Background

Transcription by eukaryotic RNA polymerase II complex (RNAPII), composed of ten or more subunits [1, 2], involves multiple steps including initiation, promoter clearance, elongation, and termination. Each step requires different activities. The initiation process involves RNAPII recognition of core promoter elements at the transcription start site (TSS) [3]. Bound RNAPII opens a bubble in the duplex DNA and begins transcribing short RNAs by a scrunching mechanism [4]. While the first steps of initiation and promoter clearance are slow and highly regulated, elongation is rapid [5, 6]. This makes elongation an ideal mechanism for rapidly responding to changing conditions. A stalled or paused RNAPII has completed the slow steps such that de-pausing allows for rapid completion of transcription.

Great progress has been made on the biochemistry of elongation, although the exact composition of complexes is unclear due to biological context, biochemical complexity, and/or methodologies used in various studies. For the majority of genes, after transcription of about 20 to 50 nucleotides, RNAPII pauses when 5,6-dichloro-1-β-D-ribofuranosylbenzimidazole (DRB) sensitivity inducing factor (DSIF, a heterodimer of Suppressor of Ty 4 and 5 (Spt4 and Spt5)) and negative elongation factor (NELF) protein complexes bind to the RNAPII and transcript [7,8,9], to form a DSIF-NELF complex promoting the pausing of RNAPII [10,11,12]. Paused RNAPII is released by the positive transcription elongation factor b complex (P-TEFb), which phosphorylates the DSIF and NELF complexes as well as the carboxy-terminal domain of RNAPII (CTD) [13,14,15,16,17]. The catalytic and regulatory components of P-TEFb are Cyclin-dependent kinase 9 (Cdk9) and Cyclin T respectively [18,19,20]. In addition to P-TEFb, Cdk9 can also form a complex with Cyclin K [21]. Recruitment of P-TEFb kinase dissociates NELF, transforming DSIF into a positive elongation factor, triggering elongation [10].

P-TEFb is often part of larger complexes (Table 1). For example, the super elongation complex (SEC), is a key regulator required for transcriptional elongation checkpoint control and rapid transcriptional induction [22,23,24]. In addition to P-TEFb, SEC contains 11-19 Lysine-rich Leukemia (ELL), encoded by Suppressor of Triplolethal (Su(Tpl)), and ALL1-Fused Gene From Chromosome 4(AF4)/Fragile mental retardation 2 (FMR2) family member 4 (AFF4), encoded by lilliputian (lilli) [23, 25, 26]. Another transcriptional elongation complex is the Dot1 complex (DotCom) [27], which contains the histone H3 Lys 79 (H3K79) methyltransferase DOT1L, encoded by grappa (gpp) [28]. H3K79 methylation by DOT1 may regulate transcription of genes by changing chromatin structure and accessibility of transcription factors [29], and has been implicated in many basic functions such as cell cycle regulation, chromosomal stability, and the DNA damage response [30]. There is also evidence that lilli and Su(Tpl) are essential for the H3K79 methylation function of DotCom during transcriptional activation, and evidence of large complexes including components of DOT1, P-TEFb, and SEC, such as the Elongation Assisting Protein complexes (EAP) and complexes of Mixed Lineage Leukemia 1 proteins (MLL) [23, 27, 31, 32]. Additional complexity includes the Little Elongation Complex (LEC), which contains Su(Tpl) but lacks Lilli and Cdk9 [33], and AF4 family/ENL family/P-TEFb complex (AEP), which contains Lilli and Cdk9 but lacks Su(Tpl) [34]. These studies indicate that there are many possible elongation complexes [25, 31]. We have explored the roles of these complexes by examining the RNA expression profiles when some of these key components are altered in Drosophila. Specifically, knockdown of lilli or Su(Tpl) should disrupt SEC, EAP, and MLL. LEC should be altered by knockdown of Su(Tpl) but not lilli, and AEP should be altered by knockdown of lilli but not Su(Tpl). The dominant-negative form of Cdk9 should disrupt P-TEFb, SEC, and AEP, but not MLL.

Table 1 Components of different elongation related complexes or protein group
Full size table

The regulatory advantage of RNAPII pausing and elongation control is that rapid responses can be achieved if earlier rate-limiting steps occur, leaving only the more rapid elongation step for completion of transcription [5, 35,36,37]. Such a rapid transcriptional response is critical when organisms experience rapid environmental changes. For example, RNAPII pausing has been extensively studied in the context of the rapid transcription of Drosophila heat shock genes [38,39,40]. The rapid response is essential because failure to adjust to rapid temperature change results in catastrophic denaturation or mis-folding of proteins [41]. Increased levels of protein chaperones encoded by the heat shock loci ameliorate this stress. The presence of DSIF and NELF results in RNAPII pausing near the heat shock gene promoters [39], but following a heat shock, P-TEFb quickly redistributes to the heat shock genes, which allows RNAPII to elongate resulting in rapid production of transcripts [42, 43]. Other types of stress requiring a quick response in Drosophila, such as arbovirus infection, are also regulated at the level of elongation [44]. Regulation of transcription elongation also plays a role during development. An example is the pausing of RNAPII downstream of the sloppy-paired-1 promoter by NELF during the rapid blastoderm-stage of development [45]. It is likely that there are other genes responding rapidly due to normal physiology. Our expression profiles might help identify them.

Elongation control also plays a role in sexual development in Drosophila. Sex is determined by counting X chromosomes [46]. In wild-type flies, XX individuals are female and XY individuals are male. The genic imbalance created by X chromosome monosomy in males is compensated by increasing X chromosome gene expression [47]. At the regulatory level, the XX karyotype results in the activation of Sex-lethal (Sxl), which controls a major branch in the sex determination network [48]. Sxl regulates the splicing of transformer (tra) pre-mRNA, which ultimately is responsible for nearly all sexually dimorphic aspects of development, physiology, and behavior [49,50,51]. One of the most important genes downstream of tra is doublesex (dsx), which encodes sex-specific isoforms of a doublesex-mab3-related-transcription factor (DMRT). DMRTs are used to regulate sexual dimorphism in most metazoans [52]. The lilli and Su(Tpl) loci are likely to be direct targets of Dsx, and while the Cdk9 locus was not among the most obvious Dsx target genes, it has a strong and conserved Dsx binding site [51, 53]. Knocking down lilli or Su(Tpl) in dsx expressing cells results in dramatic sex- and context-specific phenotypes. For example, in males, the formation of specific first leg bristles known as sex combs, shows female pigmentation and narrower female-like morphology, but male distal/proximal orientation. In females, the first legs are normal, but bristles on the genitalia are missing. In males, the genitalia are normal. Thus, these genes function in female-specific development in one tissue and male-specific development in another [51]. If regulators of elongation play a post-developmental role in other aspects of sexual dimorphism, such as physiology, then we would expect to see sex-specific effects of disrupting the function of those complexes in adult flies.

Results

Knockdown and overexpression efficiency

We were interested in measuring the effect of lilli, Su(Tpl), and Cdk9 on gene expression in adults to probe for roles of elongation. Because these genes are required for organismal viability during development [54, 55], we bypassed the developmental defects by using ubiquitously expressed Gal4 [56] along with ubiquitously expressed temperature sensitive Gal80 repressor (Gal80 ts) [57] and Gal4 responsive constructs expressing RNAi against lilli, and Su(Tpl) [58] (Fig. 1a). We also used a dominant negative Cdk9 construct (Cdk9 DN) made by substituting an asparagine for an aspartic acid codon at position 199 (D199N) in the active site of the kinase. The same substitution has been used in other studies, where it has been shown to block P-TEFb CTD kinase activity [59, 60]. Cdk9 DN under UAS control was introduced into flies using P-element mediated transformation. Expressing Cdk9 DN during development resulted in lethality (DB and DP, unpublished data). Flies bearing the Cdk9 DN transgene were viable to adulthood at 20 °C in the presence of both Gal4 and Gal80 ts.

Fig. 1
figure 1

Experimental design and efficiency of RNAi or Cdk9 DN expression. a Experimental design. The cartoons (top) illustrate the sham genotypes where Gal80ts (red) and Gal4 (green) are produced in the absence of a responding UAS transgene (left), or when the UAS transgenes are present at the uninducing (middle) or inducing (right) temperatures. We used homozygous P{tubP-GAL4} LL7, P{tubP-GAL80 ts}7 virgin females for all crosses. Sham flies were produced by crossing to lacZ (Sham1, yellow fill) or w 1118 (Sham2, gray fill) males. The schemata (bottom) shows how the shams were used as references for all induction timepoints for both females and males in results presented in subsequent figures. See Methods for further information. Transgenes are shown as bars, with regulatory sequences (open) from tubulin (tub) or the Gal4 upstream activation sequence (UAS), and encoding sequences (filled) from the Gal80 ts (red), Gal4 (green), shRNAi, and Cdk9 DN (black) labeled. Gal80ts (red hexagons), Gal4 (green ovals) proteins are shown. Active (bent black arrows) and repressed (red X) transcription are indicated. b-d Histograms showing expression of lilli (b), Su(Tpl) (c), and Cdk9 (d) transcripts based on normalized read counts (linear scale from DESeq2 [62]) across the gene models after zero (0d), one (1d), or two (2d) days of induction. Biological replicate numbers (#) are indicated. Values for females (left) and males (right) are shown in each panel. Knockdown expression levels following induction of lilli RNAi, Su(Tpl) RNAi, and Cdk9 DN (black), and shams (yellow or gray) as well as one standard deviation (bars) are shown. Significant differential expression (p adj < 0.05 from DESeq2) relative to sex and timepoint matched shams are shown (asterisk)

Full size image

We were concerned that interfering with these important transcriptional regulators would result in cell lethality and an uninformative signature of dying cells even if we bypassed developmental defects. To assess the effect of these genes on adult viability, we shifted 3 to 4 day old adult flies from the permissive temperature of 20 °C to the restrictive temperature of 29 °C to induce transgene expression. We observed no overt lethality after up to a week. The lack of adult lethality allowed us to measure the effect of Cdk9, lilli, and Su(Tpl) on the adult steady-state transcriptome in the absence of developmental defects or reduced viability. We crossed Gal4 and Gal80 ts to the experimental lines and to two control lines, which acted as shams to monitor any differential expression due to Gal4 and Gal80 ts alone (Fig. 1a). We used these sham genotypes without UAS transgenes as controls in addition to the temperature shifts. We collected sexed samples zero, one, and two days post temperature shift to 29 °C. Flies were the same age post-eclosion (7 days) when harvested. Because germline gene expression is not controlled in the expression system we used [61], we removed the gonads from the flies before preparing polyA+ RNA for stranded RNA-seq expression profiling. All experiments were biologically triplicated to quadruplicated using three flies per sample. We used a total of 99 samples, after confirming that mismatched replicate numbers did not impact conclusions (by bootstrapping, see Methods). We obtained above background RNA-seq reads for more than 8000 genes in each sample. We used the DESeq2 [62] Normalized Read Count statistical model to measure differential expression (see Methods). For each experimental sample, we made ratiometric measurements comparing replicate experimental samples to replicate sham samples as a reference. See Methods and GEO [63] (GSE77492) for detailed protocols, data, and replication statistics. A summary of all gene expression data can be found in the Additional file 1.

We first assayed the effectiveness of the adult induction experiments (Fig. 1b to d). We detected expression of Gal4 and Gal80 ts in all samples. Prior to the temperature shift, we observed no significant differential expression in lilli relative to shams (Fig. 1b), but we did observe modest but significant (Wald test, p adj < 0.05 is used throughout) reductions in target gene expression in the Su(Tpl) RNAi flies (Fig. 1c). We also observed modest but significant overexpression of Cdk9 DN (Fig. 1d), and we confirmed that the increased expression was from the transgenic Cdk9 DN allele by measuring the fraction of the reads bearing the codon substitution (~84% at 20 °C). These data indicate that the Su(Tpl) RNAi and Cdk9 DN transgenes were not fully “off” at 20 °C. Following the temperature shift to 29 °C, we observed dramatic reduction of Su(Tpl) (Fig. 1c) and lilli (Fig. 1b) steady-state expression in the appropriate RNAi bearing flies and dramatic overexpression of Cdk9 DN (~96% from the DN allele; Fig. 1d) relative to shams. These data indicate that we altered the expression levels (and the encoded activity in the case of Cdk9) of the three genes under study.

Effects on gene expression

To explore the transcriptional responses to transgene induction, we compared replicated results from experimental samples relative to each of the two types of sham samples and asked for significantly differentially expressed genes in any of these time- and temperature-matched comparisons. We observed extensive and significant differences in the gene expression profiles relative to sham controls in these 36 pairwise comparisons, especially for reduced expression in the experimental samples (e.g., Fig. 2a, b). Overall we observed that 4792 genes showed significant differential expression in at least one pairwise comparison. This constituted 27% of all genes in the genome and 58% of the genes that we scored as expressed in all samples. To identify patterns that we could use to classify the responses of this large set of differentially expressed genes in a way that would allow us to infer which elongation complexes might be affected, we used k-means to cluster the expression ratios of experimental relative to sham samples. We removed clusters driven by background genotypes (see Methods). We observed that 579 differentially expressed genes were clearly attributed to lilli RNAi, Su(Tpl) RNAi, or Cdk9 DN (Fig. 2c) treatments. There were four clusters of genes that were dominated by reduced gene expression on treatment: cluster 1 was down-regulated by lilli RNAi, Su(Tpl) RNAi and Cdk9 DN, clusters 2 and 3 were down-regulated following lilli RNAi or Su(Tpl) RNAi (cluster 3 genes were often up-regulated in Cdk9 DN flies), and cluster 4 was down-regulated only following Cdk9 DN expression. There was strong concordance between the lilli RNAi and Su(Tpl) RNAi profiles in all clusters. That there were no clear clusters of genes differentially expressed due to only one of these RNAi treatments, suggests that wild-type Lilli and Su(Tpl) positively regulate (directly or indirectly) many of the same genes. This is consistent with the hypothesis that these proteins are often obligatory components of the same functional complex. Cluster 1 showed reduced expression following Cdk9 DN induction, suggesting that wild-type lilli, Su(Tpl), and Cdk9 positively regulate these genes (directly or indirectly), consistent with the participation of all three activities at a subset of genes. This is the correlation expected for disrupted SEC or EAP. Most gene expression altered by lilli and Su(Tpl) (cluster 2 and 3) were not regulated by Cdk9, which is consistent with the composition of MLL. The Cdk9 DN expression profile was distinct (cluster 4). Therefore, there is also Cdk9-specific differential expression profile, consistent with a role for P-TEFb alone.

Fig. 2
figure 2

Differential expression and clustering. a-b Vocano-plots showing examples of differential expression for lilli RNAi (a) and Su(Tpl) RNAi (b) one day after induction in female samples relative to Sham2 versus significance (truncated at > 75). Genes with significant differential expression (DESeq2 p adj < 0.05, red) and no significant differential expression (DESeq2 p adj ≥ 0.05, black) are shown. Gene expression are shown as log2ratios of medians (X tilde) of experimental (exp) replicates divided by sham replicates. c Differential expression heatmap of experimental samples relative to shams. Values indicated in the key (bottom) were truncated at +/- 2. Reduced (blue) and increased (red) expression in experimental samples relative to sham are shown. 579 genes (rows) with coherent differential expression relative to shams are shown. Sample comparisons (columns) relative to Sham1 (yellow fill) or Sham2 (gray fill) are indicated (bottom), as are female (f) and male (m) sex, timepoint in day(s) (d), and tested transgenes. The k-means cluster ID used in the text is shown (right) and highlighted within the heatmap (open black boxes)

Full size image

Expression of genes encoding enzymes

To help determine if positive regulators of transcriptional elongation affect specific tissues or organs, we used FlyAtlas data from adults [64] to determine if differentially expressed genes had enriched expression in particular tissues. We found that the overall expression profile was not strongly enriched for genes expressed in any adult tissues (leftmost column of Fig. 3a), consistent with our use of the entire fly other than the gonads. Strikingly, the genes showing differential expression following lilli RNAi, Su(Tpl) RNAi, or Cdk9 DN induction showed coherence in implied tissue-specificity (Fig. 3a). We found that all clusters with differentially expressed genes had enriched expression in the midgut. This is intriguing, as the midgut environment is subject to rapid changes as food passes through the digestive tract.

Fig. 3
figure 3

Characteristics of differentially expressed genes. a Heatmap showing enrichment for genes expressed in tissues (rows) of wild-type flies from FlyAtlas among genes in the Fig. 2 clusters or among all genes expressed in our study (columns). The proportion of genes in a column enriched in each tissue (see key) is shown in hierarchical order. Tissues are written out except: ta ganglion = thoracicoabdominal ganglion; CNS = central nervous system; acc gland = male accessory gland; V = virgin, and M = mated. The carcass is the thorax and abdomen without gut and reproductive tract. b Heatmap showing GO biological process terms (rows) enriched in any cluster (see key). GO terms are also presented in hierarchical clustering order. c-d Browser views showing expression of two sets of related genes that are genomic neighbors and are positively regulated by lilli and Su(Tpl) (female, one day after induction). c The lysosomal α-mannosidase (LMan) genes. d The Jonah (Jon) chymotrypsin genes. Chromosome arm and genome coordinates are shown (top). The longest annotated transcript of each gene is shown. In the gene models, untranslated (gray boxes) and coding (yellow boxes) regions, and introns (gray lines) are shown. Transcription direction is shown (bent arrows). Base-level read density per million reads (RPM) (see Methods; arbitrary scale) is shown for RNAi (light green) and sham (dark green) as are test genes or sham abbreviations (right). Instances of significant induction-dependent differential expression are indicated (asterisks). e Boxplot of primary transcript length (in base pairs) by cluster (blue) or among all expressed genes (green). Clusters with significantly shorter (Wilcoxon rank test, p < 0.05) primary transcript than all expressed genes are indicated (asterisks)

Full size image

To determine if there was a functional category of genes differentially expressed due to the transgenes, we performed Gene Ontology (GO) term analysis (Fig. 3b) using expressed genes as the reference, and selected all terms that were enriched in one of the clusters. The enriched terms were associated with proteolysis, various aspects of amino acid or carbohydrate metabolism, transport, and pheromones. The genes regulated by all three of the transgenes (clusters 1) were enriched for genes encoding typsins with serine-type endopeptidase function (e.g., Serine protease 6). The genes down-regulated following lilli RNAi or Su(Tpl) RNAi but not Cdk9 DN induction (clusters 2 and 3) were also enriched in genes encoding proteolysis functions in addition to metabolic, sensory, and transport functions. We show two examples of gene groups encoding important digestive enzymes that were down-regulated by transgene induction: the Lysosomal alpha-mannosidase (LMan) genes (Fig. 3c) and chymotrypsin encoding Jonah genes (Fig. 3d). We found that these genes required lilli and Su(Tpl) (but not Cdk9) for high level expression (clusters 2 and 3). They also both showed strongly midgut-biased expression in wild-type flies [64], and their wild-type functions involve carbohydrate hydrolysis in mannose production [65] and proteolysis (serine-type endopeptidase activity) [66] respectively.

We noticed that many of these enzyme-encoding genes with differential expression had short primary transcripts. Therefore, we determined if there was an overall tendency for preferential regulation of genes with short transcripts, by plotting the distributions of annotated primary transcript length of genes in each expression ratio cluster (Fig. 3e). Interestingly, the genes regulated by all three tested genes (cluster 1, see Fig. 2c) have significantly shorter primary transcripts than all expressed genes (1517 versus 2217 bp; Wilcoxon rank test, p = 4 × 10−6). Moreover, the genes regulated by lilli and Su(Tpl) (cluster 2 and 3, see Fig. 2c) also have significantly shorter primary transcripts than all expressed genes (1679 and 1027 versus 2217 bp; Wilcoxon rank test, p ≤ 8 × 10−6). In contrast, the genes regulated by Cdk9 had insignificantly shorter primary transcript length compared to all expressed genes (2099 versus 2217 bp; Wilcoxon rank test, p = 0.6). It is worth noting that the two clusters with the shortest primary transcripts (cluster 1 and 3) also had the strongest down regulation of expression after induction (Fig. 2c), as well as concordant enrichment of genes with proteolysis function (Fig. 3b). Given that control by elongation and short transcript length both contribute to rapid production of mature transcripts, these data suggest that many genes involved in metabolism have been optimized for rapid elongation.

Sex-biased regulation

We were intrigued by the role for positive elongation factors in the expression of a large number of metabolic enzyme encoding genes expressed in the midgut, as the Drosophila midgut shows sex-biased physiology and morphology, due to factors such as the immense metabolic requirements to support egg development and the requirements for specific saturation states in lipids used for pheromone production [67,68,69]. Additionally, lilli and Su(Tpl) are required for external sexual morphology [51]. This raised the possibility that there could be different roles for positive elongation factors in the sexes. To explore this hypothesis, we measured sex-biased responses to gene expression treatments relative to shams for all 4792 differentially expressed genes. We then followed a k-means clustering procedure to identify genes with sex-biased patterns (Fig. 2c). We found three clusters (514 genes in total) showing a coherent sex difference in expression (Fig. 4a). Genes in cluster I showed higher expression in males relative to females following Cdk9 DN treatment, suggesting a role for P-TEFb. Genes in cluster II showed reduced expression in males relative to females, while those in cluster III showed higher expression in males relative to females following lilli RNAi and Su(Tpl) RNAi treatments, but not Cdk9 DN. These results suggest a role for MLL in regulating sex-biased expression.

Fig. 4
figure 4

Sex-biased regulation after knockdown of elongation. a Heatmap of induction-dependent sex-biased effects. Clusters are labeled (Roman numerals). A cluster with eight genes is included in cluster II due to its small size and similar pattern as cluster II. See Fig. 2 for labeling. b Heatmap of enrichment for tissue specific expression in the clusters. See Fig. 3 for labeling, except m = male and f = female in the key. c Heatmap of GO biological process terms enriched in any of the clusters. See Fig. 3 for labeling. d-e Browser views of two examples of sex-biased gene regulation (one day after induction) are shown - CG31091 (a possible lipase encoding gene) and CG13091 (a possible fatty acyl-CoA reductase encoding gene). Note that these two CG names differ by placement of a single digit. See Fig. 3 for labeling, except that we distinguish female (red) and male (blue) expression

Full size image

To determine where these genes with sex-biased responses might be expressed, we again compared to the FlyAtlas dataset (Fig. 4b). In addition to the digestive system, genes expressed in carcass were enriched in all three clusters. The FlyAtlas definition of carcass is the thorax and abdomen after removal of the other tissues and organs, so this category includes epidermis, musculature, and oenocytes, which perform liver-like functions in Drosophila [70]. We also found modest enrichment of genes expressed in a variety of non-gonadal tissues in the other clusters. To help determine the function of the genes showing sex-biased differential expression, we performed GO term enrichment analysis and observed that the most significant enrichment was trans-membrane transport and fatty acid elongation in genes regulated by Cdk9 (cluster I, Fig. 4c). Genes showing sex-biased expression due to lilli RNAi and Su(Tpl) RNAi treatment were enriched in secondary metabolic processes (cluster III, Fig. 4c). Despite the lack of significant enrichment for GO terms in cluster II, we did observe sex-biased differential expression for at least a few genes encoding metabolic functions. For example, CG31091 encodes a protein with similarity to lipases and requires lilli and Su(Tpl) activity for its male-biased expression (cluster II, Fig. 4d). This possible lipase shows highly midgut-biased expression in wild-type flies [64]. As another example, CG13091 encodes a protein with similarity to fatty acyl-CoA reductases and low level of expression in females requires lilli and Su(Tpl) (cluster II, Fig. 4e). The putative reductase is highly expressed in wild-type fat body, heart, and carcass but not in other tissues [64]. Thus, like the overall pattern of differential expression, the sex-biased response to elongation factors was enriched in genes encoding metabolic functions.

Discussion

Complex components

Various positive elongation factor complexes have been described. Many of those complexes have overlapping components and complex composition is variable in the literature (Table 1) [25, 27, 31]. Therefore, genetic data can be useful for helping to determine the constituent components in biologically active complexes, or in epistatic relationships. For example, SEC contains both Lilli and Su(Tpl), while LEC and AEP contain either Su(Tpl) or Lilli (Table 1). In our expression profiles, lilli and Su(Tpl) positively regulate a nearly identical set of genes, which is consistent with them being obligatory co-factors in MLL, SEC, and EAP complexes. Our data does not support a role for LEC or AEP, although we cannot rule out the existence of these complexes. We also found that about 20% of the genes regulated by lilli and Su(Tpl) were also regulated by Cdk9. This is consistent with the presence of lilli, Su(Tpl), and Cdk9 in SEC and EAP [71]. However, we also found hundreds of genes positively regulated by Cdk9 only. The genes specifically regulated by Cdk9 might be dependent on P-TEFb independently of the larger order complexes.

Metabolic responses

Promoter-proximal pausing of RNAPII is an interesting regulatory mechanism because it allows for rapid responses. RNAPII binding to the TSS, unwinding the DNA duplex and initiating transcription requires much more time in vivo (10 min [72, 73]) than elongation (about 1.5 k bases (kb) per minute [74, 75]). Such rapid responses are especially important for heat shock, pathogen defense, and proper expression of some developmentally regulated genes [36, 39, 44, 45]. We found that positive elongation factors regulate many genes encoding metabolic enzymes, especially those expressed in the midgut. Release from pausing is an attractive mechanism for rapidly producing enzymes in response to feeding. Indeed, the Jonah genes encode chymotrypsins in the midgut and are expressed in feeding, but not in non-feeding stages of development, such as the first 12 h of adult life when flies are maturing and using residual fat body stores [76]. In mosquitos, the orthologous genes are rapidly induced in the gut by a blood meal [77, 78]. We suggest that transcriptional elongation is an important regulator of rapid metabolic gene expression responses in adult Drosophila.

Rapid transcription also depends on gene length. Interestingly, genes positively regulated by the lilli and Su(Tpl) genes in our studies tend to have short transcripts, especially the ones with the highest level of differential expression after perturbation. For instance, most of the transcripts of chemotrypsin genes, such as Jonah genes and lambdaTry, are shorter than 1 kb, which enables transcription in less than 40 s. In general, the genes positively regulated by the lilli and Su(Tpl) genes produce even shorter transcripts than the heat shock genes (1751 versus 2231 bp). Although it may not be a prerequisite, short transcript length could be a feature evolutionarily favored by genes with rapid responses. We did not observe significantly shorter transcripts among genes regulated by Cdk9, suggesting that regulation of short transcripts is MLL-dependent.

Sex-biased expression

The lilli and Su(Tpl) loci contain many conserved and highly occupied Dsx binding sites, and driving lilli RNAi or Su(Tpl) RNAi in dsx expressing cells results in specific switches in sexual dimorphism in the male first leg and the female genitalia, suggesting that these genes are required in both females and males, but in different locations [51]. A sex-biased role for Cdk9 has not been described, but the locus does have a conserved consensus Dsx binding site [51]. However, we do not know if this sex- and tissue-specific role is confined to morphology, or is more general. We found that sex-biased expression changed for about 200 genes due to lilli RNAi or Su(Tpl) RNAi, and for about 300 genes due to Cdk9 DN . As was the case for morphology, the lilli and Su(Tpl) genes are required for both genes with female- and male-biased expression. This is consistent with the idea that sex-biased expression is highly context dependent and argues against strictly female- and male-biased batteries of genes that shows the same bias in all tissues. The Cdk9 regulated genes showed a more directional effect of reduced expression in females relative to males.

Many of the genes regulated by elongation factors encode metabolic functions. There is a great deal of precedent for sex-biased metabolism. For example, egg production is energy intensive and tightly linked to nutrient availability via insulin signaling [79]. At the morphological level, the Drosophila midgut is nearly twice as long in females as males and is regulated by the sexual identity pathway [67]. Finally, lipids and other metabolites also show dramatic sexual dimorphism [80]. Our work suggests that some of the sex-biased gene expression in metabolic pathways is due to indirect activity of Dsx target genes such as lilli and Su(Tpl).

Conclusions

Turning down positive elongation factors has a global transcriptional effect. Our data suggest that Lilli and Su(Tpl) are likely to be obligatory cofactors for the expression of many genes in soma, and that Cdk9 has both shared and Lilli/Su(Tpl)-independent functions. Our data also suggest that transcriptional elongation regulation by these factors is essential for rapid enzymatic and metabolic activities during digestion, a novel category of function regulated by rapid transcriptional elongation. In addition, the differential expression in females versus males suggests that positive elongation factors play roles in sex-biased gene regulation.

Methods

Summarized data is found in Additional file 1.  A table of key resources is provided in Additional file 2, which describes the reagent or resources used; including chemical and biological reagents, critical commercial kits, genotypes of the flies, deposited data, and software. We also provided the sources and identifiers for these resources.

Molecular genetics

Cdk9 DN transgenic lines were generated by PCR-amplifying the Cdk9 open reading frame from w 1118 genomic DNA and cloned into pBluescript KS. Mutant Cdk9 DN, a substitution of amino acid 199 of the active site (aspartic acid to asparagine) was obtained by the megaprimer method [81] using a mutagenic oligonucleotide. The mutant DNA was cloned in the UASp vector [82] and injected in w 1118 embryos. A transgenic line with an unmapped insertion on the 3rd chromosome was used in this study, but other lines show the same effect on development in the presence of various Gal4 drivers.

Flies

All flies were reared on the following media (per liter: 6.2 g agar, 59 g cornmeal, 59 g yeast, 5 ml propionic acid and 26 ml 99% methyl 4-hydroxybenzoate). Briefly, homozygous w*; +; P{tubP-GAL4} LL7, P{tubP-GAL80 ts}7 virgin females were crossed to homozygous males with a UAS responsive transgene or homozygous males of control lines for shams. The paternal lines were lilli RNAi, Su(Tpl) RNAi, Cdk9 DN, w 1118, and lacZ. There are additional experiments in the GEO [63] series accession GSE77492 not presented here. First, we performed profiling following gpp RNAi but did not shown them in the results due to poor knockdown. Second, we used flies with an additional copy of Cdk9 DN, but observed a non-additive effect of two copies of Cdk9 DN. Third, we ectopically expressed stand still, but these samples were for other purposes. All the parental lines were reared and crossed at 20 °C, which also applied to their progeny. We collected adult progeny at 0 to 4 h post-eclosion, aged them for 5 to 7 days in fresh vials, and shifted to the restrictive temperature (29 °C) for 0, 1 or 2 day(s). All the flies were 7 days old at the time of dissection.

RNA-seq

For each sample, three gonadectomized adults were dissected and crushed in 100 ul RNAlater solution (Life Technologies, Carlsbad CA) before sealing in deep 96-well plates at room temperature. Samples were homogenized with a mini-beadbeater (Biospec Products, Bartlesville OK) three times for 60 s each in the presence of 20 to 30 1.0 mm glass beads per sample (BioSpec Products, Bartlesville OK). We extracted RNA using RNeasy according to the manufacturer (Qiagen, Valencia CA), except that we added 600 ul RLT buffer and 700 ul 70% ethanol in each well. We used 400 ng total RNA in 50 ul nuclease-free water (Quality Biological, Gaithersburg MD) for RNA-seq library preparation. We strictly followed a previously reported polyA+ strand-specific RNA-seq library preparation protocol [83]. We conducted single-end 50 bp RNA-seq on the HiSeq2000 Sequencing System (Illumina, San Diego CA). Reads were mapped to FlyBase [84] release 6.04 using STAR (v2.4.2a) with default parameters [85]. The overall agreement between biological replicates was ≥ 0.88 (Pearson’s r). We also mapped the reads of all samples to the intergenic regions, which we used as a measure of background expression noise for calculating the number of expressed genes. Genes expressed at < 95% percentile of intergenic expression were counted as unexpressed [86]. All samples with 29 °C treatments were triplicated, and all samples without 29 °C treatments were quadruplicated except for female Sham2 samples (triplicated). To confirm that our results were robust to mismatched sample sizes, we used bootstrapping (100x) to randomly select three replicates among each sample set. The median number of re-identified differentially expressed genes was 96%, suggesting that the variable samples sizes have little impact on our results. Therefore, we used all replicates in this work.

Expression analysis

For each sample, uniquely mapped read counts from HTSeq (v0.6.1p1) (--stranded = reverse) [87] were used as inputs in DESeq2 (v1.10.1). We used Normalized Read Counts as the unit of expression except for genome browser coverage plots, where we used base-level coverage of reads (generated by bedtools v2.25.0 [88]) normalized by RNA-seq library size in millions (RPM). For Normalized Read Counts, DESeq2 calculates the geometric mean (GM) of each sample, and divides the raw read counts by the GM to obtain ratios of relative expression. For each sample, we used the median ratio as a sample-specific size factor (~1 in our experiments), and normalized read counts were calculated as raw read counts divided by the size factor. Previous studies based 726 Drosophila individuals [86] demonstrated that the DESeq2 method was a high performing normalization approach relative to total counts, upper quartile, median, trimmed mean of M-values, quantile, reads per kilobase per million mapped reads, and un-normalized count data. We also used DESeq2 to calculate differential expression. For each experimental sample within the same sex and treatment, we compared experimental relative to each of the two sham genotypes. We retained genes with significantly differential expression (Wald test p adj < 0.05) in any of the comparisons. We further categorized these genes by their induction-dependent differential expression (log2 ratio of the experiment expression relative to sham) across samples. We converted “N/A” values (divide by 0) to “0” (indicating no bias) for k-means clustering. We used the parameter k = 18 (Fig. 2c) or 20 (Fig. 4a) based on evaluation of within-cluster sum of squares [89], and also confirmed that the major clusters were robust from k = 15 to 25. We used visual examination to remove clusters driven by differences between the two sham genotypes and clusters that showed weak differential expression, or responded even without induction.

We used FlyAtlas [64] microarray expression data to infer where differentially expressed genes were expressed in our experiments. We converted microarray Oligo IDs to FlyBase [84] gene IDs using DAVID (v6.7) [90]. Some genes have multiple microarray probes, and we calculated the mean signal using 1 = “enriched”, 0 = “not different”, and -1 = “depleted” as compared to the signal in the whole body. We defined enriched expression as mean > 0. Gene Ontology (GO) [91] tests were conducted in FlyMine [92]. We used a Holm-Bonferroni cutoff of p adj < 0.05 for GO analysis and all tests were based on the background of expressed genes in any of our samples rather than all genes. We used the pheatmap package (v1.0.8) in R (v3.2.2) to generate all heat maps of tissue specificity and GO results.