could be the result of a balance between stabilizing selection toward a species stereotype, sexual selection promoting diversity, and geographic segregation alongside selection and drift. A challenge for the field is the feasibility of testing for the biological relevance of hundreds of compounds in many species, but we hope that our innovative analysis will stimulate not only further targeted functional studies of putatively important compounds, but also large chemical profile surveys in other study systems of evolutionary interest.

CO N FLI C T O F I NTE R E S T
None declared.

O PE N R E S E A RCH BA D G E
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://osf.io/28yfk/ ?view_ only=c1f7e 7a925 e74de e84fd 2229c bf3f511

A PPE N D I X 4
We compared within and between species and locality Bray-Curtis distances. We focused on H. erato and H. melpomene, as these species were collected in the most localities. We calculated a Bray-Curtis distance matrix and then used the function "dist_groups" from the package usedist to calculate distances between individuals of different groups (Bittinger, 2017). We add statistical comparisons to the violin plots using the function "stat_compare_means" from the package ggpubr (Kassambara, 2019). For both androconia and genitals, the mean chemical distance between individuals is greater between species (androconia, 0.971; genitals, 0.915) than within species (androconia, 0.554; genitals, 0.573). The mean chemical distance between individuals is also greater between localities (androconia, 0.564; genitals, 0.584) than within localities (androconia, 0.457; genitals, 0.466). However, the magnitude of this difference is much smaller, confirming the PERMANOVA and ManyGLM analyses that most variation is explained by species and not geographic location.

A PPE N D I X 5
Here, we rerun statistical analyses excluding unknown compounds or poorly sampled populations.

D O E S R EM OV I N G U N I D ENTI FI ED CO M P O U N DS O R P O O R LY SA M PLED P O PU L ATI O N S A FFEC T M O D EL S O F I NTE R S PECI FI C VA R I ATI O N I N CH E M I C A L PRO FI LE S?
We repeated the interspecific analysis without unidentified com-

WITH G E N E TI C A N D G EO G R A PH I C D I S TA N CE ?
Correlations with genetic and geographic distances were also consistent with results including all compounds. When controlling for geographic distance, genetic divergence is strongly correlated with both androconial and genital chemical divergence (Mantel test, androconia, r = .7897, p = .001; genitals, r = .5203, p = .001). When controlling for genetic distance, geographic distance is significantly but weakly correlated with chemical divergence (Mantel test, androconia, r = .06739, p = .002; genitals, r = .059, p = .003).

FEC T M O D EL S O F I NTR A S PECI FI C VA R I ATI O N I N CH E M I C A L PRO FI LE S?
Individuals of H. erato still strongly group by region when unidenti-

FEC T CO R R E L ATI O N S B E T WE E N I NTR A S PECI FI C CH E M I C A L D I V E RG E N CE WITH G E N E TI C A N D G EO -G R A PH I C D I S TA N CE ?
Results Note: The ten compounds that contribute the most to the deviance explained by a variable are listed for each variable in descending order of contribution. Compounds highlighted with * were also identified by an indicator analysis. a Naphthalene is a known flower volatile, but can also be introduced by contamination. Our blank samples never contained naphthalene, indicating the butterfly origin in our study. Note: The ten compounds that contribute the most to the deviance explained by a variable are listed for each variable. Compounds highlighted with * were also identified by an indicator analysis.

TA B L E A 9
Androconial and genital compounds that are the best indicators of different geographic groups of H. erato Note: A is a measure of group specificity of the compounds, B is a measure of group coverage, and sqrtIV is the indicator value that considers both A and B and ranges from 0 (compound not present in any individuals of that species) to 1 (compound only present in that species, and present in all individuals). Note: A is a measure of group specificity of the compounds, B is a measure of group coverage, and sqrtIV is the indicator value that considers both A and B and ranges from 0 (compound not present in any individuals of that species) to 1 (compound only present in that species, and present in all individuals). Note: The ten compounds that contribute the most to the deviance explained by a variable are listed for each variable. Compounds highlighted with * were also identified by an indicator analysis.