Setd5 is essential for mammalian development and the co-transcriptional regulation of histone acetylation

Embryogenesis requires dynamic and specific changes in gene expression patterns that involve both genetic and epigenetic mechanisms (Cantone and Fisher, 2013). Histone modifications, which include acetylation, methylation, phosphorylation and monoubiquitylation, represent one such epigenetic mechanism and play an essential role in regulating chromatin accessibility during transcription and genome maintenance (Kouzarides, 2007). Since nucleosomes present a physical barrier to gene transcription by RNA polymerase II (RNAP II), the co-transcriptional nucleosome disassembly and reassembly that occurs is not only highly dynamic but also involves rapid changes in specific histone modifications (Venkatesh and Workman, 2015).

Reversible histone lysine acetylations are installed by histone acetyltransferases (HATs) and removed by histone deacetylases (HDACs) (Verdone et al., 2005). While the acetylation of histones H3 and H4 has long been associated with transcriptionally active genes and open chromatin, more recent studies have established that histone methylations are also associated with actively transcribed genes, including histone H3 methylation at lysines 4, 36 and 79 (H3K4me, H3K36me and H3K79me) and histone H4 methylation at lysine 20 (H4K20me) (Smolle and Workman, 2013). Studies in yeast have shown that these histone modifications occur co-transcriptionally and require coordination between RNAP II, elongation complexes and regulators of histone modification (Tanny, 2014).

Polymerase-associated factor 1 complex (PAF1C), which contains Paf1, Ctr9, Leo1, Cdc73 and Rtf1 as core components, colocalizes to coding regions of genes and is involved in multiple processes during RNAP II transcription. These processes include RNA elongation, coordination between transcriptional activators and mRNA processing machinery, and recruitment and activation of histone modification factors (Crisucci and Arndt, 2011; Jaehning, 2010). Moreover, there is a crosstalk between histone methylation (at H3K4 and H3K36) and histone acetylation in the coding regions of genes, which involves methyl-lysine recognition by multiple HAT and HDAC complexes (Musselman and Kutateladze, 2011). These co-transcriptional processes, and the proteins that modulate them, are less well studied in mammalian species than in lower organisms.

Histone methylation is catalyzed by histone methyltransferases, which mediate the mono-, di- and tri-methylation of lysine ε-amine groups (Del Rizzo and Trievel, 2011). The majority of histone methyltransferases contain an evolutionarily conserved SET domain, which was originally identified based on similarities between Su(var)3-9, Enhancer of zeste and Trithorax proteins (Xiao et al., 2003). Mice that lack various SET domain-containing histone methyltransferases exhibit a wide range of developmental defects. For instance, animals lacking Setdb1, Setd8 (Kmt5a) or Nsd1 exhibit defects prior to implantation (Dodge et al., 2004; Oda et al., 2009; Rayasam et al., 2003); animals without Mll1 (Kmt2a), Ehmt2 or Setd2 die during mid-gestation (Hu et al., 2010; Tachibana et al., 2002; Yu et al., 1995); and mice lacking Setd7, Mll3 (Kmt2c) or Suv420h1/h2 (Kmt5b/c) are born but die shortly thereafter (Kurash et al., 2008; Lee et al., 2006; Schotta et al., 2008).

Our discovery of a lethal phenotype when we deleted a 5.16 kb fragment of Gt(ROSA)26Sor, a widely used locus for ubiquitous transgene expression (Soriano, 1999; Zambrowicz et al., 1997), prompted us to explore the function of the adjacent Setd5, a previously uncharacterized mammalian SET domain-containing gene. Orthologs of SETD5 include mammalian MLL5 (KMT2E), yeast Set3p and Drosophila UpSET, all of which contain a distantly conserved SET domain (Glaser et al., 2006). Interestingly, methyltransferase activity has not been associated with any of the SETD5 orthologs (Pijnappel et al., 2001; Rincon-Arano et al., 2012; Sebastian et al., 2009). However, Set3p, a key component of the yeast SET3C complex, has been shown to modulate gene transcription, repress sporulation and promote cytokinesis by altering histone deacetylation (Kim and Buratowski, 2009; Pijnappel et al., 2001; Rentas et al., 2012). Moreover, based on protein homologies, the SET3C complex has been suggested to be the yeast analog of the mammalian NCoR co-repressor complex that interacts with, and mediates the repressive activity of, unliganded nuclear receptors and other transcription factors (Pijnappel et al., 2001).

The mammalian NCOR1/NCOR2 (SMRT) co-repressors, which play a vital role in silencing gene expression, exist in a complex with histone deacetylase 3 (HDAC3) and the WD40 repeat-containing proteins TBL1X and TBL1XR1 (Guenther et al., 2000; Wen et al., 2000; Yoon et al., 2003). Consistent with a broad role in regulating gene expression, the NCoR co-repressor complex has been shown to be essential for the development of multiple organ systems, including the nervous system and heart (Jepsen et al., 2000, 2007), the maintenance of genome integrity (Bhaskara et al., 2010) and for metabolic regulation (Mottis et al., 2013). However, whereas yeast Set3p is a clearly established component of the SET3C complex, no SET domain-containing protein has, to date, been definitively established as part of the mammalian NCoR complex. Indeed, although MLL5 has been suggested to be a part of the NCoR complex, no direct evidence supporting this assertion has been reported (Kittler et al., 2007). Moreover, the phenotype of Mll5 knockout mice is surprisingly mild, with only mild hematological abnormalities and male sterility reported (Heuser et al., 2009; Yap et al., 2011).

We examined the role of Setd5 in mouse development and found that mice lacking Setd5 die by embryonic day (E) 10.5 due to cardiovascular defects, and that Setd5 is required both for normal cell cycle progression and the regulation of chromatin accessibility during transcription. Furthermore, we show that SETD5 interacts with components of the co-transcriptional PAF1 and the deacetylating NCoR co-repressor complexes and that, in the absence of Setd5, histone acetylation is not restricted to transcriptional start regions but is extended towards the 3′ end of the gene. Moreover, mouse embryonic stem cells (mESCs) lacking Setd5 have a dysregulated transcriptome, including altered expression of myogenesis and vasculogenesis genes. Finally, we show that Setd5 and ROSA26 are divergently transcribed from a bidirectional promoter, and that ROSA26, a noncoding mRNA, may help maintain Setd5 expression.

Mice that lack 5.16 kb of DNA containing the promoter and exon 1 of ROSA26 (Chen et al., 2011) were observed to be homozygous lethal. Examination of the deleted region revealed that the Setd5 and ROSA26 genes were in close proximity and in opposite orientation, with their respective transcription start sites (TSSs) separated by only 393 bp (Fig. S1). Thus, we sought to determine whether the ROSA26 deletion also impaired the expression of Setd5. To assess the function and sites of Setd5 expression in a manner that did not adversely affect ROSA26 expression, we derived mice that placed GFP coding sequences in the first exon of Setd5, upstream of the translation initiation sequence, using an RMCE-based strategy (Fig. S2A,B). The disruption of full-length Setd5 mRNA and protein expression was confirmed by qPCR, northern and western blot (Fig. S2C-E), confirming that Setd5^GFP is a null allele.

To identify sites of Setd5 expression during embryogenesis and in adult animals, we performed fluorescent stereoscopy on E8.0 to E15.5 embryos and on tissues from adult animals that were heterozygous for the Setd5^GFP allele. Fluorescence was observed in all embryos and adult tissues (Fig. 1B, Fig. S3) in a pattern that closely mimicked that described for ROSA26 (Soriano, 1999). RT-qPCR analysis confirmed that Setd5 and ROSA26 are co-expressed in all tissues (Fig. 1A). By generating double-heterozygous mice (Setd5^GFP; ROSA26^Cherry) that express GFP from the Setd5 promoter and mCherry from ROSA26 (Chen et al., 2011), we observed very similar expression patterns for both fluorescent proteins (Fig. 1B, Fig. S4).

To confirm the bidirectionality of the Setd5/ROSA26 (SR) promoter, we inserted it into a bidirectional dual luciferase reporter plasmid, which was transfected into a mouse cell line. This revealed robust transcriptional activity in both directions (Fig. 1C), which, together with the tissue co-expression data described above, convincingly indicates that Setd5 and ROSA26 are transcribed in opposite directions from the same promoter, making them a co-transcribed divergent transcript pair.

Bidirectional promoters that produce a divergent pair comprising a protein-coding and a noncoding RNA have been suggested to serve as a coupling regulatory unit in which transcriptional activity in the noncoding RNA direction, or the transcript itself, modulates expression of the protein-coding gene (Wei et al., 2011). To test whether the noncoding ROSA26 transcript modulates Setd5 expression, we examined the effects of Setd5 knockout and of several different ROSA26 knockout alleles (Fig. 2A) on the expression of both transcripts by RT-qPCR analysis of RNA isolated from E9.5 embryos. When Setd5 mRNA expression was disrupted, ROSA26 expression was not significantly altered (Fig. 2B). When ROSA26 mRNA was disrupted at the beginning of the first exon, as in ROSA26^Cherry/+ and ROSA26^{Cherry/Cherry} mice, there was a decrease in Setd5 expression (P≤0.01; Fig. 2C). However, in ROSA26^{LSL.YFP/LSL.YFP} mice, in which the first exon of ROSA26 is transcribed and spliced but a full-length transcript is not produced, Setd5 expression was not altered (Fig. 2C). When the entire ROSA26 transcript and 181 bp of preceding promoter sequence were deleted, as in the ROSA26^228.3TF.GFP allele, Setd5 transcription was decreased by about half and two-thirds in heterozygotes and homozygotes, respectively, compared with the wild-type allele (P≤0.01; Fig. 2D). Transfection of a bidirectional dual luciferase reporter driven by a similarly truncated version of the Setd5/ROSA26 promoter (SRΔ181) also exhibited a decrease in luciferase activity in both directions as compared with the full-length promoter (P≤0.01; Fig. 2E).

Taken together, these results suggest that both the ROSA26 promoter and first exon must be intact for robust transcription of ROSA26 to occur and for normal levels of Setd5 mRNA expression. However, Setd5 expression does not depend on the presence or absence of a full-length ROSA26 transcript.

Since live-born Setd5 null pups were not observed from heterozygous matings, we isolated embryos at E8.5 to 12.5 and observed no differences between wild type and heterozygotes. No viable null embryos were observed after E10.5 (Table S2) and all E10.5 embryos were underdeveloped compared with their heterozygous littermates, with many exhibiting widespread hemorrhage (Fig. S5B-D). Moreover, at E9.5, all null embryos were smaller, had fewer somites, and many had neural tubes that were not closed (Fig. 3A,B). Embryonic patterning defects also included massive pericardial effusions (or swelling) and defects in the formation of caudal structures, with the presence of tail effusions, suggesting disrupted blood flow and osmotic imbalance, and pointing to a defect in cardiovascular development. Developmental impairments were already noticeable at E8.5 (Fig. S5A).

To determine the basis of the failure of the Setd5 null embryos to thrive we examined rates of cellular proliferation and apoptosis in tissues at E9.5 (Fig. 3C-E). BrdU incorporation experiments revealed a low labeling efficiency in knockout embryos (data not shown), possibly due to impaired tissue perfusion. Ki67 staining showed a decrease in cell proliferation rates from 96% to 87% in the heterozygous versus knockout tissues (P≤0.01). At the same time, the number of apoptotic cells, as measured by the presence of activated caspase 3, was increased from 4% to 25% (P≤0.01).

We further explored the cardiovascular defects and observed that only about half of Setd5 null embryos had beating hearts at E9.5. Defects in heart development were further confirmed by whole-mount staining for two markers of cardiac differentiation: MEF2C, a transcription factor important for anterior heart field development (Lin et al., 1997); and myosin heavy chain (MHC), a structural protein enriched in cardiac muscle (Fig. S6A). At E9.5, MEF2C staining of heterozygous embryos showed a well-developed pharyngeal mesoderm/anterior heart field, cardiac muscle in a looped heart tube, and somites. By contrast, Setd5 null embryos had swollen, underdeveloped hearts with only weak staining in the anterior heart field and myocardium. Similarly, whereas MHC staining of Setd5^GFP/+ mice showed easily discernible cardiac muscle, and somites and smooth muscle in the dorsal aorta, only the heart region was stained in the knockout embryos. At a histological level, the ventricular chambers exhibited reduced trabeculation in the existing single ventricle, a thinner myocardium layer, a large gap between pericardium and myocardium, and no contact between the myocardium and endocardium (Fig. S6B).

Setd5 mutant embryos also exhibited a severe vascular phenotype in the extraembryonic tissues and embryo bodies (Fig. 4A). At E9.5, the knockout embryos had a pale yolk sac with fewer blood vessels. Whole-mount staining for the endothelial marker PECAM1 revealed that whereas heterozygous yolk sacs exhibited a well-structured hierarchical organization of large and small vessels, the yolk sacs of Setd5 null mice had a very poorly organized and abnormally dilated capillary plexus. Whole-mount staining for CD41 (ITGA2B), a marker for early hematopoietic cells, was observed in blood islands of both heterozygous and knockout yolk sacs, suggesting that definitive hematopoiesis occurred in the null embryos but that blood vessels were distended and contained fewer blood cells (Fig. S7). Vascular abnormalities in placental development might also contribute to embryonic lethality. In support of this, Hematoxylin and Eosin (H&E) staining of E9.5 placentas showed that the labyrinthine layer in heterozygous animals was well developed (Fig. 4B), whereas in knockout placentas this layer was significantly thinner, contained only maternal blood vessels with mature erythrocytes, and embryonic blood vessels remained at the periphery in the chorioallantoic region and did not invade the labyrinthine layer.

To test whether the cardiovascular defects might be due to differences in protein expression, we performed immunofluorescent staining of E9.5 wild-type embryos for endogenous SETD5 protein. We found nearly uniform expression of SETD5, with protein localization being predominantly nuclear; in particular, SETD5 expression in PECAM1⁺ and MHC⁺ cells was similar to its expression in other surrounding cells (Fig. S8A,B).

To further characterize the role of Setd5 in cell growth, we derived wild-type, Setd5 heterozygous and null mESC lines. Both the Setd5^GFP/+ and Setd5^GFP/GFP lines exhibited strong green fluorescence (Fig. S9A), but the Setd5^GFP/GFP cells exhibited slower growth rates than wild-type or heterozygous cells. Measurements of BrdU incorporation (Fig. 5A) revealed a slight decrease in cell proliferation rates in the knockout mESCs compared with the heterozygous cells (from 70% to 58%, P≤0.05), with no other differences being observed between wild-type and heterozygous cells. FACS analysis of annexin V-stained cells showed a greater than 2-fold increase in apoptosis in Setd5^GFP/GFP compared with Setd5^GFP/+ cells (P≤0.01; Fig. 5B). These results suggest that both reduced proliferation and elevated apoptosis contribute to the reduced growth rate of Setd5 null mESCs.

Next, we examined cell cycle progression in the two cell lines by FACS analysis after propidium iodide staining (Fig. 5C). Multiple defects in cell cycle progression were observed in the absence of Setd5, with a delay in the G2/M transition being the most pronounced. The knockout cells had shorter G1 and S phases (reduced from 29% to 26% and from 34% to 30% of cells, respectively; P≤0.05) and an increase in G2 phase (from 30% to 34% of cells; P≤0.01). Additionally, we observed an increase in the percentage of polyploid cells (3% for wild-type, 6% for Setd5^GFP/+ and 9% for Setd5^GFP/GFP mESCs). This observation was further supported by an increase in the percentage of cells displaying visible chromosomes (mitotic index) when nuclei were stained with DAPI (Fig. 5D,E). Although some mitotic cells showed a metaphase plate of chromosomes, many others had two sets of segregating chromosomes suggesting cellular arrest at anaphase stage and failure to complete cytokinesis. In addition, examination of the Setd5 null mESCs revealed that many had enlarged and misshapen nuclei (Fig. 5D).

To determine whether the absence of Setd5 affects the differentiation potential of mESCs, we performed directed differentiation towards cardiac progenitors through embryoid body (EB) formation in hanging drops following serum stimulation. Both Setd5^GFP/+ and Setd5^GFP/GFP mESC lines formed EBs on day 2, grew proportionally in size to day 4, and attached to the plate and formed outgrowths by day 8 of the differentiation protocol (Fig. S9B). However, only a very small number of spontaneously contracting EB outgrowths were observed in knockout cells (3% of total EBs) compared with heterozygous cells (48%). Similarly, staining for the cardiomyocyte marker MHC confirmed that there were fewer MHC-positive cells in Setd5^GFP/GFP than in Setd5^GFP/+ outgrowths, indicating that the ability of Setd5 null mESCs to differentiate into cardiomyocytes was drastically reduced (Fig. 5F,G).

To analyze global changes in gene expression profiles in the absence of Setd5 we performed RNA-seq on RNAs isolated from Setd5^GFP/+ and Setd5^GFP/GFP mESC lines. Differential expression analysis revealed that of the 22,388 expressed genes in Setd5 null mESCs, 1292 were upregulated and 1684 were downregulated compared with control cells (adjusted P-value ≤0.01; Table S3). Gene ontology analysis of the top upregulated genes showed an enrichment for genes involved in embryonic morphogenesis and pattern specification processes, including mesodermal transcription factors (Mixl1, Mesp1, Gsc, T), blood vessel morphogenic genes (e.g. Flt1, Sox17), Wnt pathway genes, growth factors and genes linked to apoptosis (Fig. 6A,B). A similar analysis of the top downregulated genes suggested they were involved in myosin complex and contractile fiber function (Smtnl1, Myh13) and myocyte differentiation (Meox1, Six1). Genes involved in stem cell maintenance (Klf4, Nanog, Sox2) were also downregulated. These findings indicate that Setd5 null mESCs have a markedly dysregulated transcriptome and impaired stem cell identity. Interestingly, the expression of Mesp1 and Meox1, two genes that might contribute to the cardiovascular phenotype, were also markedly altered. This finding was validated by RT-qPCR analysis using RNA from mESCs and embryos at E8.5 (Fig. 6C).

To analyze whether Setd5 has an effect on global chromatin state or is solely required for major histone lysine methylation marks, we performed immunoblot analysis on protein lysates from Setd5^GFP/+ and Setd5^GFP/GFP mESCs using antibodies that recognize specific modifications (Fig. 7A). H3K4me1, -me2, -me3 and H3K79me3 marks remained similar between the Setd5^GFP/+ and Setd5^GFP/GFP samples, whereas pan-H3 acetylation, H3K36me1, -me2, -me3, H4K20me3, H3K9me1, -me2 and -me3 marks were slightly increased in cells lacking SETD5. Conversely, H3K27me2 and -me3 were decreased in the Setd5 knockout mESCs (Fig. 7B). These results suggest that, in the absence of SETD5, there is an imbalance in specific chromatin modifications known to play essential roles in both activating and repressing gene expression. However, we failed to observe the total loss of any individual methylation, suggesting that SETD5 may lack methyltransferase activity.

Given the changes observed in chromatin state and prior studies demonstrating a role for SETD5 orthologs in chromatin regulatory complexes, we sought to determine the protein interaction partners of SETD5 using mass spectrometry. We immunoprecipitated FLAG-tagged SETD5 from transfected HEK 293T cells. MALDI-TOF mass spectrometry analysis detected 153 proteins that were pulled down by the FLAG-SETD5 protein but not from a control eluate (Table S4). Inspection of the most frequently detected proteins revealed that FLAG-SETD5 co-precipitated with four out of five known members of the co-transcriptional PAF1C: PAF1, LEO1, CTR9 and CDC73. Moreover, NCOR1, a member of an HDAC complex, also co-precipitated with FLAG-SETD5. Interestingly, the glycosylating enzyme O-linked N-acetylglucosamine (GlcNAc) transferase (OGT) was also co-precipitated, suggesting that SETD5 might be glycosylated, similar to its ortholog MLL5 (Fujiki et al., 2009). Other co-precipitated proteins included multiple RNAP II subunits and elongation factors, as well as many proteins known to be involved in chromatin remodeling, RNA processing, mitosis and spindle assembly, DNA replication and repair, mitotic checkpoint control and chromosome condensation and cohesion (Table S4).

To validate the key mass spectrometry results we performed anti-FLAG immunoprecipitations using FLAG-SETD5-expressing cell lysates followed by western blot analysis. The results confirmed that SETD5 co-immunoprecipitates with CTR9, PAF1, CDC73, OGT, RUVBL1 and NCOR1 (Fig. 7C). In addition to confirming interactions between SETD5 and NCOR1, we also showed that two other members of the NCoR complex, TBL1X and HDAC3, co-immunoprecipitate with SETD5. Finally, we performed reverse co-immunoprecipitations with antibodies against CTR9, PAF1 and NCOR1 and found, again, that these proteins co-precipitated with FLAG-SETD5 (Fig. 7D). Moreover, endogenous SETD5 protein partially colocalizes with HDAC3 protein in NIH3T3 cells (Fig. S8D). Together, these data indicate that SETD5 exists in close association with multiple members of the co-transcriptional PAF1C as well as with the HDAC-containing NCoR complex.

Given that SETD5 orthologs have been shown to be involved in co-transcriptional chromatin deacetylation, a process that increases the productive transcription of active genes by suppressing off-target transcription (Kim et al., 2012; Rincon-Arano et al., 2012), we sought to determine whether SETD5 has a similar function. We analyzed the chromatin acetylation status of three constitutively expressed genes using anti-acetylated histone chromatin immunoprecipitation (ChIP) followed by PCR with primers targeting different regions within and downstream of the promoter region (Fig. 8). The genes selected for this analysis were eukaryotic translation elongation factor 1 alpha 1 (Eef1a1), hydroxyacyl glutathione hydrolase (Hagh) and inositol hexaphosphate kinase 1 (Ip6k1). Eef1a1 is a housekeeping gene, whereas Hagh and Ip6k1 are two metabolic genes that have antisense transcripts near or within the gene locus and whose yeast orthologs were analyzed in studies of Set3p (Kim et al., 2012).

ChIP experiments were performed using chromatin isolated from Setd5^GFP/+ and Setd5^GFP/GFP mESC lines, as well as Setd5^GFP/GFP cells in which Setd5 expression was rescued through overexpression of FLAG-SETD5 protein under the control of the human EEF1A1 promoter. In the absence of Setd5, chromatin acetylation in all three genes was markedly increased in promoter proximal regions and remained higher in downstream regions than in the control or rescued mESCs (Fig. 8). For example, histone H3 acetylation at Eef1a1 was higher upstream of the TSS (region I, P≤0.05), at the TSS (region II, P≤0.01) and downstream of exon 1 (region III, P≤0.01) (Fig. 8A,B). The expression of Eef1a1 mRNA was also increased (P≤0.05; Fig. 8C). Similar differences were observed in both the Hagh and Ip6k1 genes (Fig. 8D,E,G,H). Interestingly, expression of Fahd1, the gene divergently transcribed with Hagh, was higher in the Setd5 null mESCs (P≤0.01; Fig. 8F). Similarly, expression of the antisense Gm38134 transcript, the promoter of which lies within the Ip6k1 gene, was also increased (Fig. 8I).

Taken together, these data suggest that SETD5 contributes to the recruitment of HDAC to locations near TSSs so as to remove chromatin acetylation marks as RNAP II proceeds to the elongation stage and moves towards the downstream region of genes.

In this study we establish that the mouse Setd5 gene plays a vital role in early development, cell replication and the co-transcriptional regulation of chromatin accessibility, possibly through interactions with protein components of the PAF1 and NCoR complexes.

Mouse embryos that lack Setd5 exhibit severe developmental delay, vascular abnormalities in the embryo and placenta, reduced cellular proliferation and increased apoptosis, and do not survive past E10.5. Although the phenotypic changes might be due to stalled development in the mutants, our findings are consistent, nonetheless, with a widespread impairment in the regulation of gene expression, as might be expected to occur with altered chromatin dynamics during transcription. Similar phenotypes with mid-gestation lethality and vascular abnormalities have been described for knockouts of several other SET domain-containing proteins (Hu et al., 2010; Jones et al., 2008), as well as enzymes regulating histone acetylation such as CBP/p300 (Tanaka et al., 2000; Yao et al., 1998).

Through likely interactions with the PAF1 and NCoR complexes, SETD5 may be involved in modulating gene transcription and, consequently, the development of multiple cell lineages. Differential RNA-seq expression profiling showed that the expression of more than 10% of mapped genes, including lincRNAs, antisense transcripts and pseudogenes, was dependent on Setd5. In mESCs lacking Setd5, upregulated genes included those important for specifying cellular lineages and for embryonic morphogenesis, gastrulation, pattern specification and regionalization. At the same time, pluripotency genes were downregulated, suggesting that Setd5 might also contribute to the maintenance of pluripotency, a finding that is consistent with the previously suggested roles for PAF1C in maintaining ESC identity (Ding et al., 2009) and for lineage specification at the blastocyst stage (Zhang et al., 2013).

Among the top differentially expressed genes in Setd5-deficient mESCs were those involved in myocyte and vascular lineage specification and function. For example, Mixl1, Mesp1 and Flt1 were markedly upregulated, whereas Meox1 and Six1 were downregulated, suggesting that their dysregulation might contribute to the cardiovascular phenotype. Similarly, different members of the zebrafish PAF1C have been shown to be important for cardiac and neural crest specification (Nguyen et al., 2010), somite segmentation (Akanuma et al., 2007) and heart tube cardiomyocyte specification (Langenbacher et al., 2011).

Impaired cell proliferation and survival of Setd5-deficient embryonic tissues and mESCs could be due to several factors. First, transcriptome profiling revealed that in the absence of Setd5 there is an increase in the expression of pro-apoptotic genes and a decrease in expression of genes involved in M-phase chromosome segregation. Second, dysregulation of histone modifications facilitated by PAF1C and NCoR could also play a role. HDAC3 is known to be involved in the control of cell cycle progression and, similar to the defects that we observed in Setd5 null mESCs, depletion of HDAC3 decreases the number of cells in S phase and leads to an accumulation of cells in G2/M phase (Bhaskara et al., 2008; Wilson et al., 2006).

The increase in cell cycle arrest during mitosis observed in cells lacking Setd5, particularly during cytokinesis, is consistent with findings that the NCoR complex localizes near the mitotic spindle, where its HDAC activity enhances mitotic progression (Ishii et al., 2008; Li et al., 2006). Similarly, human PAF1 has been shown to localize to centromeres and promote mitotic spindle formation (Moniaux et al., 2009). Moreover, mass spectrometry analysis indicates that SETD5 is capable of interacting with a number of proteins involved in chromosome cohesion and mitotic spindle assembly at the kinetochore, suggesting the direct involvement of a SETD5-containing complex in these processes.

We also found that SETD5 co-precipitates with four members of the mammalian co-transcriptional PAF1C. RNAP II is known to associate with PAF1C at or near promoters so as to maintain an association during gene transcription, and then to detach near the poly(A) site (Jaehning, 2010). Consistent with the involvement of PAF1C in multiple transcription-related processes, including facilitating the RNAP II transcription elongation machinery and the recruitment of mRNA 3′ end-processing factors, mass spectrometry analysis of SETD5-containing protein complexes identified RNAP II subunits, elongation factors, chromatin remodeling proteins and proteins involved in RNA 3′ end processing.

Our most intriguing finding among the proteomic data is, however, the interaction of SETD5 with the HDAC-containing complex NCoR. This suggests that SETD5 might be necessary for the recruitment of this chromatin-modifying complex to elongating RNAP II through interaction with PAF1C. Indeed, studies in yeast have shown that PAF1C is involved in the regulation of transcription-coupled histone modifications by promoting histone H2BK123 monoubiquitylation and recruitment of histone methyltransferase complexes that methylate histone H3K4, K36 and K79, a subset of modifications primarily associated with transcribed genes (Crisucci and Arndt, 2011; Tanny, 2014). H3K4 and K36 methylation, which modulates histone acetylation by facilitating the recruitment or activity of HAT and HDAC complexes, is also influenced by PAF1C. For instance, the yeast SETD5 ortholog Set3p is recruited near TSSs of genes through binding of its PHD domain to H3K4me2 residues. Moreover, the loss of Paf1, just as with the loss of Set3p, results in increased acetylation close to the TSS at the 5′ ends of genes (Kim and Buratowski, 2009; Chu et al., 2007). However, unlike Set3p, SETD5 appears to lack a methyl lysine-recognizing PHD domain, and might instead be recruited to transcribed genes through direct interaction with PAF1C. In turn, PAF1C may directly associate with RNAP II and/or use its different subunits to recognize native histones or post-translational histone modifications in order to facilitate further histone modification (Chu et al., 2013; Wu and Xu, 2012).

Interestingly, the lack of Setd5 in mESCs did not lead to complete loss of any of the major histone methylation marks tested. This suggests that SETD5, like its orthologs, might not possess a histone methyltransferase activity. Consistent with SETD5 being recruited with PAF1C and NCoR to actively transcribed genes, Setd5 null mESCs have slightly increased H3K36 methylation, a co-transcriptional histone modification that is enriched over the coding region of genes. Moreover, we did not observe any global changes in H3K4 methylation, which is typically associated with active promoter regions, suggesting that SETD5/HDAC3 repression might mostly target transcriptional elongation, as does the yeast SET3C complex (Weinberger et al., 2012). In mammals, several HATs and HDACs have been shown to be targeted to the transcribed regions of active genes by phosphorylated RNAP II (Wang et al., 2009).

The Drosophila ortholog of SETD5, UpSET, has also been found to be localized to transcriptionally active genes and to be specifically enriched at TSSs (Rincon-Arano et al., 2012). Both UpSET and Set3p have been shown to reset chromatin by removing acetylation marks on histones after the RNAP II machinery has passed into an elongation stage and to restrict active histone modifications to promoter regions (Kim et al., 2012; Rincon-Arano et al., 2012). These repressive actions of UpSET may reduce the probability of unmasking off-target genes or cryptic TSSs. Our data suggest that SETD5 might play a similar role in mammalian cells, since, in the absence of SETD5, there is an increase in chromatin acetylation that has spread from promoter regions downstream into the 5′ to mid regions of genes (Fig. S10).

Finally, we have shown that Setd5 and ROSA26 are co-expressed from a bidirectional promoter. Although the biological significance of divergent noncoding RNA transcripts is often elusive, it has been proposed that noncoding RNAs may regulate the transcription of the partner gene by helping to maintain a permissive chromatin environment (Kornienko et al., 2013). Such might be the case with the noncoding ROSA26 transcript, which was discovered as a result of a retroviral-mediated promoter trap but whose functional significance has long been uncertain (Zambrowicz et al., 1997). Disruption of the ROSA26 transcript by a splice acceptor trap in the intron of the gene, as occurs in ROSA26^LSL.YFP and other such alleles, did not affect the expression of Setd5, suggesting that the noncoding ROSA26 mRNA does not regulate Setd5 expression. However, when the ROSA26 locus is disrupted at the beginning of the first exon, or when adjacent sequences within the bidirectional promoter region are removed, Setd5 expression is decreased together with ROSA26 expression. These findings suggest that transcription in the ROSA26 direction might aid in creating an active and accessible chromatin environment so as to assure strong and ubiquitous expression of the functionally essential Setd5 gene. Such a notion is supported by a genome-wide modeling study in which divergent noncoding RNAs were predicted to be transcribed from promoters of essential genes in order to help maintain consistent levels of expression (Wang et al., 2011) and by the finding that chromatin signatures of active transcription are over-represented in regions around bidirectional promoters (Yang and Elnitski, 2008).

Our findings indicate that SETD5 plays a vital role in mammalian embryonic development, cell cycle progression and the co-transcriptional regulation of histone acetylation. Similar to processes that have been described in both yeast and Drosophila, the co-transcriptional regulation of histone acetylation by SETD5 may be necessary for the chromatin state of a gene to be reset after the transcription machinery has passed, in order to suppress spurious transcription initiation.

The following mouse lines were utilized: Gt(ROSA)26Sor^tm1(EYFP)Cos (ROSA26^LSL.YFP) (Srinivas et al., 2001), Gt(ROSA)26Sor^tm2Mgn (ROSA26^mCherry) (Chen et al., 2011), ROSA26^228.3TF [unpublished, contains a ʻ−228' deletion of the ROSA26 promoter similar to that described by Serup et al. (2012)] and Setd5^GFP. All experimental procedures were reviewed and approved by the Vanderbilt University Institutional Animal Care and Use Committee.

An exchange vector for performing recombinase-mediated cassette exchange (RMCE) was made beginning with pRosa26.Ex1 and inserting a GFP (Emerald)-SV40 poly(A) cassette (gift from David Piston, Vanderbilt University) between SmaI sites located in the first exon of Setd5. RMCE was performed as previously described in ROSA26^LCA mESCs (Chen et al., 2011) to generate Setd5^GFP/+ mESCs, from which mice were then derived. The resulting Setd5^GFP mouse line was backcrossed to, and maintained in, a heterozygous state on the CD-1 background (≥95%). Genotyping was performed by PCR using the primers detailed in Table S1.

Timed matings were performed, where noon of the day of the vaginal plug was considered to be embryonic day (E) 0.5. Embryos at E8.5-E11.5 were dissected out for whole-mount or section H&E staining or immunostaining and imaging as detailed, including a full description of antibodies, in the supplementary Materials and Methods. RNA was isolated from E9.5 embryos for northern blot analysis with an exon 2-6 fragment of Setdb5, or for RT-qPCR using the primers listed in Table S1 with Actb as an internal control. For details, see the supplementary Materials and Methods.

Cell proliferation was assessed in E9.5 embryos or mESCs by BrdU staining as described in the supplementary Materials and Methods.

Setd5^GFP/+ and Setd5^GFP/GFP mESCs were derived from blastocysts cultured in standard mESC media containing MEK1 inhibitor. Directed differentiation into cardiomyocytes was achieved by EB formation and subsequent culturing in high-serum media. Cell cycle progression and apoptosis were assessed in mESCs by FACS after propidium iodide or annexin V staining, respectively. For details, see the supplementary Materials and Methods.

RNA-seq of heterozygous and knockout Setd5 mESCs was performed on an Illumina HiSeq3000, with differential gene expression and Gene Ontology analysis of the top dysregulated genes as described in the supplementary Materials and Methods.

The dual luciferase assay and associated constructs used to assess promoter bidirectionality in transfected mESCs are described in the supplementary Materials and Methods.

Western blots of protein lysates from embryos or mESC lysates to detect histone modifications and anti-FLAG immunoprecipitations from HEK 293T cells to determine FLAG-SETD5 co-immunoprecipitates are described in the supplementary Materials and Methods.

Chromatin from 10⁷ mESCs was subject to ChIP analysis as described in the supplementary Materials and Methods.

FLAG peptide eluates from three immunoprecipitation experiments from FLAG-SETD5-transfected and mock transfected cells were combined and analyzed by multidimensional protein identification technology (MudPIT) as previously described (Martinez et al., 2012).

P-values were calculated using two-tailed Student's t-test for two samples with equal variance.

We thank Jean-Philippe Cartaillier for computational analysis of RNA-seq data; Judsen Schneider and Changqing Zhang for deriving the ROSA26^228.3TF.GFP mice; Katherine Boyer and Anne McGough for technical assistance; Jennifer Stancill and Hannah Clayton for critically reading the manuscript; and Scott Hiebert and Roland Stein for helpful suggestions.