The sexually dimorphic transcriptome
Nearly one-half of the genome (6,569 probe sets) exhibited significantly different transcript levels between the sexes (P(Sex) < 0.001), with 3,965 probe sets upregulated in females and 2,604 probe sets upregulated in males (the complete list is given in Additional data file 1). The greatest differences in transcript abundance between the sexes were for probe sets implicated in sex-specific functions: chorion, vitelline membrane, and yolk proteins involved in egg production were upregulated in females; and accessory gland peptides, male-specific RNAs, and protein ejaculatory bulb components were upregulated in males. However, the probe sets exhibiting sex dimorphism in expression fell into 28 biological process and 41 molecular function Gene Ontology (GO) categories; for most of these categories, differences in expression between the sexes was unexpected. We determined which GO categories contained significantly different numbers of upregulated probe sets in males and females (Table ). Genes involved in the biological process categories of cell communication, cell growth and/or maintenance, development, and cell death were upregulated more often in females than in males. Genes involved in the molecular function categories of binding, most enzymes, signal transduction, structural molecules, and regulation of transcription and translation were upregulated in females more often than in males; however, genes encoding oxidoreductase enzymes, carrier transporters and ion transporters were upregulated in males more often than in females (Table ).
Gene Ontology categories with sex-biased gene expression
The genomic distribution of sex-biased genes was not random (Figure ). There was a paucity of male-biased genes on the X and fourth chromosomes, and an excess on chromosome 2R (χ25 = 100.77; P < 0.0001). There was a deficit of female-biased genes on chromosome 4, and an excess on chromosome 2R(χ25 = 29.18; P < 0.0001).
Chromosome locations of genes differentially expressed by sex. (a) Observed (magenta) and expected (blue) number of probe sets upregulated in males. (b) Observed (magenta) and expected (blue) numbers of probe sets upregulated in females.
Transcriptional response to starvation stress
We found 3,451 probe sets with significantly different mean transcript levels between the control and starved conditions (P(treatment) < 0.001): 1,736 were downregulated (some by as much as 40-fold) and 1,715 were upregulated (at most by 7.2-fold) during starvation (the complete list is available as Additional data file 2). These probe sets fell into 24 biological process and 25 molecular function GO categories. We determined which GO categories had a significantly different number of up- and downregulated probe sets in response to starvation stress. Genes affecting the biological processes of protein and nucleic-acid metabolism (protein biosynthesis; protein catabolism, folding, localization, modification, and repair; biosynthesis of nucleic acid macromolecules and lipids) were upregulated during starvation (Table ). The expression of genes in three molecular function categories (nucleotide binding, hydrolases binding to acid anhydrides, and ribosome structure) increased during starvation; while defense/immunity proteins, peptidases, cuticle structural proteins, and carrier transport proteins were downregulated (Table ).
Gene Ontology categories with increased or decreased gene expression during starvation
The treatment × sex interaction term was significant (P < 0.001) for 817 probe sets, of which 715 had significant treatment effects for one or both sexes in the separate sex analyses (Additional data file 3). We categorized these 715 probe sets as sex-specific if significant expression changes in response to starvation occurred in one sex only; as sex-biased if expression levels changed in the same direction in both sexes, but were of different magnitude; or as sex-antagonistic if expression levels significantly changed in both sexes, but in opposite directions (Figure ). Most probe sets exhibited sex-specific or sex-biased expression, with only two genes, CG14095 and Rpd3, meeting the sex-antagonistic criterion. More probe sets exhibiting sex-specific or sex-biased expression were downregulated (454) than upregulated (263) during starvation. Starvation stress was accompanied by reduced expression of genes involved in the developmental processes of gametogenesis and sex determination as well as signal transduction in females, and of genes involved in mechanosensory and reproductive behavior in males (Table ).
Figure 2 Genetic architecture of transcription. (a-c) Sex × treatment interaction for females (magenta)and males (blue): (a) Chorion protein 38; (b) Alkaline phosphatase 4; (c) Phosphogluconate dehydrogenase. (d-k) Interactions with line. Ore (black), (more ...)
Transcript abundance versus mutations
The genes represented by probe sets with significant treatment and/or treatment × sex effects are candidate genes for starvation resistance. Previously, we screened 933 co-isogenic single P
-element insertion lines for their effect on starvation resistance [21
]. Of these insertions, 383 had significant effects on starvation resistance, while the remaining 550 did not [21
]. Of the 933 lines, we know the locations of the 385 of the inserts and that genes tagged by these inserts are represented on the array. Thus, we can directly compare the extent to which effects of P
-element mutations on the starvation phenotype correspond to changes in transcript abundance in response to starvation. This comparison allows us to assess the hypothesis that changes in transcript abundance can be used to identify candidate genes with effects on phenotype, an hypothesis implicit in previous microarray studies [5
]. Overall, there was no statistical association between the phenotypic and transcript data (χ21
= 0.0006, P
= 1). For 194 genes, there was agreement between the phenotype and the expression level. Seventy-seven genes had significant differences in both transcript profile and mutant phenotypes, and 117 genes affected neither phenotype nor expression level (Additional data file 4). There was disagreement between the expression and phenotypic analyses for 191 genes (49.6%): 108 of the genes tagged by P
-elements affected starvation resistance, but did not display differences in transcript level in response to starvation stress, and P
-element insertions in 83 genes that exhibited significant differences in transcription in response to starvation did not have significant phenotypic effects on starvation tolerance (Additional data file 4).
The genetic architecture of transcription
A total of 706 probe sets exhibited variation in expression among the six lines; 640 probe sets were significant (P < 0.001) for the main effect of line, 190 for the line × sex interaction, 200 for the line × treatment interaction, and 85 for the three-way interaction of line × sex × treatment (Additional data file 5, and Figure ). Thus, transcript abundance exhibits both genotype by sex and genotype by environment interaction.
We used post-hoc Tukey tests to group lines with similar levels of gene expression, and compared the expression clusters with the Ore and 2b genotype of the six lines. There are three possible scenarios by which genetic variation in transcript abundance could arise. First, genetic variation in regulatory regions of gene A causes variation in the expression of gene A (cis-acting regulatory variation). Second, genetic variation in regulation of gene B causes variation in expression of A, which is itself not genetically variable (trans-acting regulatory variation). Third, genetic variation in both gene A and gene B affect the transcript abundance of gene A (cis- and trans-acting regulatory variation). These two-locus interactions could be additive or epistatic. We observe whether or not expression of gene A co-segregates with markers differentiating the two parental strains. Co-segregation will always be observed in case 1. It could also be observed in cases 2 and 3 if gene B is tightly linked to gene A, such that it is not separated by recombination from A in the genotypes tested. However, co-segregation will not be observed if gene A and gene B are unlinked. The most prevalent observation was regulation of expression by unlinked genes. For example, there were unambiguous interpretations for 246 probe sets that were significant for the main effect of line only: 65 (26.4%) were regulated by linked genes and 181 (73.5%) were regulated by unlinked genes (Additional data file 6, and Figure ). We also inferred linkage of genes regulating expression levels under control and starved conditions separately. There were unambiguous Tukey interpretations for 277 probe sets under control conditions, of which 32 exhibited linked regulatory variation (11.6%) and 245 were regulated by variation at unlinked genes (88.4%). For 244 probe sets under starved conditions, 46 were regulated by polymorphism at linked loci, (18.9%) and 198 were regulated by variation at unlinked genes (81.1%) (Additional data file 7).
Association of genetic variance in transcription with QTLs
Probe sets from the three-way ANOVA that are significant for the main effect of line and/or line × sex (P
< 0.001), but not significant for the line × treatment interaction terms, exhibit genetic variation in transcription among the six lines that is independent of the starvation treatment. A total of 489 probe sets met these criteria, and we know the cytological locations of 475 of the corresponding genes. Previously, RI lines derived from Ore and 2b have been used to map QTL affecting variation in life span [22
], sensory bristle numbers [26
], ovariole number [27
], courtship signal [28
], olfactory behavior [29
], metabolism and flight [30
], as well as starvation resistance [21
]. Genes that exhibit significant differences for the main effect of line and/or line × sex which are located within QTL regions are putative candidate genes corresponding to the QTL [16
]. We identified several novel putative candidate genes affecting these traits (Additional data file 5). We examined whether probe sets with significant line and/or line × sex effects tended to cluster within regions containing QTL mapped under standard culture conditions, as would be the case if QTL regions were enriched for genes exhibiting transcriptional variation between the parental lines. We found no evidence for such clustering; indeed, the only trait showing a non-random association of probe sets with QTL that survived a Bonferroni correction for multiple tests was in the direction of a deficiency of probe sets in the QTL region (Table ).
Association of genetic variation in transcription with genetic variation in quantitative traits
The 217 probe sets with significant line × treatment and/or line × treatment × sex terms (Additional data file 5) represent genetic differences among the lines in response to the starvation treatment. Are these probe sets enriched in regions to which starvation resistance QTL map? We found that 47 of the probe sets meeting these criteria, representing 45 unique genes, fell within starvation resistance QTL regions; and the remaining 170 probe sets, representing 169 unique genes, fell outside the QTL intervals. These probe sets were not over-represented within starvation resistance QTL (χ21 = 0.26, P > 0.05).
There is significant variation in starvation half-life among the six lines (P < 0.0001; Additional data file 8). For those probe sets previously identified as having significant differences in transcript level among the lines, we assessed the extent to which variation in transcript abundance was associated with variation in starvation half-life. We found 281 probe sets with significant correlations (P < 0.05) between starvation phenotype and transcript level, for 273 of which the cytological location was known (Additional data file 5). However, 66 of the probe sets associated with starvation half-life mapped to starvation resistance QTL, and 207 did not. Again, these probe sets were not over-represented within starvation resistance QTL (χ21 = 0.45, P > 0.05).
Although there is no tendency for genes exhibiting variation in transcript abundance among lines to cluster within starvation resistance QTLs, those that do co-localize with the QTLs are candidate genes affecting variation in starvation tolerance between Ore and 2b. We found 155 probe sets, corresponding to 153 candidate genes, which met one or more of the above criteria (Additional data file 5). Most (114, 75%) were predicted genes. The remaining genes (Table ) are reasonable candidates for starvation resistance QTLs, affecting the processes of protein metabolism, defense/immune response, proteolysis and peptidolysis, and transport.
Candidate QTLs for starvation resistance
Complementation tests to mutations have implicated several candidate genes affecting variation between Ore and 2b in olfactory behavior [29
), longevity [31
] (Dopa decarboxylase
, shuttle craft
) and starvation resistance [21
] (spalt major
, Ryanodine receptor 44F
, crooked legs
, Phosphoglucose isomerase
, l(2)k17002, l(2)k00611
, and l(2)k03205
). None of these genes exhibited significant differences in transcript abundance between lines.