The BET protein FSH functionally interacts with ASH1 to orchestrate global gene activity in Drosophila

Background The question of how cells re-establish gene expression states after cell division is still poorly understood. Genetic and molecular analyses have indicated that Trithorax group (TrxG) proteins are critical for the long-term maintenance of active gene expression states in many organisms. A generally accepted model suggests that TrxG proteins contribute to maintenance of transcription by protecting genes from inappropriate Polycomb group (PcG)-mediated silencing, instead of directly promoting transcription. Results and discussion Here we report a physical and functional interaction in Drosophila between two members of the TrxG, the histone methyltransferase ASH1 and the bromodomain and extraterminal family protein FSH. We investigated this interface at the genome level, uncovering a widespread co-localization of both proteins at promoters and PcG-bound intergenic elements. Our integrative analysis of chromatin maps and gene expression profiles revealed that the observed ASH1-FSH binding pattern at promoters is a hallmark of active genes. Inhibition of FSH-binding to chromatin resulted in global down-regulation of transcription. In addition, we found that genes displaying marks of robust PcG-mediated repression also have ASH1 and FSH bound to their promoters. Conclusions Our data strongly favor a global coactivator function of ASH1 and FSH during transcription, as opposed to the notion that TrxG proteins impede inappropriate PcG-mediated silencing, but are dispensable elsewhere. Instead, our results suggest that PcG repression needs to overcome the transcription-promoting function of ASH1 and FSH in order to silence genes.


Background
Gene expression programs specify diverse cellular identities during metazoan development, ultimately allowing cells to form tissues or organs. The ability to maintain expression states is critical, since inappropriate loss or gain of gene activity may lead to developmental anomalies, tissue dysfunction, or uncontrolled cell growth such as cancer. The Trithorax group (TrxG) proteins and the Polycomb group (PcG) proteins, originally identified in Drosophila melanogaster but present in all higher eukaryotes, cooperate to sustain gene expression states by establishing and organizing information contained in the chromatin template. This epigenetic information layer assures the appropriate usage of the genetic blueprint, according to the developmental history of any given cell. TrxG proteins are required in order to maintain active states. This was discovered because of their essential trans-activating function for homeotic gene expression in flies [1]. Despite their common implication in gene activation, the TrxG encompasses a variety of different biochemical functions, ranging from chromatin remodeling and histone modification to mediator complex subunits and transcription factors [2]. Genetically, the TrxG acts antagonistic to the PcG of genes, which is essential for the maintenance of repressed expression states.
The histone methyltransferases (HMTs) absent, small, or homeotic discs 1 (ash1) and trithorax (trx) represent histone modifying activities within the TrxG. Both HMTs have attracted particular attention, because they were considered to specifically counteract PcG silencing, rather than being coactivators of transcription [3]. Studies of ASH1 enzymatic activity supporting this view show that the histone modifications catalyzed by ASH1 disfavor Polycomb Repressive Complex 1 (PRC1) binding in-vitro, whereas association of SWI/SNF chromatin remodeling complexes incorporating TrxG subunits (BRAHMA, MOIRA) is preferred [4]. However, the in-vivo relevance of this observation remains unclear, since the targeting of PRC1 seems to primarily depend on transcription factors and non-coding RNAs (ncRNAs) [5]. In addition, the assumed enzymatic specificity of ASH1 has lately been challenged by independent reports [6][7][8].
Alternative explanations on how ASH1 prevents PcGmediated silencing propose that ASH1 blocks PRC-activity downstream to chromatin recruitment. A comparison of the HOX gene Ultrabithorax (Ubx) in active and repressed conditions supports this notion: PRC1 and PRC2 binding patterns at the Ubx locus do not change between Ubx ON or OFF states, whereas ASH1 is only located downstream of the active Ubx promoter in embryonic discs [9]. In addition, Ubx repression in the absence of ASH1 is accompanied by the formation of an ectopic H3K27me3 domain, a mark associated with mature PcG-mediated repression. Taken together, these data suggest that ASH1 actively prevents PcG silencing, however not by simply impairing PRC1/2 recruitment to chromatin.
The functional relationship between PcG proteins and ASH1 has been addressed on a genome-wide scale by two recent ChIP-chip studies in Drosophila tissue culture cells. Both investigations support the proposed anti-repressor function of ASH1, either by showing that fully activated PcG target genes miss typical marks of PcG-mediated repression and instead become embedded in broad ASH1 domains [10], or by reporting that genes in 'balanced' states, characterized by simultaneous binding of PcG and TrxG proteins, are transcriptionally active [10,11]. In both studies a clear indication for a more general involvement of ASH1 in transcription is missing. In accordance to this, analysis of the wing imaginal disc transcriptome of ash1 mutant flies only identified few deregulated genes, again supporting the idea that ASH1 is only needed at a subset of developmental regulators, which need to be protected from ectopic PcG repression [12].
Curiously, studies analyzing Ubx pattern formation in embryonic disc stages suggest ASH1 to act as a general transcription-promoting factor in analogy to what has been shown for TRX. Briefly, ncRNAs, originating from the bxd region upstream of the Ubx promoter, inhibit Ubx expression in cis presumably by transcriptional interference [13]. ASH1 localization downstream of the bxd ncRNA promoters correlates with the onset of ncRNA expression [14]. This behavior is reminiscent of TRX, which was already shown to promote bxd ncRNA expression, thereby silencing Ubx indirectly [15]. In the reverse Ubx ON situation, ASH1 was found to be enriched in the coding region of Ubx and a related GFP reporter gene, but absent from bxd ncRNAs. Hence, the correlation between ASH1 binding and transcriptional activity at functionally diverse genes (ncRNAs, homeotic selector, and reporter) is clearly indicative of a more general involvement in transcription.
Another TrxG member, explicitly implicated in Ubx regulation, is female sterile (1) homeotic (fs(1)h) [16,17]. Alleles of ash1 and fs(1)h show strong genetic interactions with respect to Ubx activity. These interactions have contributed to the proposal of a 'trithorax gene set' by Shearn, following the earlier idea of a repressive gene cohort around Polycomb [18]. However, attempts to link ASH1 and FSH on a biochemical level have not been successful to date [17]. The fs(1)h gene products belong to the BET family of proteins, named after their characteristic arrangement of a tandem bromodomain and the extra-terminal domain. Mammalian BET proteins have been shown to serve as chromatin adapter proteins by binding to acetylated histone tails and to facilitate gene expression [19]. Lately, inhibition of BET function has been identified as effective strategy to treat poor-prognosis leukemia, multiple myeloma, and squamous carcinomas [20][21][22][23]. How BET proteins contribute to normal development and tissue homeostasis is still poorly understood. Loss of function phenotypes in model organisms having a single BET gene, such as C. elegans BET-1 mutants and Drosophila fs(1)h mutants, demonstrate that BET function is crucial for the establishment and maintenance of cell fates [16,24,25].
By purifying ASH1 from a stable Drosophila cell line, we discovered a biochemical interaction between ASH1 and FSH, as previously suggested from genetic studies in fly mutants. In addition, we were able to show that both proteins extensively co-localize on chromatin, especially at sites bound by PcG complexes and gene promoters. The observed co-localization at promoters is a hallmark of active genes, but independent of gene type and function. These findings are at odds with the prevalent model that TrxG proteins function primarily as PcG-specific antirepressors, but are dispensable in the absence of PcGmediated repression. Our additional finding, that a set of PcG-repressed genes displays ASH1, FSH, and TRX-C signals comparable to their active counterparts, further questions the anti-repressor hypothesis. In agreement with a general transcription promoting function, we could show that BET inhibition in Drosophila tissue culture cells leads to an immediate and widespread gene repression, as well as delayed gene activation in reaction to environmental stimuli.

ASH1 interacts biochemically with FSH
In order to screen for ASH1-interacting proteins, we purified tandem-tagged ASH1 from a stable Drosophila cell line and identified copurified proteins by tandem mass spectrometry (MS/MS). Since no full-length cDNA was available to generate bait expression constructs, we cloned the ash1 coding sequence from S2 cells by assembling four subcloned RT-PCR products (A-D) tilling the ASH1 open reading frame (ORF) from start to stop codon ( Figure 1A). We verified the resulting cDNA by transfecting cells with ASH1-GFP fusion constructs. Immunoblotting of cell lysates gave clearly detectable signals at the expected molecular weight (MW) using antibodies (kind gift from F. Sauer, UC Riverside) recognizing epitopes in the N-and C-terminal part of ASH1 ( Figure 1B). Sanger sequencing showed that our cDNA contains a micro deletion of the amino acids (aa) T1716-L1717 with respect to [Swiss-Prot:Q9VW15] ( Figure 1C). In spite of the two missing amino acids it still encodes a fully functional protein, as demonstrated by rescuing the development of ash1 null mutant flies (Steffen PA, Fonseca JP, Gänger C, Dworschak E, Kockmann T, Beisel C, Ringrose L: Quantitative in vivo analysis of chromatin binding of Polycomb and Trithorax group proteins reveals retention of ASH1 on mitotic chromatin. Nucleic Acids Research 2013, 41:5235-5010.1093/nar/gkt217Available: http://eutils.ncbi. nlm.nih.gov/entrez/eutils/elink.fcgi?dbfrom=pubmed&id = 23580551&retmode=ref&cmd=prlinks.). An ASH1 ChIPchip profile has already been published by modENCODE [11]. As a means of comparison we tested the immunoreactivity of the antibody (Q4177) used in their study towards recognizing ASH1, which according to modEN-CODE targets aa 1747-1846. To our surprise, we could not detect signals corresponding to ASH1 by probing whole cell und nuclear extracts ( Figure 1D). Conversely, control immunoblots of the same material faithfully detected ASH1 using the ASH1-C antibody.
Based on our novel ASH1 cDNA we created a doubleaffinity tagged expression construct under the control of the inducible metallothionein promoter (Figure 2A). Employing this vector construct we established a polyclonal S2-DRSC cell line, exhibiting inducible expression of 3xFLAG-8xHIS-tagged ASH1 (FH-ASH1). Following overnight induction of FH-ASH1, a three-step purification as outlined in Figure 2B was performed. In the first step, we captured tagged ASH1 from nuclear extract by immobilized metal affinity chromatography (IMAC).
Captured FH-ASH1 was afterwards subjected to ion exchange chromatography (IEX) and finally to FLAGaffinity chromatography (FLAG-AC) for polishing. In order to test if our purification scheme enriched proteins in a bait dependent manner, we simultaneously size-separated samples originating from induced and non-induced cells. Silver staining visualized many protein bands exclusively present in the experimental sample, indicating a bait protein dependent enrichment ( Figure 2C).
MS/MS analysis revealed that FH-ASH1 was strongly enriched in the experimental samples, but absent from non-induced control samples ( Figure 2D and Additional file 1). The second most enriched protein was identified to be female sterile (1) homeotic [Swiss-Prot:P13709], encoded by the bona fide TrxG gene fs(1)h. Alternative splicing of the primary fs(1)h transcript results in two polypeptides: a short isoform denoted FSH-S, and a long isoform denoted FSH-L. The second is a C-terminal extension of its shorter relative. In order to approximately map the ASH1-FSH interaction, we co-expressed the ASH1 fragments A-D in combination with FSH-S in S2-DRSC cells. Immunoprecipitation of ASH1 A proved to be sufficient to pull-down FSH-S ( Figure 2F), suggesting that the ASH1-interaction motive resides in the common N-terminal portion of the two splice products. These results agree with the FSH peptide coverage from our MS/MS analysis, since we also found peptides belonging to the long isoform.
The unique domain arrangement identifies FSH as members of the bromodomain and extraterminal (BET) family. BET proteins are conserved from humans to lower eukaryotes and have been shown to regulate gene expression by means of chromatin ( Figure 3A). Their N-terminal region contains a tandem bromodomain which attaches BET proteins to acetylated histones. The tandem bromodomain is followed by the extraterminal domain of unknown function, which is likewise present in all BET proteins. The human BET family is comprised of BRD2, 3, 4, and BRDt. Drosophila FSH-L is closest to BRD4 and BRDt, since all three proteins possess the C-terminal tail extension. This tail is characterized by an unusual amino acid composition, for instance including poly-glutamine runs, and a C-terminal motive (CTM) at the very end of the protein. FSH-S matches BRD2, 3, and BRD4-S with respect to the missing tail region.
Large-scale analysis has shown that human bromodomains cluster into eight families based on structure/ sequence similarity [26]. Family II is populated by the BET-type bromodomains of BRD2, 3, 4, and BRDt. The recent development of the highly potent, small molecule inhibitor JQ1 allows selective inhibition of BET bromodomains [20,27]. JQ1 specificity is explained by the excellent shape complementarity with the acetyl-lysine binding cavity of BET-type bromodomains. We therefore asked, if a similar selectivity can be expected in the context of the Drosophila model system. Analogous to the situation in humans, the fruit fly proteome contains several bromodomain proteins, which have been implicated in transcriptional control. We approached this question by a phylogenetic analysis including human bromodomains of known drug ability and their Drosophila homologs. We found that FSH bromodomains are highly conserved with respect to BRD proteins (approximately 80% identity, >95% similarity), but well separated from bromodomains of other families ( Figure 3B). The superimposed human JQ1 selectivity strongly argues in favor of a discriminating drug ability of FSH. Structural alignments predict that FSH bromodomains fold into the canonical left-handed bundle of four α helices (α Z -α A -α B -α C ) creating a hydrophobic acetyl lysine binding pocket ( Figure 3C). All residues creating side-chain contacts to JQ1 have been found conserved.
In order to assess whether JQ1 binding is competitive with acetyl-lysine recognition, we performed histone peptide pull-down assays. Binding of tetra-acetylated histone H4 tail peptide to a purified BD1-BD2 fragment was reduced to background levels in the presence of 10 µM JQ1 ( Figure 3D). We made similar observations using a dual acetylated H3 tail peptide. Binding of none-purified FSH-S, obtained by transfecting HEK293 cells, was likewise suppressed by JQ1 ( Figure 3D). We therefore  (D) Comparision of target and cross reactivity between the ASH1-specific antibodies ASH1-C and Q4177. Immunoblots of S2-DRSC derived whole cell (Cell) and nuclear lysate (Nuc.) were prepared using the antibody dilutions given on top.
conclude that JQ1 can be used to selectively target BET function in Drosophila.

ASH1 and FSH co-localize on chromatin
Binding sites of ASH1 to chromosomes have been mapped on a genome-scale in different Drosophila cell lines using ChIP-chip methodology, while maps of FSH chromatin occupancy are not available to date. In order to clarify whether ASH1 and FSH also interact on chromatin, we performed ChIP-seq experiments in S2-DRSC cells. In addition, we contrasted their binding profiles with genome-scale maps for polycomb (PC), polyhomeotic (PH),     Notecolors of human proteins indicate averaged temperature shifts upon binding of 10 µM JQ1 measured by differential scanning fluorimerty [20]. (C) Protein sequence alignment of human and Drosophila BET bromodomains. Boxed residues reside in α helices as determined by structural analysis [26].
(E) Similar histone peptide pull-down assay using nuclear extract from HEK293 cells transfected with V5-tagged FSH-S.
posterior sex combs (PSC), and the C-terminal TaspaseI cleavage product of Trithorax (TRX-C), generated previously by us [28]. We enriched for ASH1-bound chromatin by using the antibodies ASH1-N and ASH1-C. FSH-bound chromatin was precipitated by antibody preparations from immune sera kindly provided by Igor B.
Dawid (NICHD). Antibody ID166, raised against an immunogen in the common N-terminal part of FSH, is reactive against both FSH isoforms ( Figure 4A, B). ID173 can be used to exclusively detect FSH-L. We verified antibody specificity by performing fs(1)h knockdown experiments in S2-DRSC cells ( Figure 4C). Since ASH1 and FSH are known trans-activators of homeotic gene expression, with central importance for Ubx activation and maintenance [16,29], we first focused on the bithorax complex (BX-C). We found that ASH1 and FSH co-localize at several discrete sites throughout the BX-C ( Figure 5A). The great majority of bound sites coincide with known boundary elements, which divide the gene cluster into regulatory domains [30]. But also known cis-regulatory elements like the Ubx enhancers bx and bxd show strong ASH1, FSH co-enrichments. Despite their presence throughout the BX-C, all genes of the homeotic cluster are silent in S2-DRSC cells. Considering their trans activating function, this finding is contrary to expectation.
Our comparison with the additionally mapped PcG/ TrxG factors revealed that ASH1, FSH co-localization throughout the BX-C is highly coordinated with PRC1 and TRX-C binding to chromatin ( Figure 5A). We defined PRC1-bound chromatin as genomic intervals co-enriched for the PRC1 core subunits PC, PH, and PSC (see Material and Methods for details). Again, this finding is rather surprising since TrxG and PcG proteins implement antagonistic gene regulatory functions. We decided to further Mcp Fab-6 Fab-7 Fab-8 bx investigate these phenomena on a genome scale and found that the observed co-localization is not restricted to the homeotic clusters. In total, we identified about 10,700 regions enriched for ASH1 and about 8,800 occupied by FSH. By using the combined chromatin maps for PC, PH, and PSC we identified 800 sites bound by PRC1 with high confidence. More than 80% of these sites are shared with ASH1 and the same is true for FSH (Table 1). In agreement with our ChIP-seq data, immunostaining of polytene chromosomes visualized hundreds of FSH-bound regions ( Figure 5B).
We previously showed that gene promoters are major recruitment sites for TrxG/PcG proteins [28]. We therefore examined the spatial relationship between transcription start sites (TSS) and ASH1, FSH bound genomic regions, by measuring the distance between ASH1, FSH peaks and the closest annotated TSS (see Materials and methods for details). As shown in Figure 5C, a majority of peaks are closer to TSS then expected by chance, in fact >50% are within 1 kb of distance (random peak distribution was modeled by shuffling peaks along chromosomes, P value <2.2 × 10 -16 ). In order to further fine-map ASH1, FSH binding relative to TSSs, we calculated binding profiles across non-overlapping 1 kb windows centered at TSSs. For both proteins we obtained unimodal coverage distributions with a maximum very close to the aligned TSS positions ( Figure 5D). These localization profiles demonstrate that ASH1 and FSH prefer binding nearby TSSs.
In summary, ASH1 and FSH binding to chromatin is strongly correlated and seems to happen at many sites in dividing cells, as well as post mitotic tissue. Our high-resolution analysis suggests that these sites coincide with the starting point of transcription. But also intergenic elements bound by the PRC1 complex attract ASH1 and FSH.

ASH1 and FSH co-localization is a hallmark of active genes
The fact that ASH1 and FSH both bind to promoter regions, prompted us to examine the relationship between gene activity and ASH1, FSH binding in more detail. We did this by inspecting genomic regions containing a mix of silent and transcribed genes. A representative example is shown in Figure 6A. In agreement with our previous results, ASH1 and FSH ChIP-seq signals peak closely to TSSs. A striking difference in ASH1, FSH localization was observed when we compared active and silent genes: All active genes -determined by detectable mRNA expression (RNA-seq) and high H3K4me3 levels at the promoter (ChIP-seq) -displayed prominent ASH1 and FSH peaks close to their TSS. Silent genes, by contrast, lacked these characteristic signals.
We further examined the observed correlation between transcription and ASH1, FSH localization to promoters, by comparing metagene profiles of genes grouped according to their expression level. We refer to a metagene profile as the mean read density over gene bodies scaled to a common length. As reference point for the analysis, we included our H3K4me3 ChIP-seq data, because this chromatin modification is known to be highly correlated with gene expression. In analogy to H3K4me3, the total amount of ASH1, FSH ChIP-seq signal correlated with the magnitude of expression ( Figure 6B). The maximum ASH1, FSH signal intensity collapsed with the TSS position when inspecting profiles for active gene bins. To rule out the possibility that our analysis was confounded by a sequencing bias at highly expressed loci, we inspected metagene profiles for our input sample using identical binning, but could not detect indications for such systematic errors (data not shown).
Coverage profiles suggested that ASH1 and FSH normally bind close to the TSS of active genes, but are absent from promoter distal parts of the gene body. Deviating from this canonical observation, we found signals within gene bodies at a subset of active genes: predominantly, these genes fall into the category of intron-rich, active genes. Similar observations have been described earlier [11]. However, the same study did not pick-up the correlation between ASH1 binding and transcription at intron poor genes. Representative examples illustrating the difference between intron-rich and intro-poor active genes are shown in Figure 6C. At the active, intron-rich Dad gene, ASH1 and FSH signals seem to spread along the gene body. In contrast, the active intron-poor gene Cyp12c1 does not show signals distal to the TSS. Here, the ASH1, FSH signal is tightly restricted to the promoter proximal region.
Combined TrxG/PcG binding patterns predict the regulatory state of genes Genetic experiment using double-mutants for TrxG and PcG genes, suggested that TrxG proteins such as ASH1 are primarily needed to prevent inappropriate PcGmediated silencing, instead of having a general function  during transcriptional activation [3]. Since our data clearly favor the second possibility, we decided to examine the combinatorial patterns of TrxG/PcG proteins found at chromatin. The fact that all our ChIP-seq experiments have been carried out using identical experimental procedures/materials ensures high comparability across chromatin maps.
In brief, we identified combinatorial binding patterns by fitting a three component sparse Gaussian mixture model (sGMM) to the ChIP-seq enrichments at promoters (see Materials and methods for details). This approach is related to k-means clustering. Our choice to solely focus on the promoter region was based on our previous finding that all of the TrxG/PcG factors mapped by our laboratory preferred binding close to the TSS. In addition, this special constrain allows for an intuitive correlation between ChIP signals and gene expression.
The fitted sGMM partitioned genes as follows: class 1 genes do not show enrichments for any of the considered TrxG/PcG proteins at their promoters ( Figure 7A,   ). Judged by mRNA production the great majority of class 1 genes is not expressed and accordingly lacks the 'active' histone modification H3K4me3 ( Figure 7C). In total, approximately 55% of all analyzed genes fall into this class. Therefore, we simply termed this class 'Inactive'. Class 2 genes display strong signals for ASH1 and FSH at their promoters together with intermediate TRX-C enrichments ( Figure 7A, B). These genes are clearly transcribed, matching our previous results that ASH1 and FSH presence at promoters correlates with gene expression ( Figure 7C). We named this group, containing approximately 43% of all analyzed genes, the 'Active' class. A representative gene belonging to the active cluster is shown in Figure 8A. In agreement with its cluster assignment Spf45 displays strong ASH1 and FSH ChIP-seq signals at the transcription start site. The RNA-seq and H3K4me3 ChIP-seq profiles reflect high expression as expected for a splicing factor. Class 3 genes are characterized by extraordinary high signals in the PC, PH, PSC, and TRX-C dimensions, but their promoters are also occupied by ASH1 and FSH, notably to the same extend as class 2 genes ( Figure 7A, B). Approximately 1.5% of all analyzed genes belong to this class. Inspection of their activity showed undetectable expression with very few exceptions ( Figure 7C). In combination these findings suggested that class 3 contains genes being actively repressed by the PcG. We therefore termed class 3 'PcG-repressed'. A representative example is given in Figure 8B. The H15 TSSs is bound by ASH1, FSH, PRC1, and TRX-C. According to RNA-seq and H3K4me3 signals it is completely repressed. This observation makes sense, since H15 encodes for a transcription factor involved in cardioblast cell fate commitment.
To answer the question how accurately our classification predicts gene expression in a simplified binary manner (ON or OFF), we labeled genes according to their mRNA score as silent or transcribed (log 2 FPKM cut-off = 2.5) and inspected label frequencies within classes 2 and 3. The cut-off was chosen according to the bimodal distribution of expression values as shown in Figure 7C. For the active class we found that approximately 88% of the genes are expressed according to our cut-off. In case of the PcGrepressed class the inverse is true, meaning approximately 85% of examined genes are silent. Taken together, these frequencies demonstrate that our unsupervised clustering approach has identified patterns that can be used to predict simplified gene expression states with high accuracy. In contrast to former attempts to predict gene expression based on chromatin binding profiles, we did not use any cotranscriptional histone modifications [31,32].
Our results suggest that the class 2 TrxG/PcG signature is a good predictor for active promoters. In order to confirm this, we simply divided the genome into overlapping tiles of 1 kb and examined the correlation between RNA production and class 2 affiliation based on our fitted model (see Materials and methods for details). The genome-wide correlation coefficient of 0.7 indicates that the appearance of class 2 entities and transcription are strongly coupled. To visualize the spatial correlation we co-plotted smoothed density estimates for mRNA production and class 2 predictions along chromosomes ( Figure 7D). As suggested by the overall correlation coefficient, both estimates are highly correlated.

Pharmacologic inhibition of FSH function causes global down-regulation of gene expression
Our chromatin profiling suggests that ASH1 and FSH operate as global transcriptional co-activators irrespective of gene type and function. In order to test this hypothesis, we perturbed FSH binding to chromatin by using the BET inhibitor JQ1 [20]. To test the effect of FSH inhibition on gene expression, we treated exponentially growing S2-DRSC cells with JQ1 for 1, 2, 4, and 20 h and monitored gene expression changes over time by RNA-seq. At 4 h post JQ1 addition approximately 2,500 genes were downregulated to <50% of their initial level ( Figure 9A, C, and D). In sharp contrast, only very few genes reacted to the treatment by elevating transcription. Most genes that were down-regulated at 4 h were already repressed at earlier time points and stayed repressed during the monitored time frame, suggesting a fast and sustainable reduction of gene activity by JQ1 treatment ( Figure 9C). Some genes were progressively repressed over time and some reverted to their initial steady-state level at 20 h. The last observation might indicate that these genes find ways to compensate the loss of FSH activity. An alternative explanation could be the propagation of secondary effects in the regulatory network. Plotting gene expression fold-changes versus FSH enrichments at promoters visualized that downregulated genes are bound by FSH in the absence of drug treatment ( Figure 9B). We next asked how the withdrawal of FSH from chromatin affects the localization of ASH1. We addressed this question by treating S2-DRSC cells with JQ1 for 4 h and monitoring the chromatin association of both proteins to a selection of active promoters using ChIP-qPCR. In agreement with the previous result the treatment induced a graduate loss of FSH from active promoters. At most examined sites, FSH is reduced to approximately 50% of the initial level, but there are also insensitive loci, which show no or minor changes upon BET inhibition ( Figure 9F). The reduction of FSH occupancy at promoters is accompanied by a limited loss of ASH1 from the same locations. This correlation might suggest that FSH helps to stabilize the interaction of ASH1 with target promoters. Alternatively, altered ASH1 levels at promoters simply reflect reduced transcription.
The reported experiments focused on the relevance of FSH function for steady-state transcription. In addition, we wondered whether FSH is needed for the effective activation of gene expression in response to stimuli. In order to test this, we exposed S2-DRSC cells to lipopolysaccharide (LPS), which is known to trigger the expression of genes coding for antimicrobial peptides (AMPs). Cells pretreated with JQ1 showed a slower accumulation of the AMP transcripts CecA1 and CecB in response to LPS induction ( Figure 9E).

RNAi-mediated knockdown of FSH induces minor gene expression changes with repespect to pharmacological treatment
Our expression profiling of JQ1-treated Drosophila cells showed a fast and widespread down-regulation of transcription. In contrast to mammalian cell culture systems, Drosophila offers the opportunity to knockdown BET function by targeting a single gene. We therefore decided to compare the transcriptional response between FSH knockdown and small molecule inhibition. RNAi effectively depleted FSH isoforms from dsRNA-treated S2-DRSC cells within 3 days ( Figure 10A). By targeting either both isoforms, or exclusively FSH-L, we intended to partially deconvolute their contribution to transcriptional control. This is of special interest, since the CTM connects FSH-L to the pTEF-b pathways, whereas FSH-S lacks this protein interaction domain [37]. For both knockdowns we monitored gene expression changes using RNA-seq. Subsequently we identified differentially expressed genes by comparing FSH knockdown conditions (two replicates each) with control treatments (eGFP specific dsRNA). For both knockdowns only few genes showed >2-fold differential expression ( Figure 10B). In addition, we observed comparable numbers of up-and down-regulated genes. Overall, the selected targeting of FSH-L still showed a weaker effect on transcription, than a complete depletion of both isoforms.
We therefore conclude that FSH knockdown does not induce the drastic transcriptional changes that we observed by small molecule inhibition. We believe that the deviating results are caused by the limited time resolution inherent to RNAi experiments, which in general do not allow measuring the primary response of the gene regulatory network (see Discussion for details).

Discussion
Experiments using double mutants of the TrxG genes ash1 and fs(1)h demonstrated a strong genetic interaction, whereas attempts to demonstrate a biochemical interaction remained unsuccessful so far. By purifying tagged ASH1 from a stable cell line, we discovered the missing physical connection between the two transcriptional activators. Indeed, the study by Chang et al. [17] attempting to demonstrate this interaction by FSH-S purification might have failed because of inappropriate purification conditions, or deviating starting material with respect to our study.
In line with our interaction proteomics results we show that ASH1 and FSH extensively co-localize on chromatin. Both proteins jointly target active gene promoters irrespective of gene function. Intron-rich, active genes tend to display ASH1and FSH signal distal to the TSS. Contrasting to our results, two recently published ChIP-chip studies have reported that ASH1 forms broad domains encompassing fully activated PcG target genes, or is bound at TSSs maintained in a 'balanced' state [10,11]. This balanced state is distinguished by the co-existence of PcG and TrxG proteins as well as production of fulllength mRNA. The investigators deduced from these results that ASH1 cannot be a general transcriptional activator, but instead functions as specific antagonist of PcG-mediated repression. This conclusion was backed up by earlier genetic experiments showing that ASH1 function is dispensable for Ubx activation in the absence of PcG-mediated repression [3]. Our ASH1 chromatin maps unequivocally identify ASH1 binding to genes as being highly predictive for transcriptional activity. In addition, we cannot confirm that this behavior is in any sense restricted to a defined set of target genes, such as genes under PcG-control. On the contrary, we detected strong ASH1/FSH signals at housekeeping genes, for instance Cyp12c1 ( Figure 6C) and Spf45 ( Figure 8A). We could extend the list by further components of cell metabolism like ribosomal subunits or histone genes (not shown). These findings strongly argue in favor of a general involvement of ASH1 and FSH in transcription. Support for this notion comes from a study examining the chromatin localization of ASH1L in a panel of human cell lines. By applying ChIP-qPCR the authors could show that ASH1L associates with the transcribed region of all active genes examined [33]. In addition, the distribution of ASH1L in transcribed chromatin was found to resemble that of H3K4me3. Our genome-wide, high resolution mapping study in fly cells comes to similar conclusions.
Missing ASH1 signals at active transcription units have been used to argue against a general involvement of ASH1 in transcription. Our analysis of the S2-DRSC chromatin landscape suggests that previous mapping studies failed to detect these signals and therefore jumped to conclusions based on incomplete evidence. A reasonable explanation for the enhanced sensitivity of our mapping study might be the employment of next-generation sequencing technology instead of microarrays. Direct comparison of ChIP-chip and ChIP-seq has shown that ChIP-seq generally produces profiles with better signalto-noise ratio and allows detection of more and narrower peaks [34]. Still, the lower measurement sensitivity/ resolution is not sufficient to explain the magnitude of differences. Our investigation of the antibody Q4177, generated on behalf of the modENCODE consortium, suggests that poor target reactivity of the immunoreagent might be the major explanation for missed ChIP signals in the study by Kharchenko et al. [11] (Figure 1D). Schwartz et al. [10] however partially relied on the same antibodies as we did.
Regarding the absolute number of reported binding sites we would like to note that we used a peak detection algorithm (MACS) that was developed to map human transcription factor (TF) binding sites in ChIP-seq data (see Materials and methods for details). Sequence specific TFs typically generate sharply restricted, Gaussian-like ChIP-seq signals. Therefore, the algorithm was ideally suited to detect ASH1, FSH signals at TSS. At the same time, it tends to break up more widely distributed signals, like the ones we described for intro-rich active genes, into closely spaced peak clusters. Hence, the absolute number of peaks reported in this study does not reflect the number of functional elements bound by ASH1 and FSH in a one-to-one correspondence.
The second major argument against a general involvement of ASH1 in transcription was provided by the Ubx activity in ash1/E(z) double-mutant mitotic clones, located in-and outside of the endogenous Ubx expression domain [3]. In our opinion, this observation does not directly contradict the co-activator hypothesis favored by our data. It is conceivable that ASH1 and other TrxG proteins only modulate the transcriptional output, rather than being absolutely required for transcription. The incomplete, stochastic loss of Ubx activity in embryonic tissues devoid of ASH1 protein already argues against such a strict requirement [3,29], and studies of heat shock response have already demonstrated a transcription modulating function for TRX [35]. The latter finding is especially intriguing, because trx/E(z) double-mutant clones show Ubx reactivation patterns similar to their ash1/E(z) counterparts [3].
Our unsupervised clustering of genes according to their combined TrxG/PcG binding profiles identified a class of transcription units with concurrent high PRC1 and TrxG enrichments at the promoter. In contrast to the 'balanced' state introduced by Schwartz et al. [10], these genes do not generate a transcript. We argue that the existence of this 'PcG-repressed' class provides additional evidence against the anti-repressor hypothesis. Instead, we favor the idea that fully installed PcG-repression needs to overcome the general co-activator function of TrxG proteins. We have previously identified Polycomb (Pc) interacting proteins by in-vivo biotinylation tagging [36]. Unexpectedly, this approach revealed FSH to interact with Pc in the Drosophila embryo. Our current analysis of FSH chromatin occupancy now shows that PRC1 and FSH co-localize at  repressed promoters and intergenic gene regulatory elements, although they are known to fulfill antagonistic functions. Mechanistically this might suggest that PRC1 silences genes by blocking the elongation-promoting functions of FSH, for instance via the pTEF-b pathway. In the light of these deliberations it is sensible that PcG-bound promoters do show elevated levels of promoter proximal pausing, as we have reported earlier [28]. Future experiments using defined in-vitro transcription systems in conjunction with reconstituted PRC1 and FSH might shed light on the mechanistic details of this interaction.
Our data provides evidence that FSH promotes the transcriptional activity of genes. How might this work at a mechanistic level? The human FSH-L homolog BRD4 proteins contain a conserved pTEF-b interacting domain in the very C-terminus, which is sufficient to directly recruit pTEF-b to target genes [37]. Following recruitment, pTEFb triggers the transition from early elongation (promoterproximal pausing) to productive elongation by phosphorylating Serine 2 residues in the RNA polymerase II CTD [38]. In addition, pTEF-b has been shown to phosphorylate DISF and NELF which releases the repressive factor NELF and converts DISF into a positive elongation factor. Studies in Drosophila have shown that pausing represents a widespread gene regulatory strategy. Our data are in line with this possible mechanism, since displacement of FSH leads to a widespread down-regulation of transcription. A recent study has shown that BRD4 is also able to directly phosphorylate Serine 2 of the Pol II CTD [39]. A kinase activity has also been demonstrated for FSH-S [17]. Using a defined in-vitro transcription system, it has been shown that the human BRD2, BRD3 render nucleosomes marked by acetylation permissive to the passage of elongating RNA polymerase II, thereby bypassing FACT requirement [40]. Taken together, FSH might therefore support the passage of two central checkpoints of the transcription cycle: first, it might be critical to release RNA polymerase II into productive elongation by initiating CTD phosphorylation and pausing factor release. Second, it potentially removes nucleosomal barriers in front of the elongating polymerase. In addition, the loss of ASH1 from promoters after FSH removal might indicate that the ASH1-FSH interplay stabilizes the association of ASH1 with chromatin. Additional evidence for cooperation between ASH1 and FSH during gene activation is provided by our recent study, applying regression models in order to predict gene expression based on chromatin binding profiles. In this quantitative modeling framework, the two proteins form an interaction pair [31].
Our JQ1 treatment of Drosophila tissue culture cells provides evidence, that transcriptome homeostasis is heavily affected by small molecule inhibition of BET proteins. This observation corresponds to our FSH chromatin map, since we observed a high correlation between FSH localization and gene activity. Exactly the same results have recently been described with respect to BRD4 in human CD4+ T cells and ES cells. ChIP-seq revealed that BRD4 localizes to active promoters and enhancers, and that expression levels are strongly correlated with BRD signal intensity [41]. BRD4 profiles across active human promoters display that BRD4 binds at the TSS. Here we show the same for FSH in Drosophila cells. The degree of correlation between BRD4 localization and transcriptional activity even encouraged the authors to predict that 'BRD4 could potentially be used as a genome-wide hallmark of active or poised genes'. Our gene clustering according to TrxG/PcG promoter binding supports this idea. Interestingly, our clustering into active and PcG-repressed genes parallels the human classification into active and poised genes in a sense that poised genes are occupied by BET proteins, but do not produce mRNA. Disruption of BRD4 chromatin binding by JQ1 in CD4+ T cells resulted in reduced expression of more than a thousand genes in CD4+ T cells [41]. Again, we have reporter similar observation using JQ1 treatment in fly cells. In contrast, knockdown of FSH did not result in comparable transcriptional repression. We argue that the deviating results are due to the different natures of the treatment. Small molecule inhibition happens almost instantaneously. By monitoring the system in a timeresolved manner it is possible to untangle primary and secondary effects. Effective depletion of target proteins by RNAi can only be achieved within time scales of days. Therefore RNAi experiments rather disclose a novel steady-stead of the reduced network than primary effects. Similar thoughts should be considered when interpreting expression profiles obtained from zygotic mutants, since maternal gene products are present during development. Like in RNAi-mediated knockdowns it becomes impossible to define a time point that corresponds to functional null situation, but is not governed by secondary effects.
A recent study in erythroid cells shows, that BRD3 is recruited to target promoters by reading an acetyl mark on the transcription factor GATA1, instead of being primarily dependent on histone acetylation [42]. The concentration of FSH signals around TSSs fits the idea that acetylated transcription factors might also contribute to BET protein recruitment in flies.

Conclusions
There is substantial evidence that TrxG proteins play global roles in transcription. In the future it will be interesting to find out, why this group of proteins genetically scores as being important for the maintenance of the active state of HOX genes, or in other words, what discriminates them from other factors in functional proximity to RNA polymerase. More mechanistic insight how TrxG complexes are connected to transcription machinery will be necessary to answer this question. It might indeed turn out that TrxG complexes 'are among those that have emerged as being important for maintenance for a relatively uninteresting reason e.g., because even relatively subtle changes in the expression of Drosophila HOX genes cause homeotic transformations' [1].

Materials and methods
Generation of ASH1 expression construct and stable cell line The ash1 coding sequence was PCR-amplified from S2 cell derived cDNA in four overlapping parts (A-D) using the primers listed in Additional file 2, Table 1. Subsequently, each PCR product was inserted into the pENTR/D-TOPO vector for the Gateway Cloning System as outlined in the manufacturer's instructions (pENTR/D-TOPO cloning kit, Invitrogen). The full-length ash1 coding sequence was reassembled from these primary clones by three directional subcloning steps, utilizing the restriction sites indicated in Figure 1A. Sanger sequencing confirmed that the derived full-length cDNA matches the ASH1 SwissProt entry [Swiss-Prot:Q9VW15] with the exception of codons T1716 and L1717, which seems to be absent from ash1 mRNA in S2 cells ( Figure 1C).
In order to generate the ASH1 expression construct outlined in Figure 2A, the Hygromycin resistance cassette (AccI/SapI fragment taken from pCoHygro plasmid, Invitrogen) was inserted into the AccI/SapI-cut pMT/V5-His A vector backbone (Invitrogen). In addition, the sequence between EcoRV/HpaI encompassing MCS to SV40 polyA was replaced by the EcoRV/PmeI-excised fragment from pAFW (Drosophila Gateway Collection, available through DGRC). In a subcloning step, this fragment was previously modified by inserting the 8xHis-TEV coding sequence (generated by hybridizing the oligo nucleotides listed in Additional file 1, Table 4 into the AgeI site between 3xFLAG and gateway cassette. Finally, the ash1 coding sequence was transferred to the inducible pMT/FHW HygRes expression vector by standard gateway cloning according to manufacturer's instructions (Invitrogen).
In order to generate the stable cell line, S2-DRSC cells grown in Schneider's medium incl. 10% FCS were lipotransfected using pMT/FH-ASH1 HygRes according to standard Effectene protocols (Quiagen). Two days posttransfection transformants were selected by Hygromycin B addition to the growth medium (500 µg mL −1 ) and passaged for 4 weeks in the presents of the selecting agent.

FH-ASH1 purification and identification of co-purified proteins
Expression of tandem tagged ASH1 was induced by adding 500 µM copper(II) sulfate to the growth medium for approximately 16 h. Afterwards cells were scraped in ice cold PBS, pelleted by low speed centrifugation (500 × g) peptides were separated using online reverse-phase chromatography and electrosprayed into a LTQ Orbitrap Velos tandem mass spectrometer (Thermo Scientific). MS spectra were recorded at 60,000 R and most intense peptide ions were selected for CID fragmentation and recording of product ion spectra. MS/MS spectra were matched against UniProtKB release 2011.08 using the Mascot search engine (Matrix Science), assuming the digestion enzyme Trypsin and up to one missed cleavage. Mascot was searched with a fragment ion mass tolerance of 0.5 Da and parent ion tolerance of 5 ppm. Iodoacetamide derivative of cysteine was specified in Mascot as a fixed modification. S-carbamoylmethylcysteine cyclization (N-terminus) of the n-terminus, deamidation of asparagine and glutamine and oxidation of methionine were specified in Mascot as variable modifications. Scaffold 3.3.1 (Proteome Software) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at >95.0% probability as specified by the Peptide Prophet algorithm [43]. Protein identifications were accepted if they could be established at >99.0% probability and contained at least three identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm [44]. For complete results see Additional file 1.

Generation of FSH expression constructs and coimmunoprecipitation
The FSH-S coding sequence was PCR-amplified from the Drosophila Gold Collection clone LD26482 (available through DGRC) using primers listed in Additional file 2, Table 1. The PCR-product was afterwards inserted into pENTR/D-TOPO as outlined in the manufacturer's instructions (pENTR/D-TOPO cloning kit, Invitrogen). In order to generate the HA-tagged expression construct the FSH-S ORF was transferred to pAHW (Drosophila Gateway Collection, available through DGRC) using standard gateway cloning. In addition, the ASH1 fragments A-D were transferred to the 3xFLAG-tagged pAFW plasmid (Drosophila Gateway Collection). For co-immunoprecipitation S2-DRSC cells were co-transfected with pair-wise combinations of pAH-FSH-S and pAF-ASH1 A-D by using standard Effectene (Qiagen) lipotransfection protocols. Two days post-transfection whole cell lysates were prepared using RIPA lysis buffer and tagged ASH1 fragments were immunoprecipitated by FLAGM2 antibody (Sigma) coupled to Protein G Dynabeads (Invitrogen). Precipitated proteins were eluted by boiling Dynabeads at 70°C in 1xLDS sample buffer (Invitrogen) for 10 min. Immunoblotting of eluted proteins and nuclear extracts was conducted using Novex TA 3-8% gradient gels and iBlot transfer system according to manufacturer's instructions (Invitrogen). For immunodetection of HA-tagged FSH-S HA.11 antibody (Covance) was used at 1:2,000 in PBS, 0.1% Tween-20, 5% dry milk powder followed by anti-mouse IgG-HRP (GE Healthcare) at 1:10,000 in PBS, 0.1% Tween-20, 5% dry milk powder. ECL detection of proteins was conducted using ECL substrate and X-ray films from GE Healthcare.

FSH knockdown by RNAi
PCR products serving as templates for in-vitro transcription were generated from pENTR/D-TOPO FSH-S using T7 recognition sequence flanked primers listed in Additional file 2, Table 3. PCR amplicons DRSC18778, DRSC29017 were taken from the DRSC reagent database [45]. The FSH-L specific amplicon was designed using E-RNAi [46]. Purified PCR products were transcribed to dsRNA and cleaned using the MEGAscript RNAi Kit according to manufacturer's instructions (Ambion). RNAi was carried out according to standard bathing protocols in 6-well plate format (Drosophila RNAi Screening Center, Harvard Medical School) by adding 15 µg of dsRNA to each well before platting S2-DRSC cells in serum free Schneider's medium and starving cells for 30 min.

FSH antibody preparation
Serum from rabbits immunogenized with FSH-specific antigens was kindly provided by Igor Dawid (National Institute of Child Health and Human Development). The antigen used for rabbit ID166 corresponds to the XbaI/ PstI fragment from clone e1.20 and the antigen injected into rabbit ID173 is derived from the EcoRI/HindIII fragment of clone e4.1, both described in [16]. Generation of β-Gal/FSH-S fusion constructs, protein expression, and immunization was done in the laboratory of I. Dawid as described in [17]. Immunoglobulin G purification from rabbit serum using Protein A Sepharose CL-4B (GE Healthcare) was carried out as described in [47], Basic Protocol 2. In order to validate the antibody specificity, we performed immunoblotting of S2-DRSC lysates depleted of FSH protein by RNAi as described before ( Figure 4C). In order to reduce high background signals, FSH-specific antibodies were diluted in high-salt PBS (300 mM NaCl), 1% Tween-20, 5% dry milk powder.

Quantification of LPS-inducible gene expression under JQ1 treatment
One hour before lipopolysaccharide (LPS) was added to the growth medium S2-DRSC cells were pretreated using either JQ1 or DMSO as described before. After LPS addition (5 µg mL -1 ) cells were lysed at indicated time points using Trizol (Invitrogen). In order to quantify the transcriptional response of known LPS-inducible genes (CecA1, CecB) total RNA was extracted, reverse transcribed (First Strand cDNA Synthesis kit, Fermentas), and TURBO DNase (Ambion) treated as outlined in the manufacturer's instructions. Quantification of cDNA was done by qPRC using SYBR Green chemistry and primers listed in Additional file 2, Table 2 on the LightCycler 480 thermocycler (Roche Diagnostics). For calculating target gene levels relative to the house keeping gene ATPsyn-Cf6 the ΔΔct method including efficiency correction was used. Determination of c t values was done by calculating the second derivative maximum of the SYBR green signal.
The BD1-BD2 coding sequence was PCR amplified from the vector LD26482 and cloned into pNIC28-Bsa4 (Structural Genomics Consortium) applying ligation independent cloning (LIC). The resulting plasmid was used to express 6xHis-tagged BD1-BD2 in BL21(DE3) E. coli. Cell lysate was prepared by sonicating cells in 20 mM Tris-HCl pH 8.0, 1 mM PMSF, 1mg/ml Lysozyme, 0.05% Triton X-100 for 15 min (30 s ON/30 s OFF) in a water bath sonicator followed by high speed centrifugation. BD1-BD2 was purified from cleared lysate by IMAC on Ni-NTA Superflow (Qiagen) deploying an imidazole step-gradient (50, 100, 150, 250 mM). BD1-BD2 containing eluate fractions were monitored by Coomassie-stained SDS-PAGE and concentrated in 25 mM HEPES pH 8.0, 150 mM Nacl using 10 NMWL centrifugal filter devices (Millipore). For BD1-BD2 pull-downs approximately 3 µg of purified protein was incubated with 35-40 µM of histone tail peptide in 25 mM HEPES pH 8.0, 150 mM NaCl in the presence or absence of 10 µM JQ1. After 1 h of incubation at 4°C biotinylated peptides were captured on streptavidin magnetic beads and washed three times in binding buffer. Precipitated proteins were eluted by boiling magnetic beads in 1x LDS buffer. Eluted proteins and input samples were immunoblotted using anti His-probe (H-3) HRP (Santa Cruz Biotechnology) and standard ECL reagents.

Analysis of protein binding to chromatin by ChIP-seq
Chromatin immunoprecipitation followed by nextgeneration sequencing (ChIP-seq) was essentially done as described in [28] using formaldehyde cross-linked chromatin from 2.5 × 10 7 S2-DRSC cells per ChIP. In order to assure data comparability across ChIP-seq experiments, all chromatin samples were taken from the same chromatin batch. This chromatin batch is identical to the one we used previously to generate ChIP-seq data for PC, PH, PSC, and TRX-C.
ASH1-bound chromatin was enriched using two antibody preparations provided by the lab of F. Sauer (University of California, Riverside). Anti-ASH1-N recognizes an antigen located in the amino-terminus of ASH1, while the anti-ASH1-C specific antigen resides in the carboxyterminal portion of the protein. Both polyclonal antibodies have been raised in rabbits. Since different dsRNA treatments of S2-DRSC cells did not induce detectable ASH1 knockdown (data not shown), we investigated antibody specificity by expressing ASH1-GFP fusion proteins in S2-DRSC cells and immunoblotting of cell lysates ( Figure 1B). FSH-bound chromatin was enriched by the anti-FSH and anti-FSH-L antibody preparations described above.
Sequencing libraries were prepared from 10 ng of immunoprecipitated DNA using the ChIP-seq DNA Sample Prep Kit (Illumina), including size selection of preamplified fragments on agarose gels (200 bp +/-30 bp). For quality control, the size distribution of the final libraries was assayed on the Agilent BioAnalyzer 2100 using High Sensitivity DNA microfluidic chips. Each library was sequenced for 36 cycles in a single-end run on the Genome Analyzer IIx (Illumina). Sequencing yielded approximately 20 × 10 6 quality filtered reads per library (Illumina Chastity Filter), corresponding to a mean genome coverage of approximately 5x. Short reads (36 bp) were aligned to the Drosophila reference genome (BDGP Release 5) using Bowtie 0.12.7 and the following parameters: -n 2, -m 20, -k 1, -best [48]. Since anti-ASH1-N and anti-ASH1-C target the same protein isoform, we decided to merge the corresponding alignments prior to downstream analysis. The same was done for the FSH-specific alignments, since we could not detect mayor differences between these datasets (data not shown). Merging and indexing of alignments was done using SAMtools 0.1.9 [49]. Regions showing significantly enriched read coverage compared to input chromatin (also referred to as 'peaks') were calculated using MACS 1.4.0 and the following parameters: band width, 300; model fold, 10 to 30; P value cut-off, 1 × 10 -5 [50].
Read coverage profiles were calculated from aligned reads by shifting and extending reads according to the fitted MACS models. The distance calculation between peak intervals and TSS was carried out using the R/Bioconductor package ChIPpeakAnno (Release 2.9) and basic R functions [51]. Interval based co-localization analysis was done by comparing MACS peak lists with BEDtools 2.10.0 [52]. The compilation of high-confidence PRC1 sites was constructed by intersecting peak lists for PC, PH, and PSC using BEDtools. PC, PH, and PSC specific peak list were compiled using MACS on our previously published PcG ChIP-seq dataset available through NCBI's Gene Expression Omnibus [53]. Metagene profiles were created using the R/Bioconductor package GenomicRanges (Release 2.9). In short, based on alignments, coverage vectors for each chromosome were calculated. From these vectors, all subvectors were extracted that correspond to ORFs of known protein coding genes having well separated TSSs (min. distance to next TSS = 1 kb). Next, kernel density functions were calculated from these subvectors and sampled at 500 equally spaced points in order to create read density estimates at positions relative to ORFs. Finally, the position-wise mean densities were calculated after grouping genes according to expression level and plotted along relative positions in order to obtain metagene profiles. In order to create promoter profiles, coverage vectors spanning all non-overlapping 1-kb windows centered at known TSS were calculated. Position-wise mean read coverage was calculated and plotted along the relative position to obtain coverage profiles. Quality-filtered reads, MACS peak lists, and coverage profiles have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Super Series accession number GSE36450 [54].

Gene expression analysis by RNA-seq
For JQ1 treatment of exponentially growing S2-DRSC cells lyophilized JQ1 was dissolved in DMSO (10 mM) and added to the culture medium at a concentration of 10 µM. RNA extraction from S2-DRSC cells was done using Trizol reagent according to manufacturer's instruction for adherent cells (Invitrogen). Barcoded sequencing libraries were prepared from 4 µg of total RNA according to Illumina's TruSeq protocol including polyA-enrichment, mixed in equal proportions and sequenced together on a single HiSeq2000 lane (Illumina) using a 50-cycle single-end run.
Alignments of RNA-seq reads were generated using the splice-junction-aware aligner TopHat [55] and Drosophila gene models originating from Ensembl release 64 (default parameters). All reads uniquely mapping to gene models were counted by the HTSeq-count script (HTSeq developed by Simon Anders, EMBL Heidelberg) and the count statistic was forwarded to bioconductor package DESeq [56]. Since the standard size factor estimation of DESeq produced inappropriately adjusted counts, an alternative normalization procedure was applied, assigning the 100 most abundant genes in the control condition as normalization index. The principle behind this strategy is comparable to qPCR normalization using a house keeping gene index.
Demultiplexed RNA-seq reads and raw gene counts have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Super Series accession number GSE36450 [57].

Computation of ChIP-seq enrichments at promoters
ChIP-seq and input reads were counted on 1kb windows +/-500 bp around N = 13,254 unique TSS defined in the ensGenes table of the UCSC browser with the pysam python module. Subsequently, the relative frequencies p i = X i /m of ChIP counts X i on TSS i were normalized by a rescaling of the library size m to maximize the number of TSS where the relative ChIP frequency was identical to that of the corresponding value q i = Y i /n in the input experiment. The symbol Y i denotes the read count in interval i in the input experiment and n the input library size. For each TSS and ChIP experiment, a binomial likelihood ratio test was computed testing the null hypothesis p i = q i versus the alternative p i ≠ q i , and the sum over all TSS of the resulting P values was maximized numerically over m. Normalized enrichments were then computed as p i /q i . Data normalization and enrichments were computed in R.

Gene clustering according to TrxG/PcG ChIP-seq profiles at promoters
The sparse Gaussian mixture model (sGMM) is a probabilistic mixture model of multivariate Gaussian distributions, in which the covariance matrices are estimated under an L 1 -penalty favoring sparse solutions [59]. The model was fitted to the set of logarithmic TrxG/PcG ChIP-seq enrichments at known transcription start sites defined above.
Let x i Î R n be a vector of ChIP-seq enrichments from different antibodies in TSS window i and j(x i ; µ, Σ) denote the density of a multivariate Gaussian with mean μ and covariance Σ. An sGMM with m mixture components (classes) is defined by the likelihood function where ρ ≥ 0 defines the penalization strength, ||Σ k -1 || 1 is the L 1 -norm of the inverse covariance matrix, and φ k x i = φ x i ; μ k , k exp −ρ −1 k