Methionine Adenosyltransferase II-dependent Histone H3K9 Methylation at the COX-2 Gene Locus*

Background: MATII biosynthesizes AdoMet, which supplies methyl group for methylation of molecules, including histone. Results: MATII interacts with histone methyltransferase SETDB1 and inhibits COX-2 gene expression. Conclusion: AdoMet synthesis and histone methylation are coupled on chromatin by a physical interaction of MATII and SETDB1 at the MafK target genes. Significance: MATII may be important for both gene-specific and epigenome-wide regulation of histone methylation. Methionine adenosyltransferase (MAT) synthesizes S-adenosylmethionine (AdoMet), which is utilized as a methyl donor in transmethylation reactions involving histones. MATIIα, a MAT isozyme, serves as a transcriptional corepressor in the oxidative stress response and forms the AdoMet-integrating transcription regulation module, affecting histone methyltransferase activities. However, the identities of genes regulated by MATIIα or its associated methyltransferases are unclear. We show that MATIIα represses the expression of cyclooxygenase 2 (COX-2), encoded by Ptgs2, by specifically interacting with histone H3K9 methyltransferase SETDB1, thereby promoting the trimethylation of H3K9 at the COX-2 locus. We discuss both gene-specific and epigenome-wide functions of MATIIα.

Methionine adenosyltransferase (MAT) 2 catalyzes the synthesis of S-adenosylmethionine (AdoMet) (1). AdoMet is an intermediate product in the methionine cycle (2) and is utilized as a methyl donor in the transmethylation of diverse substrates, including histones, by specific methyltransferases (3). MAT is present in all living organisms, and three isozymes of MAT are known in mammals (1,4). MATII is composed of ␣ and ␤ subunits and is widely expressed (1). The catalytic subunit MATII␣ is encoded by the MAT2A gene, and its catalytic activity is enhanced or inhibited by the regulatory subunit MATII␤, which is encoded by the MAT2B gene (5,6).
The transcription factor Bach1 represses genes related to heme function and metabolism such as globin and heme oxygenase-1 (HO-1) by forming heterodimers with the small Maf oncoproteins (7,8) MATII␣ serves as a transcriptional corepressor of MafK-Bach1 (9). Together, these proteins repress the expression of HO-1 gene (10), an enzyme implicated in iron reutilization, anti-inflammation, and cytoprotection. MATII␣ and -␤ form a nuclear AdoMet-integrating transcription regulation module that further interacts with histone methyltransferase activities and other chromatin regulators (9). However, the identities of the histone methyltransferase(s) associated with MATII␣ are unclear. In addition, little is known about the downstream target genes of MATII␣ other than HO-1.
The gene expression tends to be silenced in heterochromatin and to be activated or poised for expression in euchromatin (11). These distinct chromatin structures are generated depending, in part, on histone modifications, including methylation. For example, methylation of lysine 4 of histone H3 (H3K4) is related to transcriptional activation (12), whereas methylation of histone H3K9 is related to transcriptional repression (13). Each methylation reaction is catalyzed by a specific methyltransferase. For example, SETDB1 and Suv39h1 catalyze the trimethylation of histone H3K9 for transcriptional repression (11).
In this report, we demonstrate that the Ptgs2 gene encoding cyclooxygenase 2 (COX-2/prostaglandin-endoperoxide synthase 2) is a direct target of MATII␣. COX-2 is an inducible enzyme required for the biosynthesis of prostanoids and contributes to inflammation, angiogenesis, and tumorigenesis (14,15). The COX-2 gene locus is epigenetically regulated, and its chromatin structure changes to activate the transcription of the gene (16). We found that MATII␣, MATII␤, MafK, and SETDB1 were recruited to the enhancer and promoter of the COX-2 gene to repress its expression in part by promoting H3K9 methylation.
Purification of MATII␣-associated Proteins-The purification of MATII␣-associated proteins was carried out as described previously (9). The nuclear fraction from iMEFs was extracted as described previously (21).
Immunoblot Analysis-For histone modification analysis, histones corresponding to 1 ϫ 10 6 cells were purified by acid extraction (22). Whole cell extracts and SDS-PAGE were performed using the cells as described previously (23).
Microarray Expression Profiling-Preparation of total RNAs from iMEFs was carried out using a total RNA isolation mini kit (Agilent Technologies). Agilent whole mouse genome (4 ϫ 44K, G4122F) arrays were used for this study, as described previously (23). The analysis and clustering of genes were performed using the GeneSpring software package (Agilent Technologies).

RESULTS
Identification of MATII␣-regulated Genes-We performed a transient knockdown (KD) of Mat2a using iMEFs to identify MATII␣-regulated genes (Figs. 1, A and B). Using a previously reported RNAi sequence (9), we could achieve roughly 70% reduction of the Mat2a mRNA expression. We compared the gene expression profiles between the control and MATII␣ KD cells using a DNA microarray analysis (Fig. 1C). Of the 41,252 gene probes included on the arrays, the expression levels of 1,597 gene probes were altered by Ͼ2-fold in the MATII␣ KD cells compared with control cells. Among the 1,597 gene probes affected, 776 and 821 probes showed up-and down-regulation, respectively. These affected genes were not enriched for specific gene ontology terms, thus suggesting that MATII␣ is involved in diverse cellular processes. We noticed that enzymes in the arachidonic acid cas-cade, namely COX-2 (Ptgs2) and prostaglandin E synthase (Ptges), were up-regulated in MATII␣ KD cells. We also found that p53 target genes such as Noxa, Fas, Igfbp3, and Gadd45b were up-regulated (Table 1). Using a qRT-PCR, we found that MATII␣ KD resulted in significant elevation of the mRNAs of Noxa, Fas, and COX-2, whereas Igfbp3, Gadd45b, and Ptges were only modestly affected (Fig. 1D). We also confirmed by an immunoblot analysis that MATII␣ repressed COX-2 expression (Fig. 1E). We therefore focused on the COX-2 gene for further analysis because the effect of MATII␣ depletion was most remarkable, and its expression was induced in response to oxidative stress or diethylmaleate, similar to HO-1 (Fig. 1, F and G) (7).
Repression of COX-2 Expression by MATII-To investigate whether the MATII regulates the COX-2 gene in the context of inflammation, we carried out MAT2A knockdown using THP-1 cells, which are derived from human acute monocytic leukemia (Fig. 2, A and B) (18,19). Using qRT-PCR, we confirmed MATII␣ KD ( Fig. 2A). The mRNA levels of COX-2 modestly but reproducibly increased upon MATII␣ KD (Fig. 2B). These results suggest that MATII may regulate the COX-2 expression in the monomacrophage system. To further elucidate the regulation of COX-2 by MATII in different cells, we carried out MAT2A knockdown using Hepa-1 cells (Fig. 2C). The mRNA levels of COX-2 also increased upon MATII␣ KD (Fig. 2D).
MATII Recruitment to the HO-1 Locus-We determined whether MATII was recruited to the COX-2 gene locus. We generated new anti-MATII␣ and anti-MATII␤ antibodies and performed immunoprecipitation and immunoblot analyses, confirming their reactivity with the respective endogenous proteins in murine erythroleukemia cells (Fig. 3A). Using these antibodies in ChIP assays, we found that endogenous MATII␣ and -␤ were specifically recruited to the E1 enhancer and promoter of HO-1 in iMEFs (Fig. 3, B and C). They tended to bind to the E2 enhancer, but the results were not statistically significant. MafK was specifically recruited to both of the two enhancers of HO-1 (Fig. 3C). These observations were consistent with our previous findings using epitope-tagged MATII␣ in Hepa1 cells (9).
MATII Recruitment to the COX-2 Locus-Having established the ChIP conditions using the new antibodies, we next examined the recruitment of MATII␣, MATII␤, and MafK at the COX-2 gene locus in iMEFs (Fig. 4). We identified one putative MARE sequence at 1.6 kbp upstream of the COX-2 gene promoter (Fig. 4A). MATII␣ and -␤ were both recruited to the putative MARE and promoter regions of the COX-2 gene (Fig.  4, B and C). In contrast, their binding to the further upstream region (i.e. up 3 kb) was less apparent. MafK was specifically recruited to the putative MARE (Fig. 4D). These data indicated that both MATII␣ and -␤ were recruited to the putative COX-2 gene MARE, together with MafK. The binding of MATII␣ and MATII␤ to the promoter region suggests that they might also interact with the basal transcription machinery. This possibility was supported by the finding that the purified MATII␣ complex interacted with the basal transcriptional machinery (9).
Modification of H3K9 and H3K4 at the COX-2 Locus-The recruitment of MATII␣ to the COX-2 gene locus suggested that MATII␣ might affect the methylation of histones around the locus. To investigate this possibility, we examined the levels of trimethylation at the COX-2 locus in iMEFs treated with control or MATII␣ siRNA (Fig. 5). Trimethylation of histone H3K9 was clearly observed at the COX-2 locus in control cells (Fig. 5, A-C). This modification was higher in the 3 kb upstream region than in the other two regions. This finding is in accord with observations made by Zhu et al. (13) showing that H3K9 trimethylation-rich regions flank enhancers of inactive genes. Upon MATII␣ KD, the trimethylation of histone H3K9 was decreased in these regions (Fig. 5, D-F). These observations suggest that MATII␣ is required for the maintenance of H3K9 trimethylation around not only the MARE and promoter regions where it was recruited but also the upstream region of the COX-2 locus where its recruitment was not apparent.
We observed higher levels of trimethylation of H3K4 at the MARE and promoter regions than in the upstream region (Fig.  5, D-F). The depletion of MATII␣ tended to affect the levels of this modification in the MARE and upstream regions, but the effects were not statistically significant (Fig. 5, D-F). It is worth noting that the trimethylation of H3K4 at the promoter remained high upon MATII␣ KD. Taken together, these results suggest that MATII␣ is required for H3K9 trimethylation but is dispensable for H3K4 trimethylation for the regulation of the COX-2 locus.
To investigate whether the histone acetylation, which is known to correlate with gene activation (11), was affected, we examined H3K9 and K27 acetylation levels at the COX-2 locus in iMEFs treated with control or MATII␣ siRNA (Fig. 5G). Upon MATII␣ KD, the levels of the H3K9 acetylation were increased at the promoter region but not at the upstream region (Fig. 5G). K27 acetylation also tended to be increased at the promoter region but it did not reach statistical significance (Fig.  5G). These observations suggest that MATII␣ is involved in deacetylation of histone H3K9 at the COX-2 promoter.
Impact on Modification of H3K9 and H3K4 of MATII␣-The region-and modification-specific effects of MATII␣ KD suggest that histone methylation might not necessarily be dependent on MATII␣. An immunoblotting analysis of the chromatin isolated from the control and MATII␣ KD iMEF cells revealed that the trimethylation of histones H3K9 and H3K4 was both decreased, whereas the mono-and dimethylation of these residues were not affected by MATII␣ KD (Fig. 6). The incorporation of histones into the nucleosome was not grossly affected by MATII␣ KD (Fig. 6). These results suggest that, in contrast to the COX-2 locus, the trimethylation of H3K9 and K4 at the epigenome-wide level was dependent on MATII␣. These results suggest that MATII may contribute to histone

Candidates for MATII␣ target genes
Shown is a list of MATII␣ target genes that were up-regulated in MATII␣ knockdown iMEFs in comparison with control iMEFs. The same GenBank TM accession no. genes are not shown.
Association of H3K9 Methyltransferases with MATII␣-One model of these pathways, based on the above observations, would be that among the many histone methyltransferases, some of the H3K9 methyltransferases might be highly dependent upon MATII␣. Considering that such histone methylatransferases may interact with MATII␣, we tried to identify the putative specific histone H3K9 trimethyltransferases associated with MATII␣. We first generated iMEFs stably co-expressing FLBio-MATII␣ (tagged with FLAG and biotinylation sequences) and biotin ligase BirA (Fig. 7A). Biotinylated MATII␣ was purified from the nuclear extracts by avidin affinity chromatography. As a control, we performed a mock purification from cells expressing only BirA (Fig. 7A, left lane).
Using the immunoblotting analyses, we found that MATII␣ associated with SETDB1 and Suv39h1, both of which are able to catalyze histone H3K9 trimethylation. These interactions appeared to be specific because MATII␣ did not interact with RIZ1, another enzyme involved in H3K9 trimethylation, or Ash1 H3K4 methyltransferase in iMEFs (Fig. 7B).
Regulation of the COX-2 Locus by SETDB1 and MATII␣-We next examined the involvement of SETDB1 and Suv39h1 in the regulation of COX-2. SETDB1 was efficiently reduced in iMEFs; however, the Suv39h1 silencing was not successful (Fig.  8A, data not shown). The levels of COX-2 mRNA were elevated by the SETDB1 KD in comparison with the control cells (Fig.  8B). In ChIP assays, SETDB1 was recruited to the MARE region but not to the 3 kb upstream or promoter regions (Fig. 8C). As a negative control for SETDB1 binding, we examined the 34th exon at the collagen type XI ␣2 (Col11a2) gene, which was used as a negative control for SETDB1 binding in a previous report (25) and confirmed that SETDB1 was not recruited to this region. In contrast to SETDB1, we could not detect binding of Suv39h1 to the COX-2 locus (Fig. 8D). Based on these results, we concluded that the expression of COX-2 was repressed by MATII␣ and SETDB1 in iMEFs (Fig. 8E). These results do not exclude the possible involvement of histone methyltransferases other than SETDB1.

DISCUSSION
The AdoMet-integrating transcription regulation module, which is constituted by MATII␣ and -␤ in the nucleus, has been proposed to couple histone methylation and transcriptional regulation by forming a complex that contain DNA binding transcription factors Bach1 and MafK, and histone methyltransferase(s) (9). However, other target genes for this system other than HO-1 are not known. Although the catalytic activity for the AdoMet synthesis of MATII␣ is required for the repression of HO-1 gene (9), the identity of methyltransferase(s) interacting with MATII␣ has been unknown. We have attempted to address these issues in this study, identifying COX-2 and SETDB1 as new components of this system, leading us to suggest a model depicted in Fig. 8E. In this model, AdoMet synthe-  sis and histone methylation are coupled on chromatin by a physical (direct or indirect) interaction with MATII␣ and SETDB1 on the subset of MafK target genes.
However, there are several issues that still need to be resolved in further studies. First, while this model explains the genespecific role of MATII␣, our results also indicated that MATII␣ was involved in the trimethylation of both H3K9 and K4 when viewed on a larger, epigenome-wide scale (Fig. 6). Such an epigenome-wide function of MATII␣ may not be dependent on the local recruitment of this enzyme. In this mode of action, MATII␣ may simply supply nuclear AdoMet without interacting with the chromatin and/or methyltransferases. For example, MATII may be involved in the methylation of newly biosynthesized, non-chromosomal histones. This is suggested by the finding that SETDB1 catalyzes the methylation of nonchromosomal histone H3K9 prior to its incorporation into chromatin in the S phase (26). It will be important to examine whether the nuclear localization of MATII␣ (9) is required for this epigenome-wide function. Second, the physical interaction of MATII␣ with transcription factors and methyltransferases may not be restricted to transcriptional repression. The expression of a substantial number of genes was reduced upon MATII␣ KD (Fig. 1C). Consistent with this idea, we recently found that MATII␣ interacted with transcription activators (9). Recently, it has been reported that the chromatin regulators partition into six modules correlated with binding patterns. One of the six modules, which include SETDB1, co-localizes with not only repressed genes but also active and competent promoters (27). MATII␣ may be involved in this regulation. Third, it may be surprising that a limited knockdown of only ϳ70% of Mat2a mRNA was sufficient to cause the dramatic increase in COX-2 expression in iMEFs. This observation suggests that MATII␣ is rate-limiting in cells and is consistent with previous reports showing signal-responsive induction of MATII␣ expression (28 -30). Mutation of MATII␣ may show haploinsufficiency. Knockdown of Mat2a (MAT2A) mRNA in several cell lines resulted in up-regulation of COX-2 (Figs. 1 and 2). Therefore, the involvement of MATII␣ in the repression of COX-2 is not restricted to particular types of cells. Furthermore, although SETDB1 was recruited to the COX-2 locus (Fig.  8C), the enrichment was only 1.5-fold. It remains unclear whether this level of differential binding would account for the H3K9 methylation at the COX-2 locus. One possibility is that, if there is any cooperative function between SETDB1 and other methylation-related factors, a small increase in SETDB1 binding would result in higher levels of methylation than expected. Indeed, a cooperative function of SETDB1, Suv39h1, G9a, and G9a related protein has been reported (31). Of course, cooperativity between SETDB1 and MATII␣ may explain the differential H3 methylation. Alternatively, SETDB1 may be redundant with other methyltransferases. Suv39h1 was apparently not recruited to this locus (Fig. 8D). However, epitope availability on this locus may be the cause of this observation. It still remains possible that methyltransferase(s) other than SETDB1, including Suv39h1, is involved in the MATII␣-dependent COX-2 repression.
Upon MATII␣ knockdown, we found that H3K9 acetylation was increased at the COX-2 promoter region. There are several  possibilities to explain these observations. First, the reduction in the H3K9 trimethylation would allow acetylation at the same residue by histone acetyltransferases. Second, because MATII␣ complex includes histone deacetylase-1 (9), its recruitment to the locus would be decreased by MATII␣ knockdown, leading to an increase in the acetylation level. histone acetyltransferases contain a large number of members such as p300/CBP and GCN5/PCAF (32,33). GCN5/PCAF may be involved in the expression of the COX-2 gene because it has been reported that they catalyze H3K9 acetylation in MEFs (33).
Our present and previous observations suggest that there is a coupling of AdoMet synthesis and histone methylation within nuclei, which may confer several advantages. First, the local synthesis for local consumption of AdoMet may ensure effective utilization of limited resources, because it would be possible to lower the overall AdoMet levels and to maintain locally sufficient levels of the substrate for methylation reactions. In addition, this could help avoid the putative genotoxic effect of AdoMet (34). H3K4 trimethylation was less dependent on MATII␣ compared with H3K9 trimethylation at the COX-2 locus (Fig. 5), raising the possibility that there may be a subset of histone methyltransferases, which are less dependent on MATII␣. Such enzymes may possess higher affinity for AdoMet. Second, the association and dissociation of MATII␣ with histone methyltransferases may provide a mechanism for dynamic regulation of histone methylation. To further explore these interesting possibilities, it will be important to understand how the nuclear accumulation of MATII␣ and its interaction with other nuclear/chromatin proteins are regulated.
COX-2 catalyzes biosynthesis of prostaglandin G2 and prostaglandin H2 in the arachidonic acid cascade (14). COX-2 also contributes to tumorigenesis (15,35). Taking together these previous and current observations, MATII may play an important regulatory role in inflammation and tumorigenesis. Indeed, its connections with diseases have been reported in several model systems (36 -38).