Polymorphisms in a desaturase 2 ortholog associate with cuticular hydrocarbon and male mating success variation in a natural population of Drosophila serrata

Introduction

Sexual selection is a source of strong directional selection in natural populations (Kingsolver et al., 2001; Hereford et al., 2004) that is thought to be responsible for some of nature's more spectacular phenotypes such as elaborate male armaments and ornamental display traits (Darwin, 1871; Andersson, 1994). Beyond the evolution of phenotypic novelty, sexual selection may contribute to speciation (Ritchie, 2007), the maintenance of sexual reproduction (Agrawal, 2001; Siller, 2001), accelerating rates of molecular evolution (Swanson & Vacquier, 2002) and generating intragenomic conflicts between sexes (Bonduriansky & Chenoweth, 2009). Although sexual selection has been studied in depth from phenotypic (Andersson, 1994) and quantitative genetic perspectives (Chenoweth & McGuigan, 2010), there are very few species in which the basis of natural variation in sexually selected traits has been established at the level of DNA sequence variation (Wilkinson et al., 2015). Candidate gene association studies have been more commonly applied to the analysis of post-copulatory (e.g. Fiumera et al., 2007; Zhang et al., 2013) rather than precopulatory sexual selection (Chenoweth & McGuigan, 2010). Understanding the genetic basis of sexually selected traits may be informative of the selective processes affecting standing variation, mutational target sizes and the assumptions of models of sexual selection (Wilkinson et al., 2015).

A molecular understanding of sexually selected trait variation requires empirical approaches that can detect the genomic variants underlying not only trait variation but also the fitness components of sexual selection. Sexually selected traits evolve, in the first instance, via their genetic covariance with a component of relative sexual fitness such as pre- or post-copulatory mating success. Ultimately, this must also covary genetically with total fitness in some way (Chenoweth & McGuigan, 2010). For example, although a gene may be involved in the construction of a sexually selected trait during development, it may not produce variation in the trait or if it does, it may not produce variation in a way that generates differences in sexual fitness between individuals of the same sex. Thus, to understand the responses to sexual selection at the DNA level requires the detection of variants with pleiotropic effects on sexual traits and sexual fitness. Although classic quantitative genetic approaches have been used to estimate genetic correlations between sexually selected traits and components of both sexual and nonsexual fitness (Brooks, 2000; Hine et al., 2002), these summary measures are not informative of the specific evolutionary processes maintaining variance at given loci. For example, we may wish to understand whether the variants underlying sexually selected traits conform to a mutation–selection balance model (Haldane, 1927) or whether they sometimes are affected by balancing selection between opposing sexual and natural selection (Johnston et al., 2013). By dissecting the genetic variance in both trait and sexual fitness to the level of individual variants and their population frequencies, we may be able to test these ideas in new ways (e.g. Park et al., 2011).

The approach outlined above requires experimental systems where both sexual fitness and sexually selected traits can be assayed simultaneously, ideally in an outbred population. A well-studied system of sexually selected traits is the cuticular hydrocarbons of Drosophila serrata. Cuticular hydrocarbons are waxy long-chained molecular synthesized in oenocyte cells under the insect cuticle. At the molecular level, D. serrata CHCs are a suite of monoenes, dienes and methyl-branched alkanes (Howard et al., 2003). Female D. serrata express genetically based preferences that generate sexual selection on male CHCs (Hine et al., 2002; Rundle et al., 2009). Recent physical manipulation of male CHCs has demonstrated experimental links between male mating success and CHC phenotype (Chung et al., 2014), and multigenerational artificial selection on ‘attractive’ CHC blends produced a correlated response of increased male attractiveness to females (Hine et al., 2011). Although D. serrata represents a tractable system for understanding the genetic architecture of sexually selected trait variation, like many systems for studying sexual selection, it remains largely unexplored at the molecular level. One notable exception is the recent study of the mFAS gene which, when manipulated in transgenic males, reduced production of two methyl-branched CHCs important for male mating success (Chung et al., 2014).

The biosynthetic pathway for CHCs in insects is not completely understood (Howard & Blomquist, 2005); however, it is clear that two processes in hydrocarbon synthesis involve the desaturation and elongation of precursor molecules. In the biosynthetic pathway, the enzymes responsible for these processes are desaturases and elongases, respectively (Howard & Blomquist, 2005). In D. melanogaster, two desaturase genes, desat1 and desat2, are of particular interest (Dallerac et al., 2000; Labeur et al., 2002). Both genes code for ∆9 fatty acyl-CoA desaturases; however, in species where both genes are expressed, the two enzymes produce desaturases from different fatty acyl-CoA precursors, altering the double-bond placements of the final CHC products (Ferveur, 2005; Howard & Blomquist, 2005). In geographically distinct populations of D. melanogaster, a 16 base-pair deletion mutation in the desat2 gene has been associated with the differing expression of CHC dienes in females (Dallerac et al., 2000; Coyne & Elwyn, 2006), and this CHC difference may also contribute to sexual isolation between geographical races (Fang et al., 2002).

Here, we perform a candidate gene association study of male CHCs and male mating success in a sample of 100 families derived from an outbred natural population of D. serrata. We cloned and sequenced a D. serrata ortholog of the desat2 gene, which unlike in D. melanogaster, is expressed in both sexes rather than only in females. Our experimental design allowed us to simultaneously estimate the genetic variation in mating success and CHCs and to associate this variation with single nucleotide polymorphism (SNP) variation at the desat2 locus. Our analyses suggest that some variants at this locus may have pleiotropic effects on sexually selected traits and on the fitness component of precopulatory sexual selection, whereas other variants may influence only one of these.

Materials and methods

We employed an indirect association study design where multiple families are founded from outbred parents who are then genotyped, and samples of their offspring, are phenotyped. Such indirect designs have been successfully used for candidate gene association studies in D. melanogaster (Weeks et al., 2002; Kennington et al., 2007; Rako et al., 2007). The test of association is performed between family mean phenotype and the expected frequency of the minor allele in the progeny. This approach offers two advantages over the creation of inbred lines or directly genotyping and phenotyping wild-caught flies. First, the traits are not subject to inbreeding depression, and second, the designs produce significantly more power per genotype for low heritability traits compared to genotyping and phenotyping all individuals (Chenoweth & Visscher, 2009). This is an issue of relevance when dealing with fitness component traits (e.g. male mating success), which tend to have lower heritabilities than morphological traits (Mousseau & Roff, 1987; Roff & Mousseau, 1987). We collected 200 inseminated female D. serrata from a natural population in St Lucia, Qld, Australia, in 2008 and allowed them to lay their eggs in vials in the laboratory. From each brood, we sexed either a single male (100 lines) or a single female (100 lines). These virgin males and females were then randomly paired and allowed to mate, creating 100 completely independent full-sib families. Both parents from each family were sequenced at an ortholog of the D. melanogaster desat2 locus using Sanger sequencing, and a maximum of 10 male offspring per family were scored for both mating success and CHC phenotype.

Statistical analysis

We followed standard transformation of CHC peak areas using logcontrasts, which avoid the unit-sum constraint that is typically associated with compositional data types such as CHC proportions (Aitchison, 1986). The diene Z,Z-5,9-C_24 : 2 was the common divisor in all analyses to allow comparison with other D. serrata CHC studies. The CHC data were screened for multivariate outliers prior to any analysis using the Mahalanobis distance method (Sall et al., 2005). Outlier analysis is performed across the entire data set ignoring any family relationships. Typical outliers do not run in families but reflect errors in CHC extraction or peak integration. Broad-sense heritabilities for CHCs and male mating success were calculated with a univariate random-effects model: CHC = line + error. Each male offspring was raised singly in an individual vial; thus, environmental variance due to any rearing vial effect is contained with the error term in this model. We tested for significant sexual selection on male CHCs using standard Lande & Arnold (1983) least squares multiple regression. Linear models were fitted using the MIXED procedure of sas version 9.2 (SAS Institute, Cary, NC, USA).

The test for association in the indirect design consists of a regression of expected minor allele frequency (MAF) in the progeny against family mean phenotype (Chenoweth & Visscher, 2009). For full-sib families, the expected minor allele frequency in the progeny can have values 0, 0.25, 0.5, 0.75 and 1, reflecting the different types of parental matings possible. The CHCs of D. serrata are highly genetically correlated, and previous work has demonstrated that these traits are best analysed within a multivariate framework (Blows et al., 2004). Consistent with this analytical approach, we implemented a multivariate test of association based on canonical correlation analysis (Ferreira & Purcell, 2009). In addition to making use of information in the data that are invisible to univariate analyses – cross-trait phenotypic and genetic covariances – the approach has the added benefit of reducing Type 1 error rates (Ferreira & Purcell, 2009). We implemented the test using the MIXED procedure of sas (Version 9.2), which allowed us to assume an unstructured variance–covariance matrix among the eight CHC traits. The linear model for this analysis is:

(1)

where Y is a n = 100 × 8 matrix of family means for the eight CHCs scores, X is an 100 × 2 element design matrix with 1s in the first column and the coding variables for the progeny SNP frequency in the second column, and B is 2 × 8 element matrix containing the eight intercept (row 1) and the slope estimates (row 2) that represent the additive effects of the SNP on each CHC. The second row of B contains the ‘effects vector’ for the additive effects of the SNP on the eight traits. The residual error is E. We fitted model 1 to the data using an unstructured error variance–covariance matrix in the error term.

We tested all SNPs with a MAF ≥ 0.02 for associations with CHCs and male mating success. We observed significant linkage disequilibrium between some pairs of SNPs within the gene, thereby violating the assumption of independence among association tests for each SNP. Therefore, we assessed statistical significance using permutation tests. For the mating success data, we implemented a permutation test using R (R Core Team ) shuffling the relationship between mating success and each SNP genotype 1000 times. The permutation test for the multivariate test of association was implemented in a similar fashion but using SAS, again with 1000 permutations of the data. False discovery rate (FDR) q-values were estimated from the permutation-based P-values using the q value package in R (Dabney & Storey 2009). For this study, a nominal significance level was q-value ≤0.05 for CHCs and ≤0.1 for male mating success owing to the fact that CHCs have a higher heritability than male mating success.

When we observed SNPs having significant associations with male CHCs, we compared the direction of their vectors of additive effects to the multivariate direction of sexual selection in the population, β, by calculating the angle between the two vectors:

(2)

where a_chc is the vector of additive CHC effects from 1. Small values of θ indicate that SNP effects are closely aligned with the direction of sexual selection (i.e. may increase sexual fitness through a CHC effect), whereas larger angles indicate poor alignment.

Results

Quantitative genetic and selection analyses

We detected standing genetic variation for both the sexually selected traits (CHCs) and the fitness component of precopulatory sexual selection (male mating success). CHCs and mating success were scored on 785 male flies from 99 families. There was significant genetic variance for all CHCs and male mating success (Fig. 1), with likelihood ratio tests confirming line level variance components significantly greater than zero at P < 0.05. Broad-sense heritabilities for CHCs ranged from 0.52 for the logcontrast diene Z,Z-5,9-C_25 : 2 to values exceeding 1.0 for the methyl-branched alkane 2-Me-C₃₀ and the diene Z,Z-5,9-C_29 : 2 (Fig. 1). As expected, the broad-sense heritability of male mating success was lower (H² = 0.10).

We confirmed that CHCs were under directional sexual selection in this sample of flies using multiple linear regression (Lande & Arnold, 1983) (anova: F_7,776 = 4.76, P < 0.001, R² = 0.05). Consistent with previous work in this population, directional sexual selection was strongest on the methyl-branched alkane, 2-MeC₂₈, and the diene, Z,Z-5,9-C_29 : 2 (Table 1)_.

Table 1. Vector of directional sexual selection gradients, β, and additive effects vectors, a, for the three CHC-associated SNPs, q-value < 0.05. Also indicated are the angles between additive effects vectors and β for each SNP. All vectors have been normalized to a unitary length.

CHC	β	a 132	a 371	a 580
Z,Z-5,9-C24 : 2	−0.3150	−0.6143	−0.3660	−0.6627
Z,Z-5,9-C25 : 2	0.2982	−0.2617	−0.5959	−0.3096
Z-9-C25 : 1	−0.1096	−0.1225	−0.5283	−0.4195
2-Me-C26	−0.0326	−0.2617	0.2729	0.0445
Z,Z-5,9-C27 : 2	−0.1692	−0.4785	−0.0934	−0.3820
2-Me-C28	0.4736	−0.3513	0.0145	−0.2662
Z,Z-5,9-C29 : 2	0.7389	−0.0294	0.3463	0.2518
2-Me-C30	−0.0003	−0.3426	0.1686	0.0825
Angle ax-β		88°	75°	73°

Genotype–phenotype associations

We genotyped 36 SNPs at the desat2 locus (MAF ≥ 0.02). Typically, linkage disequilibrium was low among the 36 SNPs, although some LD blocks were apparent (Fig. S1). For CHCs, no individual SNPs survived experimentwide permutation test significance (Doerge & Churchill, 1996). However, three SNPs had significant marker-wise associations (q-value < 0.05, Fig. 2). These were at positions 132 (5′ UTR), 371 (Exon 1) and 580 (Exon 2). The SNP s132 was in weak linkage disequilibrium with the other two SNPs, which were more closely linked to each other (Fig. S1). A strength of our approach is that it allows allelic effects on multiple CHCs to be directly compared with the direction of multivariate sexual selection in this population. The estimated angles between additive effects vectors and β, the direction of sexual selection, were all quite large (Table 1, Fig. 3a). SNP s132 had the largest angle, a near-orthogonal value of 88, whereas s371 and s580 had angles of 74° and 73, respectively.

For male mating success, 15 of the 36 markers tested had q-values < 0.1 (Fig. 2, Table 2). The estimated effect sizes of these were moderate (ranging from 2 to 6% of the among-line variance). The adjusted R² value of a model fitting all of these 15 SNPs simultaneously was 0.11, which suggests that the linkage disequilibrium among SNP markers inflates the effect size estimates of individual SNPs. Two of the SNPs with significant CHC associations also had nominally significant mating success associations (s371 and s580), whereas there was no suggestion of a mating success association for s132.

There was a directional skew in minor allele effects on male mating success. Of the 15 significant male mating success associations, 13 were found to reduce male mating success (Fig. 3b). This result was significant using a binomial test, which assumes an equal number of positive and negative associations under the null hypothesis (P = 0.007). However, a binomial test does not account for linkage disequilibrium between markers and so the test is likely anticonservative. To address this, a permutation test that preserved the linkage disequilibrium structure of the SNP data was performed to determine how often 13 or more negative associations might be expected under the null hypothesis of no association between SNPs and male mating success. This test was still significant (10 000 permutations; P = 0.043), suggesting that, at least at the broad level, minor-frequency alleles within this gene tend to be associated with a reduction in male mating success.

Table 2. Significant associations between SNPs at the desat2 locus and male cuticular hydrocarbons and male mating success (MMS) in an outbred population of D. serrata.

Position	MAF	Location	Base change	AA change	CHC q-value	MMS q-value	MMS R2
s132	0.02	5′ UTR	T/G		0.0228b	0.4021
s350	0.15	Exon 1	C/T		0.1580	0.0929a	0.020
s371	0.25	Exon 1	G/A		0.0114b	0.0706a	0.045
s401	0.02	Exon 1	C/T		0.1580	0.0829a	0.035
s438	0.18	Intron1	T/A		0.1580	0.0706a	0.060
s439	0.18	Intron1	G/T		0.1580	0.0706a	0.058
s500	0.02	Exon 2	C/T		0.2991	0.0829a	0.030
s580	0.11	Exon 2	C/T		0.0114b	0.0794a	0.033
s630	0.22	Exon 2	C/T		0.2088	0.0929a	0.024
s826	0.15	Intron 2	G/C		0.2088	0.0706a	0.043
s849	0.23	Intron 2	A/T		0.2088	0.0929a	0.023
s914	0.02	Exon 3	C/T		0.2972	0.0929a	0.023
s984	0.16	Exon 3	G/T	A → S	0.1580	0.0706a	0.044
s1022	0.14	Exon 3	C/T		0.1580	0.0807a	0.033
s1159	0.05	Intron 3	T/G	I → M	0.2169	0.0829a	0.030
s1354	0.04	Exon 4	T/C		0.1580	0.0929a	0.022

• ^a q < 0.1.
• ^b q < 0.05.

Discussion

We provide some of the first evidence of the relationship between nucleotide variation and sexually selected traits in a model system for the study of precopulatory sexual selection, the cuticular hydrocarbons of Drosophila serrata. Focusing on a gene confirmed to be involved in CHC biosynthesis in other species, we found evidence for associations with both CHC phenotype and male mating success, a fitness component of precopulatory sexual selection.

Three variable sites in the desaturase 2 gene had multivariate associations with CHCs. Consistent with its role in affecting dienes in D. melanogaster, the strongest loadings in the additive effects vectors in D. serrata were for diene CHCs. Like D. melanogaster, D. serrata dienes have double bonds at the 9th position, and thus, these effects are consistent with that expected of a Δ9 desaturase. One association was in the 5′ UTR, whereas the other two were in Exons 1 and 2, respectively. Although the association at s132 (5′ UTR) may have regulatory activity, it is less clear what the functional consequence of the two exonic SNPs might be as they lead to synonymous changes. However, results from other genomewide association studies report quantitative trait associations for synonymous sites (Magwire et al., 2012) as have candidate gene studies of postcopulatory sexual selection in D. melanogaster (Reinhart et al., 2015). SNPs s371 and s553 were in linkage disequilibrium (Fig. S1), and therefore, these two sites may not reflect independent effects. Although linkage disequilibrium was generally quite low within the desaturase 2 locus in our sampled population, we cannot exclude the possibility that these associated sites are linked to causal SNPs outside the sequenced region.

Our multivariate association testing approach allows us to express SNP effects in the same trait space as multivariate sexual selection and compare their orientations. The two CHC-associated SNPs, s371 and s553, were also associated with male mating success, with angles of 74° and 73, respectively. These sites may have pleiotropic effects on CHCs and male mating success and are thus likely to contribute to the genetic covariance between CHCs and male mating success in this population. It is this class of variant that would be predicted to contribute to a phenotypic response to sexual selection. In contrast, the variant at s132, although associated with CHCs, was not associated with male mating success. This becomes clearer when considering the nearly orthogonal angle (88°) between its estimated effects vector and the direction of sexual selection, β, which suggest they are uncorrelated. These relatively weak overall associations between SNP effects on CHCs, and the direction of sexual selection are consistent with previous quantitative genetic studies of D. serrata that report a poor alignment between β and the major axes of CHC standing variance (Hine et al., 2004). As the above analyses consider only a tiny fraction of the D. serrata genome, future experiments, conducted on a genomewide scale, may provide a more complete picture of the orientation of the effects of individual genomic variants with the direction of sexual selection.

In providing estimates on direction of effect on male mating success and naturally occurring allele frequencies, our association study data can be used to evaluate possible models for the maintenance of sexually selected variation at this locus. Broadly, a model of mutation–selection balance predicts that lower frequency alleles should be those that reduce fitness (Haldane, 1927). We detected a strong skew in the direction with which the minor allele affected male mating success, because in 13 of 15 cases, the minor allele decreased male mating success, a result that was robust to linkage disequilibrium between SNPs. Similar patterns have been observed in studies of early-onset human diseases where the minor-frequency allele more often increases disease risk (Park et al., 2011). However, in this study we have not investigated ‘fitness’ per se, but rather a component of male sexual fitness. Thus, the detection of a pattern consistent with mutation–selection balance for sexual fitness can go some way towards informing us of the orientation between sexual and nonsexual fitness. The patterns here suggest a positive association between sexual and nonsexual fitness, at least in this gene.

Although we may expect mating success to be a large part of male sexual fitness, these results contrast with artificial selection results in this system, which suggest that some of the variants affecting attractive CHC blends reside at low frequency because of negative pleiotropic effects with nonsexual fitness (Hine et al., 2011). However, the Hine et al. (2011) study was conducted on a long-term laboratory-adapted population rather than a recently derived natural population as we have studied here. Long-term laboratory populations are more likely to have reached a local adaptive peak, and therefore, we may expect different equilibrium frequencies of alleles with pleiotropic effects on sexual and nonsexual fitness (Long et al., 2012). Variants where sexual and nonsexual fitness effects are closely aligned will have reached high frequencies if adaptive or very low frequencies if deleterious, leaving those with opposing pleiotropic effects segregating and therefore contributing to much of the genetic variance for sexual fitness (Connallon & Clark, 2012). It should be noted that the minor-frequency alleles in this study are not particularly rare when considering the expected frequencies of alleles that are highly correlated with total fitness (Eyre-Walker, 2010). Thus, it is likely that, despite detecting an overall bias in the frequency–effect relationship suggesting male mating success contributes positively to total fitness, total selection is likely to have been weak, and therefore, opposing pleiotropic fitness effects remain likely. Sources of counter-selection on CHCs in D. serrata could include sexually antagonistic effects (Gosden et al., 2012), or effects mediated through desiccation or temperature resistance (Gibbs et al., 1997; Chung et al., 2014).

There are some important caveats to note concerning our association testing approach. Association testing should be viewed as an initial step towards understanding links between genotype and sexually selected phenotypes in D. serrata as it is not a demonstration of causality and, although suggestive, these associations will ultimately need to be replicated in other samples. Although the sophisticated tools available in D. melanogaster are not available for D. serrata, careful transgenic approaches are possible (Chung et al., 2014) and new advances in genome engineering technology in nonmodel insects may ultimately facilitate experimental validation in this species (Kistler et al., 2015). Our approach of estimating both genetic variation in sexually selected traits and male mating success within the same experimental design as individuals are genotyped has permitted confirmation that the traits are indeed under sexual selection and, more importantly, the identification of variants that influence traits in ways that actually covary with sexual fitness itself. With rapidly falling sequencing costs, our approach could potentially be extended to a genomewide scale with the use of whole-genome resequencing of parents. Such genomewide approaches may be useful in exposing variation across the genome in the models that can best explain the maintenance of genetic variation in sexually selected traits.