The expression data from both wild type and tud progeny males and females was first analyzed to identify genes expressed in somatic and germline tissues during metamorphosis. Here, 8,482 and 9,725 genes of the 13,820 genes examined had expression data in the tud and wild type experiments, respectively, demonstrating that ~70% (9,725 of 13,820 genes in wild type flies) of the predicted Drosophila genes are expressed during metamorphosis (see Methods for details). This also suggests that approximately 1,200 additional genes are expressed during metamorphosis due to the presence of the germline in wild type males and females. Our previous study examining gene expression in wild type flies found a larger percentage of genes (~94% of genes represented on arrays) expressed during metamorphosis (3,784/4,028 genes; 7). Our previous study employed a cDNA microarray platform representing about one-third of the genes in Drosophila. As such, it was biased for genes with high expression levels, which might account for the differences in the two studies 7.

To identify genes whose transcript abundances differ between the two sexes in somatic tissues through metamorphosis, the tud progeny gene expression data was analyzed using F-statistics, conducted using LIMMA contrasts with sex and time as independent factors (see Methods and Additional file 1, for details). For the F-test analyses, lists of P values were converted to q values, an estimate of the false discovery rate 13. Two-hundred-fifty-eight genes were identified with significant somatic, sex-differential expression (q < 0.15 for sex or sex-time interaction term; see Methods for details; Additional files 2 and 3). Similar numbers of genes were identified with male- and female-biased expression (124 and 134, respectively). Overall, the percentage of genes with somatic, sex-differential expression during pupal stages (1.9%) is similar to the number of genes displaying somatic, sex-differential expression at adult stages (1.7% of genes [68/4028] in 7, 2.5% genes [301/11893] in 14, and 1.4% of genes [167/11893] in 15). In contrast, thousands of genes show sex-differences in transcript levels in the male and female germline tissues, at both pupal and adult stages (see germline section below; 714).

For the 258 genes with somatic, sex-biased expression, moderated-t-tests were performed 16, comparing gene expression in tud progeny males and females to determine at which stage the gene displays somatic, sex-differential expression (Table 2). The five time points (0, 24, 48, 72 and 96 hour APF) do not have large differences in the numbers of genes with somatic, sex-biased transcript levels (q < 0.15; 75, 181, 79, 131, and 152 genes, respectively; Figure 2), with the data from the 24 hour time point containing the largest number of genes.

At the statistical threshold used for the t-tests (q < 0.15), 242 of the 258 genes identified by F-tests showed significant, somatic, sex differential-expression at a minimum of one time point. Close to half of these genes (119 genes) displayed somatic, sex-differential expression at only one or two time points, suggesting that they are likely to mediate discrete, sex-specific, developmental events. RNA on the X 1 (roX1) is male-biased and one of only thirteen genes that were sex-differentially expressed at all five time points examined. roX1 and a gene RNA on the X 2 (roX2) produce non-coding RNAs that are components of the dosage compensation macromolecular structure 17; roX2 was either expressed exclusively in males or four times higher in males than females at each of the five time points examined. Dosage compensation is the process in which genes on the single X chromosome in males undergo increased transcription, which results in roughly equal amounts of mRNA product produced by the two X chromosomes in females [reviewed in 18]. Of the additional 12 genes (three and nine with male- and female-biased expression, respectively) with sex-differential expression at all five time points examined (Microsomal glutathione S-transferase-like, Glutathione Synthetase, CG4586, Larval serum protein 1 alpha, CG15369, CG15347, CG31775, Ilp6, CG1702, Succinate dehydrogenase B, CG7430, and cabut), eight are located on the X chromosome. Sex-lethal (Sxl), the gene at the top of the sex-determination hierarchy (see Figure 1, [reviewed in 3]), displays female-biased expression at four of the five time points examined. Interestingly, CG3056, a gene with male-biased expression at four of the five time points, is a paralog of Sxl 19; the product of this gene may underlie additional sex-differential splicing that regulates sex-specific development.

The differences in transcript abundances observed at each stage are due to biological differences and not poor quality data, as the microarray data showed high correlation among experimental replicates and similar numbers of genes had expression data at each time point in both sexes (see Additional file 4 for all microarray correlations and number of genes expressed in each experiment).

One of the primary goals of this study was to identify genes that direct the patterning and morphogenesis of sexually dimorphic, somatic tissues. Hierarchical clustering, an algorithm that groups genes based on the similarity of their expression profiles 20, was performed using the five time points of tud progeny expression data to identify genes with similar expression profiles in somatic tissues. Because the numbers of expected clusters were unknown, other clustering methods, including K-means clustering and self-organizing maps [reviewed in 21], were not employed. Thirty-eight clusters, each with a greater than 0.80 average Pearson's correlation in gene expression profiles and containing 15 genes or more, were identified and further analyzed (Figure 3 and Additional file 5). Combined, these clusters contained 4410 genes, including 82 genes with alternative transcripts expressed in two clusters and four genes with alternative transcripts expressed in three clusters. There is a high degree of separation among the clusters, with an average correlation of 0.026 between all of the clusters, demonstrating that the expression profiles for genes within a cluster are not similar to expression profiles for genes in other clusters and thus each should be considered separately in this study (see Methods).

An analysis of the expression data across the five time points for the genes in each of the 38 clusters identified 51 peaks and 49 troughs of gene expression within the clusters (Figure 3; Methods). Based on a re-sampling analysis, each of these 51 peaks and 49 troughs was significantly different than the average expression at all time points in the cluster from which they were identified (P < 0.01; see Methods). When wild type expression data was incorporated into the cluster analyses, gene expression profiles appear similar in wild type and tud progeny experiments, based on visual inspection (Additional file 6); only genes with expression in tud progeny were included in the cluster analyses. Furthermore, a statistical comparison of the average tud progeny expression data to the average wild type expression data within each cluster showed that 31 of the 38 clusters had correlations >0.70, demonstrating a high degree of similarity of expression for genes in tud progeny and wild type animals. The seven clusters with correlations <0.70 between the average tud progeny and wild type expression data contained only 172 of the 4410 genes present in the 38 clusters. Therefore, for most genes examined, the pattern of gene expression in somatic tissues during metamorphosis does not appear to be largely influenced by the presence of the germline.

To functionally analyze these clusters, the sets of genes from each cluster were examined using the program DAVID, which identifies overrepresented functional groups among the genes in each cluster, as compared to all the genes represented on the array platform (see Methods; Additional file 7; DAVID is the Database for Annotation, Visualization and Integrated Discovery 22). To confirm the DAVID results, an independent tool that searches for enrichment of Gene Ontology terms (FlyMine; 23) was used to assess overrepresented functional categories and gave very similar results (see Methods for details).

At the white pre-pupal stage (wpp; 0 hour APF), the fly is transitioning from a wandering larva into an immobile pre-pupa. The pre-pupal animal initiates the major larval-to-adult transition in several discrete ways: 1) strictly larval tissues are destroyed and replaced by corresponding adult tissues [reviewed in 24], 2) imaginal discs and rings begin to give rise to adult structures including eyes, antennae, wings, legs, and genitalia [reviewed in 25], 3) histoblast nests proliferate in number and give rise to non-imaginal disc derived adult epidermal structures 26, and 4) the larval central nervous system is remodeled through the destruction of some larval neurons, proliferation of neuroblasts to generate new neurons, and remodeling of some larval neuronal projections [reviewed in 27]. The 24 hour APF time point is between the pre-pupal pulse of ecdysone, which peaks at 12 hour APF and triggers head eversion, and the large pupal pulse of ecdysone that initiates around 24 hour APF [reviewed in 1]. By 24 hour APF the majority of larval-specific tissues are degraded and adult development is triggered [reviewed in 28]. During the time between the 24 to 48 hour APF stages, the imaginal discs are still undergoing morphogenesis, but are close to their final adult form. The wings, leg muscles, abdominal bristles, abdominal muscles and internal genital ducts are all well formed, while further development of the eyes, legs, wings, thorax, and abdomen is occurring [reviewed in 2]. During the later stages of metamorphosis (72 hour APF), many of the tissues and structures developing in the pupae are close to their final adult form [reviewed in 2729]. By 96 hour APF, the pupa is within a few hours of eclosion, or emergence of the adult fly [reviewed in 30].

To understand the transcriptional basis of these complex developmental events, expression data was analyzed in the following ways: first, genes with similar expression patterns in both male and female somatic tissues were identified based on the hierarchical cluster analyses (Figure 3). Second, clusters were identified that contained genes that either had a peak or trough of their transcript abundance at each time point (see Methods). At the 0 hour APF stage, Cluster 5 (586 genes) has genes with peak expression and was enriched for genes that encode proteins that function in the proteosome (34 genes; P = 6.8 × 10^-31), have cell death activities (35 genes; P = 4.2 × 10^-7) or peptidase activities (90 genes; P = 5.4 × 10^-18) and thus likely function in the histolysis of larval tissues. Cluster 21 (254 genes) has genes in a trough of expression and is enriched for genes whose products function in development (60 genes; P = 4.2 × 10^-6), differentiation (25 genes; P = 4.6 × 10^-4), and cell communication (54 genes; P = 9.8 × 10^-5), suggesting that a large fraction of genes that function in these patterning and developmental processes are at low transcript levels immediately after pre-pupal formation.

At the 24 hour APF stage, the largest cluster identified, Cluster 13 (1,121 genes), shows peak transcript abundance (Figure 3). Cluster 13 is overrepresented with genes encoding products that are annotated as functioning in imaginal disc morphogenesis (73 genes; P = 2.1 × 10^-15), neurogenesis (58 genes; P = 3.9 × 10^-14), programmed cell death (62 genes; P = 9.9 × 10^-11), nervous system development (122 genes; P = 2.4 × 10^-18), and transcription (183 genes; P = 2.8 × 10^-16), demonstrating that at about 24 hour APF, many genes that drive morphogenesis and patterning have reached a peak in their transcript abundance, marking this period as critical for patterning and morphogenesis.

At the 48 hour APF stage, Cluster 18 and Cluster 21 contain 138 and 254 genes, respectively, and show peak levels only at this stage (Figure 3). Cluster 18 is enriched with genes whose products function in cell organization and biogenesis (21 genes; P = 0.016), appendage morphogenesis (5 genes; P = 0.032), and pupal development (8 genes; P = 0.041), suggesting that although the rudimentary adult structures are formed, there are still many structural changes taking place. Consistent with this idea, Cluster 21 is enriched with genes whose protein products function in development (60 genes; P = 4.2 × 10^-6), cell communication (54 genes; P = 9.8 × 10^-5) and morphogenesis (30 genes; P = 1.1 × 10^-4).

At the 72 hour APF stage, genes in Cluster 31 (737 genes, Figure 3), which are in a trough at the 0 and 24 hour APF stages, quickly increase in transcript levels to ultimately peak at 72 hour APF. Cluster 31 is overrepresented with genes that encode products that function in the mitochondria (153 genes; P = 1.4 × 10^-77).

One large cluster, Cluster 32 (278 genes; Figure 3) contains genes that showed a sharp rise in transcript levels at 96 hour APF, but no peaks early in metamorphosis. This cluster is enriched for genes encoding products that function in the response to light stimulus (7 genes; P = 6.6 × 10^-4), visual perception (7 genes; P = 0.0050), and rhabdomere function (6 genes; P = 3.0 × 10^-5), all of which are critical for proper vision and development of the adult eye. On the other hand, the largest cluster, which is enriched for genes that encode products functioning in developmental processes (all P = 3.9 × 10^-14) and peaked in transcript abundance at 24 hour APF, is in a trough of transcript levels at 96 hour APF (Cluster 13, 1,193 genes), consistent with the idea that morphogenesis is largely complete by the pharate adult stage.

Genes with expression during metamorphosis in either male or female germline tissues were identified using two independent F-statistic analyses (q < 0.15 for each test; see Methods; see Additional file 2 for numbers of genes with significant expression differences). This statistical approach also identifies genes whose expression in somatic tissues is dependent on the presence of male or female germline tissues, respectively; these two classes of genes cannot be distinguished in this study. The gene sets that are expressed within or dependent on the presence of the germline in males and females are referred to as the pupal male- and female-biased germline sets, respectively. Sets of 659 and 342 genes with male- and female-biased expression in the germline, respectively, were identified (Additional files 8 and 9). Both the male- and female-biased pupal germline sets had significant overlap with genes previously identified as highly expressed in adult male and female germline tissues, respectively (P < 1.1 × 10^-4, hypergeometric test for both sets; 14). Five-hundred-forty-three genes identified in the pupal male germline set were on the previous study's array platform 14. Of those 543 genes, 392 were highly expressed in the adult male germline. Similarly, 305 genes identified here as being expressed in the pupal female germline were present on the previous study's array platform. Forty-seven of those 305 genes were also highly expressed in the adult female germline.

Gene expression in the male germline has already initiated at the start of metamorphosis, and by 24 hour APF has reached its peak level of gene expression; this high level of expression lasts throughout metamorphosis (Figure 4A). Sixty-nine of the 659 genes in the pupal male germline set encode products that function in the mitochondria (P = 5.8 × 10^-19; DAVID analysis), consistent with the essential role for mitochondria in spermatid development and adult function [reviewed in 31]. One-hundred-fifty-one genes (of 543 genes, 28%) were identified that are expressed in the male germline during metamorphosis, but were not previously identified as expressed in the adult male-germline (present on platform of previous study, but not significantly differentially expressed 14), suggesting that there is pupal-specific, male-germline gene expression that might underlie male germline development.

Previously, it was observed that most genes expressed in the female germline showed the first post-embryonic peak of transcript abundance during adult stages 7. However, because our previous study did not have data from pupal stages examining gene expression in each of the sexes separately, or in male and female tud progeny, we were unable to definitively identify the genes expressed in the female germline at pupal stages. The data presented here demonstrate that are a substantial number of genes with female-biased germline expression during pupal stages (Figure 4B). Between the 48 and 72 hour APF stages, the structures derived from the female genital disc establish connections with female gonadal tissues to form the female reproductive system [reviewed in 32]. The development of female reproductive structures likely requires gene expression in both somatic and germline tissues. This idea is consistent with the functions of genes with pupal female germline expression, as this set is overrepresented with genes that function in developmental processes (76 genes, P = 0.0031; DAVID analysis).

Interestingly, the transcript level of genes expressed in the pupal female-germline also peaks in both wild type males and females and tud progeny males and females at the early stages of metamorphosis (Figure 4B), suggesting they also play a non-sex-specific role in pupal somatic tissues. Several genes encoding products annotated as functioning in the female germline (found in Cluster 21) peak in transcript abundance in male and female somatic tissues at 48 hour APF (Figure 3). However, by the later stages of metamorphosis, the levels of these transcripts remain high only in wild type females and drop to trough levels in wild type males, and tud male and female progeny.

The chromosomal distribution of genes with sex-biased expression in the male and female germlines was additionally analyzed. Genes expressed in the pupal male germline are underrepresented on the X chromosome and overrepresented on the left arm of the second chromosome (P = 3.5 × 10^-7 and 0.0072, respectively, hypergeometric test), both of which have been shown for genes expressed in adult male germline tissues 14. Interestingly, genes expressed in the pupal male germline are also overrepresented on the right arm of the third chromosome (P = 0.024, hypergeometric test).

Hierarchical clustering was performed using all the data from each microarray experiment from the time course study, rather than using the data from each gene, to determine how similar global expression patterns are between males and females. When the tud progeny expression data was analyzed, the global expression profiles of males and females from each pupal time point were most similar to each other (Figure 5A). This was expected because very few genes with somatic, sex-differential expression were identified (see above). A clear distinction between overall gene expression at early stages (0–48 hour APF pupae) and late stages (72–96 hour APF pupae) was observed (Figure 5A). This is consistent with our cluster analyses (Figure 3), where many genes appear co-regulated at either early or late stages of metamorphosis, but not at both early and late stages: 1,745 genes shared either peaks or troughs of transcript levels at multiple early stages (0–48 hour APF) or late stages (72–96 hour APF), while only 513 genes shared peaks or troughs of transcript levels at an both an early and a late time point.

When the wild type expression data and the tud progeny expression data were analyzed together, the clear distinction between early and late metamorphosis remained, as with the tud progeny expression data alone (Figure 5B). As expected, male-germline gene expression has a large effect on how the global transcriptional profiles cluster, with wild type males expression data always clustering separately from wild type females and from the male and female tud progeny. This effect appears to be less substantial at 0 hour APF, as the global expression profile of wild type males clusters closest to the data from the other three genotypes, suggesting that at the start of metamorphosis, many genes expressed in the male germline are not as abundant during early time points. The wild type female array experiments also cluster closely to, but separately from, the tud progeny array data, with the largest differences seen at 72 and 96 hour APF. This is consistent with gene expression in the female germline increasing at the end of metamorphosis (Figure 4B).

Next, genes regulated by the sex hierarchy during pupal developmental stages were identified. Nearly all of the sexually dimorphic tissues are either patterned or undergoing morphogenesis to bring about the adult sexual dimorphisms during pupal stages. The 48 hour APF pupal stage was chosen because previous studies showed that FRU^M peaks at this stage 33 and DSX shows high expression at this time 34. For these experiments, the array hybridizations were performed as direct comparisons using RNA from the two genotypes (see Table 1 and Methods). Genes were first identified that had sex-biased transcript levels between wild type males and females by analyzing expression data from two different wild type strains, Canton S and Berlin (q < 0.15), and from tud progeny males and females (one-tailed t-test, q < 0.15). This resulted in a set of 420 genes (320 and 100 genes with female- and male-biased expression, respectively; Additional file 10). This is substantially more than was identified in the time course analysis for this time point (79 genes, see above); however this difference is likely due to the increased number of replicates (four replicates for each comparison versus three replicates in time course) and the decreased statistical error by directly comparing gene expression on the same array, as opposed to using a common reference RNA sample.

Given the larger number of somatic, sex-differentially expressed genes identified by this approach, it could be determined if there was a bias for chromosomal positions in these gene sets. Genes with somatic male-biased transcript levels at the adult stage are known to be underrepresented on the X-chromosome 14. There was not a similar bias for the 100 genes with somatic, male-biased transcript levels at the 48 hour APF pupal stage, but rather there was an equal distribution across all chromosomes (hypergeometric test, P > 0.05). Interestingly, the genes with somatic female-biased transcript levels at this pupal stage were overrepresented with genes located on the X chromosome (90 genes, hypergeometric test, P < 0.001). The previous study did not find any significant over- or underrepresentation on any chromosome for genes with adult somatic female-biased expression in the adult 14.

Next, genes were identified that were differentially expressed as a consequence of tra, a gene in the sex hierarchy that encodes a pre-mRNA splicing factor required for the production of the sex-specific dsx mRNA splice variants 35. Transcript levels in chromosomally XX flies mutant for tra (hereafter called tra pseudomales) were compared to wild type female flies. The tra pseudomales produce DSX^M and look very similar to wild type males. Of the 420 genes that showed somatic sex-biased transcript levels, 95 genes were identified (72 female-biased and 23 male-biased) that are also significantly different between tra pseudomales and wild type females (one-tailed t-test, q < 0.15 for each test, Additional file 10). In this experimental design genes were required to be differentially expressed in three different genotype comparisons of male and female gene expression (CS, Berlin, and tud), precluding the identification of genes that are differentially expressed in only one strain. This is also true for the set of DSX-regulated genes identified below. As a validation of our experimental approach, Sxl, tra, roX1, and roX2 are all sex-differentially expressed in somatic tissues (wild type and tud progeny comparisons). Only tra is differentially expressed in the chromosomally XX, tra pseudomale and wild type female comparison. This is expected as Sxl, roX1, and roX2 are not regulated downstream of tra in the sex determination hierarchy, but are regulated downstream of the primary determinate of sex, the X chromosome to autosome ratio (Figure 1; [reviewed in 36]).

Of the 326 genes that are not regulated by TRA, a large portion (169) may be false negatives as their expression values are significant or close to significantly different (q < 0.30) in microarray experiments identifying genes regulated by dsx (see below). A gene regulated downstream of dsx should also be regulated downstream of tra. Another 19 have a q value close to the cutoff for significance in the tra microarray expression data (q < 0.30). Removing these 188 genes from consideration still leaves a large number of genes (138 genes) that are sex-differentially expressed independently of tra. A significant number of these 138 genes (45 genes; P < 0.05, hypergeometric test) are located on the X chromosome. It is possible that differences in transcript levels of these genes is due to differences downstream of Sxl, or X chromosome composition in males and females, suggesting that the dosage compensation process does not completely normalize expression between males and females for all genes on the X chromosome.

Next, gene expression between pupae that are transheterozygous for the dsx^D allele 11 – an allele that only produces the male-specific isoform (DSX^M) – and a dsx null deletion allele (dsx^m+r15) was compared to gene expression in wild type females. These chromosomally XX, dsx^D/dsx^m+r15 pseudomales look very similar to wild type males, as they only produce DSX^M. Of the 95 genes that are sex-differentially expressed in somatic tissues and downstream of tra, 66 genes were identified as being regulated downstream of dsx (one-tailed t-test, q < .015). Forty-six and 20 genes are more highly expressed in females and males, respectively, downstream of tra and dsx (Table 3 and Additional file 10).

Aside from the tra gene itself, 28 genes were differentially regulated by TRA, but were not differentially expressed between dsx^D and wild type females. If genes that are close to the significance level (q < 0.30; 6 genes) and genes with expression data in only one or two dsx^D comparisons (3 genes) are removed, 19 genes remain that are downstream of tra, but not dsx. Interestingly, none of these genes are significantly differentially expressed in similar experiments examining FRU^M regulation at this stage (data not shown). This suggests either an alternate branch of the sex-hierarchy downstream of tra, possibly through dissatisfaction 37, or the possibility of additional genes on which TRA acts to sex-specifically splice their pre-mRNAs, leading to differential abundance of transcripts due to differences in mRNA stability.

Requiring a gene to show statistical differences in expression in all direct microarray experiments yields a high confidence set of true positives regulated downstream of dsx, but will likely generate false negatives. To identify additional dsx regulated genes that might have been missed because of the stringency of having to pass multiple tests, genes were included that showed sex-differential, somatic expression, but which were not differentially expressed in the tra microarray comparisons. These genes were required to be significantly differentially expressed in the dsx^D comparisons at a more stringent level (q < 0.05) to avoid false positives. This yielded an additional 107 genes, with 75 and 32 showing female- and male-biased expression, respectively (Table 3 and Additional file 10). This study thus identified 173 genes regulated as a consequence of dsx (DSX set; 121 and 52 genes with female- and male-biased expression, respectively).

Several of the genes with male- and female-biased expression in the DSX set with the highest fold change include those with products that might be involved in epithelial morphogenesis, imaginal disc morphogenesis or cuticle formation, based on their sequence identity. The 52 genes with male-biased expression contains seven such genes, including ecdysone inducible ImpE1 (FC = 4.8), miniature (FC = 3.8), and dusky-like (FC = 3.1). Among the 15 genes with the highest female-biased expression, four encode proteins with cuticular domains (Cuticular protein 97Eb, 50Ca, 97Ea, and 51A; FC = 5.9, 3.5, 3.3, and 2.8, respectively), as well as obstructor-A (FC = 2.5) and abdominal A (FC = 1.6). While it has long been recognized that cuticle deposition is tied to tissue morphogenesis and both are developmental events occurring during the middle of metamorphosis, the identification of several genes likely involved in sex-specific aspects of this process had not been determined until this study.

Six genes with female-biased expression regulated downstream of DSX at 48 hour APF have products with functions in the muscle or muscle differentiation (flightin, Limpet, CG31781, Tropomyosin 1, abdominal-A and Sarcoplasmic calcium-binding protein 73839404142). This suggests that aspects of pupal muscle development occur in temporally distinct manner between males and females, and that this differential timing is regulated by DSX. It is not clear is if this is due to the development of sex-specific muscles or due to differences in the developmental rate of non-sex-specific muscles between males and females.

In our previous microarray study examining modes of DSX-regulated gene expression at the adult stage in head tissues, a large number of sex-biased genes were found that were either activated or repressed as consequence of dsx activity in both males and females, but the extent of activation or repression was sex-specific 4344. This mode of regulation was distinct from the previous descriptions of DSX-regulated gene expression based on the only known direct targets of DSX, Yolk protein 1 (Yp1) and Yolk protein 2 (Yp2). DSX^F activates Yp1 and Yp2 expression in the female fat body and DSX^M represses Yp1 and Yp2 expression in the male fat body tissues 45.

To test if the set of genes regulated as a consequence of DSX activity contains genes that may be directly regulated by DSX, it was determined if the known DSX binding site sequences are present in the regulatory region of these genes 46. We searched for the presence of two DSX binding sites in the DNA sequence 2000 base pairs upstream from the transcription start and within the first intron, for each gene in our list (see Methods for search criterion) 45. Two DSX-binding sites were identified in 46 of the 165 (28%) DSX-regulated genes, a statistical overrepresentation as compared to all genes in the genome (P = 0.0002, hypergeometric test), suggesting that a fraction of the genes identified here may indeed be direct targets of DSX. We note that in this study regulation by DSX may be direct or indirect.

To determine the modes of regulation by DSX in pupal stages, gene expression was compared between chromosomally XX and XY dsx null flies and wild type females and males, respectively (hereafter called dsx null comparisons). Data was examined for the 173 genes we identified here as being downstream of dsx (DSX set; Table 3 and Additional file 10). Fifteen genes did not have enough expression data for statistical analysis or were not significantly differential expressed in either dsx null comparison and were therefore not considered for further analysis. Of the remaining 158 genes, 151 show significant differential expression (q < 0.15) in both dsx null comparisons, which suggests regulation downstream of both DSX^F and DSX^M activity. The remaining seven genes (2 and 5 male- and female-biased genes, respectively) only showed significant differential expression in one of the dsx null comparisons; these seven genes may possibly be regulated downstream of one isoform of DSX, a method of DSX regulation that was previously proposed for some genes with sex-differential expression in the adult 4344.

Of the 151 genes regulated as a consequence of dsx in both sexes, 104 of the genes had female-biased expression and were more highly expressed in wild type females and males as compared to dsx null females and males, respectively. This suggests that these genes are activated downstream of DSX in both females and males, but that DSX^F activity results in more potent activation. Forty of the 151 genes were male-biased and more highly expressed in male and female dsx null flies than in wild type males and females, suggesting these genes are repressed as a consequence of DSX activity in both males and females, but DSX^F activity results in more potent repression. Thus, the majority of genes that are regulated as a consequence of dsx are not regulated in the Yp-like mode of regulation, but rather are regulated similarly in both sexes, with gene expression downstream of one isoform resulting in more potent activation or repression, as previously described in our studies of adult head tissues 44. Interestingly, Yp-like regulation was observed for only seven genes in our pupal dataset: the genes with male-biased expression CG8086 and CG14995 and the genes with female-biased expression abdominal-A (abdA), LpR1, CG10802, CG1441, and CG9485. It is possible that the Yp-like mode of regulation may be the more common method of regulation for a particular class of genes (i.e., direct targets) or might be revealed to be the primary mode of dsx regulation when higher resolution analyses are performed.

abdA, which appears to be activated downstream of DSX^F in females and repressed downstream of DSX^M in males, is a well-characterized homeotic selector gene that was shown to be important for specifying segment identity 4748. In the time course analyses above, abdA was found to have female biased expression at 0, 24, 48, and 72 hour APF, with the highest expression difference between the sexes at 48 and 72 hour APF. Previous research examining 40–45 hour APF, suggested that ABD-A and DSX, along with Abdominal-B, act to regulate the expression level of a downstream target, bric-a-brac, and lead to differential abdominal pigmentation between males and females 49. In that study it was shown that abdA transcript levels in the abdominal epidermis do not vary between dsx null animals and wild type animals, thus suggesting that abdA is not regulated by DSX in this tissue. The dsx-dependent differential expression of abdA that we observed could be due to expression in other tissues, since here whole pupae were analyzed. Indeed, abdA has been shown to be expressed and functional in several distinct tissues and cell types, including abdominal neuroblasts and the female genital disc 5051.

The proposed modes of regulation were validated by additional microarray experiments in which the male and female isoforms of DSX were over-expressed (data not shown). Of the 158 genes in the DSX-regulated set for which DSX modes of regulation was examined, 47 did not show significant differential expression in the experiments when we either over-expressed DSX^F in females or DSX^M in males. Of the remaining 111 genes, 35 were male-biased; these genes showed decreased expression when DSX was over-expressed, either in one or both of the DSX isoform over-expression experiments. Similarly, of the remaining 76 female biased genes, 74 showed increased expression levels when DSX was over-expressed, either in one or both of the DSX isoform over-expression experiments. Only two genes with female-biased expression (CG4484 and CG3244) showed decreased expression levels when DSX^F was over expressed in females compared to control females, opposite of the predicted effect from our model.

Male and female flies were collected at the white pre-pupal stage between Zeitgeber time (ZT) 1 and ZT 4 and aged at 25°C for the following hours: 0, 24, 48, 72, and 96 (time course) or 48 (sex hierarchy mutants), and then snap frozen in liquid nitrogen. Wild type flies were Canton S (CS) and Berlin. Animals that lack germline tissues (called tud progeny) were the progeny of female tud¹bw¹sp¹ and male y, w, P[w^+cMUBI-GFP];ID-1 P [FRT(w^hs)]101. Female tud progeny had GFP expression and could be distinguished from males by fluorescence microscopy. Below, chromosomal sex for sex hierarchy mutant flies is indicated in parentheses. Chromosomally XX sex hierarchy mutants were identified based on the GFP marker on the X chromosome, derived paternally. tra pseudomales were the genotype y, w, P [w^+cMUBI-GFP]/+;tra¹/Df(3L)st-j7 (XX) and were compared to CS females. dsx pseudomales were y, w, P[w^+cMUBI-GFP]/+;dsx^D, Sb¹, e¹/dsx^m+r15 (XX), and compared to CS females. For the dsx null analyses, P [w^+cMUBI-GFP]/+; dsx^d+R3/dsx^m+r15 (XX) flies were compared to CS females and dsx^d+r3/dsx^m+r15 (XY) flies were compared to CS males. All flies were kept at 25°C in a 12:12 hour light-dark cycle and grown using standard food media.

All time course microarray experiments were conducted with three replicates; for every experiment, cDNA from the experimental genotype contained incorporated Cy3-labeled dUTP and cDNA from the reference sample contained incorporated Cy5-labeled dUTP. The common reference was comprised of RNA from CS flies of both sexes from all pupal stages. Microarray comparisons using RNA derived from sex determination hierarchy mutants were conducted with four replicates, with a dye-swap design; i.e. cDNA from each genotype contained incorporated Cy3-labeled dUTP in two experiments and contained incorporated Cy5-labeled dUTP in the other two experiments. RNA was isolated from ~30 pupae by homogenization and extraction using the TRIzol^® protocol (Invitrogen, Carlsbad, CA) and resuspended in 20 μL diethylpyrocarbonate (DEPC)-treated H₂O.

cDNAs were directly labeled with Cy5 or Cy3 during the reverse transcription reaction; 30 μg of total RNA was used as a starting template. The reverse transcription reaction was performed for two hours at 42°C using the following reagents (values in parentheses are final molarity or final concentration): oligo dT primer (Operon, 3.75 μM), dithiothreitol (Invitrogen, 10 mM), First Strand Buffer (Invitrogen, 1×), dNTPs minus dTTP (Invitrogen, 0.5 mM), dTTP (Invitrogen, 50 μM), Cy-labeled dUTP (Perkin-Elmer, 0.625 nM), and Superscript II reverse transcriptase (Invitrogen, 10 U/μL). The reaction was stopped and RNA was hydrolyzed by a 20 minute incubation at 65°C with NaOH (167 mM) and EDTA (83 mM). After neutralizing by adding HEPES buffer (pH8.0; 294 mM) and sodium acetate (pH5.2; 228 mM), cDNA samples were purified (Gel-Purification Kit, Qiagen, Valencia, CA). cDNA samples were then dried and resuspended in formamide (56%), sodium citrate buffer (SSC, 3.37×), SDS (1.12%), Denhardts (5.62×), and Polyadenylic acid potassium salt (0.9 mg/ml; Sigma-Aldrich, St. Louis, MO). Samples were boiled for 2 minutes and then applied to microarray slides underneath a LifterSlip (Erie Scientific, Portsmouth, PA). Microarrays were hybridized at 42°C for 14–18 hours, then washed in a solution of 1.5% SDS and 1 × SSC for 5 minutes, a solution of 0.20 × SSC for 5 minutes, and two solutions of 0.05 × SSC for 10 minutes each.

The oligonucleotide set that was printed on the glass slides consisted of 15,156 oligonucleotides, representing the full predicted set of transcribed regions of the D. melanogaster genome, including 14,454 known and predicted open reading frames and an additional 702 control spots. The 14,454 oligonucleotides represent 13,820 unique genes. Throughout the paper, the number of genes identified is reported, while the Additional files contain information on the specific transcripts for each gene, including multiple transcripts identified for the same gene. The oligomer set was designed by the International Drosophila Array Consortium (INDAC; http://www.indac.net/) based on release 4.1 of the D. melanogaster genome using a custom implementation of OligoArray2 57. The oligonucleotides were designed with sizes ranging between 65–69 nucleotides, a minimal Tm window, bias towards the 3'-ends of transcripts, and minimal sequence similarity to other genes 58. The oligonucleotides were synthesized by Illumina (San Diego, CA); sequences can be downloaded from FlyMine: http://www.flychip.org.uk/services/core/FL002/. Additionally, microarrays used for the time course analyses and the dsx null analyses contained additional array elements for Sxl, tra, the female-specific splice form of dsx (dsx^F), the male-specific splice form of dsx (dsx^M), the fru transcript that is sex-specifically spliced to produce FRU^M (fru P1), and each of three of the fru DNA-binding domains (fru^A, fru^B, and fru^C). Sequences for these oligonucleotides can be found in Additional file 11. All microarrays were printed in the laboratory of Dr. Eric Johnson at the University of Oregon (Eugene, OR) using slides coated with aldehyde chemistry and were postprocessed using the Nunc SuperChip Aldehyde protocol (Thermo Fisher Scientific, Waltham, MA).

All arrays were scanned using the GenePix 4100A scanner and GenePix Pro 5.0 software from Axon Instruments (Molecular Diagnostics, Sunnyvale, CA). Visual inspection of the microarray images filtered out florescence most likely not due to labeled cDNA binding; the data from these array elements was discarded. Array elements were only considered for further analysis if at least one channel (Cy3 or Cy5) had greater than 75% of the pixels with intensity values one standard deviation above background levels. All microarray normalization and statistical analyses were performed using the LIMMA package of BioConductor in the program R 16596061. Global-loess normalization was used for all arrays, and significance was converted to q values using the q value application for R 13.

To determine the experimental reproducibility for the time course study, correlation values between replicates of each experiment were determined using Pearson's correlation on the logarithm of the ratio values for every oligonucleotide on the microarray. For each experiment, pair-wise Pearson correlation comparisons were performed on the three microarray replicates. All comparisons between replicates for the tud progeny experiments of the time course microarray study had correlations >0.75 and 27 of the 30 replicate comparisons had correlations >0.80. All comparisons between replicates for the wild type experiments in the time course microarray study had correlations >0.68 and 28 of the 30 replicate comparisons had correlations >0.80. See Additional file 4 for correlation values.

To identify the number of genes expressed during metamorphosis, we required a gene to have statistically analyzable expression data for at least one time point, in either of the sexes. See Additional file 4 for numbers of genes with expression during metamorphosis in males and females for both wild type (CS) pupae and tud progeny pupae.

Microarray data can be accessed at NIH GEO database with the following GEO accession number: GSE11316.

Clusters were generated using the tud progeny data and the program Cluster 20. Genes in the cluster must have had expression values for at least three of the five time points from both the tud progeny females and male datasets to be included in the cluster. The logarithms of the ratio values for each gene in each sex were median-centered and clustered. For clustering, we used Pearson correlation as the distance measure and defined similarity between clusters using average-linkage clustering. The cluster files outputted were imaged and analyzed using Java TreeView 62. Co-regulated clusters were defined as any group of genes for which the average correlation was at least 0.80 and contained at least 15 members. All correlations between the clusters are <0.80 and only 31 of 703 pair-wise comparisons between clusters have correlations >0.70, demonstrating a high degree of separation between the clusters. In Additional file 6, the average expression for the genes in each cluster from the wild type male and female experiments is included; the data from each wild type sex is also median-centered separately, but was not used to mathematically generate these clusters.

The graphical images in Figure 3 were generated by averaging the logarithm of the ratio data for all member genes of each cluster at each time point. The cluster intensities represent the combined average for male and female tud progeny. Peaks and troughs for gene expression in each cluster were determined as follows: for each cluster, the mean of gene expression at all time points was calculated. A time point was considered to be at a peak if the average gene expression at a particular time point was greater than 2/3 of a standard deviation above the mean. Conversely, a time point was considered to be in a trough if the average gene expression at a particular time point was 2/3 of a standard deviation below the mean. To determine if the mean expression values of the declared peaks and troughs were significantly different than expected at random, a re-sampling analysis was conducted as follows: for each cluster, let n be the number of genes in the cluster. For each time point, n expression values were randomly selected with replacement from the list of all expression values from all time points in the cluster and the mean expression value was determined for each time point. This was repeated with 10,000 permutations and significance was declared if a peak or trough had a mean expression value greater than or less than, respectively, 99% of randomly permuted averages.

The dendrograms shown in Figure 5 were produced using the logarithms of the ratio data for every annotated gene spotted on the microarray, not including control spots. Clusters were generated using the program Cluster 20, Pearson correlation as the distance measure, and average-linkage clustering to define similarity between clusters. The outputted cluster files were then imaged using Java TreeView 62 and the resulting dendrograms were exported. Confidence values for each node were generated using the statistical package Pvclust in R 63. Confidence values using the approximately unbiased P value were generated from multi-scale bootstrap resampling.

For all of the statistical analyses below, significance was declared if P < 0.05. Significant overrepresentations of functional annotations were generated with the program DAVID 22. Unique GenBank accession number identifiers for the gene list and the whole set of unique GenBank accession numbers for all possible transcripts in the full array set (13,614 genes total) were used. When searching for overrepresented functional annotations using DAVID, the following categories were selected: all levels (or only Level 4) of each of the three Gene Ontology (GO) categories, Uniprot Sequence Features, Swiss-Prot Keywords, KEGG metabolic pathways, InterPro domains, PIR superfamily names, and SMART domains 646566676869. The gene ontology enrichment function in FlyMine, an integrated database for Drosophila and Anopheles genomics 23, was used as a secondary method to confirm the DAVID results. All functional categories presented in Figure 3 and all functional categories discussed in the text were found to be significantly overrepresented (P < 0.05) by FlyMine. Furthermore, the results from DAVID were additionally confirmed with FlyMine by identifying the ten most significant GO terms as determined by DAVID for each of the eight clusters described in detail in the text (a total of 80 functional terms). Sixty-four of these 80 significant functional groups (84%) identified by DAVID were also found to be significantly overrepresented (P < 0.05) by FlyMine.

Additionally, our own program was used to determine overrepresentation of chromosomal location and conserved genes in the clusters. This code used a gene's Flybase number as the unique identifier and the background set consisted of the whole set of unique Flybase numbers for all possible transcripts in the full array set (13558 genes total). Code is written in Java and is available upon request. Significance was declared using the binomial approximation of the hypergeometric test on the list of unique identifiers, with all possible unique identifiers in the full array set as background. Genes whose coding regions overlap with the 2,500 most conserved DNA elements, as generated by PhastCons 70 was determined. These highly conserved sequences range from hundreds to thousands of base pairs in length. In the previous study 70, multiple alignments of the genomic sequences of four insect species were used to determine conservation.

To identify genes with significant differential expression between the sexes or across time, a two-tiered approach using F-statistics was implemented (see Additional file 1 for more details). First, using contrasts in LIMMA 16, genes were identified with significant differences in expression according to time, sex, or the sex-time interaction (q < 0.15). Genes with significant differential expression in sex or the sex-time interaction were considered to be sex-biased, identifying 258 genes with sex-differential expression in the somatic tissues during metamorphosis. The 258 genes identified as having sex-differential expression according to the above criteria were further analyzed to determine at which stage(s) they show sex-differential expression. These analyses were conducted using moderated t-tests in LIMMA (q < 0.15) comparing the mean expression values of tud progeny males and females at each of the five time points for which expression data was generated (0, 24, 48, 72, and 96 hour APF).

To identify genes that are expressed in the male and female germlines during metamorphosis, a similar approach was implemented (see Additional file 1 for more details). First, genes were identified with male- or female-biased expression in the wild type tissues using LIMMA contrasts with sex as the independent factor (q < 0.15). We expect genes with germline expression to be highly expressed, and as such we required genes with significant differential expression to also have at least a 1.2 fold difference in expression between the sexes. To identify genes expressed in the male germline, a LIMMA contrast analysis on the wild type male and male tud progeny data was performed, with genotype as the independent factor. Genes expressed in the male germline were those male-biased genes that additionally had significantly greater expression in the wild type males than tud progeny males (genotype as independent factor;q < 0.15). These genes were additionally required to show at least a 1.2 fold increase in expression in wild type males as compared to tud progeny males. In addition, to avoid false negatives, genes were included that are expressed in wild type males for at least four of the five time points examined and which have no expression data in the wild type female and male tud progeny experiments. The set of genes with expression in the female germline was defined in a similar manner, using the female wild type and tud progeny expression data.

To identify genes with sex-differential expression in the somatic tissues downstream of the sex hierarchy, genes with significant expression differences between male and females flies were first identified (the CS and Berlin experimental data was analyzed together; q < 0.15). The list was refined by requiring a gene to also be significantly, differentially expressed (q < 0.15; one tailed t-test) between male and female tud progeny (sex-differential expression was required to be in the same direction for the wild type and tud progeny experiments). The list was further refined to identify genes regulated downstream of tra, by identifying genes that were differentially expressed (q < 0.15; one-tailed t-test) between females and tra pseudomales. Here, it was required that genes with female-biased expression have higher expression in wild type females and genes with male-biased expression have higher expression in tra pseudomales.

Finally, genes expressed downstream of dsx in somatic tissues, were identified by refining the list of genes expressed downstream of tra, and requiring a gene to be significantly differentially expressed (q < 0.15; one-tailed t-test) between females and dsx^D pseudomales. Here it is assumed that genes with female-biased expression have higher expression in wild type females and genes with male-biased expression have higher expression in dsx^D pseudomales. To avoid false negatives in the identification of genes regulated downstream of dsx, which might have resulted from genes that failed to be significantly, differentially expressed in the tra pseudomale comparison, the DSX-regulated gene set was extended to include genes that were differentially expressed between wild type males and females (CS and Berlin; q < 0.15), tud males and females (q < 0.15), and female and dsx^D pseudomales at a more stringent statistical threshold (q < 0.05), thus excluding the requirement that the gene was differentially expressed in the tra experiments. To examine DSX modes of regulation, the genes regulated downstream of dsx were examined that displayed significant differential expression (q < 0.15) in either of the dsx null comparisons.

To identify the presence of DSX binding sites in the regulatory regions of Drosophila genes, we used the known DSX binding site (ACAAWGT) 46. We searched for the presence of two DSX binding sites in a region containing 2000 base pairs upstream from the transcription start and the first intron for each gene. In total, 13,449 genes were included in the search; the genomic sequence was determined from Release 4.0 of the Drosophila genome. Overrepresentation of genes with the DSX-binding was determined using a hypergeometric test compared against all genes in the Drosophila genome.