Introduction

Identifying genes contributing to adaptive divergence and reproductive isolation between species is a crucial step for understanding the process of speciation. Reproductive isolation can be caused by mechanisms acting before or after mating. One contributing factor to premating isolation is behavioral or sexual isolation, but unlike postmating isolation, few behavioral genes influencing speciation have been identified (Noor, 2003; Orr et al., 2004; Noor and Feder, 2007). Mating signals in multiple modalities may influence sexual isolation. Chemosensory reception, which includes olfaction and gustation, may have a large role so that differences in pheromones functioning as mating signals can influence sexual isolation (Smadja and Butlin, 2009).

Evidence exists for the effects of both single and multiple genes on pheromonal differentiation. A change in the expression of a single desaturase in the Asian Corn Borer results in a novel pheromonal blend compared with the European Corn Borer (Roelofs et al., 2002), and there is evidence for a single gene change in sibling species of Helicoverpa moths resulting in a pheromonal blend change (Wang et al., 2008). Although in both of these cases single genes are implicated, the allelic difference between species results in a change in the blend of compounds, not just a single pheromone. As there are multiple compounds requiring different biosynthetic pathways, there is the potential for the interaction of multiple genes through both additive and epistatic effects: some enzymes may be required for the synthesis of precursor molecules, whereas other enzymes may modify precursors to final products. In Drosophila, multiple gene effects and epistatic interactions have been found for interspecific differences between D. simulans and D. sechellia, as well as between D. pseudoobscura and D. persimilis (Coyne, 1996; Noor and Coyne, 1996; Gleason et al., 2005).

Drosophila cuticular hydrocarbons (CHCs) can function as pheromones and affect mate recognition (Cobb and Jallon, 1990; Coyne et al., 1994; Etges and Ahrens, 2001; Ferveur, 2005). CHCs are derived from long-chain fatty acids and influence desiccation (Toolson, 1982; Lockey, 1988; Rouault et al., 2004), cold-tolerance (Ohtsu et al., 1998) and starvation resistance (Hoffmann et al., 2001). Species within the D. melanogaster subgroup are either sexually monomorphic or dimorphic for CHCs. Monomorphic species (such as D. simulans and D. mauritiana) have high levels of monoenes, especially 7-tricosene (7-T, which has 23 carbons and one double bond, C23:1). Males of sexually dimorphic species (D. melanogaster and D. sechellia) have monoenes, but the females have dienes, primarily 7,11-heptacosadiene (7,11-HD, C27:2, Cobb and Jallon, 1990). Males of monomorphic species rarely court females of dimorphic species (Cobb and Jallon, 1990). Although it has not been demonstrated definitively that 7-T and 7,11-HD are the active compounds involved in sexual isolation between D. simulans and D. sechellia, 7,11-HD has been shown to be the major stimulant for D. melanogaster courtship (Antony et al., 1985).

7-T and 7,11-HD are not the only CHCs differing between these two species, as other CHCs are present that differ in their number of carbons and double bonds (Table 1). The overall pattern is that CHCs in D. sechellia have longer chains and are less saturated than those found in D. simulans. In a previous study we used a quantitative trait locus (QTL) approach to examine differences in the amount of 7-T and 7,11-HD in females of D. simulans and D. sechellia (Gleason et al., 2005). Through composite interval mapping (CIM), we found at least four QTLs for the monoene 7-T, three on chromosome 3 and one on the X, and two QTLs for the diene 7,11-HD on chromosome 3. Although genes have been identified that may influence CHC biosynthesis (candidate genes), very few were included as markers in that study, though the approximate locations of these genes were determined (Table 6 in Gleason et al., 2005). Of the possible candidate genes, only those on chromosome 3 were close to QTL. Gleason et al. (2005) did not study the other CHCs that differ between these two species.

Table 1 Cuticular hydrocarbons studied and parameters from composite interval mapping
Full size table

All Drosophila CHCs are synthesized de novo from acetate- and linear-saturated fatty acids by elongation, decarboxylation and desaturase pathways (Figure 1, Chan Yong and Jallon, 1986; Pennanec’h et al., 1997). Analysis of multiple CHCs may give an insight into the genetic basis of the two major compounds influencing mate recognition in these two species through establishing the relative contributions of various enzymes associated with CHC biosynthesis (Legendre et al., 2008). Very little is known about the genes encoding carboxylases, but the elongases and desaturases have received more attention. Recent work has shown that elongase F (eloF), expressed in female D. melanogaster and D. sechellia, but not D. simulans, is involved in the elongation of long hydrocarbons, especially diene pheromones (Chertemps et al., 2007). RNA interference knockdown of eloF in oenocytes is sufficient to reduce the expression of long dienes (Chertemps et al., 2007). The role of other elongases in Drosophila CHC expression has not been explored.

Figure 1
figure 1

Hydrocarbon biosynthesis in Drosophila melanogaster. The first ω7 desaturation, carried out by Desat1, is common to males and females. The ω5 desaturation, carried out by Desat2, is only expressed in D. melanogaster African females (those that have 5,9-heptacosadiene (HD)). D. simulans synthesis for both males and females follows the D. melanogaster male pathway. D. sechellia hydrocarbon biosynthesis is presumably similar to that of D. melanogaster, though this has not been shown. Adapted from Legendre et al., (2008).

Full size image

Much more is known about the desaturases, which comprises a gene family with effects on pheromones (Roelofs and Rooney, 2003). Desaturases add double bonds to existing compounds and, hence, may influence the relative amounts of dienes, monoenes and saturated compounds. There are at least 10 fatty acid desaturases in D. melanogaster, seven of which are on chromosome 3. Three of these third chromosome desaturases, desat1, desat2 and desatF, have been shown to be involved in pheromone biosynthesis in D. melanogaster (Figure 1 for example, Dallerac et al., 2000; Legendre et al., 2008).

In D. melanogaster, induced mutations in desat1 have reduced levels of all unsaturated CHCs (including 7,11-HD and 7-T) but increased levels of saturated hydrocarbons (Labeur et al., 2002); one mutation also influences males perception of pheromones (Marcillac et al., 2005). Genetically adjacent to desat1 on chromosome 3, desat2 has been implicated in geographical differentiation in CHCs (7,11-HD vs 5,9-HD) in D. melanogaster (Coyne et al., 1999; Dallerac et al., 2000; Takahashi et al., 2001), though molecular transformation experiments are in disagreement regarding the contribution of desat2 to ecological adaptation in D. melanogaster (Greenberg et al., 2003, 2006; Ritchie and Noor, 2004; Coyne and Elwyn, 2006).

Also on chromosome 3 is desatF (syn. Fad2). The RNA interference knockdown of this gene reduces diene production to <5% of normal and increases the level of monoenes (Chertemps et al., 2006). Males with ‘feminized’ hydrocarbons through the targeted expression of transformer, express desatF and are courted by wild-type males. desatF is expressed in D. sechellia females (but not D. simulans), and hybrids between D. melanogaster and D. simulans have dienes only if the desatF allele from D. melanogaster is present (Legendre et al., 2008). desatF is thus a strong candidate gene for increased levels of dienes.

In addition to genes directly in biosynthesis pathways, genes influencing their regulation may be potential candidate genes. As CHCs are sexually dimorphic in D. sechellia, genes in the sex determination pathway are potential candidates. These include Sexlethal (Sxl), transformer (tra), intersex (ix), doublesex (dsx) and fruitless (fru). In our previous work (Gleason et al., 2005), Sxl, tra and ix were not near QTL. The remaining two genes are both on chromosome 3. dsx, which influences the production of female dienes in D. melanogaster, (Jallon et al., 1988) was on the edge of a QTL (Gleason et al., 2005). The location of fru was not determined in that study.

In this study, we expand our previous investigation (Gleason et al., 2005) using the same individuals to include more CHCs in addition to 7–T and 7,11-HD, and refine the mapping with the addition of marker loci in candidate genes. By examining six CHCs with varying saturation and carbon chain length, we examine the covariance of CHC production to help identify common genetic pathways among CHCs. In addition, by scoring markers in the potential candidate genes desat1, desatF, doublesex (dsx) and fruitless (fru), we are able to further assess if these contribute to QTL.

Materials and methods

Mapping population

This study used the same individuals as Gleason et al. (2005), with the introduction of new markers and additional CHCs in the analysis. Strain maintenance, crosses, molecular markers, cuticular hydrocarbon extraction and genetic map construction are described in Gleason et al. (2005). For the mapping population, female D. simulans were crossed to male D. sechellia and backcrossed to D. simulans to obtain backcross individual females for analysis.

Cuticular hydrocarbons were isolated by washing 5-day-old females in hexane containing 800 ng hexacosane, as a control for extraction efficiency. The quantities of the CHCs (listed in Table 1) were calculated through gas chromatography (described in Gleason et al., 2005) as the area under each peak on the chromatograms. By dividing the CHC peak areas by the quantity of hexacosane recovered, we were able to account for differences in extraction and gas chromatography analysis. Eight extreme data outliers were removed and one was added before the resulting values were natural log transformed to remove a right skew. Remaining residuals were normally distributed. All subsequent analyses were done with the transformed variables. In addition to the backcross females, D. simulans, D. sechellia and F1 hybrid females were also measured. Final sample sizes are given in Table 1.

After hexane washes, DNA was isolated from the backcross females in a 100 μl single fly DNA prep (Gloor and Engels, 1992). For each individual, five morphological markers and 37 molecular markers were scored (described in Gleason et al., 2005). For this study, four new markers in the candidate genes desat1, desatF, dsx and fru, all located on chromosome 3, were also scored using the same procedures (Supplemental Table 1) on the same DNA that had been stored at −20 °C. A fifth candidate gene, desat2 is not presented because the gene is only 7 kb from desat1 in D. simulans and D. sechellia (Tweedie et al., 2009), and is not resolvable as different from desat1 in this analysis. Markers were scored by PCR amplification and visualization with gel electrophoresis. For markers that did not differ in size between D. simulans and D. sechellia, restriction digests were used to distinguish the two alleles (Supplemental Table 1).

Genetic mapping and QTL analysis

A genetic linkage map was assembled from the 46 markers using Antmap with the Kosambi map function (Iwata and Ninomiya, 2006). The map was subsequently used in QTL analyses using QTL Cartographer v 1.17j (Basten et al., 2005). A marker on the fourth chromosome (eyeless) was not significantly associated with any trait and thus results for this chromosome, which comprises about 1% of the genome, are not shown. Composite interval mapping was carried out for the rest of the genome using all significant background markers and a window size of 10 cM. Significance levels at P=0.05 (Table 1) for all CHC were calculated from 1000 permutations of the trait data among marker classes (Churchill and Doerge, 1994). Boundaries of QTL were determined by finding the highest peak and then extending the QTL for 2-LOD (logarithm of odds) around the peak, which is a conservative estimate of the location (van Ooijen, 1992; Manichaikul et al., 2006). When peaks were adjacent, if there was not a 2-LOD drop between them, the QTL region was combined. Nearest markers to the peaks were determined from the map.

Statistical tests

Differences in quantities for each CHC were compared among pure species, hybrids and backcross hybrids using analysis of variance with a Tukey post-hoc test with a 5% family error rate. Epistatic interactions between QTL were detected using generalized linear analysis of variance models with loci as covariates and all first order epistatic interactions. Genotype state (homozygous D. simulans or hybrid) was taken from the marker nearest to the likelihood peak of each QTL. As we examined six traits, an appropriate Bonferroni corrected significance level is 0.008.

Results

QTL analysis

Six of the most abundant CHCs were measured (Table 1). These varied greatly in abundance both within and between D. simulans and D. sechellia females (Figure 2). These differences were not likely to be an effect of a difference in body size between the two species because most of these CHCs were present in one species and absent in the other. For most CHCs, the hybrid mean was intermediate. 7,11-HD showed a dominance effect of D. simulans alleles and 7-T of D. sechellia alleles, and there was transgressive segregation (in the form of overdominance) for 7-H and 7,11-PD. The backcross mean was more similar to D. simulans than to D. sechellia and not statistically significantly different from D. simulans for 7-T, 7-pentacosene (P), 7-heptacosene (H) and 7,11-HD, though 7-P showed transgressive segregation. The backcross mean was more like D. simulans for tricosane, although the backcrosses had significantly less of the compound. For 7,11-PD, the backcross mean was similar to the hybrid mean and D. sechellia.

Figure 2
figure 2

Amounts of cuticular hydrocarbons (CHCs) per individual. Nanogram quantities of each CHC are given with standard errors for Drosophila simulans and D. sechellia females, hybrids of the cross between D. simulans females with D. sechellia males and backcrosses of hybrid females to D. simulans. All values were corrected for the amount of standard recovered for each individual. The sample sizes are given below each bar. Different letters above each bar indicate statistically significant differences between pairs of female types by Wilcoxon's signed-ranks tests with Bonferroni correction for multiple tests.

Full size image

QTL analysis showed 2–5 QTL for each CHC (Figure 3 and Table 2). Effects for tricosane and 7-T were all in the expected positive direction relative to D. sechellia and for 7,11-PD and 7,11-HD were all negative. For the other two CHCs, the directions of the effects were mixed (Table 2). The total phenotypic variance explained by all QTL for individual CHCs ranged from 27.95–85.93%.

Figure 3
figure 3

Quantitative trait loci (QTL) mapping of all six cuticular hydrocarbons (CHCs). Composite interval mapping (CIM) was carried out with a walking speed of 2 cM, a window size of 10 cM and forward/backwards regression on all significant background markers (see Table 1 for the number of significant markers for each CHC). The locations of the marker loci are indicated on the x axis in the conventional left to right order for Drosophila. The experiment-wide significance level of P<0.05 obtained by 1000 permutations of the trait data among marker classes is shown with a horizontal line. Only the highest significance level, for 7,11-heptacosadiene (HD) is shown. Values for the significance levels are given in Table 1. Candidate loci, which are new markers in this study, are in pink. Across the top of the figure are the boundaries of the 2-LOD intervals for each QTL.

Full size image
Table 2 Locations and effects of QTLs for each cuticular hydrocarbon and ratios
Full size table

The ranges of the 2-LOD drop intervals for the QTL were either unique to a CHC (three for 7,11-PD and one for T) or shared among 4–6 CHCs (Figure 3). The unique QTL included the only QTL found on the second chromosome, whereas all the others were on the third chromosome. The QTL regions shared by multiple CHCs were in three distinct areas. The first was on the X chromosome, closest to the forked (f) gene and was shared among the most saturated compounds. No known candidate genes were in this region. The second region was on the left arm of the third chromosome and was shared by both dienes (7,11-PD and 7,11-HD), as well as 7-T and tricosane. The two dienes were very close to desatF, suggesting that desatF was a relevant candidate gene for the dienes. No other candidate genes were in this region.

The final shared QTL was on 3R and influenced all of the CHC. The bounding markers for this region are prospero (pros) and Metallothionein A (MtnA). The two candidate genes desat1 and desat2 were adjacent to this interval and were within the 2-LOD drop QTL interval for tricosane and 7,11-PD, but were not within the QTL for the two CHC which were the most abundant CHCs in D. simulans and D. sechellia, 7-T or 7,11-HD.

Epistasis

Pairwise analyses of QTL for each compound was carried out to identify epistatic interactions (Table 3). QTL for only three compounds had epistatic interactions; all of these compounds were more abundant in D. sechellia than D. simulans (Table 1). For 7,11-PD and 7,11-HD, all epistatic interactions were between a QTL on the left arm of chromosome 3 and one either on the right arm of chromosome 3 or on chromosome 2. 7,11-HD is present in large amounts only when both the QTLs on chromosome 3 are heterozygous (Figure 4).

Table 3 Tests for Epistasis
Full size table
Figure 4
figure 4

Epistatic interaction for 7,11-heptacosadiene (HD). The amount of the cuticular hydrocarbon (CHC) is given for each allelic state (heterozygote or homozygote Drosophila simulans) at the Sod and Metallothionein A (MtnA) loci in the recombinant backcross females. Error bars are 95% confidence intervals.

Full size image

In the case of 7-H, which lacks a QTL on the left arm of chromosome 3, the epistatic interaction detected was between the QTL on the X chromosome and one on the right arm of chromosome 3. Without Bonferroni correction, these two QTL would have a significant (P=0.013) interaction for 7-P as well (data not shown).

Covariance of traits

Quantities of CHC compounds were related as the amounts present showed high covariance and strong positive or negative correlations (Table 4). The potential for covariance to reflect common QTL is an important and rarely tested issue (Gardner and Latta, 2007), and here the greatest covariance (Table 4) was for the traits sharing the most QTLs (3 peaks in common between 7-T and tricosane; Figure 3). 7-P and 7-H were also strongly correlated (and share two QTL peaks; Figure 3). A compound found only in D. sechellia, 7,11-HD, was strongly correlated with all the other CHCs, and was negatively correlated with those CHCs that are most abundant in D. simulans.

Table 4 Covariances and correlation coefficients between quantities of CHC compounds a
Full size table

Analysis of covariance between pairs of traits showed a unique, non-linear relationship for 7-T and 7-P, which formed two distinct clusters with different slopes (Figure 5). The majority of the individuals comprising the cluster with the steeper of the two slopes (that is, a higher ratio of 7-P to 7-T than the other cluster) were heterozygous for pros (Figure 5) and MtnA (data not shown). The majority of the individuals comprising the second cluster were homozygous for these two loci. The mean value of this ratio for D. simulans falls within the cluster for the homozygotes (Figure 5). The difference between homozygotes and heterozygotes was consistent with an elongase function in the QTL region near pros and MtnA, as 7-P is longer than 7-T.

Figure 5
figure 5

The relationship between 7-pentacosene (P) and 7-tricosene (T). Natural log transformed quantities of 7-P vs. 7-T is plotted for all the backcross individuals, as well as the mean values for Drosophila simulans, D. sechellia, their hybrids and the backcrosses.

Full size image

To examine this relationship further, the ratio of 7-P to 7-T was used in composite interval mapping for QTL (Supplementary Figure 1). The ratio of 7-P to 7-T was used to compare a difference in chain length (25 vs. 27 carbons), but not saturation (one double bond each), between CHCs. In all, 3 QTL were identified that account for nearly 94% of the phenotypic variance (Table 2). These QTL were the same as the intervals found for multiple QTL in the single trait mapping, one on the X chromosome and two to either side of MtnA. No epistatic interactions were found for the ratio of 7-P to 7-T.

To study the difference in saturation (two vs. one double bond), but not chain length (both 25 carbons), the ratio of 7,11-PD to 7-P was used in QTL mapping (Supplementary Figure 1). In this case 4 QTL were found that explain only approximately 28% of the phenotypic variance. Again, the X-chromosome QTL was present, as was one near MtnA. In addition, the largest peak was near desatF. A fourth peak is unique, located between Catalase (Cat) and Glucose dehydrogenase (Gld). The QTL nearest desatF had strong epistatic interactions with both the QTL on the X chromosome and the major QTL on the right arm of chromosome 3 (Table 3).

Discussion

Drosophila CHCs have important roles in behavior and ecological adaptation (Lockey, 1988; Cobb and Jallon, 1990; Coyne et al., 1994; Ohtsu et al., 1998; Hoffmann et al., 2001; Fang et al., 2002; Rouault et al., 2004). CHCs result from the action of enzymes in biochemical pathways that start from a common substrate and, therefore, may be phenotypes that are likely to have shared QTL, covariance and epistasis because the underlying enzymes influence multiple traits or relationships between traits. By examining six CHC compounds differing between D. simulans and D. sechellia females, we have found evidence for the following three effects: common QTL regions, covariance and epistasis.

QTL, candidate genes and QTL function

In our previous study (Gleason et al., 2005), we identified QTL for the two CHC, 7-T and 7,11-HD, that have a large influence on sexual isolation between D. simulans and D. sechellia (Cobb and Jallon, 1990; Coyne, 1996). By including additional CHC, here we were able to assign functions to QTL through inferences about shared structure (chain length and saturation). Associations with candidate genes reinforce the hypothesized functions. In addition to the four new markers in candidate genes, we identified other candidate genes through the literature and a search of Flybase (Tweedie et al., 2009; Wicker-Thomas and Chertemps, 2010) for genes with the molecular processes related to CHC production (for example, stearoyl-CoA desaturase activity and fatty-acid elongase activity). There are 10 desaturases and 20 elongases, many of which occur in tandem pairs or in clusters (Supplementary Table 2).

Three chromosome arms had major QTL that overlapped for multiple CHCs: 3L, 3R and the X chromosome. The total proportion of the phenotypic variance explained by QTL was high for the two 25 carbon compounds, both of which show transgressive segregation in the F1 or F2 backcross populations. A major inference from these shared QTL is that the same pathways are involved in the production of multiple CHCs and that phenotypic covariance is reflected in these shared QTL.

Starting with 3L, QTL for the two dienes were located between Sod and Est6 (Figure 3, Table 2). These QTL overlapped with broad QTL located near Acp790A for tricosane and 7-T. The proportion of the phenotypic variance (Vp) explained by the QTL for the dienes greatly exceeded that of the other two compounds (Table 2). The best candidate gene for this region was the desaturase desatF, given that saturation differences mapped most strongly here (Supplementary Figure 1), as shown by the ratio of 7,11-PD to 7-P, two compounds that differ in saturation but not chain length. However, two elongases are adjacent to desatF (Table 2, Supplementary Table 2). As neither of these elongases is expressed in D. melanogaster females (Chertemps et al., 2005), neither are likely to be involved in female CHC biosynthesis and are probably not the causative gene(s) here if expression patterns have not changed for these species, though this remains to be confirmed for D. simulans and D. sechellia.

A QTL on 3R was shared by all of the CHCs implying that the QTL covers one or more fundamental genes involved in the production of CHCs. The most well-studied candidate genes among our markers were the pair desat1 and desat2; however, these fell outside of the QTL for most of the CHCs. The 3R QTL explained a large proportion of Vp for the ratio of 7-P to 7-T, two compounds differing in chain length and not saturation implying that there was an elongase function for this QTL. Two clusters of genes, all elongases, were located in this region. The first cluster containing five genes fell between pros and MtnA, and included eloF, which has female-biased expression in D. melanogaster. Furthermore, eloF affects the relative production of dienes of different chain lengths and is not expressed in D. simulans (Chertemps et al., 2007), thus is an excellent candidate gene. The second cluster of four genes fell between MtnA and slo; one or more of these elongases is likely to be important in the difference between D. simulans and D. sechellia. Most of the literature discussing genetic control of pheromonal differences among species has focused on the role of desaturases (e.g., Coyne et al., 1999; Rosenfield et al., 2001; Fang et al., 2002; Roelofs et al., 2002; Greenberg et al., 2003; Roelofs and Rooney, 2003; Coyne and Elwyn, 2006; Legendre et al., 2008). Although a desaturase has been implicated for 3L, there is at least an equal role in CHC for elongases in the difference between D. simulans and D. sechellia.

The other major QTL region identified affected tricosane and all the monoenes (on the X chromosome, Figure 3) implying a common pathway for these compounds. This QTL was not associated with any known candidate genes. Examination of the region between nonA and f in Flybase (Tweedie et al., 2009) identified 148 genes, the majority of which do not have a known function. Of the genes with inferred or tested functions, none have functions that are known to be involved in CHC production. Thus, as yet undescribed loci are implicated here.

The sex determination genes used here as candidate genes are not implicated in the shared pathways of these CHCs, as these genes are not associated with the shared QTL. dsx was within the bounds of the unique tricosane QTL (Figure 3) that overlapped with a QTL for the ratio 7,11-PD to 7-P (Supplementary Figure 1), although neither of these compounds have a QTL in this location when mapped individually. Our search for additional candidate genes on Flybase identified a fatty-acid elongase, CG2781, tightly linked to dsx (Supplementary Table 1) and thus may be the relevant gene. In the original report of a fru allele affecting male CHC production (Gailey and Hall, 1989), the fru allele was actually an inversion and the CHC effect mapped not to what was later shown to be fru locus, but to a breakpoint that in our map (Figure 3) is located just to the right of gl. This area is implicated in QTL for some compounds and thus fru is not the candidate gene for CHCs.

Sex determination pathway genes are not likely to have a role in the species differences here because most mutations in sex determination genes that affect pheromone production change the pheromone from that of one sex to that of another (for example, Savarit and Ferveur, 2002; Tompkins and McRobert, 1995). In the evolution of these two species, it is not clear if differentiation of D. sechellia female CHCs were gained or if dimorphism were lost in D. simulans (Jallon and David, 1987). If dimorphism were lost, sex determination might have a role in CHC production by ‘masculinizing’ D. simulans females. If sexual dimorphism were gained and the sex determination pathway were involved, then the pathway would be responsible for de novo synthesis of new compounds, which is unlikely. Regardless, there is no strong evidence for the role of sex determination pathway genes here.

Epistasis and covariance of traits

Strong epistatic interactions are prevalent between the left arm of chromosome 3 and other loci for the dienes. In general, QTL for the monoenes do not have strong epistatic interactions, with the exception of 7-H, which is marginally more abundant in D. sechellia (Figure 2). One implication of the epistatic interactions for the dienes is that we have identified QTL in a single pathway such that the production of the dienes is dependent upon multiple enzymatic reactions, and thus, dienes are not produced in large quantities unless multiple genes have a D. sechellia allele (Figure 4).

Complex interactions are further exemplified by those of 7,11-HD, as it was correlated with all the other CHCs (Table 4). As the predominant CHC in D. sechellia, it was negatively correlated with the D. simulans CHCs (with the exception of 7-P) and positively correlated with the other D. sechellia CHCs. Thus, there is more evidence for common pathways for these CHCs.

All the epistatic interactions detected involved QTL of CHCs, which are more abundant in D. sechellia than in D. simulans. As these compounds are both longer and less saturated than the predominant ones of D. simulans, they are products from further along the biosynthesis pathway (Figure 1). Thus, the difference in epistatic interactions between these two species may reflect the more complicated biosynthesis required for D. sechellia rather than D. simulans. Detection of epistatic interactions for CHCs more abundant in D. sechellia than in D. simulans may be an artifact of the experimental design in that we used backcross progeny, and thus, cannot contrast all possible genotypes (no individuals were homozygous for D. sechellia alleles). However, the epistatic interactions were detected for the D. sechellia alleles in the heterozygous state, thus it is unlikely that we have missed epistatic interactions that would be evident only in a homozygous state.

Multiple candidate genes per QTL region, the presence of epistasis and the correlation of multiple CHCs may imply that isolating individual genes involved in these species differences may be difficult. Testing a single gene (for example, in a complementation test) may fail to reveal effects, although a particular candidate gene may be an important component of the biosynthetic pathway for a CHC. Thus, there may be support for a profound effect of genetic background on alleles that affect CHCs as suggested by Greenberg et al., 2006.

Major models for the evolution of post-mating isolation between species rely on epistatic interactions within co-adapted gene complexes, such as the Dobzhansky–Muller models of hybrid incompatibilities (thought to underly Haldane's Rule; Turelli and Orr, 2000). Genes affecting sexual isolation have been argued to be more likely under direct rather than indirect selection, and thus may be less prone to such interaction effects than genes affecting post-mating isolation (Ritchie and Phillips, 1998). However, in this QTL study, we find that genes responsible for important species-specific divergence in pheromones are also affected by epistasis. The nature of the production of these important mating signals (by alternative biosynthetic pathways from a common precursor) may make CHCs prone to few genes of large effect with strong epistasis.