Trithorax monomethylates histone H3K4 and interacts directly with CBP to promote H3K27 acetylation and antagonize Polycomb silencing

Polycomb group (PcG) and Trithorax group (TrxG) proteins play evolutionarily conserved roles in the epigenetic regulation of gene expression. Originally identified in Drosophila by their role in maintaining the spatially restricted patterns of homeotic (HOX) gene expression during development, they regulate many other genes that encode transcription factors and signaling factors that act as ‘master’ regulators of the many distinct cell identities found in multicellular organisms (Schwartz and Pirrotta, 2007; Schuettengruber et al., 2011). The mutually antagonistic activities of PcG and TrxG proteins promote the stable, mitotically heritable maintenance of repressed and active transcriptional states, respectively.

Maintenance of transcriptionally silent states of PcG-regulated genes requires Polycomb repressive complexes 1 and 2 (PRC1 and PRC2) (Margueron and Reinberg, 2011), which are recruited to Polycomb response elements (PREs) (Müller and Kassis, 2006). PRC2 trimethylates histone H3 lysine 27 (H3K27me3), a mark that is distributed in broad domains over inactive PcG-regulated genes, encompassing promoters, flanking regulatory regions, including PREs and enhancers, as well as transcribed regions (Schwartz et al., 2006). PRC1 binds H3K27me3 via its PC subunit, and has several other activities essential for silencing (Margueron and Reinberg, 2011).

Drosophila Trithorax (TRX) is best known for its role in antagonizing transcriptional silencing by PcG proteins, stimulating enhancer-dependent transcription (Poux et al., 2002), and maintaining a mitotically heritable ‘cellular memory’ of prior transcriptional activity of PcG-regulated genes (Schuettengruber et al., 2011). TRX binds constitutively to PREs, apparently even through DNA replication (Petruk et al., 2012), which are thus also TRX response elements (TREs). TRX is also found at promoters of PcG-regulated genes (Schwartz et al., 2010; Enderle et al., 2011). Tethering a GAL4-TRX fusion protein to a Ubx reporter transgene revealed that TRX can boost enhancer-dependent reporter expression but has no intrinsic transcription-activating activity in the absence of enhancers. Stimulation of enhancer-dependent transcription by endogenous TRX requires the presence of a PRE/TRE (Pirrotta et al., 1995; Poux et al., 2002). Thus, although TRX is recruited to PRE/TREs, surrounding enhancers may be important targets of TRX catalytic activity.

TRX is a large multifunctional protein with a SET domain, four PHD fingers, and FYRN and FYRC domains (Ringrose and Paro, 2004), which are conserved in its mammalian orthologs MLL1 and MLL4. The TRX SET domain has been shown to have lysine methyltransferase activity with substrate specificity for histone H3K4 (Smith et al., 2004); however, its product specificity has not been definitively demonstrated (Smith et al., 2004; Ardehali et al., 2011). H3K4 is present in mono-, di- and tri-methylated isoforms in vivo. H3K4me3 is a signature mark of active promoters, whereas H3K4me1 is a signature mark of enhancers (Heintzman et al., 2007). Both TRX and its mammalian ortholog MLL1 were, until recently, thought to be responsible for deposition of the H3K4me3 mark at promoters of active TRX-regulated genes (Petruk et al., 2012). However, Drosophila SET1 (and mammalian SET1A and SET1B, also known as SETD1A and SETD1B) now appears to be the principal H3K4 trimethyltransferase responsible for the H3K4me3 at promoters of active genes (Ardehali et al., 2011; Hallson et al., 2012). This prompted us to reexamine the intrinsic catalytic activity of the TRX SET domain. We report here that both the TRX and TRR SET domains (and their mammalian orthologs) have only robust H3K4 monomethyltransferase activity in vitro and that a recombinant TRX core complex [TRX SET domain + WRAD (WDR5, RBBP5, ASH2L, DPY30)] has only a greatly enhanced H3K4 monomethyltransferase activity. Moreover, the genome-wide distribution of TRX is highly correlated with H3K4me1 (but not H3K4me3) at PcG-regulated genes. Consistent with this, the catalytically inactive trx^Z11 and trr³ mutants have reduced H3K4me1 levels but normal H3K4me3 levels in vivo. Furthermore, trx^Z11 suppresses the Polycomb phenotype of Pc³/+ mutant adults, implicating its catalytic activity in antagonizing Polycomb silencing.

We previously reported that the histone acetyltransferase (HAT) CBP (encoded by the nejire gene) antagonizes Polycomb silencing by acetylating histone H3K27 (H3K27ac), which prevents trimethylation of H3K27 by PRC2 (Tie et al., 2009). We also showed that H3K27ac levels are reduced in trx¹ mutants and elevated in TRX overexpressers (Tie et al., 2009), suggesting that TRX might promote acetylation of H3K27 by CBP at PcG-regulated genes. H3K27ac is highly correlated with actively transcribed genes, including many that are not PcG regulated, and is found at both their enhancers and promoters (Wang et al., 2008; Karlić et al., 2010). A TRX complex purified from Drosophila embryos was previously reported to contain CBP (Petruk et al., 2001). However, CBP was not found in TRX (or TRR) complexes subsequently purified from Drosophila S2 cells (Ardehali et al., 2011; Mohan et al., 2011), or in the orthologous human MLL1 complex (Dou et al., 2005). However, human CBP has been shown to bind directly to MLL1 and this appears to be required for MLL1-dependent transcription (Ernst et al., 2001; Goto et al., 2002; De Guzman et al., 2006; Arai et al., 2010). We show that CBP binds directly to TRX near its SET domain, physically coupling their H3K4 monomethylation and H3K27 acetylation activities and suggesting a molecular basis for TRX-dependent H3K27 acetylation. We discuss the implications of these new findings for the function of TRX in antagonizing silencing, stimulating enhancer-dependent transcription and maintenance of cellular memory.

To determine the effect of selective loss of TRX catalytic activity on H3K4 methylation in vivo, we examined H3K4me1 and H3K4me3 levels in two trx mutants: the catalytically inactive trx^Z11 (Smith et al., 2004) and the hypomorphic temperature-sensitive trx¹ (Ingham and Whittle, 1980), a spontaneous mutation associated with a 9 kb insertion in the first intron (Breen and Harte, 1991). When raised at 29°C, trx¹ homozygotes have a substantially reduced amount of TRX protein (Tie et al., 2009) and so are expected to be impaired in all TRX functions. The catalytically inactive trx^Z11 missense mutation (G3601S) alters a single invariant SET domain residue (Stassen et al., 1995). It does not appear to affect the stability of the TRX protein (Chinwalla et al., 1995; Smith et al., 2004).

In late third instar salivary glands from trx¹ homozygotes (raised at 29°C) the H3K4me1 level was reduced (by 30%) in bulk histones compared with wild-type controls (29°C) (supplementary material Fig. S1A). The H3K27ac level was also reduced (by 24%), as was the H3K4me3 level (by 26%) (supplementary material Fig. S1A), consistent with our previously reported results from trx¹ adults (Tie et al., 2009). Conversely, the H3K4me1 level was increased by 40% in TRX overexpressers, as were the H3K27ac (by 28%) and H3K4me3 (by 34%) levels (supplementary material Fig. S1B). RNAi knockdown of TRX in Drosophila S2 cells has similar effects on bulk H3K4me1 and H3K4me3 levels (Ardehali et al., 2011).

trx^Z11 homozygotes, which can survive into the pupal stage (Mortin et al., 1992; Breen, 1999), also have a reduced H3K4me1 level (by 26%) (Fig. 1A), but, in contrast to trx¹ mutants, their H3K4me3 level is not appreciably different from that of wild type (Fig. 1A). The H3K4me1 level was also substantially reduced on salivary gland polytene chromosomes of trx^Z11 homozygotes and increased on those of TRX overexpressers (Fig. 1B).

Previous temperature shift experiments with the trx¹ temperature-sensitive mutant identified a critical requirement for trx expression in 0- to 4-hour embryos (Ingham and Whittle, 1980). Consistent with this, a reduced H3K4me1 level (but normal H3K4me3 level) is also readily detectable in trx^Z11 embryos by stage 10 (4.3-5.3 hours) (Fig. 1C), around the same time that trx null mutants begin to show reduced levels of homeotic gene expression (Breen and Harte, 1993; Sedkov et al., 1994). This suggests that TRX catalytic activity is required for at least a portion of the total H3K4me1 in vivo. The apparently normal H3K4me3 level in trx^Z11 embryos (Fig. 1C) further suggests that the monomethyltransferase activity of TRX might be its principal activity in vivo.

Drosophila TRR is another SET domain protein with H3K4 methyltransferase activity, which functions as a co-activator of the Ecdysone receptor (EcR) (Sedkov et al., 2003). We examined H3K4me1 and H3K4me3 levels in vivo in two trr nonsense mutants: trr¹ and trr³, both of which are embryonic lethal. trr¹ appears to be a protein-null allele, and trr³ truncates the SET domain and is expected to be catalytically inactive (Sedkov et al., 2003). The trr¹/Y null mutant embryos had decreased levels of both H3K4me1 and H3K4me3, like trx¹ (supplementary material Fig. S1C). The trr³/Y mutant embryos had a decreased H3K4me1 level but apparently normal H3K4me3 level (Fig. 1D), like trx^Z11.

Since H3K4me1 is only partially reduced in trx and trr mutants, we examined trr³/Y; trx^Z11 double mutants to determine whether H3K4me1 is further reduced. Although these embryos arrest earlier and appear to have more severe defects than either single mutant, the H3K4me3 level in individual nuclei appears to be similar to that of wild-type controls, whereas H3K4me1 was undetectable (Fig. 1E). These data strongly suggest that TRX and TRR are the principal H3K4 monomethyltransferases in vivo and that together they are responsible for the bulk of the H3K4me1 present in embryos. They further suggest that the reduced H3K4me3 in trx¹ and trr¹ null mutants is not due to the loss of their methyltransferase activities, but to the loss of some other function of TRX and TRR that remains intact in the trx^Z11 and trr³ mutant proteins. One possibility is the third PHD finger of TRX, which binds to H3K4me3 (Chang et al., 2010; Park et al., 2010; Wang et al., 2010) and might shield it from demethylation.

We previously showed that TRX-dependent H3K27 acetylation by CBP antagonizes Polycomb silencing by preventing H3K27 trimethylation by PRC2 (Tie et al., 2009). Whether the H3K4 methyltransferase activity of TRX is also required for this antagonism has not been determined. We compared the ability of the catalytically inactive trx^Z11 and the protein-null trx^B11 mutations to suppress the extra sex combs phenotype of Pc³/+ mutant adults, a characteristic PcG mutant phenotype due to derepression of Scr in mesothoracic (T2) and metathoracic (T3) legs of adult male flies. We found that trx^Z11 and trx^B11 mutations suppress the T2 extra sex combs of Pc³/+ similarly (Table 1), indicating that TRX catalytic activity does play a role in antagonizing Polycomb silencing. Consistent with this, both TRX and H3K4me1 (but not H3K4me3) are present at all known PRE/TREs in the Bithorax and Antennapedia complexes (supplementary material Fig. S2 and see below).

The reduced H3K4me1 and normal H3K4me3 levels in the trx^Z11 mutant suggest that TRX might be predominantly an H3K4 monomethyltransferase. To examine this, we tested GST-TRX-C fusion proteins (see constructs in Fig. 2D) containing the C-terminal SET domain and additional adjacent residues (that mediate interaction with WRAD core subunits) in histone methyltransferase (HMT) assays. Reaction products were examined by western blot with antibodies that specifically recognize H3K4me1, H3K4me2 or H3K4me3. As stringent controls for the specificity of these antibodies, on each blot we included full-length recombinant H3 proteins with single chemically introduced methyl-K analogs of K4me1, K4me2 or K4me3 (Simon et al., 2007). These serve as sensitive controls for false-positive western signals caused by cross-reactivity that can appear as a result of using excess antibody or sample overloading.

We found that larger SET domain-containing fragments (≥251 residues) have only a robust H3K4 monomethyltransferase activity on a recombinant (unmodified) free histone H3 substrate (Fig. 2A, lanes 2-4 and 2C, lane 2; supplementary material Fig. S6). Deletion of the SET domain from TRX-C751 abolished its enzymatic activity (Fig. 2A, lane 5). The TRX SET domain also monomethylated H3K4 on an H3(1-20) peptide, which was confirmed by MALDI TOF mass spectrometry (supplementary material Fig. S3). We also examined the activity of TRX-C751 in complex with WRAD core subunits. A recombinant human MLL1-WRAD core complex assembled in vitro was previously shown to monomethylate but not trimethylate H3K4 by quantitative MALDI-TOF mass spectrometry. When TRX was substituted for MLL1 in this complex, we detected exclusively H3K4me1 and no H3K4me3 product in the HMT assay (supplementary material Fig. S4) indicating that, although the WRAD subunits may greatly increase the reaction rate, they do not convert the TRX SET domain to a trimethyltransferase.

The trx^Z11 missense mutation (G3601S) alters a residue that is invariant in all SET domains (supplementary material Fig. S5) and is part of the conserved SAM-binding site (Cheng et al., 2005). It was previously reported to abolish catalytic activity (Smith et al., 2004). As expected, it abolishes the H3K4 monomethyltransferase activity of the TRX SET domain (Fig. 2B, lane 4), which was confirmed by MALDI-TOF mass spectrometry (supplementary material Fig. S3). An E3616S substitution, which alters an invariant Glu residue that lies next to an invariant Tyr in the predicted catalytic site (supplementary material Fig. S5), also severely impairs the H3K4 monomethyltransferase activity (Fig. 2B, lane 5).

The product specificity of SET domain methyltransferases (e.g. H3K4me1/2/3) depends on several key residues in the active site, including a so-called ‘Phe/Tyr switch’ position. Di- and trimethyltransferases tend to have a Phe at this position, whereas monomethyltransferases tend to have the bulkier Tyr residue; changing Tyr (Y) to Phe (F) can convert monomethyltransferases to di- and trimethyltransferases (Collins et al., 2005). To confirm that the absence of H3K4 trimethylation by the TRX SET domain is not due to the reaction conditions used or to loss of enzyme activity during the reaction, we changed the corresponding Tyr in TRX-C to Phe (Y3701F), which converted it to a di- and trimethyltransferase, yielding H3K4me2 and H3K4me3 at the expense of H3K4me1 (Fig. 2C, lanes 4 and 5). This indicates that Y3701 prevents H3K4 di- and trimethylation by the isolated TRX SET domain and further suggests that its principal intrinsic activity is H3K4 monomethylation. The corresponding substitution in the MLL1 SET domain (Y3942F) also converts it to a di- and trimethyltransferase (Patel et al., 2009). The results of the in vitro HMT assays are summarized in Fig. 2D.

RNAi-mediated knockdown of the H3K4 methyltransferase TRR in S2 cells and wing imaginal discs was reported to reduce bulk H3K4me1 (Ardehali et al., 2011; Herz et al., 2012) and the TRR SET domain was also shown to have H3K4 monomethyltransferase activity in vitro (Ardehali et al., 2011). We confirmed that purified GST-TRR-C421 (C-terminal 421 residues) has only a robust H3K4 monomethyltransferase activity (Fig. 3A, lane 2). A G2304S substitution, corresponding to the trx^Z11 mutation (supplementary material Fig. S5), also abolished TRR methylation activity (lane 4) and the Y2404F substitution at the TRR Phe/Tyr switch position also converts it to an H3K4 di- and trimethyltransferase (Fig. 3A, lane 5). We conclude that the TRR SET domain also has only an intrinsic H3K4 monomethyltransferase activity.

MLL1 and MLL4 are human orthologs of TRX, and MLL2/ALR and MLL3 are human orthologs of TRR (Ardehali et al., 2011; Chauhan et al., 2012) (supplementary material Fig. S5). We compared human MLL1-C225 with TRX-C251 in our HMT assay and found that MLL1-C225 also has a robust H3K4 monomethyltransferase activity (Fig. 3B, lanes 2 and 3) as previously reported (Patel et al., 2009). The SET domain of MLL2 also has only a robust H3K4 monomethyltransferase activity (Fig. 3C, lane 4), indicating that this activity has been evolutionarily conserved in the human TRX and TRR orthologs.

We previously showed that in vivo some H3K27 acetylation by CBP depends on TRX (Tie et al., 2009). Given the conflicting claims about the presence of CBP in TRX complexes, we re-examined the association of CBP with TRX (and TRR) by co-immunoprecipitation (co-IP) assay. TRX and MLL1 are cleaved by Taspase 1 into N- and C-terminal fragments that remain physically associated in vivo (Hsieh et al., 2003a; Hsieh et al., 2003b; Capotosti et al., 2007). Using antibodies that specifically recognize TRX-N, we found that TRX-N and TRR (Fig. 4A, lane 3) were present in anti-CBP IPs from embryo nuclear extract (NE). CBP was also present in reciprocal anti-TRX-C IPs (Fig. 4B, lane 4). We conclude that TRX and TRR are each independently physically associated with CBP in vivo.

To determine whether TRX and CBP interact directly, we initially used GST-TRX-C751 in a GST pulldown assay with embryo NE. CBP (but not BRM, UTX or TRR) was strongly pulled down by GST-TRX-C751 (Fig. 5, lane 3). As a positive control, SNR1, which is a subunit of Drosophila BRM (SWI/SNF) complexes that was previously shown to bind to the TRX SET domain (Rozenblatt-Rosen et al., 1998), was also pulled down (lane 3 at bottom). The trx^Z11 missense mutation (G3601S) in the SET domain did not abolish pulldown of CBP or SNR1 (lane 4). Interestingly, GST-TRR-C421 also pulled down SNR1 (Fig. 5A, lane 5) but not CBP.

To map the region of TRX-C751 that mediates its physical association with CBP, we tested a set of subfragments of TRX-C751 (see constructs in Fig. 5D). The C-terminal fragments TRX-C343 and C251, which contain the SET domain, failed to pull down CBP but retained SNR1 binding (Fig. 5B, lanes 3 and 4), suggesting that the deleted region (TRX2976-3383) is required for the interaction with CBP. Consistent with this, deletion of the SET domain from TRX-C751 severely impaired SNR1 binding but retained CBP binding (lane 5). TRX-ΔC416, which deletes both the SET domain and FYRC (Hsieh et al., 2003b) from TRX-C751, also did not affect the CBP interaction (Fig. 5C, lane 3). However, deletion of an additional 141 residues (TRX-ΔC557) abolished CBP binding (lane 2). We assayed the GST-TRX-C pulldown in Fig. 5 and found that it has both robust H3K4 monomethylation and H3K27 and H3K18 acetylation activities in vitro (supplementary material Fig. S6).

To further delimit the CID, we used an additional set of TRX fragments (bottom four constructs in Fig. 5D) in GST pulldown assays. TRX-ΔC467, a deletion of just 51 additional C-terminal residues from TRX-ΔC416, abolished CBP binding (Fig. 5C, lane 4). TRX3146-3310 retained CBP binding (Fig. 5C, lane 7) and was defined as the CID. Deletion of 25 (CIDΔ1) and 63 (CIDΔ2) residues from the N-terminus of CID respectively impaired and abolished CBP binding (lanes 5 and 6). GST pulldown results are summarized in Fig. 5D. We conclude that TRX3146-3310 is the CID.

To map the TRX-interacting region in CBP, we used GST-CBP fragments to pull down purified TRX CID in vitro. The TRX CID binds to the KIX domain (Fig. 6A, lane 5) but not to other conserved domains of CBP (lanes 3, 4, 6 and 7). These results are summarized in Fig. 6B.

We identified four KIX-binding motifs [ΦXXΦΦ (Lee et al., 2009; Wang et al., 2012)] in the TRX CID (Fig. 6C). The KIX domain has two distinct binding surfaces that can simultaneously bind two ΦXXΦΦ motifs (De Guzman et al., 2006; Wang et al., 2012) and this capability mediates cooperative binding of MLL1 and CREB to CBP (Ernst et al., 2001). As shown in Fig. 5C, deletion of motif 1 (CIDΔ1) reduced binding by more than two-thirds, and deletion of motifs 1 and 2 (CIDΔ2) or motifs 3 and 4 (ΔC467) abolished binding to the KIX domain. These data suggest that multiple KIX-binding motifs are required for robust binding of the TRX CID to CBP.

We previously showed that some H3K27 acetylation is dependent on TRX (Tie et al., 2009) and some has also recently been shown to depend on TRR (Herz et al., 2012). This suggests that monomethylation of H3 by TRX and TRR might influence H3K27 acetylation by CBP. To test this, we compared unmodified and K4me1-containing H3 as substrates for in vitro acetylation by recombinant CBP(1603-2678) (Tie et al., 2012). H3K27 (and H3K18) acetylation was enhanced 2- to 3-fold on the H3K4me1-containing substrate in this assay (Fig. 7A, compare lane 2 with lane 1). Other H3 modifications (K4me2/3, K9me1, K36me1) had no enhancing effect (Fig. 7B, lanes 2, 3, 5 and 6).

TRX overexpression in vivo elevates both H3K4me1 and H3K27ac levels (supplementary material Fig. S1B) and perturbs Polycomb silencing, causing homeotic derepression phenotypes characteristic of PcG mutants. Moderate RNAi-mediated knockdown of CBP, using two independent da-GAL4-induced UAS-CBP RNAi transgenic lines, suppresses the extra sex comb phenotype of TRX overexpressers (Table 2). This indicates that CBP is required for the impaired silencing induced by TRX overexpression. Together with the suppression of Pc³ phenotypes by trx^Z11 (Table 1), it also is consistent with the possibility that the elevated H3K4me1 in TRX overexpressers may antagonize silencing by locally enhancing H3K27 acetylation by TRX-associated CBP.

To test whether the H3K27ac level depends specifically on H3K4 monomethylation, we independently altered the H3K4me1 level by genetic manipulation of the Drosophila H3K4me1/2 demethylase LSD1 [encoded by the Su(var)3-3 gene] and examined H3K4me1 and H3K27ac levels in adults. As previously reported (Di Stefano et al., 2007), the H3K4me1 level was increased (by 30%) in protein-null lsd1^ΔN mutants compared with wild-type controls (Fig. 7C, top panel, lane 2) and decreased (by 25%) in LSD1 overexpressers (lane 3), while the H3K4me3 levels in both genotypes were little changed (Fig. 7C, third panel). The H3K27ac level was also increased in lsd1^ΔN mutants and decreased in LSD1 overexpressers (Fig. 7C, second panel), consistent with an enhancing effect of H3K4me1 on H3K27 acetylation.

Chemically inhibiting LSD1 activity had a similar effect on the H3K27ac level. Incubating Drosophila S2 cells for 3 days in media containing 40 μM and 80 μM bisguanidine polyamine analog, which inhibits human LSD1 in breast cancer cells (Zhu et al., 2012), significantly increased their H3K4me1 level, as expected, and also increased their H3K27ac level, while CBP and H3K4me3 levels were unchanged (Fig. 7D, lanes 4 and 5). Together, these results further suggest that the TRX- and TRR-dependence of the H3K27ac level might, at least in part, reflect dependence on the H3K4 monomethylation activities of TRX and TRR.

To determine the extent to which TRX, TRR, CBP, H3K4me1 and H3K27ac colocalize on chromatin, we used our previous ChIP-chip data for CBP, H3K27ac, H3K27me3 and PC (Tie et al., 2012) and recent TRX-C, TRR, H3K4me1 and H3K4me3 ChIP-seq data (Enderle et al., 2011; Herz et al., 2012) from Drosophila S2 cells and generated an unbiased heat map clustered by all peaks. As shown in Fig. 8A, strong TRX-C peaks are highly correlated with CBP and TRR peaks (left three columns) in regions highly enriched for H3K4me1 (clusters 1 and 2), consistent with the H3K4 monomethyltransferase activity of TRX and TRR and with the observed physical association of both TRX and TRR with CBP. The cluster 2 profile, containing CBP, H3K4me1 and H3K27ac (absence of H3K27me3), is similar to the signature of so-called ‘active’ enhancers. The cluster 1 profile, containing CBP, H3K4me1 and H3K27me3 marks (absence of H3K27ac), is similar to the signature of so-called ‘poised’ enhancers in mammals. Note that the H3K4me1 peaks in cluster 2 form broad domains spread over a much larger region than in cluster 1, and these are accompanied by similarly broad H3K27ac peaks, suggesting that this is a feature associated with genes in a transcriptionally active state. Known and presumed PREs (marked by very high sharp peaks of PC) and also enhancers of homeotic genes and other PcG-regulated genes are found in these two clusters (supplementary material Figs S7-S9). Some TRX binding sites correspond to functionally defined enhancers recently identified in a novel genome-wide screen by STARR-seq (supplementary material Fig. S2) (Arnold et al., 2013). TRX-C and TRR are also present at active promoters marked by very strong H3K4me3 and H3K27ac peaks (Fig. 8A, cluster 3), but their signal intensities are weaker than in clusters 1 and 2. These data suggest that TRX and TRR are associated predominantly with H3K4me1-marked cis-regulatory elements including PREs and enhancers.

We generated an additional heat map (Fig. 8B) from clusters 1 and 2 (TRX- and TRR-enriched, H3K4me1-marked enhancers and PREs) sorted by the intensity of TRX-C peaks, which reveals that TRX-C is highly correlated with PC peaks and a portion of TRR, CBP and H3K4me1 peaks (Fig. 8B, top). TRR peaks also coincide with CBP, H3K27ac and H3K4me1 peaks, but some of them lack TRX-C and PC peaks (Fig. 8B, bottom), suggesting that TRR regulates both PcG and non-PcG target genes (supplementary material Fig. S10) (Herz et al., 2012), whereas TRX only regulates PcG target genes. We propose a model for epigenetic regulation of PcG gene expression by TRX and CBP in Fig. 8C.

The major findings presented here are: (1) TRX and TRR are monomethyltransferases and together account for the bulk of the H3K4me1 in vivo; (2) the catalytic activities of both TRX and CBP are required to antagonize PcG silencing; (3) TRX and TRR are physically associated with CBP in vivo and TRX binds directly to the CBP KIX domain via a region that contains multiple KIX-binding motifs; (4) TRX and TRR colocalize genome-wide with H3K4me1 and CBP at PREs and enhancers; and (5) H3K4me1 enhances histone acetylation by CBP. Together, these data suggest that the primary target of TRX monomethyltransferase activity is not promoters but PREs and neighboring enhancers. They suggest a new model (Fig. 8C) for how TRX antagonizes Polycomb silencing, stimulates active enhancers, and establishes a cellular memory of active transcription. This differs significantly from the previous view that TRX trimethylates H3K4 (Petruk et al., 2012).

The evidence presented here indicates that the SET domains of TRX, TRR and their human orthologs possess intrinsic H3K4 monomethyltransferase activities (Figs 2, 3) and are prevented from being trimethyltransferases by the presence of the bulkier Tyr residue at their respective F/Y switch positions, as previously shown for MLL1 (Patel et al., 2009). Although these data do not rule out the possibility that TRX and TRR complexes might have some H3K4 trimethylation activity in vivo in some chromatin contexts, the reduced H3K4me1 and apparently normal H3K4me3 levels in the catalytically inactive trx^Z11 and trr³ mutants (Fig. 1) strongly suggest that H3K4 monomethylation is the predominant activity of TRX and TRR in vivo. Moreover, the absence of detectable H3K4me1 in trr³; trx^Z11 double-mutant embryos (Fig. 1E) suggests that they are the principal H3K4 monomethyltransferases in vivo, consistent with their genome-wide colocalization with H3K4me1 at PREs and enhancers. Suppression of Pc³ mutant phenotypes by trx^Z11 further suggests that the monomethyltransferase activity of TRX plays a role in antagonizing Polycomb silencing.

We found that TRX and TRR are physically associated with CBP in embryo extracts, confirming a previous report for TRX (Petruk et al., 2001). The direct binding of TRX to the CBP KIX domain (Fig. 6) and the genome-wide correlation of H3K27ac with H3K4me1 on active genes suggests that their activities are coupled in vivo. Consistent with this, TRX-CBP complexes pulled down from embryo extracts have both H3K4 monomethyltransferase and H3K27 acetyltransferase activities (supplementary material Fig. S6). Moreover, the impaired Polycomb silencing caused by TRX overexpression in vivo (which elevates both H3K4me1 and H3K27ac levels) requires CBP (Table 2) and presumably the TRX-CBP interaction. Mutating the CID will be required to show this conclusively. No direct interaction between CBP and the TRR C-terminus was found, but we previously reported that CBP interacts directly with the H3K27 demethylase UTX (Tie et al., 2012), which is another subunit of the TRR complex (Mohan et al., 2011). Together, these data suggest that these direct interactions are required for TRX- and TRR-dependent H3K27 acetylation and further suggest that TRX and TRR complexes function by fundamentally similar mechanisms.

The enhanced in vitro acetylation of H3K27 on K4me1-containing recombinant H3 substrates (Fig. 6) suggests that H3K4me1 might be a preferred CBP substrate in vivo. Consistent with this, altering the H3K4me1 level in vivo by manipulating LSD1 causes concordant changes in H3K27ac in adults (Fig. 7C). Moreover, a genome-wide analysis of hundreds of bona fide enhancers in purified mesodermal cells from Drosophila embryos revealed that H3K27ac is not present on enhancers without H3K4me1, whereas H3K4me1 is present without H3K27ac prior to enhancer ‘activation’ (Bonn et al., 2012). This suggests that the presence of H3K4me1 might be a prerequisite for the deposition of H3K27ac at enhancers. Interestingly, some of the catalytically inactive trx^Z11 mutants survive until the late pupal period and exhibit strong homeotic transformations. This suggests that TRX catalytic activity might be more important for stimulating enhancers that drive robust homeotic gene expression, whereas the physical association of TRX with CBP, which is intact in trx^Z11, is more important for preventing silencing of normally active PcG-regulated genes in the embryo.

H3K4me1 and CBP are part of a conserved chromatin ‘signature’ of enhancers and H3K27ac marks ‘active’ enhancers (Heintzman et al., 2007; Wang et al., 2008). Our data strongly suggest that TRX and TRR are responsible both for the H3K4me1 on enhancers and, via their physical association with CBP, for the H3K27ac on active enhancers. Determining which H3K4me1 is TRX dependent will require ChIP-seq analysis of trx^Z11 mutant cells.

Like TRX, H3K4me1 and CBP are also present at PRE/TREs of both active and inactive genes, suggesting that PRE/TREs have a functional connection to enhancers. Functional analyses of the strong bxd PRE/TRE in vivo suggest that PRE/TREs are distinct from enhancers, do not possess enhancer activity, but can boost enhancer-dependent transcription in a TRX-dependent manner. A GAL4-TRX fusion protein tethered to a transgene reporter exhibits these same properties (Pirrotta et al., 1995; Poux et al., 2002).

TRR was recently shown to occupy many presumed enhancers (Herz et al., 2012). We find that TRR binds more sites than TRX and also co-occupies most TRX binding sites genome-wide, including PRE/TREs. This raises the possibility that both TRX and TRR regulate many PcG-regulated genes, perhaps in different contexts or in response to different signals. The presence of UTX in the TRR complex suggests that TRR can facilitate switching of PcG-regulated genes from silent to active, whereas TRX might only be capable of maintaining the expression of genes activated prior to the onset of Polycomb silencing in the early embryo, or genes subsequently derepressed by the UTX activity associated with the TRR complex. This might explain the previously reported critical requirement for TRX in early embryogenesis (0-4 hours; i.e. prior to the onset of Polycomb silencing) for later robust expression of the homeotic genes in imaginal discs. Absence of TRX in 0- to 4-hour embryos cannot be compensated by its subsequent restoration (Ingham and Whittle, 1980; Poux et al., 2002). Further investigation will be required to determine whether and in what contexts there is functional collaboration or division of labor between TRX and TRR.

Although it is required continuously, the critical early requirement for TRX might provide an important clue to its function. This suggests that TRX and CBP, bound to PRE/TREs, might be required for de novo ‘priming’ of surrounding enhancers with H3K4me1 and H3K27ac in the early embryo (Bonn et al., 2012), prior to the onset of Polycomb silencing and perhaps even prior to transcriptional activation of the zygotic genome.

There is little detectable H3K27me3 in 0- to 4-hour embryos, whereas the H3K27ac level is already high relative to later embryonic stages (Tie et al., 2009). H3K4me1 is already present during syncytial stages (not shown). We speculate that before zygotic genome activation, TRX and CBP are constitutively bound to PRE/TREs and deposition of H3K4me1 and H3K27ac might initially be restricted to nucleosomes adjacent to PRE/TREs. Binding of activators to early-acting enhancers promotes spreading of H3K4me1 and H3K27ac from PRE/TREs across adjacent cis-regulatory regions to form broad domains, perhaps facilitated by interactions between activators and TRX/CBP complexes. Spreading of H3K27ac initially proceeds unchecked by H3K27me3, encompassing all surrounding enhancers, including those that will be ‘activated’ later (e.g. the imaginal disc enhancers) and protects them from subsequent deposition of H3K27me3 by PRC2 at the onset of Polycomb silencing. PcG-regulated genes that are not activated in the early embryo become subject to deposition/spreading of H3K27me3 in similar broad domains, blocking subsequent H3K27 acetylation. There might also be some active removal of pre-existing H3K27ac by PRC1/PRC2-associated RPD3 (Tie et al., 2009). Subsequent activation requires removal of H3K27me3 by UTX, and thus might require TRR, which is also present at PRE/TREs and so is poised to respond to the binding of TRR-dependent activators, such as EcR.

Other functions of H3K4me1 and H3K27ac at PREs and enhancers are not yet understood, but they might (1) recruit H3K4me1 and H3K27ac ‘readers’ that further stimulate/maintain the active transcriptional state, (2) facilitate the targeting of enhancers to promoters (Ong and Corces, 2011) and (3) perpetuate the broad domains of H3K4me1 and H3K27ac by enhancing their own deposition by TRX and CBP, as suggested by the enhancing effect of H3K4me1 on H3K27 acetylation in vitro. Perpetuation of the broad domains of H3K4me1 and possibly H3K27ac through replication and mitosis could also constitute the elusive cellular memory of past transcriptional activity.

Rabbit anti-TRX-N antibodies were generated against TRX residues 670-997. Rabbit anti-TRX-C antibodies were made previously (Chinwalla et al., 1995). Rabbit anti-CBP antibodies were generated against CBP residues 1-286 and purified on CBP(1-286)-coupled Sepharose. Rabbit anti-TRR and rat anti-SNR1 antibodies were kindly provided by A. Mazo (Sedkov et al., 2003) and A. Dingwall (Dingwall et al., 1995). Rabbit anti-H3K27ac (Abcam, ab4729), anti-H3K4me1 (Abcam, ab8895), anti-H3K4me2 (Abcam, ab11946), anti-H3K4me3 (Active Motif, 39159) antibodies were used in this study. Guinea pig anti-CBP and other antibodies were described previously (Tie et al., 2009; Tie et al., 2012).

The cDNA clones of TRX (LD39445) and TRR (SD13650) were obtained from the Drosophila Genomics Resource Center (DGRC). The SET domain-containing TRX-C751 (residues 2976-3726) (and others) and TRR-C421 (residues 2011-2431) were generated by PCR and inserted into modified pGEX-2T vector by NdeI and NsiI sites. Constructs pGEX-TRX(2976-3574), deleting the TRX SET domain (ΔSET), and pGEX-TRX(2976-3310) (ΔC416), were generated from pGEX-TRX-C751 by removing the C-terminal NcoI-NsiI and XbaI-NsiI fragments of TRX, then end-filling to produce blunt ends and recircularizing. Construct pGEX-TRX(2976-3169) (ΔC557) was generated from pGEX-TRX-C751 by removing the C-terminal EcoRI-EcoRI fragment of TRX and recircularizing. TRX mutations, trx^Z11 or G3601S (GGC/AGC), E3616S (GAA/TCA), Y3701F (TAC/TTC), the trx^Z11 corresponding substitution in TRR G2304S (GGC/AGC) and Y2404F (TAC/TTC) were generated using the QuikChange Site-Directed Mutagenesis Kit (Agilent). All mutations are confirmed by sequencing. PCR fragments coding wild-type histone H3 and mutant H3K4R were inserted into modified pET-11 vector (with a six-histine tag at C-terminus) by NdeI and NsiI sites.

Human MLL1 construct pGEX-MLL1-C225 (residues 3745-3969) was described previously (Patel et al., 2009) and pGEX-MLL2-C219 (residues 5319-5537) was generated similarly.

GST-TRX-C proteins (wild type and mutants) were induced by 0.2 mM IPTG for 2 hours at gradual decreasing temperature from 35°C to 28°C in BL-21 cells and were purified on Glutathione Sepharose 4B (GE Healthcare). Histone H3 and H3K4R (with a 6xHis tag at C-terminus) were induced by 0.5 mM IPTG for 2 hours in BL-21 cells and were purified on Ni-NTA His Bind Resin (Qiagen) in 6-8 M urea. Purified histones were dialyzed in PBS and stored at -80°C.

GST-TRX-C (1 μg) in 2× HMT buffer (Patel et al., 2009) (0.1 M Tris pH 8.2, 0.4 M NaCl, 6 mM DTT, 10 mM MgCl₂, 10% glycerol) was mixed with recombinant histone H3 (1-4 μg) and 0.5 mM S-adenosyl-methione (SAM) in a total volume of 30 μl. The reactions were incubated at 30°C (or 28°C) for 1 hour and stopped by adding one volume of 2× SDS sample buffer. H3 products were separated by SDS-PAGE (15%) and analyzed by western blot.

Drosophila S2 cells (DGRC) were cultured at 25°C in Schneider’s Drosophila Medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS). S2 cells were seeded in 25 cm² flasks with 5 ml medium. When they had grown to 50-60% confluency, LSD1 inhibitor (a bisguanidine polyamine analog from Calbiochem, catalog number 489476) was added to the medium at a final concentration of 10, 20, 40 or 80 μM. Cells were grown for 3 days and harvested for analysis of histone H3 modifications and proteins.

For TRX mutants and over-expressers, protein extracts were prepared from larval salivary glands of trx¹, trx^Z11, lsd1^ΔN mutants and UAS-lsd1/da-GAL4 and trx^EP3541/da-GAL4, which overexpress LSD1 and TRX (Tie et al., 2009). Homozygous trx^Z11 larvae from a trx^Z11/TM6B, Tb stock were identified as non-Tubby (Tb). Third instar larval salivary glands were dissected in PBS and homogenized in equal volumes of PBS plus 8 M urea and 2× SDS sample buffer. For SDS-PAGE, 5-10 μl of the extracts were used. Quantitative western analyses were performed (Tie et al., 2009) using various anti-H3 antibodies. A GE Typhoon Trio system was used for data acquisition. Data were exported to Excel and the ratios of signals from different histone modifications to total H3 level were calculated.

trr³ and trr¹ stocks were obtained from A. Mazo (Sedkov et al., 2003). UAS-lsd1 and lsd1^ΔN stocks were obtained from N. Dyson (Di Stefano et al., 2007). CBP-RNAi lines 1 and 2 were from S. Smolik (#32577) (Ludlam et al., 2002) and J. Kumar (Kumar et al., 2004). The GAL4-inducible trx^EP3541 line used for TRX overexpression and other stocks were obtained from the Bloomington Drosophila Stock Center (BDSC).

To determine whether the catalytic activity of TRX is required for antagonizing Polycomb silencing, we compared the suppression of the dominant extra sex comb phenotype of Pc³/+ males by the protein-null trx^B11 and the catalytically inactive trx^Z11 mutations. trx^Z11/TM6B and trx^B11/TM6B females were crossed to Pc³/TM3 males. Adult male progeny were scored for the presence of extra sex combs on the second thoracic (T2) leg, a Pc³ phenotype resulting from derepression of Scr in T2 and indicative of homeotic transformation of T2 to T1 leg.

To determine whether moderate RNAi knockdown of CBP suppresses the Polycomb phenotypes caused by GAL4-induced overexpression of TRX, trx^EP3541/trx^EP3541 females were crossed to wild-type males (control) and males from two homozygous CBP RNAi lines that target different portions of the CBP transcript: Hsp70-Gal4; CBP RNAi-1 and Hsp70-Gal4; CBP RNAi-2. Adult male progeny were scored for the presence of extra sex combs on the T2 leg. Crosses were at 29°C, which promotes substantial leaky expression of Hsp70-Gal4.

For LSD1 overexpression, UAS-lsd1 flies were crossed to da-GAL4 (BDSC stock 8641) and maintained continuously at 25°C.

trx^Z11, trr¹, trr³ single and trr; trx^Z11 double-mutant embryos were obtained from the following stocks: (1) trx^Z11/TM3, ftz-lacZ, (2) trr¹ or trr³/FM7c, [Kr-GAL4, UAS-GFP] and (3) trr¹ or trr³/FM7c, [Kr-GAL4, UAS-GFP]; trx^Z11/TM3, ftz-lacZ. Hemizygous trr/Y, homozygous trx^Z11 and trr³; trx^Z11 double-mutant embryos were identified by the absence of GFP, β-galactosidase (β-gal), or both. Anti-H3K4me1 and anti-H3K4me3 antibodies (1:1000 dilution) were used for embryo staining as previously described (Tie et al., 2009).

We thank A. Mazo, S. Smolik, J. Kumar, N. Dyson, A. Dingwall and the BDSC for fly stocks and antibodies; and Wei Wang and Geoff Parker for assistance with plasmid construction.