1. Genetics and Genomics
Download icon

Genetic architecture of natural variation in cuticular hydrocarbon composition in Drosophila melanogaster

  1. Lauren M Dembeck
  2. Katalin Böröczky
  3. Wen Huang
  4. Coby Schal
  5. Robert R H Anholt
  6. Trudy F C Mackay  Is a corresponding author
  1. North Carolina State University, United States
Research Article
  • Cited 36
  • Views 2,010
  • Annotations
Cite this article as: eLife 2015;4:e09861 doi: 10.7554/eLife.09861

Abstract

Insect cuticular hydrocarbons (CHCs) prevent desiccation and serve as chemical signals that mediate social interactions. Drosophila melanogaster CHCs have been studied extensively, but the genetic basis for individual variation in CHC composition is largely unknown. We quantified variation in CHC profiles in the D. melanogaster Genetic Reference Panel (DGRP) and identified novel CHCs. We used principal component (PC) analysis to extract PCs that explain the majority of CHC variation and identified polymorphisms in or near 305 and 173 genes in females and males, respectively, associated with variation in these PCs. In addition, 17 DGRP lines contain the functional Desat2 allele characteristic of African and Caribbean D. melanogaster females (more 5,9-C27:2 and less 7,11-C27:2, female sex pheromone isomers). Disruption of expression of 24 candidate genes affected CHC composition in at least one sex. These genes are associated with fatty acid metabolism and represent mechanistic targets for individual variation in CHC composition.

https://doi.org/10.7554/eLife.09861.001

eLife digest

The outermost layer of an insect’s body is called the epicuticle and is made of a blend of fat molecules. “Cuticular hydrocarbons” (or CHCs) are the most common fat molecules in the epicuticle, and play an important role in protecting the insect’s body from harsh, dry habitats. CHCs also have other roles in insect behavior. For example, these molecules act as chemical cues when insects search for mates (i.e. pheromones), and they can even contribute to camouflage.

Insects are amongst the most diverse groups of animals on Earth, and different species have different blends of CHC molecules in their epicuticles. Fruit flies are a useful model to understand the genetics of CHC production, including CHCs that act as sex pheromones. Previous research has analyzed the CHCs made by both sexes in several fruit fly strains. However this work was unable to uncover which genes influence how much of a given CHC an individual fly will make.

Dembeck et al. have now looked into CHC production in a collection of 205 different fly strains, all of which have already had their total genetic material sequenced and studied. Comparing these known sequences and looking for associations between genetic differences and particular CHCs uncovered 24 genes that may be involved in CHC manufacture. Only six of the genes had been identified previously. Dembeck et al. found that interfering with the activity of any of the 24 genes had a knock-on effect on many other CHCs present in the flies’ epicuticle.

These 24 genes could to be pieced together in a network that is needed to make and recycle CHCs. The complexity and flexibility of this network can explain in part how insects have been able to build epicuticles for almost every environment. These data set the stage for future work directed towards understanding the evolutionary significance of variation in CHC composition in many fruit fly populations.

https://doi.org/10.7554/eLife.09861.002

Introduction

Insects comprise the most species-rich class in the animal kingdom. They evolved about 480 million years ago and their fecundity and rapid evolutionary adaptations have made them successful in populating almost every ecological niche on our planet. The evolution of mechanisms for desiccation resistance was critical for insects to colonize dry land. To prevent desiccation insects evolved the ability to produce and accumulate species-specific blends of fatty acid-derived apolar lipids on the epicuticle; the most prominent of these are cuticular hydrocarbons (CHCs) (Jallon et al., 1997). CHCs are produced continuously by specialized cells called oenocytes, and are transported through the hemolymph and then to the cuticular surface through specialized pore canals (Romer, 1991; Schal et al., 1998; Blomquist and Bagnères, 2010). The primary role of CHCs is desiccation resistance (Gibbs, 1998; 2002), but they have been co-opted to serve as chemical signals and cues mediating intra- and inter-specific social interactions (Venard and Jallon, 1980; Jallon, 1984; Ferveur, 2005). These interactions include species and nest-mate recognition, assessment of reproductive status, and mate choice, and CHCs play a prominent role in camouflage and mimicry that mediate inter-specific parasitic relationships (Stanley-Samuelson and Nelson, 1993; Blomquist and Bagnères, 2010).

Adaptive evolution acts through selection on phenotypic variation within a population. Thus, understanding the genetic basis for individual variation in CHC composition will shed light on evolutionary mechanisms of assortative mating (Noor et al., 1996; Ishii et al., 2001; Rundle et al., 2005; Gleason et al., 2005) and evolution of social organization (Howard and Blomquist, 2005; Richard and Hunt, 2013).

The extensive genetic resources available for Drosophila melanogaster make it a valuable model for studying the genetic basis of CHC production and natural variation in CHC composition. Mature D. melanogaster have sexually dimorphic CHCs ranging from chain lengths of 21 to 31 carbons (C21–C31) (Antony and Jallon, 1982; Jallon and David, 1987). Males produce predominantly shorter-chain CHCs (< C26) and they use two of the CHCs, 7-C23:1 and 7-C25:1, as sex pheromones. Females produce predominantly longer-chain dienes, among which 7,11-C27:2 and 7,11-C29:2 act as the primary female sex pheromone components (Antony and Jallon, 1982; Cobb and Jallon, 1990; Arienti et al., 2010).

CHCs are produced from the fatty acid biosynthetic pathway, which begins with an acetyl-CoA. Acetyl-CoA carboxylase (ACC) then catalyzes the synthesis of malonyl-CoA, and the multifunctional protein fatty acid synthase (FASN) successively incorporates malonyl-CoA units onto the acetyl-CoA, elongating the chain by two carbons each time and forming long chain fatty acids (LCFA). RNAi-knockdown of ACC in the oenocytes completely eliminates CHCs in both male and female D. melanogaster (Wicker-Thomas et al., 2015). Products of insect FASN, including Drosophila FASN, can be 14, 16 or 18 carbon fatty acids. A thioesterase that is part of the multienzyme FASN removes the elongated chain as a free fatty acid, and fatty acid elongation and desaturation use the CoA derivative and take place in the microsomal fraction (endoplasmic reticulum). Members of a family of tissue-specific elongases (ELOVL) catalyze the incorporation of malonyl-CoA units to form very long chain fatty acids (VLCFA). After condensation of the malonyl-CoA with the fatty acyl-CoA the next three steps in each elongation cycle include reduction of a carbonyl to an alcohol (by a 3-keto-acyl-CoA-reductase; KAR), dehydration (by a 3-hydroxy-acyl-CoA-dehydratase; HADC), and reduction of the carbon-carbon double bond by a trans-enoyl-CoA-reductase (TER). The VLCA as the CoA derivative is readuced to an aldehyde by a fatty acyl-CoA reductase (FAR). The one-carbon chain-shortening conversion of aldehydes to hydrocarbons is catalyzed by a cytochrome P450 enzyme, Cyp4G1 (Qiu et al., 2012). Desaturation reactions to introduce double bonds, leading to unsaturated fatty acids, appear to occur on the 16 or 18 carbon fatty acyl-CoAs.

A comprehensive analysis of CHCs in both sexes segregating in a panel of recombinant inbred lines derived from a natural population identified 25 quantitative trait loci (QTL) in females and 15 in males contributing to variation in CHCs (Foley et al., 2007), but this study did not have the power to resolve QTLs to individual genes. QTL mapping analyses have also identified genomic regions called small monoene quantities (smoq) and seven pentacosene (sept), respectively, associated with variation in the proportions of 7-C23:1 and 7-C25:1 in males (Ferveur and Jallon, 1996). In a mutagenesis study, another genomic region, nerd, drastically reduced 7-C23:1 production in males and altered courtship behavior (Ferveur and Jallon, 1993). However, none of these male loci have been resolved to specific genes. Several additional genes affecting CHC biosynthesis have been described in D. melanogaster: an acetyl-CoA carboxylase (ACC) (Wicker-Thomas et al., 2015); two fatty acid synthases, FASN2 (CG3524) which is involved in the synthesis of precursors of 2-methylalkanes (Chung et al. 2014), and FASN1 (CG3523) which is expressed in the fat body but nonetheless contributes to the CHC pool (Wicker-Thomas et al., 2015); an elongase (EloF, Wicker-Thomas et al., 1997; Chertemps et al., 2007); KAR (CG1444), TER (CG10849) and HADC (CG6746) whose knockdown eliminates or reduces CHC synthesis (Wicker-Thomas et al., 2015); a cytochrome P450 (Cyp4G1 [CG3972], Qiu et al., 2012); and three desaturases, Desat1, Desat2 and DesatF (Labeur et al., 2002; Marcillac et al., 2005; Chertemps et al., 2006; Fang et al., 2009).

Here, we used the sequenced, inbred lines of the D. melanogaster Genetic Reference Panel (DGRP) (Mackay et al., 2012; Huang et al., 2014) to perform genome wide association (GWA) analyses for nearly all detectable CHCs in both sexes in a scenario where all common genetic variants are genotyped and local linkage disequilibrium (LD) is sufficiently low to identify candidate genes and causal polymorphisms. We found considerable heritable genetic variation in a majority of male and female CHCs, distilled the axes of genetic variation into several principal components (PCs), and performed GWA analyses on each PC. We identified 24 candidate genes plausibly associated with CHC biosynthesis and for all of them disruption of their expression altered CHC profiles in males, females, or both sexes. Surprisingly, we also found that the DGRP lines are segregating for the ancestral and deletion alleles in Desat2, previously associated with CHC profiles thought to be unique for African flies. Finally, our results provide a new perspective and highlight the complexity of the biosynthetic and catabolic pathways that regulate the dynamics of CHC composition and provide the stage for adaptive evolution. In agreement with other recent studies of complex traits, our results demonstrate that the genetic architecture underlying potentially adaptive traits can consist of many, even hundreds, of polymorphic loci with small effects affecting different aspects of the phenotype.

Results

CHCs in the DGRP

In this article, linear alkanes are referred to by the abbreviation n-Cx, where ‘‘x’’ is the total carbon number. For example, tricosane is designated n-C23. Methyl-branched alkanes are referred to with a y-Me-Cx prefix, where 'y' is the carbon onto which the methyl group is bound. For example, 9-Me-C23 is a 23 carbon chain with a methyl on the 9th carbon. For monoenes (z-Cx:1) and dienes (z,z-Cx:2) the number of double bonds is indicated after the colon and the double bond position/s are 'z' or 'z,z"). For example, 7-C23:1 has one double bond between the 7th and 8th carbons. We identified and quantified 71 female CHCs and 42 male CHCs in 169 and 157 DGRP lines, respectively (Table 1, Figure 1, Supplementary file 1). Sixteen of these CHCs have not been described previously in D. melanogaster. Eight of the new compounds were methyl-branched CHCs, seven were dienes, and one was a monoene. Nine of these compounds were detected only in females.

Table 1

Cuticular lipids identified by GC-MS in DGRP males and females. NI = not identified; nd = not detected; bold typeface = not previously identified in D. melanogaster.

https://doi.org/10.7554/eLife.09861.003
#Cuticular componentRetention index#Cuticular componentRetention index
1n-C212100210033NI2516nd
2x-C22:1 (quantified only in ♂)2179217934NI2521nd
3x-C22:1nd218435 13-Me-C25
11-Me-C25
2533nd
ccis-vaccenyl acetatend218936 5-Me-C25 2550nd
4n-C222200220037 8,12-C26:2 2555nd
57,11-C23:22259nd387,11-C26:22560nd
62-Me-C2222632263392-Me-C2525622562
7NInd2267406,10-C26:22566nd
89-C23:12273227341 9-C26:1 (only in ♀)
3-Me-C25
25722572
97-C23:122802283427-C26:125772577
106-C23:122852286436-C26:1 + impurity (i)25812581
115-C23:12291229144n-C2626002600
12x-C23:12294nd459,13-C27:22652nd
13n-C2323002300467,11-C27:2 (only in ♀)
2-Me-C26
26642663
14 11-Me-C23
9-Me-C23
23362336475,9-C27:2 (only in ♀)
9-C27:1
26752675
157,11-C24:22355nd487-C27:126822682
16 x,y-C24:2 23632363495-C27:12693nd
173-Me-C23
9-C24:1 (only in ♀)
2373237350n-C2727002700
187-C24:1 (quantified only in ♂)2377237751 8,12-C28:2 2756nd
196-C24:123802380522-Me-C27
7,11-C28:2 (only in ♀)
27612761
20NI (Variable in DGRP lines but not detected in GC-MS samples)ndnd53 6,10-C28:2
3-Me-C27
9-C28:1
2768nd
215-C24:12386238654NI27722772
22n-C242400240055n-C2828002800
239,13-C25:22451nd569,13-C29:22852nd
247,11-C25:22460nd572-Me-C28
7,11-C29:2 (only in ♀)
28642862
252-Me-C2424632463589-C29:1
5,9-C29:1 (only in ♀)
28752875
26 6,10-C25:2 24682468597-C29:128822882
275,9-C25:2 (only in ♀)
9-C25:1
2474247460n-C2929002900
288-C25:12478nd61 8,12-C30:2 (only in ♀)
2-Me-C29
29612961
297-C25:124822483622-Me-C3030603060
30 6-C25:1 2485nd637,11-C31:23065nd
315-C25:12492249264n-C3131003100
32n-C2525002500ISn-C3232003200
Representative male and female chromatograms from the DGRP.

Male cuticular lipids of DGRP_38 are shown on the top (blue) and female CHCs of DGRP_786 are mirrored below (red). All peaks for both sexes were assigned a unique number based on its corresponding compound determined by GC-MS; thus compounds shared between the sexes carry the same number. See Table 1 for the list of compound names. Compounds not previously described in D. melanogaster are shown in bold typeface. Some CHC isomers were not resolved by conventional GC, so a few chromatogram peaks contain more than one CHC. pA = picoAmperes, c = cis-vaccenyl acetate, * = contaminants from CHC extraction, IS = internal standard (n-C32).

https://doi.org/10.7554/eLife.09861.004

We assessed the extent to which the CHCs were genetically variable in the DGRP. All but three female CHCs and one male CHC (female peaks 26, 39, and 52 and male peak 59) showed significant among-line variation in a univariate ANOVA (Supplementary file 2). Broad sense heritabilities in females ranged from 0.98 for 7,11-C25:2 (peak 24) to 0.22 for 6-C25:1 (peak 30). Broad sense heritabilities for males ranged from 0.97 for 7-C25:1 (peak 29) to 0 for 9-C29:1 (peak 58; Supplementary file 2).

African female CHC phenotypes

The Desat locus was the first desaturase gene sequence described in insects (Wicker-Thomas et al., 1997) and has been implicated in the biosynthesis of pheromones. It consists of two desaturase genes, Desat1 and Desat2. Desat1 is expressed in both sexes and encodes a Δ-9 desaturase that catalyzes the synthesis of palmitoleic acid, an ω-7 fatty acid and precursor to 7-monoene and the first double bond of 7,11-dienes (Labeur et al., 2002; Marcillac et al., 2005). Desat2 has a female-specific effect on CHC production and has been associated with adaptive divergence of African and Cosmopolitan races of D. melanogaster (Greenberg et al., 2003, but see Coyne and Elwyn, 2006). Desat2 encodes a functional desaturase in African females, but is inactive in Cosmopolitan females due to a 16-bp deletion in the promoter region (Coyne et al., 1999; Dallerac et al., 2000, Takahashi et al., 2001). Females with an intact Desat2 gene produce altered CHC profiles which are high in the pheromonal CHC isomers, 5,9-C27:2 and 5,9-C29:2, and low in the 7,11-isomers. African D. melanogaster females also exhibit a strong behavioral bias against non-African males (Fang et al., 2002; Takahashi and Ting, 2004; Grillet et al., 2012).

Surprisingly, females of 15 DGRP lines expressed the African phenotype; i.e., they had high levels of 5,9-C27:2 and low levels of the primary female sex pheromone, 7,11-C27:2 (Figure 2A). Since Desat2 (cytological position 87B10) is located within the common African inversion In3R(K) (computed breakpoints: 86F1-86F11;96F11-96F14 (St. Pierre et al., 2014) we expected the Desat2 allele status (ancestral African or a Cosmopolitan 16-bp deletion in the promoter), CHC phenotype and inversion status to perfectly co-segregate in the DGRP lines. However, this was not the case. A total of 17 DGRP lines with female CHC phenotypes contained the ancestral, functional Desat2 allele (Table 2, Figure 2B). There was significant variation in the 7,11- and 5,9-C27:2 peaks among these lines (5,9-C27:2 & 9-C27:1 F = 35.17, p<0.0001; 7,11-C27:2 & 2-Me-C26 F = 16.09, p<0.0001; Supplementary file 3). We obtained female CHC data for two African lines, Z53 and Z30, for comparison. Females had significantly less 7,11-C27:2 according to deletion status and correspondingly more 5,9-C27:2 (Figure 2C). Females from DGRP lines that were either homozygous or heterozygous for the ancestral Desat2 sequence had intermediate amounts of the 7,11- and 5,9-C27:2 compared to the DGRP lines homozygous for the deletion or the African lines, respectively.

DGRP lines segregate for the female African CHC phenotype, Desat2 allele, and In3R(K) inversion status.

(A) Overlaid chromatograms of African D. melanogaster CHCs (Z30 and Z53), a DGRP line with an African-like CHC phenotype (DGRP_235), and a Cosmopolitan DGRP line (DGRP_714). (B) DGRP lines with at least one ancestral Desat2 allele exhibit natural variation in the percentage of each CHC peak for the isomeric sex pheromones 7,11-C27:2 (2-Me-C26 co-elutes with 7,11-C27:2) (gray) and 5,9-C27:2 (9-C27:1 co-elutes with 5,9-C27:2) (red). (C) Box-plots of the proportion of each sex pheromone peak for DGRP and African lines according to Desat2 allele and In3R(K) genotypes. DGRP_105 and DGRP_551, which have more Cosmopolitan-like phenotypes despite having the functional ancestral Desat2 allele, are indicated.

https://doi.org/10.7554/eLife.09861.005
Table 2

Phenotypes and In(3R)K genotypes for females from DGRP lines with functional Desat2 alleles. Red text indicates "mismatched" Desat2 genotype ('+' = ancestral; '-' = 16-bp deletion) and inversion status (‘INV’ = In(3R)K; ‘ST’ = Standard karyotype). Blue background indicates "mismatched" Desat2 genotype and phenotype.

https://doi.org/10.7554/eLife.09861.006
DGRP lineDesat 2 genotype% 7,11-C27:2% 5,9-C27:2RatioIn(3R)K status
DGRP_31 + / -23.718.71.27INV / ST
DGRP_38 + / -19.424.90.78INV / ST
DGRP_48 + / -19.125.00.76INV / ST
DGRP_100 + / +22.332.30.69INV / INV
DGRP_136 + / -14.733.70.44INV / ST
DGRP_309 + / -19.117.81.07INV / ST
DGRP_440 + / -20.014.01.43INV / ST
DGRP_559 + / -23.510.52.24INV / ST
DGRP_646 + / +17.324.60.70INV / INV
DGRP_802 + / -18.523.40.79INV / ST
DGRP_732 + / -19.317.31.12INV / ST
DGRP_367 + / +12.636.80.34ST/ ST
DGRP_776 + / +16.78.981.86ST / ST
DGRP_509 + / +17.827.50.65ST / ST
DGRP_235 + / +10.322.20.46ST / ST
DGRP_551 + / -26.64.296.20ST / ST
DGRP_105 + / +21.45.104.20INV / INV

The co-segregation of the functional allele and In3R(K) was not perfect: five of the lines with the ancestral Desat2 sequence were homozygous for the standard karyotype (Figure 2C, Table 2). Furthermore, two lines with the ancestral and presumably functional Desat2, DGRP_105 and DGRP_551, did not exhibit the African CHC phenotype. In total, six of the 17 lines with the ancestral sequence had mismatched inversion and CHC status in females. It is possible that the lines with homozygous standard karyotypes are actually segregating for the inversion at low frequency and thus only the homozygous standard flies were sampled for karyotyping. Alternatively, the Desat2 deletion may have occurred prior to the inversion event.

We next checked Desat2 for potentially damaging genetic variants (Mackay et al., 2012; Huang et al., 2014). We identified five Desat2 alleles unique to DGRP_105. Two variants are synonymous coding (3R_8262545_SNP and 3R_8263020_SNP), one is a deletion causing a frameshift (3R_8263023_DEL), and two are nonsynonymous coding (3R_8263031_MNP and 3R_8263048_SNP). The frameshift and nonsynonymous variants are potentially damaging and could explain why this line produced the Cosmopolitan phenotype despite having the functional Desat2 allele. We did not find such evidence for DGRP_551, suggesting this line may contain unknown genetic variants that inhibit the production of 5,9-C27:2.

CHC correlations and principal components analyses

Most of the CHCs belong to homologous series in which the chain length increases by two carbons; thus, these compounds may be genetically correlated due to a shared biosynthetic pathway and the data may be confounded by multi-colinearity (Martin and Drijfhout, 2009). We visualized the correlations between CHCs using modulated modularity clustering (MMC) (Stone and Ayroles, 2009). The MMC algorithm clusters highly correlated variables based on the Spearman's rank correlation coefficients (ρ). As expected, some CHCs were highly correlated within each sex (Figures 3 and 4; Supplementary file 4).

MMC modules of DGRP female CHCs based on Spearman's rank correlation coefficients (ρ).

Correlations are color-coded from +1 (dark red) to -1 (dark blue). Correlated CHCs are clustered into groups (modules). Modules (outlined in black) are arranged along the diagonal according to the average strengths of the correlations within each cluster; the most strongly correlated modules are on the top left and the weakly correlated modules are on the bottom right.

https://doi.org/10.7554/eLife.09861.007
MMC modules of DGRP male CHCs based on Spearman's rank correlation coefficients (ρ).

Correlations are color-coded +1 (dark red) to -1 (dark blue). Correlated CHCs are clustered into groups (modules). Modules (outlined in black) are arranged along the diagonal according to the average strengths of the correlations within the groups; the most strongly and weakly correlated are on the top left and bottom right, respectively.

https://doi.org/10.7554/eLife.09861.008

In the first two female modules, seven dienes had strong positive correlations with each other. There was also one peak (47 – 5,9-C27:2 and 9-C27:1) that strongly negatively correlated with those dienes and the shorter-chain (≤ C25) alkanes of module 3 (Figure 3, Supplementary file 4). Similarly, the module 3 shorter-chain alkanes had strong positive correlations with each other, some dienes of modules 1 and 2, and monoenes and dienes in module 5. These four modules (1, 2, 3, and 5) all had weak to moderate negative correlations with module 7, which consisted of strongly intercorrelated longer-chain (≥ C25) alkanes and dienes.

We found similar trends in the male CHCs (Figure 4; Supplementary file 4). Module 1 consisted of longer-chain alkanes that negatively correlated with the shorter-chain alkanes of module 2. This could also be seen between module 2 and the longer-chain monoenes in module 7. These negative correlations were a consistent trend between other long- and short-chain CHCs and exemplify the tradeoff between short- and long-chain compounds, as the latter are produced through the elongation of fatty acids that serve as precursors for the former.

The correlations among CHCs are plausible given the biology of CHC production. We computed PCs of genetically variable CHCs within each sex to reduce the dimensionality of the data to orthogonal PCs. The first seven and five PCs accounted for 98.00% and 98.12% of the total variation among the DGRP lines for female and male CHCs, respectively (Table 3, Figure 5, Supplementary file 5). We hypothesized that the GWA results would provide insights into the genetic architecture underlying the MMC trends.

Principal component biplots for PC1 and PC2 of DGRP CHCs.

(A) Female and (C) male PC1 and PC2. (B) Female and (D) male PC1 and PC2 eigenvectors. The percent of variance explained by each PC is indicated on the x- and y-axes. In (A) and (C) DGRP lines are color-coded (Supplementary file 5).

https://doi.org/10.7554/eLife.09861.009
Table 3

Percent of CHC variation in the DGRP explained by PCs.

https://doi.org/10.7554/eLife.09861.010
SexNumberEigenvaluePercentCumulative percent
Female 10.006141.1641.16
20.004329.4770.63
30.002114.5085.13
40.00096.2291.35
50.00053.0794.42
60.00032.3096.72
70.00021.2998.01
Male 10.017075.5275.52
20.003314.5790.10
30.00104.5994.69
40.00052.0496.73
50.00031.3998.12

GWA analyses

We performed GWA analyses using the PCs of natural variation in individual chromatographic peaks to identify novel components of the CHC metabolic pathways in D. melanogaster. None of the female or male PCs were affected by the presence of the endosymbiont Wolbachia pipientis in some of the DGRP lines (Supplementary file 6). Female PC1 and PC2 and male PC1 were affected by the In(2L)t inversion; female PC1, PC2 and PC3, and male PC5 were affected by the In(3R)K inversion; and the In(3R)P inversion only affected male PCs (PC1, PC3, PC5) (Supplementary file 6). We corrected for these effects prior to conducting the GWA analyses. Although these inversions contain genetic variants affecting CHC production, they cannot be resolved by GWA analysis due to elevated linkage disequilibrium within the inversions (Mackay et al., 2012; Huang et al., 2014) and are thus excluded from consideration.

We identified genetic variants in or near 305 (173) genes nominally (P≤10–5) associated with female (male) PCs (Supplementary file 7). Although all of the top variants did not reach individual Bonferroni-corrected significance levels, most quantile-quantile plots indicated no systematic inflation of the test statistic and a clear departure from random expectation below P<10–5, justifying our choice of this reporting threshold and suggesting that the top associations were enriched for true positives (Figure 6).

QQ-plots of CHC PCA GWA P-values.

(A–G) Female PCs. (H–L) Male PCs.

https://doi.org/10.7554/eLife.09861.011

Several of the top variants associated with each PC are in or near candidate genes with plausible roles in CHC metabolism (Supplementary file 7). In females, these include a single nucleotide polymorphism (SNP) in the fourth intron of Lipase2 (Lip2) associated with variation in PC1; a SNP 641 bp upstream of the cytochrome P450 gene, Cyp49a1, associated with PC2; SNPs in and near Desiccate (Desi), a gene previously shown to contribute to desiccation resistance in D. melanogaster, associated with PC4 and PC6; variants in and near two peroxidase genes, Immune-regulated catalase (Irc) and Peroxidase (Pxd), associated with PC5; and variants in three fatty acid metabolism genes (CG14688, CG9458, CG8814) associated with variation in PC6. Several notable candidate genes were also implicated by variants associated with PC7; CG9801 contained the 3 most significant SNPs. One of the top associated SNPs causes a missense mutation in Cyp6w1. Cyp4s3 and a predicted dihydrolipoamide branched chain acyltransferase (St. Pierre et al., 2014), CG5599, were down- and up-stream, respectively, of associated variants. We did not find any associated variants within or near Desat1 or Desat2. In the case of Desat2 the ancestral allele was not tested for association, because there were only six DGRP lines homozygous for the insertion, so its minor allele frequency (MAF = 6/169 = 0.035) did not reach the MAF cutoff (0.05) for evaluation. Further, any effect of this variant would have been minimized by correcting for the effect of the In(3R)K inversion.

In males, nearly all of the variants associated with variation in PC1 were in or near CG13091, a putative fatty acyl-CoA reductase, of which one, a nonsynonymous coding variant (2L_8521314_SNP), was the most significant variant in this study (P = 2.19E-11) and passes the Bonferroni-correction for multiple tests (Supplementary file 7). Many of the variants in CG13091 were in perfect or near-perfect linkage disequilibrium, and thus it is not possible to discern which was/were the causal variant(s). Variants in CG13091 were also associated with variation in PC4 and PC5. Genes tagged by variants associated with PC2 include approximated (app), a palmitoyltransferase; PHGPx, a peroxidase; and CG16979, a thiolester hydrolase. Top variants in the PC3 GWA analysis included two genes predicted to be fatty acid elongases (CG18609, CG30008) and an NADH dehydrogenase (CG8680). The most significant variant in the PC5 GWA analysis was a nonsynonymous SNP (3R_8220563_SNP, P = 1.97E-09) in CG10097, which also encodes a fatty acyl-CoA reductase. Interestingly, there are two cytochrome P450 genes located upstream of CG10097, one functional (Cyp9f2) and one pseudogenized (Cyp9f3Psi).

Functional validation of candidate genes

We selected 24 candidate genes with plausible contributions to various stages in the biosynthesis or turnover of CHCs and tested the effects of disruption of expression of these genes on CHC composition. All but one of these genes had publicly available transgenic UAS-RNAi lines (Dietzl et al., 2007). We used an oenocyte-specific GAL4 driver line, PromE(800)-GAL4 (Billeter et al., 2009), to restrict reduction in candidate gene expression to the CHC-producing cells. For the gene for which no RNAi line was available, CG10097, we used a PiggyBac insertion line to study the effects of this mutation on CHC composition (Supplementary file 8) (Thibault et al., 2004).

The candidate genes implicated by our GWA analyses that we selected for further functional assessment are annotated to encode a palmitoyl transferase (app), fatty acyl reductases (CG13091 and CG10097), a thiol hydrolase (CG16979), thioredoxin peroxidases (PHGPx and Prx6005), fatty acid elongases (CG30008, CG18609, and CG9458), cytochrome P450s (Cyp49a1, Cyp9f2, and Cyp4s3), peroxidases (Irc, Pxd and Pxn), an NADH dehydrogenase (CG8680), and a dihydrolipoamide branched chain acyltransferase (CG5599), which is involved in the metabolism of branched chain amino acids leucine, isoleucine and valine, which serve as precursors for methyl-branched alkanes (Blomquist and Bagnères, 2010). Disruption of expression of any of these genes by targeted RNA interference resulted in altered CHC compositions, often with sexually dimorphic effects. The alterations in CHC composition as a consequence of gene disruption were often complex and unexpected (Figures 7 and 8, Figure 7—figure supplements 124).

Figure 7 with 24 supplements see all
Summary of RNAi and mutant experiments for female CHCs.

UAS-RNAi target gene and the CG10097e00276 mutant are indicated on the horizontal axis. CHC names and numbers are listed on the y-axis. Data are color coded to represent P-values (P ≤ 0.05) from t-tests for the mean differences of the experimental and the control lines. Black = no significant change; blue = significant decrease; green = significant increase; gray = not applicable (peaks 46 and 57 split into two peaks for the CG10097 mutant).

https://doi.org/10.7554/eLife.09861.012
Summary of RNAi and mutant experiments for male CHCs.

UAS-RNAi target gene and the CG10097 e00276 mutant are indicated on the horizontal axis. CHC names and numbers are listed on the y-axis. Data are color coded to represent P-values (P ≤ 0.05) from t-tests for the mean differences of the experimental and the control lines. Black = no significant change; blue = significant decrease; green = significant increase; gray = not applicable (peaks 46 and 57 split into two peaks for the CG10097 mutant).*

https://doi.org/10.7554/eLife.09861.037

In insects, the evidence to date suggests that the elongated fatty acyl-CoA is reduced to an aldehyde prior to oxidative decarbonylation, the latter step being catalyzed by CYP4G1 in Drosophila (Qiu et al., 2012). While it is possible that the elongated fatty acyl-CoA is reduced from acyl-CoA to aldehyde to alcohol and then reoxidized to aldehyde before oxidative decarbonylation, to date there is no evidence for this. CG13091 and CG10097 encode fatty acyl-CoA reductases, which are both expressed at high levels in males and at low levels in females. FARs reduce fatty acyl-CoA to an alcohol (Howard and Blomquist 2005). Inhibition of the production of these FARs promoted the production of longer chain CHCs. Males in which CG13091 expression was targeted with RNAi had higher relative amounts of longer-chain CHCs in general and reduced shorter-chain monoenes and methyl-branched CHCs. In particular, 7-C25:1 and 7-C27:1 and 2-MeC26, and 2-MeC28 increased substantially (Figures 8 and 9). The increase in 7-C25:1 is of particular interest, as this compound acts as a male sex pheromone that mediates female mate choice. Females showed a similar trend but to a lesser extent; they also had some increased longer-chain dienes and, like the males, increased 7-C27:1, 7,11-C27:2 & 2-MeC26, and 2-Me-C28 & 7,11-C29:2 (Figures 7 and 9). Both sexes of the CG10097 mutant were similar, with higher relative amounts of longer-chain CHCs. They also had lower amounts of shorter-chain monoenes and methyl-branched CHCs. In the CG10097 mutant and control females, we were able to separate the 7,11-C27:2 and 2-Me-C26 peaks as well as the 2-Me-C28 and 7,11-C29:2 peaks (Figure 9), enabling us to infer decreased 7,11-C27:2 and 7,11-C29:2 and increased 2-Me-C28 relative to the control. Our results suggest that the FARs encoded by CG13091 and CG10097 may be specifically associated with alkane and monoene synthesis.

Example chromatograms of oenocyte-specific RNAi knockdowns and mutants – CG13091 and CG10097.

(A) and (B) PromE(800)-GAL4 x UAS-CG13091. (C) and (D) Exelixis mutant CG10097 e00276. pA = picoAmperes, IS = internal standard, CHCs significantly increased or ↓ decreased according to the individual t-tests.

https://doi.org/10.7554/eLife.09861.038

The complexity of effects on CHC composition through RNAi-targeting is illustrated by the diverse effects on CHC composition of the three different fatty acid elongases encoded by CG30008, CG18609 and CG9458. In males, CG30008 may play a role in the elongation of precursors of n-alkanes and monoenes, and CG18609 may elongate precursors of longer-chain n-alkanes and 2-methlyalkanes (Wicker-Thomas and Chertemps, 2010). Disruption of CG30008 expression in males resulted in greatly reduced amounts of total CHCs (~2-fold). The effect was less severe in females, but the overall trend was also towards reduced CHCs (Figure 10). Disruption of CG18609 decreased longer-chain CHCs in both sexes but increased 2-Me-C24 in females (Figure 7) (Wicker-Thomas and Chertemps, 2010). Males also had increased total CHCs and large increases in shorter-chain CHCs (Figure 8). This effect was mimicked by interference with expression of Cyp9f2, which also resulted in fewer longer-chain monoenes, alkanes, and many methyl-branched CHCs. However, as for CG18609, 2-Me-C24 increased in females and other shorter-chain methyl-branched CHCs were trending upward. In males, there were overall increases, but especially in the shorter-chain CHCs. These observations suggest that CG18609 may function in the biosynthesis of CHCs in coordination with Cyp9f2. Interestingly, the CG9458 knockdown had sexually dimorphic effects on CHC production, increasing male CHCs and decreasing nearly all n-alkanes and monoenes in females. This suggests that CG9458 is critical for the biosynthesis of n-alkanes and monoenes in females.

Example chromatograms of oenocyte-specific RNAi knockdowns – CG8680 and CG30008.

(A) and (B) PromE(800)-GAL4 x UAS-CG8680. (C) and (D) PromE(800)-GAL4 x UAS-CG30008. pA = picoAmperes, IS = internal standard, CHCs significantly increased or ↓ decreased according to the individual t-tests.

https://doi.org/10.7554/eLife.09861.039

The three cytochrome P450s that we tested (Cyp49a1, Cyp9f2, Cyp4s3) all affected overall amounts of CHCs. RNAi knockdown of Cyp49a1 and Cyp9f2 in males led to general increases in CHCs, with a few exceptions. The increase in CHCs in males with compromised Cyp49a1 function was accompanied by a decrease in many female CHCs. Reduced expression of Cyp4s3 resulted in increased CHC abundance in both sexes. The link between these CYPs and CHC production and maintenance is unclear. However, we speculate that oxidation reactions mediated by these CYPs may regulate CHC degradation and turnover. The specific functions of most insect CYPs are still unknown (Chung et al., 2009). Given the role of CYP4G1 in CHC production (Qiu et al., 2012), we believe it is possible for these CYPs to have similar functions that are perhaps specific to particular subsets of CHCs.

The complex interrelationships that give rise to variation in sexually dimorphic CHC profiles is further illustrated by RNAi interference of Irc, Pxd, and Pxn, which all have corresponding alleles associated with variation in CHC composition in the DGRP and encode peroxidases. However, interference with their expression through targeted RNAi resulted in different shifts in CHC composition. Disruption of Irc reduced the amounts of monoenes and increased the amounts of dienes in females, while increasing male CHCs, reminiscent of the effects of disruption of Cyp49a1. The effects of interference with Pxd were more complex. In females, nearly all CHCs increased, while in males many odd-chain CHCs increased, but there were decreases only in even-chain CHCs. A similar phenomenon was observed with disruption of Cyp4s3. Pxn may be important for diene synthesis because the knockdown females had decreased dienes and corresponding increased levels of longer-chain monoenes. However, in males there were decreased longer-chain monoenes, methyl-branched alkanes, and n-alkanes but increased shorter-chain CHCs. CG7724 is inferred to contribute to oxidation-reduction processes and steroid synthesis. Disruption of CG7724 expression in females resulted in a decrease in longer-chain monoenes and methyl-branched CHCs and an increase in 2-Me-C22, 2-Me-C24, 2-Me-C25, 2-Me-C26 and 7,11-C27:2, x,y-C24:2 and x,y-C26:2. Males also showed increases in 2-methyl alkanes but also in shorter-chain alkanes and monoenes. These results reveal a complex and dynamic network of oxidative enzymes of which the summed activity determines the sexually dimorphic composition of CHCs.

Disruption of the NADH dehydrogenase CG8680 and of CG5599, which is predicted to have dihydrolipoamide branched chain acyltransferase activity, resulted in remarkable increases in the total amount of CHCs in both males and females (Figures 7, 8 and 10, Figure 7—figure supplements 2, 4, Supplementary file 9). Individual control female and male flies produced ~1.5–2 µg and ~1–1.5 µg of CHCs, respectively; in contrast, the RNAi-CG8680 females and males produced ~5.5 µg and ~3 µg, respectively, of CHCs per fly (Figure 7—figure supplement 4). The CG5599 knockdown caused a >4 µg increase per fly in each sex, representing ~3-fold (~7 µg/fly) and 4-fold (~5.5 µg/fly) increase for females and males, respectively. Thus, it is possible that these genes play roles in catabolic pathways thus resulting in an increase in CHCs. Alternatively, RNAi knockdown of these genes may lead to reduced or abnormal grooming behavior leading to an accumulation of CHCs (Böröczky et al., 2013). Inhibition of expression of Pxd, CG8814 and Cyp4s3 also resulted in increased levels of total CHCs in both sexes, with notable decreases in 2-Me-C26 in Pxd and Cyp4s3 males (Figures 7 and 8, Figure 7—figure supplements 5, 14, 23).

The gene products of app, PHGPx, and Prx6005 may be involved in the release of fatty acids from the fatty acid synthase complex. The app palmitoyltransferase may be specific to methyl-branched fatty acids since we observe decreases in methyl-branched CHCs in males and females when its expression is disrupted. Male and female PCAs were here clearly separated from the controls and in both sexes the total amounts of CHCs were slightly increased (Figures 7 and 8, Figure 7—figure supplement 1). Additionally, males had elevated longer-chain methyl-branched alkanes, n-alkanes and monoenes, while females had increased abundances of n-alkanes, methyl-branched alkanes, monoenes and dienes of both longer- and shorter-chain lengths. RNAi targeting of Prx6005 increased shorter-chain monoenes and alkanes in both sexes. In females, longer-chain dienes and methyl-branched CHCs were trending downward, and in males levels of 2-Me-C26 and 2-Me-C24 decreased. Disruption of PHGPx increased the dienes 7,11-C25:2 and 9,13-C25:2, and some longer-chain n-alkanes in females. Levels of many short-chain monoenes and nearly all 2-methyl alkanes were elevated in males. Furthermore, disruption of CG16979, which encodes a gene product annotated as having thiolesterase activity, had no effect in females but caused many CHC increases in males, predominantly in monoenes. Total CHCs were also elevated (Figures 7 and 8). Thus, there appears to be functional specialization among thiolesterases in the biosynthesis of CHCs.

We also assessed the effects of two candidate genes, Desi and Lip2, associated with variation in CHC composition in the DGRP that affect desiccation resistance. Expression of Desi fluctuates in wandering D. melanogaster larvae in response to environmental conditions and RNAi knockdown of Desi in larvae leads to higher mortality (Kawano et al., 2010). However, the nature of the resistance to desiccation is unknown and CHCs were not phenotyped in the larvae. In contrast to the GWA results, knocking down Desi had no effect on female CHCs, while the male CHCs clearly separated from the controls in the PCA (Figure 7—figure supplement 17). Males had increased alkanes, monoenes, and 2-Me-C28 (Figure 8). Lip2 has been associated with clinal variation of life history traits in D. melanogaster populations in the eastern United States (Fabian et al., 2012), and has triglyceride lipase activity that could regulate release of free fatty acids for CHC synthesis. Females in which expression of Lip2 was targeted with RNAi had increased n-alkanes and increased total CHCs, whereas the males had increased 2- and 3-methyl alkanes (Figure 7).

Finally, we examined the effects of CG14688 and CG9801; little is known about the function of both of these genes (St. Pierre et al., 2014). Suppression of gene expression resulted in sexually dimorphic increases and decreases in overall CHCs. RNAi targeting of CG14688 increased female CHCs but had more complex effects in the males, increasing shorter-chain n-alkanes and methyl-branched alkanes and decreasing longer-chain monoenes. Males expressing RNAi against CG9801 had increased total CHC levels, while females of these lines had overall decreases in CHCs. In mammals, alpha-oxidation is used to chain-shorten 3-methyl-fatty acyl-CoAs by one carbon. We are not aware of any examples of alpha-oxidation in insects, but it could be involved in CHC metabolism.

Discussion

Since CHCs represent the boundary between the organism and its environment and mediate social interactions while offering protection against adverse environmental effects, variation in CHC profiles may present a target for natural selection and adaptive evolution. Using the DGRP, we report one of the most comprehensive characterizations of natural variation in insect CHCs to date. We provide unambiguous evidence for extensive heritable individual variation in the relative abundances of CHC components. MMC analysis shows that CHC components are not independent, but can be organized as correlated modules, which likely reflect common biosynthetic origins. Variation in the CHC profiles can be captured with a limited number of PCs, which were used as composite phenotypic values for GWA analyses. This resulted in the identification of candidate genes associated with variation in CHC composition in males and females, including several genes which could be plausibly associated with CHC biosynthesis. Despite a lenient significance threshold for reporting associations, targeted gene disruption of each of these 24 candidate genes indeed affected CHC composition, often in a sexually dimorphic manner. Our results highlight the complex interactions among genes that directly affect the composition and accumulation of CHCs and regulatory elements that affect CHC indirectly through general lipid metabolism and nutrition (see also Wicker-Thomas et al., 2015).

Identification of novel CHC components

We found substantial heritable natural variation in CHC composition for males and females of the inbred, sequenced DGRP lines. Several of the epicuticular compounds identified in this study have not been reported previously for D. melanogaster (Jallon and David, 1987; Foley et al., 2007; Everaerts et al., 2010; Dweck et al., 2015). These compounds separated in our GC analyses because we used a thin high-resolution column and a relatively long temperature program. Several of the newly identified monoenes and dienes had double bonds in an even carbon position, an unusual configuration in insects. While most of these new compounds represented a very small fraction of the total CHCs, two (peak 51 = 8,12-C28:2 and peak 53 = 6,10-C28:2, 9-C28:1 and 3-Me-C27) were female-specific and could potentially play a role in sexual communication.

A segregating African CHC phenotype in a Cosmopolitan population

The African CHC phenotype has an abundance of 5,9-C27:2 and lower levels of 7,11-C27:2, and is present only in populations from Sub-Saharan Africa and the Caribbean (Coyne et al., 1999). Based on genotyping the 16-bp deletion/ancestral Desat2 allele, Caribbean populations are known to have spread northward into the southern United States; however, populations north of Alabama and Mississippi were thought to be nearly fixed for the Cosmopolitan deletion (Yukilevich and True, 2008). Surprisingly, we found that the DGRP is segregating for the African CHC phenotype. Thus, to the best of our knowledge the DGRP progenitor population at the North Carolina Farmers Market represents the northern most population of D. melanogaster with the African Desat2 allele documented to date.

While 17 DGRP lines contain the ancestral allele that confers a functional Desat2, only 15 of these lines exhibited the African phenotype. On average, females of DGRP lines that were heterozygous or homozygous for the functional allele, regardless of inversion status, had intermediate amounts of the sex pheromone CHC peaks relative to the African lines or DGRP lines with the deletion. However, one DGRP line, DGRP_367, had more 5,9-C27:2 than either African line. Two DGRP lines (DGRP_105, ins/ins, INV/ INV and DGRP_551, ins/del, ST/ ST) had the functional Desat2 allele, yet exhibited the Cosmopolitan phenotype. These results are in contrast to previous reports of complete association of the ancestral allele with the African female CHC phenotype (Dallerac et al., 2000; Takahashi et al., 2001). Further analyses showed that other Desat2 polymorphisms in DGRP_105 likely result in a non-functional protein. However, the other Cosmopolitan-like line, DGRP_551, is puzzling because it is heterozygous for the functional allele, but homozygous for the standard karyotype. It is possible that this line is segregating for the In(3R)K inversion at low frequency and individuals with the ST/INV karyotype were not detected, but the lack of correspondence between the functional allele and CHC status remains unexplained since Desat2 itself does not harbor a potentially damaging mutation in this line. Perhaps a polymorphism in an unknown gene in DGRP_551 interacts with the Desat2 functional allele to suppress its effects. Finally, there were four DGRP lines that exhibited the African phenotype and were homozygous for the ancestral allele, but they were ST/ST in karyotype. One possible explanation for this 'mismatching' of phenotype and genotype with the karyotype is that the Desat2 16-bp deletion occurred prior to the inversion event. This would mean that the inversion may be segregating for the Desat2 16-bp allele.

GWA analysis and functional assessment of candidate genes reveals a complex enzymatic network for CHC biosynthesis and turnover

We used principal component analysis to reduce dimensionality of the data, which was motivated by the high co-linearity among the CHCs. However, PCA may cause genetic variation for a certain CHC to be distributed among multiple PCs and therefore dilute its association with QTLs. Surprisingly, we did not detect variants in the DGRP in previously identified CHC biosynthesis genes (FASN1, Desat1, eloF, DesatF, Cyp4G1; and several genes reported in Wicker-Thomas et al., 2015) associated with CHC variation. However, we did find many novel candidate genes, and functional tests showed that disruption of expression of all tested candidate genes had significant effects on the amount of CHC on the cuticular surface. While the mechanistic relationships between any of these genes are unknown, some share commonalities in their phenotypes when their expression is disrupted with RNAi. However, the majority of genes which encode gene products with similar molecular functions result in different shifts in CHC profiles when disrupted. Thus, variation in the CHC profile arises as an emergent phenotype from the dynamics of complex interrelated biosynthetic and catabolic pathways.

The RNAi-induced shifts in CHC profiles are frequently sexually dimorphic. This could reflect different expression levels of metabolic enzymes associated with CHC production. However, we cannot exclude the possibility that differential effectiveness of RNAi in males and females may contribute to apparent sexual dimorphism. In addition, we note that phenotypic changes associated with knocking down expression of a target gene with RNAi may not be causal, but a consequence of off-target effects of RNAi or the GAL4 driver. Further studies are needed to clarify the effects on CHC production of these genes and their interactions, and to test specific mechanisms and enzymatic activities through which they exert these effects. Furthermore, it is important to note that the RNAi and mutant experiments test for effects at the level of genes and are only proxies for the effects of segregating natural variants in these genes.

We recognize that the 24 candidate genes on which we focused represent only a subset of all candidate genes associated with variation in CHC composition. Many of these genes may directly or indirectly affect CHC composition through as yet unknown mechanisms. GWA studies of glucosinolates in Arabidopsis thaliana (Chan et al., 2011), flowering time in maize (Buckler et al., 2009), human height (Yang et al., 2010), and other traits in D. melanogaster such as body pigmentation (Dembeck et al., 2015) have also shown that the genetic architecture underlying variation in potentially adaptive traits includes many polymorphic loci with small effect sizes. Thus there is no reason to assume that variation in adaptive traits is controlled by few, large effect loci. Our results provide a framework for future studies of the mechanisms that regulate CHC composition and their adaptive potential regarding cold/heat tolerance and desiccation resistance, and pleiotropic effects on chemical communication and mate choice.

Materials and methods

Drosophila stocks and phenotyping

Request a detailed protocol

DGRP and African (Z30 and Z53) lines were reared in vials containing cornmeal-molasses-agar medium at 25°C, 75% relative humidity, a 12:12-h light-dark cycle, and a controlled adult density of 10 males and 10 females. The parental generation was allowed to lay eggs for three days. Upon eclosion, virgin males and females were separated and placed into new vials containing the same medium and aged for four days.

The flies from each line were separated into at least two samples of five flies each per sex. On average, three samples were collected for each. To avoid cross-contamination of cuticular lipids a fresh tissue paper was placed on the carbon dioxide pad and the flies were handled with acetone-washed titanium forceps at each round of sorting. Flies were placed in 2 mL glass auto-injection vials with a Teflon cap and were flash frozen. All samples were stored at -30°C until CHC extraction. We collected samples from 169 and 157 DGRP lines for females and males, respectively (1,078 total samples). All lines were reared simultaneously and DGRP lines that did not produce sufficient offspring for CHC analysis were excluded to avoid any block effects of rearing. For the two African lines, Z30 and Z53, we reared and collected 5 samples for each sex.

Quantification and identification of cuticular hydrocarbons

Request a detailed protocol

Cuticular lipids were extracted from each sample using 200 µl of hexane containing an internal standard (IS, 1 µg n-C32) with gentle swirling for five minutes. The flies were briefly extracted a second time with 100 µl of hexane (free of internal standard). After each wash the extract was transferred to a 300 µl conical glass insert. The extract was dried using a gentle stream of high-purity N2 and re-suspended in 50 µl of hexane. The samples were immediately processed using gas chromatography or stored at 4°C (no longer than one day) until processing.

The cuticular lipid extracts were analyzed using an Agilent 7890A gas chromatograph with a DB-5 Agilent capillary column (20 m x 0.18 mm x 0.18 µm) and a flame ionization detector (FID) for quantification. We introduced 1 µl of sample using an Agilent 7683B auto injector into a 290°C inlet operated in splitless mode. The split valve was turned on after 1 min. The oven temperature program was as follows: 50°C for one min, increased at 20°C/min to 150°C, and increased at 5°C/min to 300°C followed by a 10 min hold. Hydrogen was used as the carrier gas at constant flow (average linear velocity = 35 cm/sec) and the FID was set at 300°C.

Selected samples were analyzed for chemical identification in a 6890N GC system (Agilent) coupled with a 5975 mass selective detector (MSD) (Agilent) and equipped with a DB-5 (20 m x 0.18 mm x 0.18 μm) column (Agilent). Helium was used as carrier gas at 33 cm/s average linear velocity. Injection and temperature settings were identical to the settings described above, and the transfer line was maintained at 300°C. Positive electron ionization at 70 eV with default temperature settings (ion source at 150°C, quadrupole at 230°C) were used for the MSD. Ions were detected in scan mode in the range of 33–650 m/z at 1.23 scan/s scan rate. Compounds were identified based on their mass spectra in comparison to those in the reference library (Wiley 7th/NIST 05) and based on comparison of their retention indices and fragmentation patterns to already published Drosophila CHCs (Howard et al., 2003; Everaerts et al., 2010; Dweck et al., 2015).The position of the double bonds was not confirmed by performing microderivatization reactions and chirality was not determined for any of the CHCs. More details on how the putative structure of previously unpublished D. melanogaster CHCs was deduced (based on mass spectra and retention index data) are given in Supplementary file 10.

All chromatograms were analyzed using Agilent ChemStation software. For quantification individual peak areas were obtained for 42 and 60 male and female CHC containing peaks, respectively (some of the peaks contained multiple CHCs). Response factors were not determined for individual components. To account for natural variation in body size and absolute amounts of CHCs between the lines, the data were represented as proportions by dividing each peak area by the sum of all integrated peaks.

Statistical and quantitative genetic analyses

Request a detailed protocol

We partitioned variation of each CHC peak into genetic and environmental components using an ANOVA model of form Y = µ + L + ε, where Y is phenotype, µ is the overall mean, L is the random effect of line, and ε is the residual. We estimated variance components using restricted maximum likelihood and computed the broad-sense heritability (H2) of each CHC peak as H2 = σ2L/(σ2L + σ2ε), where σ2L is the among-line variance component and σ2ε is the error variance. All analyses were performed with version 9.3 of the SAS System for Windows (2013 SAS Institute Inc.).

A majority of CHCs belong to homologous series in which the chain length increases by two carbons; thus these compounds may be genetically correlated due to shared biosynthetic pathways and the data may be confounded with multi-co-linearity (Martin and Drijfhout, 2009). We visualized the correlations between CHCs using modulated modularity clustering (MMC) (Stone and Ayroles, 2009). The MMC algorithm clusters highly correlated variables based on the Spearman's rank correlation coefficients (ρ). In order to take these correlations into account, we conducted principal components (PC) analysis on the variance-covariance matrices for the male and female CHC line means. For each analysis we included only CHC peaks that had an estimated H2 ≥ 0.25. We retained PCs explaining greater than 1% of the variation for subsequent GWA analysis. PCA was conducted in JMP Pro10 (2013 SAS Institute Inc.).

Genome-wide association (GWA) analyses

Request a detailed protocol

We conducted a GWA analysis for each CHC PC, separately for males and females. The DGRP lines are segregating for Wolbachia infection status and for the following common inversions: In(2L)t, In(2R)NS, In(3R)P, In(3R)K, and In(3R)Mo. We performed GWA studies in two stages. In the first stage, we adjusted the line means for the effects of Wolbachia infection and major inversions. We then used the adjusted line means to fit a linear mixed model in the form of Y = Xb + Zu + ε, where Y is the adjusted phenotypic value, X is the design matrix for the fixed SNP effect b, Z is the incidence matrix for the random polygenic effect u, and ε is the residual. The vector of polygenic effects u has a covariance matrix in the form of Aσ2, where σ2 is the polygenic variance component. We fitted this linear mixed model using the FastLMM program (version 1.09) (Lippert et al., 2011). We performed these single marker analyses for the 1,883,938 (females) and 1,912,894 (males) biallelic variants (SNPs and indels) with minor allele frequencies ≥ 0.05 whose Phred scale quality scores were at least 500 and genotypes whose sequencing depths were at least one and genotype quality scores at least 20 (Huang et al., 2014). All segregating sites within lines were treated as missing data.

RNAi and mutant candidate gene validation experiments

Request a detailed protocol

We selected candidate genes with available mutations and RNAi knockdown constructs to test for effects on CHC production based on FlyBase annotations. We obtained lines with RNAi knockdown constructs from the Vienna Drosophila RNAi Center (VDRC) and crossed them to the oenocyte-specific GAL4 driver, PromE(800)-GAL4 (Dietzl et al., 2007; Billeter et al., 2009). We tested knockdown constructs and their co-isogenic controls (F1 individuals from crosses of the empty vector strain to PromE(800)-GAL4) for 23 genes (Supplementary file 8). Since no RNAi knockdown line was available from VDRC for CG10097 identified in the male GWA analysis, we obtained and tested for this gene a PiggyBac insertion mutant from the Harvard Exelixis Collection (Thibault et al., 2004) along with the w1118 control line. From each cross and mutant line, we collected and aged both male and female virgins and analyzed the CHCs in the same manner as described for the DGRP flies. The analysis of CHCs using GC was also the same. However, for these lines, instead of calculating the proportion that each peak contributed to the total chromatogram, we used the internal standard to calculate the amount of each CHC present in the sample (ng/fly) with the assumption that body sizes between the control and RNAi knockdown or mutant were not significantly different. These measures provide a more quantitative measure of CHCs and capture differences that proportion data may not resolve. PC analyses and t-tests pairing the test lines with the controls were conducted on these data. We also calculated the mean total amount of CHCs (ng/fly); we used the more conservative Cochran and Cox test, which assumes unequal group variances to assess statistically significant effects on CHC composition. We also present these data as µg/fly in Figure 7—figure supplement 124. t-tests were conducted in SAS v. 9.4 (2013 SAS Institute Inc.).

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
    Insects groom their antennae to enhance olfactory acuity
    1. K Böröczky
    2. A Wada-Katsumata
    3. D Batchelor
    4. M Zhukovskaya
    5. C Schal
    (2013)
    Proceedings of the National Academy of Sciences of the United States of America 110:3615–3620.
    https://doi.org/10.1073/pnas.1212466110
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
    Characterization of Drosophila melanogaster cytochrome P450 genes
    1. H Chung
    2. T Sztal
    3. S Pasricha
    4. M Sridhar
    5. P Batterham
    6. PJ Daborn
    (2009)
    Proceedings of the National Academy of Sciences of the United States of America 106:5731–5736.
    https://doi.org/10.1073/pnas.0812141106
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
    A mutation in the promoter of Desaturase 2 is correlated with sexual isolation between Drosophila behavioral races
    1. S Fang
    2. A Takahashi
    3. CI Wu
    (2002)
    Genetics 162:781–784.
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
    Morphogenetic Hormones of Arthropods
    1. F Romer
    (1991)
    The oenocytes of insects: Differentiation, changes during molting, and their possible involvement in the secretion of moulting hormone, Morphogenetic Hormones of Arthropods, New Brunswick, New Jersey, Rutgers University Press.
  50. 50
  51. 51
  52. 52
  53. 53
    Insect Lipids: Chemistry, Biochemistry, and Biology
    1. DW Stanley-Samuelson
    2. DR Nelson
    (1993)
    Lincoln, Nebraska: University of Nebraska Press.
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
    Insect Hydrocarbons: Biology, Biochemistry and Chemical Ecology. 1st ed
    1. C Wicker-Thomas
    2. T Chertemps
    (2010)
    Molecular biology and genetics of hydrocarbon production, Insect Hydrocarbons: Biology, Biochemistry and Chemical Ecology. 1st ed, Cambridge, Cambridge University Press.
  61. 61
  62. 62
  63. 63

Decision letter

  1. Daniel J Kliebenstein
    Reviewing Editor; University of California, Davis, United States

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

Thank you for submitting your work entitled "Genetic architecture of natural variation in cuticular hydrocarbon composition in Drosophila melanogaster" for peer review at eLife. Your submission has been favorably evaluated by Detlef Weigel (Senior editor) and three reviewers, one of whom, Daniel J Kliebenstein, is a member of our Board of Reviewing Editors.

The reviewers have discussed the reviews with one another and the Reviewing editor has drafted this decision to help you prepare a revised submission.

We felt that the manuscript was very interesting but that some key factors need amendment to reflect insect biochemistry and editorial changes to increase the potential audience of this nice genetic analysis. We therefore ask you to do the following:

1) Rework the manuscript to properly reflect the specifics of FA biosynthesis in insects rather than rely upon the plant model of the pathway.

2) Strive to present the broader picture and impacts of the results found in this study beyond CHC metabolism in Drosophila.

3) Clarify that the candidate genes found may influence the CHC accumulation but further work needs to be done to test the specific enzymatic activities.

4) Discuss issues on chemical identification and quantification.

Explicit reviews can be found below.

Reviewer #1:

The authors investigate the quantitative genetic variation in CHC hydrocarbons within Drosophila using a GWA approach and identify a highly polygenic underpinning to this trait. They then validate the potential role of 24 candidate genes in altering this trait. This analysis was very nicely conducted but the writing was solely built around CHC composition in Drosophila which will make it difficult for this paper to attract the potential readership that it should obtain. From my reading, there are two broad messages that all researchers using GWA should be able to obtain from this manuscript.

1) It is possible to being dissecting highly polygenic GWA traits that have 100s or more potential candidate genes down to the functional level. There is no need to assume that there are few causal genes for potentially adaptive traits.

2) Polygenic adaptive traits may be structured in a manner that will allow data decomposition approaches to allow significant progress in understanding complex traits like metabolic mixtures.

The manuscript would likely get broader visibility if there was more general information about the bigger picture presented in the manuscript that could pull in researchers studying GWA or metabolic traits in other systems as this manuscript has information for those researchers even if they are studying plants or fungi or humans. While motivated researchers will see the broader impacts of this manuscript on the GWA/metabolite field it would help to work and bring them into the research without this effort by presenting the broader ramifications of the research in the Introduction and Discussion.

While I appreciate the use of PCs to reduce the dimensionality, this is an inherently linear approach that can have trouble with non-linear processes such as epistatic variation in biosynthetic pathways. This can be illustrated by Figure 5A where the PC1 x PC2 plot shows two linear relationships. One going NW to SE and the other from NE to SW, this suggests that PC1 and PC2 are capturing variation at potentially biosynthetic loci that work within the pathway to control structural variation. Did the authors attempt to parse out these potential biosynthetic clues to synthesize a pathway? Another reason to consider this is that the PC approach can hide key biosynthetic genes at the intersection of the two PCs. This is true in the glucosinolate work within Arabidopsis wherein the PC approach utilized by Bergelson and colleagues was unable to identify all of the key biosynthetic enzymes that could be identified on a more metabolite by specific metabolite analysis or alternatively using ratios of precursors to products (Wentzell et al., 2007; Chan et al., 2011; Brachi et al., 2015). While I don't feel this is necessary to conduct, it would be beneficial to discuss this difficulty that is imparted by the use of linear trait decomposition for GWA.

Brachi, B. et al.(2015). Coselected genes determine adaptive variation in herbivore resistance throughout the native range of Arabidopsis thaliana. Proceedings of the National Academy of Sciences of the United States of America 112, 4032-4037.

Chan, E.K. et al. (2011). Combining genome-wide association mapping and transcriptional networks to identify novel genes controlling glucosinolates in Arabidopsis thaliana. PLoS Biol 9, e1001125.

Wentzell, A.M. et al. (2007). Linking metabolic QTLs with network and cis-eQTLs controlling biosynthetic pathways. PLoS Genet 3, 1687-1701.

Reviewer #2:

This manuscript represents a great deal of very good work and can be made into an excellent publication. However, much of the biochemistry upon which it is based needs extensive correcting and rewriting.

The identification of new components in this much studied species is well done and important.

In the subsection “Functional validation of candidate genes”, second paragraph: Fatty acid biosynthesis begins with an acetyl-CoA and then elongation occurs with a malonyl-CoA in which the energy is supplies by decarboxylation of the malonyl-CoA as two carbons incorporated into the growing chain. The next three steps are a reduction of a carbonyl to an alcohol, dehydration, and reduction of the carbon-carbon double bond. The authors are making too extensive use of plant lipid metabolism data. The FAS in animals, including flies, occurs on the multifunctional protein fatty acid synthetase (FAS) in which the growing chain is attached to an acyl carrier type unit that is part of the multifunctional (7 enzyme activities on one protein, act in a dimer). It is not typically called an acyl carrier protein complex. Products of insect FASs, particularly Drosophila FAS, can be 14, 16 or 18 carbons. A thioesterase that is part of the multienzyme FAS removes the elongated chain as a free fatty acid. I do not know what is meant by conjugating and the acyl group is never removed by a thioreductase. In all cases studied in insects, desaturation and elongation take place in a microsomal fraction (endoplasmic reticulum) and not the mitochondrial fraction (many references).

In the subsection “Functional validation of candidate genes”, fourth paragraph: Fatty acids are elongated and desaturated attached to an ACP in plants. In insects, the parallel reactions occur with the acyl group attached to CoASH. The evidence to date suggests that the elongated fatty acyl-CoA is reduced to an aldehyde prior to oxidative decarbonylation, the latter step catalyzed by Cyp4G1 in Drosophila (Qiu et al., 2012). While it is possible that the elongated fatty acyl-CoA is reduced from acyl-CoA to aldehyde to alcohol and then re-oxidized to aldehyde before oxidative decarbonylation, there is no evidence for this to date in insects. (See Qiu et al. and the Reed et al. papers referenced there.)

In the subsection “Functional validation of candidate genes”, end of fourth paragraph: An esterase cleaves an ester bond and does not reduce the carboxylic ester group on the fatty acyl-CoA (not ACP) to an alcohol. A FAR (fatty acyl-CoA reductase) does the reduction.

In the subsection “Functional validation of candidate genes”, sixth paragraph: I am not convinced the CYPs studied here are involved in CHC turnover.

In the subsection “Functional validation of candidate genes”, last paragraph: The authors are comparing 3-methyl fatty acids with a 3-methyl hydrocarbon derived from a n-3 methyl fatty acid.

When stating “Our results provide a new perspective and highlight the complexity of the biosynthetic and catabolic pathways that regulate the dynamics of CHC composition”, the authors are blithely referring to the biosynthetic pathways of hydrocarbons (which are somewhat known) with the catabolic pathways (which to my knowledge almost nothing is known). This sentence is illustrative of several times in the manuscript where the authors make sweeping statements of biochemistry for which there is no basis.

Reviewer #3:

This manuscript by Dembeck et al. is the latest in a series of fine papers from the Mackay lab using the sequenced inbred lines from the DGRP to understand the genetic architecture of phenotypic trait variation, in this case, the variation in cuticular hydrocarbons (CHC) in Drosophila melanogaster. This manuscript reports three main findings: 1) the discovery of 16 previously unidentified CHCs in D. melanogaster, 2) that reported Desat2 alleles do not match reported female CHC phenotypes; and 3) the identification of previously unidentified candidate genes in the biosynthesis of CHCs (using genome-wide association analysis to identify candidates and their functional validation by either RNAi knockdown or transposon insertion).

The main strengths of this manuscript are its systematic approach to gene identification and functional validation. Previously, much work in the area of CHC biosynthesis has relied on biochemical approaches and the field has largely not taken advantage of potential genetic approaches as demonstrated here. The results here provide concrete functional evidence of the roles of numerous genes in CHC biosynthesis. This paper is the product of a tremendous amount of work that will provide good leads for future biological studies. The main weakness of the manuscript is that it does not provide new mechanistic insights into any compelling biological question.

Detailed comments are as follows:

In the first section of the paper, Dembeck et al. claim to have detected 16 previously unidentified CHCs from Drosophila melanogaster, including dienes with double bonds in the even-numbered configurations as well as methyl-branched CHCs which are not 2-methyl-branched. The authors stated that they were able to identify these compounds because they use a thin high res column (DB-5) and a relatively long temperature program. I am concerned about the use of this column for the separation of D. melanogaster CHCs as this column cannot separate two major compounds in female Drosophila. Peak 46 contains both 2Me-C26 and 7,11-C27:2. This is a concern for the second part of the paper (the Desat2 allele and female CHC phenotyping) because 2Me-C26 is a major compound. If they cannot separate these two compounds, then it would be impossible to quantify accurately the amount of 7,11-C27:2 in the fly. Another recent paper (Dweck et al., 2015, PNAS) also used a different method to identify CHC in D. melanogaster. While they identified some methyl-branched CHCs which are not 2-methyl-branched, they did not identify the dienes with the even double bonds presented in this study. The authors stated that these compounds are identified based on comparison to the reference library and to already published D. melanogaster CHCs and the position of the double bond was not confirmed for any of the CHCs. While most of these compounds exist in low quantities, I suggest the authors proceed with caution on claiming that they have identified new (and unusual) compounds that many other groups using different methods over the years have not been able to identify.

In the second part of the manuscript, Dembeck et al. report that the African CHC phenotype (predominantly 5,9-C27:2) is segregating in the DGRP. This section reports the mismatch between what is called the "16-bp deletion / ancestral alleles" of Desat2 and female CHC phenotype. Two DGRP lines, DGRP_105 and DGRP_551 have the African allele ("functional Desat2") but the cosmopolitan phenotype. This analysis is based on protein sequence alone and does not take into account the possibility that unique polymorphisms in the Desat2 regulatory region of these lines changed the expression of Desat2. Therefore, without properly quantifying whether Desat2 is expressed or not in these lines, the authors should be careful in determining which alleles are functional based on protein sequence analysis alone (and a 16-bp deletion in the promoter that was not properly characterized previously).

In the third section which forms the bulk of the paper, the authors identify numerous genes potentially involved in CHC synthesis in D. melanogaster. The same authors have used similar methods to identify novel pigmentation genes in Drosophila (Dembeck et al., 2015, PLoS Genetics). In this section, using GWA, the authors identified genetic variants in hundreds of genes which could potentially affect CHC composition. The authors selected 24 candidate genes for functional studies. They used an oenocyte specific driver (although this driver also expresses GAL4 in some other tissues) to knockdown candidate genes in the oenocytes, where CHCs are synthesized. This is a very comprehensive section that entails a tremendous amount of work in both genomic analysis and functional characterization. The authors found that oenocyte-driven RNAi knockdown of 23 candidate genes individually showed quantitative changes in CHC profiles. These results provide very strong evidence that these candidate genes are involved in CHC biosynthesis, but there are two caveats to consider.

First, some of these genes might not be expressed in adult oenocytes. A brief look on the FlyAtlas website showed that the transcripts of some of these genes are not present or expressed at very low levels in the "adult carcass" where the oenocytes are located. Therefore, it is not known whether the phenotypes that the authors observed are off target effects of RNAi or altered physiology of the whole fly due to the oenocyte driver that they use, which may be driving GAL4 in other parts of the fly.

Second, while some of the effects of the candidate genes can be explained by the biological/molecular functions of the gene (i.e. fatty acid reductase influencing chain length), others are not (i.e. P450s affecting CHC abundance, although Qiu et al. 2012 PNAS identified the exact mechanism for another P450, Cyp4g1 in CHC synthesis). How would a P450 (Cyp9f2) be involved in elongation of CHCs? As far as is known, that is not one of the reactions that a P450 could catalyze.

https://doi.org/10.7554/eLife.09861.050

Author response

We felt that the manuscript was very interesting but that some key factors need amendment to reflect insect biochemistry and editorial changes to increase the potential audience of this nice genetic analysis. We therefore ask you to do the following:

1) Rework the manuscript to properly reflect the specifics of FA biosynthesis in insects rather than rely upon the plant model of the pathway.

We have addressed these points in the revised manuscript.

2) Strive to present the broader picture and impacts of the results found in this study beyond CHC metabolism in Drosophila.

We have revised the manuscript to address this comment in the Introduction and Discussion. We note that it is possible to dissect highly polygenic traits down to the functional level, in accordance with the suggestion of Reviewer 1. We also note in the Discussion that there is no need to assume that there are few causal genes for potentially adaptive traits. However, we are not the first to demonstrate that the genetic basis of high-dimensional polygenic traits can be assessed using data decomposition approaches, so we did not stress this point.

3) Clarify that the candidate genes found may influence the CHC accumulation but further work needs to be done to test the specific enzymatic activities.

We have added the following statement to the Discussion: “Further studies are needed to clarify the effects on CHC production of these genes and their interactions, and to test specific mechanisms and enzymatic activities through which they exert these effects.”

4) Discuss issues on chemical identification and quantification.

We have addressed this issue by adding a new Supplementary file 10 which incorporates our responses to the reviewer’s comments.

Explicit reviews can be found below.

Reviewer #1:

The authors investigate the quantitative genetic variation in CHC hydrocarbons within Drosophila using a GWA approach and identify a highly polygenic underpinning to this trait. They then validate the potential role of 24 candidate genes in altering this trait. This analysis was very nicely conducted but the writing was solely built around CHC composition in Drosophila which will make it difficult for this paper to attract the potential readership that it should obtain. From my reading, there are two broad messages that all researchers using GWA should be able to obtain from this manuscript.

1) It is possible to being dissecting highly polygenic GWA traits that have 100s or more potential candidate genes down to the functional level. There is no need to assume that there are few causal genes for potentially adaptive traits.

2) Polygenic adaptive traits may be structured in a manner that will allow data decomposition approaches to allow significant progress in understanding complex traits like metabolic mixtures.

We appreciate these suggestions and have revised the manuscript to address them in the Introduction and Discussion. We explicitly note that it is possible to dissect highly polygenic traits down to the functional level and that there is no need to assume that there are few causal genes for potentially adaptive traits.

The manuscript would likely get broader visibility if there was more general information about the bigger picture presented in the manuscript that could pull in researchers studying GWA or metabolic traits in other systems as this manuscript has information for those researchers even if they are studying plants or fungi or humans. While motivated researchers will see the broader impacts of this manuscript on the GWA/metabolite field it would help to work and bring them into the research without this effort by presenting the broader ramifications of the research in the Introduction and Discussion.

Again, we appreciate the suggestion but feel that in this case the broader impacts with respect to other complex traits and organisms will be readily understood by a general reader, and that our CHC analysis will be of broad interest in and of itself. Since we are not the first to demonstrate that the genetic basis of high-dimensional polygenic traits can be assessed using data decomposition approaches, we did not stress this point.

While I appreciate the use of PCs to reduce the dimensionality, this is an inherently linear approach that can have trouble with non-linear processes such as epistatic variation in biosynthetic pathways. This can be illustrated by Figure 5A where the PC1 x PC2 plot shows two linear relationships. One going NW to SE and the other from NE to SW, this suggests that PC1 and PC2 are capturing variation at potentially biosynthetic loci that work within the pathway to control structural variation. Did the authors attempt to parse out these potential biosynthetic clues to synthesize a pathway? Another reason to consider this is that the PC approach can hide key biosynthetic genes at the intersection of the two PCs. This is true in the glucosinolate work within Arabidopsis wherein the PC approach utilized by Bergelson and colleagues was unable to identify all of the key biosynthetic enzymes that could be identified on a more metabolite by specific metabolite analysis or alternatively using ratios of precursors to products (Wentzell et al., 2007; Chan et al., 2011; Brachi et al., 2015). While I don't feel this is necessary to conduct, it would be beneficial to discuss this difficulty that is imparted by the use of linear trait decomposition for GWA.

Brachi, B. et al.(2015). Coselected genes determine adaptive variation in herbivore resistance throughout the native range of Arabidopsis thaliana. Proceedings of the National Academy of Sciences of the United States of America 112, 4032-4037.

Chan, E.K. et al. (2011). Combining genome-wide association mapping and transcriptional networks to identify novel genes controlling glucosinolates in Arabidopsis thaliana. PLoS Biol 9, e1001125.

Wentzell, A.M. et al. (2007). Linking metabolic QTLs with network and cis-eQTLs controlling biosynthetic pathways. PLoS Genet 3, 1687-1701.

We appreciate this insightful and thought-provoking comment from the reviewer. We added the following text to the Discussion regarding the motivation of using PCs and caveats of losing the ability to detect genes that control variation of individual CHCs: “We used principal component analysis to reduce dimensionality of the data, which was motivated by the high co-linearity among the CHCs. However, PCA may cause genetic variation for a certain CHC to be distributed among multiple PCs and therefore dilute its association with QTLs.”

Regarding the specific concerns raised by the reviewer:

A) PC1 and PC2 are orthogonal (uncorrelated) by definition. The appearance of certain linear relationships with subsets of lines may be a result of projecting multi(>2)-dimensional data onto the 2-D plane created by PC1 and PC2. Such patterns may or may not be biologically meaningful. However, we are not aware of a good approach that would use such patterns to synthesize a biosynthetic pathway. On the other hand, albeit incomplete and lacking hierarchical information, our approaches using individual PCs did arrive at key genes involved in the CHC biosynthetic pathways.

B) We agree with the reviewer that analysis using PCs may miss genes controlling the biosynthesis of CHCs. However, given the severe co-linearity of the data, we think the gain by reducing dimensionality is greater than the loss. Following the reviewer’s suggestion, we acknowledge this limitation in the Discussion.

Reviewer #2:

This manuscript represents a great deal of very good work and can be made into an excellent publication. However, much of the biochemistry upon which it is based needs extensive correcting and rewriting. The identification of new components in this much studied species is well done and important. In the subsection “Functional validation of candidate genes”, second paragraph: Fatty acid biosynthesis begins with an acetyl-CoA and then elongation occurs with a malonyl-CoA in which the energy is supplies by decarboxylation of the malonyl-CoA as two carbons incorporated into the growing chain. The next three steps are a reduction of a carbonyl to an alcohol, dehydration, and reduction of the carbon-carbon double bond. The authors are making too extensive use of plant lipid metabolism data. The FAS in animals, including flies, occurs on the multifunctional protein fatty acid synthetase (FAS) in which the growing chain is attached to an acyl carrier type unit that is part of the multifunctional (7 enzyme activities on one protein, act in a dimer). It is not typically called an acyl carrier protein complex. Products of insect FASs, particularly Drosophila FAS, can be 14, 16 or 18 carbons. A thioesterase that is part of the multienzyme FAS removes the elongated chain as a free fatty acid. I do not know what is meant by conjugating and the acyl group is never removed by a thioreductase. In all cases studied in insects, desaturation and elongation take place in a microsomal fraction (endoplasmic reticulum) and not the mitochondrial fraction (many references).

We appreciate the reviewer's expertise on this topic and re-wrote the entire paragraph as the reviewer suggested to accurately describe fatty acid biosynthesis in insects.

In the subsection “Functional validation of candidate genes”, fourth paragraph: Fatty acids are elongated and desaturated attached to an ACP in plants. In insects, the parallel reactions occur with the acyl group attached to CoASH. The evidence to date suggests that the elongated fatty acyl-CoA is reduced to an aldehyde prior to oxidative decarbonylation, the latter step catalyzed by Cyp4G1 in Drosophila (Qiu et al., 2012). While it is possible that the elongated fatty acyl-CoA is reduced from acyl-CoA to aldehyde to alcohol and then re-oxidized to aldehyde before oxidative decarbonylation, there is no evidence for this to date in insects. (See Qiu et al. and the Reed et al. papers referenced there.)

As the reviewer suggested, we have removed the reference to plant fatty acid elongation and changed the paragraph to accurately state what is known in insects.

In the subsection “Functional validation of candidate genes”, end of fourth paragraph: An esterase cleaves an ester bond and does not reduce the carboxylic ester group on the fatty acyl-CoA (not ACP) to an alcohol. A FAR (fatty acyl-CoA reductase) does the reduction.

We have removed this sentence from the manuscript.

In the subsection “Functional validation of candidate genes”, sixth paragraph: I am not convinced the CYPs studied here are involved in CHC turnover.

We appreciate the reviewer's skepticism. This line is our speculation based on the RNAi knockdown results. We propose it merely as a hypothesis for future functional studies of these CYPs. We have modified this statement to explicitly state that it is speculation.

In the subsection “Functional validation of candidate genes”, last paragraph: The authors are comparing 3-methyl fatty acids with a 3-methyl hydrocarbon derived from a n-3 methyl fatty acid.

We understand the reviewer’s point that for the 3-methylhydrocarbons the methyl branch is inserted right at the beginning and then elongation happens with the methyl-branched end becoming farther and farther away from the CoA “head”, whereas in a 3-methyl fatty acid the methyl group is on carbon 3 counted from the carboxyl “head”. We have removed our previous speculation from the manuscript.

When stating “Our results provide a new perspective and highlight the complexity of the biosynthetic and catabolic pathways that regulate the dynamics of CHC composition”, the authors are blithely referring to the biosynthetic pathways of hydrocarbons (which are somewhat known) with the catabolic pathways (which to my knowledge almost nothing is known). This sentence is illustrative of several times in the manuscript where the authors make sweeping statements of biochemistry for which there is no basis.

We have modified this sentence to say "Our results provide new perspective and highlight the complexity of the CHC dynamics."

Reviewer #3:

In the first section of the paper, Dembeck et al. claim to have detected 16 previously unidentified CHCs from Drosophila melanogaster, including dienes with double bonds in the even-numbered configurations as well as methyl-branched CHCs which are not 2-methyl-branched. The authors stated that they were able to identify these compounds because they use a thin high res column (DB-5) and a relatively long temperature program. I am concerned about the use of this column for the separation of D. melanogaster CHCs as this column cannot separate two major compounds in female Drosophila. Peak 46 contains both 2Me-C26 and 7,11-C27:2. This is a concern for the second part of the paper (the Desat2 allele and female CHC phenotyping) because 2Me-C26 is a major compound. If they cannot separate these two compounds, then it would be impossible to quantify accurately the amount of 7,11-C27:2 in the fly. Another recent paper (Dweck et al., 2015, PNAS) also used a different method to identify CHC in D. melanogaster. While they identified some methyl-branched CHCs which are not 2-methyl-branched, they did not identify the dienes with the even double bonds presented in this study. The authors stated that these compounds are identified based on comparison to the reference library and to already published D. melanogaster CHCs and the position of the double bond was not confirmed for any of the CHCs. While most of these compounds exist in low quantities, I suggest the authors proceed with caution on claiming that they have identified new (and unusual) compounds that many other groups using different methods over the years have not been able to identify.

Separation of the odd-carbon 7,11-dienes and the preceding even-carbon 2-methyl-alkane: Both separate on our column (as in Dweck et al. 2015 PNAS) by just a couple of Kovats units. So, when the diene is absent the alkane can be readily separated and vice versa. However, when the diene is a major component (e.g., 7,11-C27:2 in females, where it serves as a sex pheromone component), the alkane represents a tiny shoulder preceding the diene and they cannot be reliably separated. See our Figure 1: Peak 57 in males is 2Me-C26 and it elutes just slightly ahead of peak 57 in females, which is mainly 7,11-C27:2. Therefore, this peak can be highly variable depending whether it is in males or females. In order to treat this variable peak consistently in all chromatograms, we considered these 2 compounds together as a single peak for integration, with the recognition that it represents almost exclusively the diene in females and the alkane in males. The paper referenced by the reviewer (Dweck et al. 2015 PNAS) in fact did not separate 2Me-C26 and 7,11-C27:2 (their peaks #74 and 75 in Table Supplement 1 and Figure Supplement 1). While the table shows them separated by 3 Kovats units, the figure shows them co-eluting in females because of the large amount of #75. Only when males are examined can #74 be separated because males lack #75 (7,11-C27:2). The same is true for shorter chain compounds. In fact, Dweck et al. (2015 PNAS) confirm our findings that the 2-methyl-branched compounds are found in males (their peaks #59, 68) and either not found at all in females or in very small amounts. So, our Peak 46 contains both 2Me-C26 and 7,11-C27:2 (as in Dweck et al. 2015 PNAS peaks 74 and 75), but the alkane represents a tiny fraction of the diene that serves as sex pheromone. Dweck et al. (2015 PNAS) used a HP5 column (30 m, 0.25 mm, 0.25 µm) and ran >45 min chromatograms and could not separate these compounds in females. We ran an identical GC phase (DB-5) but in a column (20 m, 0.18 mm, 0.18 µm) that allowed high resolution in much shorter run times of 32 min to optimize the analysis of thousands (hundreds?) of chromatograms. Others (e.g., Frentiu et al. 2010 Evolution) also show no evidence of separating these compounds in females.

The reviewer is rightly concerned that in the second part of the paper (the Desat2 allele and female CHC phenotyping) “because 2Me-C26 is a major compound”. As stated above, in normal females 2Me-C26 represents a tiny fraction of 7,11-C27:2. In the African phenotype and in females with an intact Desat2 gene we found high 5,9-C27:2 and 5,9-C29:2 and low amounts of 7,11-dienes. The reviewer is correct that 2Me-C26 now represents a larger fraction of this combined peak. See our Figure 2A, where the leading shoulder, representing 2Me-C26 (retention time 26.25 min), becomes more apparent as peak size decreases (7,11-C27:2 decreases) and 5,9-C27:2 and 5,9-C29:2 increase. Nevertheless, we contend that the potential confounding influence of 2Me-C26 is minimal because our main objective in this section was to quantify and separate the Cosmopolitan and African phenotypes, and this is readily apparent by the relationship of 7,11-C27:2 and 5,9-C27:2.

Chemical identification of methyl-branched hydrocarbons where the methyl group is near to the chain terminus was based on the relatively high abundance of indicating fragments near the molecular ion. These are produced in the EI ion source by α-cleavage at the methyl branch on the side of the near chain terminus. Hydrogen rearrangement also leads to low-weight radical ion fragments with an even weight corresponding to fragmentation on the other side of the methyl branch. Mass spectra and characteristic ions are shown in the figures in Supplementary file 10, see A and B for 3-methyl- and 5-methylpentacosane, from a male and a female in this study, respectively. Methyl branch position on carbon 3 was also confirmed by the retention index of these hydrocarbons: they all have a retention index increment of 68–73 on the 5% phenyl-methylpolysiloxane column (peaks 17, 41, and 53 in Table 1 in the manuscript) due to the methyl branch at that particular position (Schulz, Lipids 2001, 36:637-647).

Though the position of double bond in monoalkenes was not confirmed with microderivatization reactions the retention indices were matched to that of already published monoalkenes in D. melanogaster. Even positions were tentatively assigned when there was a gas-chromatographic peak with a molecular ion and fragmentation pattern of a monoalkene between two peaks of monoalkenes with double bonds at consecutive odd positions, i.e. 6- was assigned in between 7- and 5- (peak 30 between 29 and 31 in Table 1).

As for alkadienes with the double bonds in even positions just a few carbons apart, we followed the same principle as described in Howard et al. (Journal of Chemical Ecology 2003, 29:961-976), namely that these compounds (as the ones with the double bond in odd positions) have characteristic fragments resulting from the fragmentation of the molecular ion where the double bonds are and from α-cleavage at the outer side of the double bonds. An example on Figure C demonstrates the above with 6,10-hexacosadiene from one of the females analyzed in this study.

We have included a new Supplementary file 10 to summarize these points.

In the second part of the manuscript, Dembeck et al. report that the African CHC phenotype (predominantly 5,9-C27:2) is segregating in the DGRP. This section reports the mismatch between what is called the "16-bp deletion / ancestral alleles" of Desat2 and female CHC phenotype. Two DGRP lines, DGRP_105 and DGRP_551 have the African allele ("functional Desat2") but the cosmopolitan phenotype. This analysis is based on protein sequence alone and does not take into account the possibility that unique polymorphisms in the Desat2 regulatory region of these lines changed the expression of Desat2. Therefore, without properly quantifying whether Desat2 is expressed or not in these lines, the authors should be careful in determining which alleles are functional based on protein sequence analysis alone (and a 16-bp deletion in the promoter that was not properly characterized previously).

For the analysis of DGRP lines DGRP_105 and DGRP_551, we studied the DNA sequence data. While DGRP_551 does not have any obvious mutations that could disrupt Desat2 expression, we found that DGRP_105 has many unique polymorphisms. We recently completed a gene expression analysis of 185 DGRP lines using Affymetrix tiling arrays – the manuscript reporting these results is in press at the Proceedings of the National Academy of Sciences of the USA (Huang et al., 2015). This study was done on whole young adult flies. We checked the expression of Desat2 from this study and found that it is highly genetically variable in females (Line FDR = 1.39 E-34, H2 = 0.766) but not in males (Line FDR = 1.83E-01, not significant). We present these results below, classifying the DGRP lines as homozygous for the derived deletion (Del/Del), the ancestral insertion (Ins/Ins) and heterozygous (Ins/Del). The expression level of DGRP_105 is shown by the red data point. It is clear that the Desat2 expression of this line overlaps that of the Del/Del genotypes, completely consistent with our CHC analysis and DNA sequence analysis showing that variants in the Desat2 gene in this line likely result in a non-functional protein. The genotype of this line indeed matches the Cosmopolitan CHC phenotype despite the ancestral ‘functional’ insertion. Unfortunately, DGRP_551 was not among the 181 DGRP lines included in the recent gene expression analysis. We feel that further characterization of the nature of the functional variation at Desat2 in these lines warrants many additional experiments and is outside the scope of the current manuscript.

In the third section which forms the bulk of the paper, the authors identify numerous genes potentially involved in CHC synthesis in D. melanogaster. The same authors have used similar methods to identify novel pigmentation genes in Drosophila (Dembeck et al., 2015, PLoS Genetics). In this section, using GWA, the authors identified genetic variants in hundreds of genes which could potentially affect CHC composition. The authors selected 24 candidate genes for functional studies. They used an oenocyte specific driver (although this driver also expresses GAL4 in some other tissues) to knockdown candidate genes in the oenocytes, where CHCs are synthesized. This is a very comprehensive section that entails a tremendous amount of work in both genomic analysis and functional characterization. The authors found that oenocyte-driven RNAi knockdown of 23 candidate genes individually showed quantitative changes in CHC profiles. These results provide very strong evidence that these candidate genes are involved in CHC biosynthesis, but there are two caveats to consider.

First, some of these genes might not be expressed in adult oenocytes. A brief look on the FlyAtlas website showed that the transcripts of some of these genes are not present or expressed at very low levels in the "adult carcass" where the oenocytes are located. Therefore, it is not known whether the phenotypes that the authors observed are off target effects of RNAi or altered physiology of the whole fly due to the oenocyte driver that they use, which may be driving GAL4 in other parts of the fly.

The reviewer is correct: formally, one can never disprove the potential for off-target effects of RNAi or GAL4 driver. Only one off-target effect is reported for the RNAi constructs we tested (Irc has a reported off-target effect on CG6170). We added the following text to the Discussion to acknowledge this caveat: “In addition, we note that phenotypic changes associated with knocking down expression of a target gene with RNAi may not be causal, but a consequence of off-target effects of RNAi or the GAL4 driver. Further studies are needed to clarify the effects on CHC production of these genes and their interactions, and to test specific mechanisms and enzymatic activities through which they exert these effects.”

Second, while some of the effects of the candidate genes can be explained by the biological/molecular functions of the gene (i.e. fatty acid reductase influencing chain length), others are not (i.e. P450s affecting CHC abundance, although Qiu et al. 2012 PNAS identified the exact mechanism for another P450, Cyp4g1 in CHC synthesis). How would a P450 (Cyp9f2) be involved in elongation of CHCs? As far as is known, that is not one of the reactions that a P450 could catalyze.

We removed the sentence suggesting that Cyp9f2 may be involved in the elongation of precursors of longer chain n-alkanes and monoenes in females. We later state that: “The link between these CYPs and CHC production and maintenance is unclear. However, we speculate that oxidation reactions mediated by these CYPs may regulate CHC degradation and turnover.”

https://doi.org/10.7554/eLife.09861.051

Article and author information

Author details

  1. Lauren M Dembeck

    1. Department of Biological Sciences, North Carolina State University, Raleigh, United States
    2. Genetics Program, North Carolina State University, Raleigh, United States
    3. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    Present address
    Ecology and Evolution Unit, Okinawa Institute of Science and Technology, Onna-son, Japan
    Contribution
    LMD, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  2. Katalin Böröczky

    1. Genetics Program, North Carolina State University, Raleigh, United States
    2. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    3. Department of Entomology, North Carolina State University, Raleigh, United States
    Present address
    Department of Neurobiology and Behavior and Department of Ecology and Evolutionary Biology, Cornell University, Ithaca, United States
    Contribution
    KB, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  3. Wen Huang

    1. Department of Biological Sciences, North Carolina State University, Raleigh, United States
    2. Genetics Program, North Carolina State University, Raleigh, United States
    3. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    Contribution
    WH, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  4. Coby Schal

    1. Genetics Program, North Carolina State University, Raleigh, United States
    2. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    3. Department of Entomology, North Carolina State University, Raleigh, United States
    Contribution
    CS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  5. Robert R H Anholt

    1. Department of Biological Sciences, North Carolina State University, Raleigh, United States
    2. Genetics Program, North Carolina State University, Raleigh, United States
    3. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    Contribution
    RRHA, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  6. Trudy F C Mackay

    1. Department of Biological Sciences, North Carolina State University, Raleigh, United States
    2. Genetics Program, North Carolina State University, Raleigh, United States
    3. W. M. Keck Center for Behavioral Biology, North Carolina State University, Raleigh, United States
    Contribution
    TFCM, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    trudy_mackay@ncsu.edu
    Competing interests
    The authors declare that no competing interests exist.

Funding

National Institutes of Health (R01 GM59469)

  • Robert R H Anholt
  • Trudy F C Mackay

Blanton J Whitmore Endowment

  • Coby Schal

National Institutes of Health (R01 GM45146)

  • Robert R H Anholt
  • Trudy F C Mackay

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Rick Santangelo for technical support. We also thank Drs. Akihiko Yamamoto and Richard Lyman for maintenance of fly lines and Drs. Joel Levine and Pat Estes for sharing the PromE(800)-GAL4 line. We acknowledge the Vienna Drosophila RNAi Center and the Exelixis Collection at the Harvard Medical School.

Reviewing Editor

  1. Daniel J Kliebenstein, University of California, Davis, United States

Publication history

  1. Received: July 3, 2015
  2. Accepted: November 12, 2015
  3. Accepted Manuscript published: November 14, 2015 (version 1)
  4. Accepted Manuscript updated: November 20, 2015 (version 2)
  5. Version of Record published: January 15, 2016 (version 3)

Copyright

© 2015, Dembeck et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,010
    Page views
  • 562
    Downloads
  • 36
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Genetics and Genomics
    2. Neuroscience
    Rebecca Delventhal et al.
    Research Article
    1. Cell Biology
    2. Genetics and Genomics
    Felix Kessler, Paolo Longoni
    Insight