MAPKAP kinase 3pK phosphorylates and regulates chromatin association of the polycomb group protein Bmi1.

Polycomb group (PcG) proteins form chromatin-associated, transcriptionally repressive complexes, which are critically involved in the control of cell proliferation and differentiation. Although the mechanisms involved in PcG-mediated repression are beginning to unravel, little is known about the regulation of PcG function. We showed previously that PcG complexes are phosphorylated in vivo, which regulates their association with chromatin. The nature of the responsible PcG kinases remained unknown. Here we present the novel finding that the PcG protein Bmi1 is phosphorylated by 3pK (MAPKAP kinase 3), a convergence point downstream of activated ERK and p38 signaling pathways and implicated in differentiation and developmental processes. We identified 3pK as an interaction partner of PcG proteins, in vitro and in vivo, by yeast two-hybrid interaction and co-immunoprecipitation, respectively. Activation or overexpression of 3pK resulted in phosphorylation of Bmi1 and other PcG members and their dissociation from chromatin. Phosphorylation and subsequent chromatin dissociation of PcG complexes were expected to result in de-repression of targets. One such reported Bmi1 target is the Cdkn2a/INK4A locus. Cells overexpressing 3pK showed PcG complex/chromatin dissociation and concomitant de-repression of p14(ARF), which was encoded by the Cdkn2a/INK4A locus. Thus, 3pK is a candidate regulator of phosphorylation-dependent PcG/chromatin interaction. We speculate that phosphorylation may not only affect chromatin association but, in addition, the function of individual complex members. Our findings linked for the first time MAPK signaling pathways to the Polycomb transcriptional memory system. This suggests a novel mechanism by which a silenced gene status can be modulated and implicates PcG-mediated repression as a dynamically controlled process.

many biological processes, such as adaptation to environmental changes, differentiation, immune activation, inflammatory responses, cell cycle modulation, transformation, and apoptosis (1). These phosphorylation cascades may function as switching mechanisms that modulate gene activity (2). 3pK, also known as MAPK-activated protein kinase 3 (MAPKAPK3), belongs to a growing family of kinases that are activated by one or more members of the MAPK family (3). To better understand processes regulated by 3pK, we aimed to identify interactors and potential novel substrates of 3pK.
Here, we report the finding of a group of novel interaction partners of 3pK, the Polycomb group (PcG) proteins, including the mammalian PcG proteins HPH2 and Bmi1 (4,5). PcG⅐protein complexes maintain a transcriptionally repressed gene status in dividing and differentiating cells throughout developing eukaryotes (6 -8). The mechanisms that contribute to PcG-mediated silencing are beginning to unravel. PcG proteins form large multimeric protein complexes that associate with chromatin (reviewed in Refs. 9 and 10). The current view is that PcG-mediated silencing in mammals involves generation and recognition of histone modifications; a subset of PcG complexes, the EED⅐EZH complex (hPRC2 complex), associates with histone deacetylases, implicating histone tail deacetylation in PcG-mediated repression (11). Within this complex, EZH2 was identified as a histone methyltransferase with preference for histone H3 lysine 27 (12). A biochemically distinct Polycomb-repressive complex (hPRC1 in human, PRC1 in Drosophila (13,14)) recognizes and binds the epigenetic trimethyl(3m)lysine 27 mark produced by the (h)PRC2 complex; recognition and maintenance of repressed chromatin may involve additional histone modifications, such as histone ubiquitination or methylation (12,(15)(16)(17)(18). The current view is that PcG-mediated repressional maintenance involves a combination of (associated) catalytic activities responsible for posttranslational modifications on histone tails and extensive interactions among PcG members and other chromatin (-bound) factors, which generate a repressive higher order chromatin structure inaccessible to transcriptional activators and chromatin remodeling factors (8, 19 -21).
Little is known about how chromatin association and, potentially, function of PcG complexes is controlled. In the present study, we identified Polycomb group (PcG) proteins as in vivo interaction partners of the MAPKAP kinase 3pK. We have provided evidence that MAPK signal transduction cascades target PcG⅐protein complex/chromatin interaction through phosphorylation. 3pK acts as a Bmi1 kinase in vitro and in vivo. Of relevance, 3pK overexpression causes PcG/protein dissociation from chromatin and de-repression of the Cdkn2a/ INK4A locus. Our data support a model in which Polycombmediated repression is modulated by stress-and mitogenactivated protein kinase cascades. These findings point to a molecular mechanism by which a transcriptionally silenced gene status can be reprogrammed and implicates PcG-mediated repression as a dynamically controlled process. These observations are expected to have important implications in understanding epigenetic mechanisms in development and disease.
Yeast Two-hybrid System-A kinase-inactive 3pK mutant (K73M) was used as bait in two-hybrid screens against a human heart MATCH-MAKER cDNA library cloned into the pGAD10 vector (Clontech). Yeast strain CG-1945 (Clontech) was manipulated according the manufacturer's instructions. Positive clones were monitored by growth on SD/-Trp/-Leu/-His plates and activity of the lacZ reporter gene in filter assays. In direct two-hybrid assays, reselected SD/-Trp/-Leu/-His filters were incubated for 12-24 h at 30°C before the ␤-Gal assay was performed. The liquid culture ␤-Gal assay with o-nitrophenyl-␤-D-galactopyranoside (Sigma) as substrate was performed according to the manufacturer's instructions using the yeast strain Y190.
Cell Culture and Cell Cycle Synchronization-Indicated cell lines were cultured at 37°C, 5% CO 2 , 100% humidity in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS), antibiotics (100 units/ml penicillin and 100 g/ml streptomycin), 200 mM L-glutamine and 4.5 g/liter L-glucose. The TIG3-Bmi1.2PY has been described elsewhere (26). Human U2OS osteosarcoma cells and TIG3 primary human fibroblasts expressing the murine ecotropic receptor were provided by Dr. D. Shvarts (Utrecht Medical Center, Utrecht, The Netherlands) and Dr. D. Peeper (Netherlands Cancer Institute, Amsterdam, The Netherlands), respectively. Titers of LZRS⅐IRES⅐GFP (Bmi1, 3pK, or control) viral supernatants were sufficiently high to achieve near 100% infection; infection with pBABE-GST (3pK or control) virus was followed by selection with puromycin. G 0 /G 1 arrest of cells was established by contact inhibition and/or a 48-h serum starvation period (in Dulbecco's modified Eagle's medium supplemented with 0.1% FCS). M phase enrichment was achieved by adding colcemid (0.01 g/ml) to S phase-synchronized cells, acquired by a double thymidine block (2 ϫ 2 mM; 15-15 h) essentially as published previously (27). M phase enrichment was accomplished by mitotic shake-off.

RESULTS
3pK Binds the C Terminus of HPH2 in Vitro-In two independent yeast two-hybrid screens using an inactive 3pK (K73M) variant as bait, 20 overlapping clones containing cDNA of the human Polyhomeotic protein (HPH2) were isolated (Fig.  1A). These clones represented 73, 145, 198, or 209 C-terminal amino acid fragments of HPH2 (Fig. 1A). In a direct yeast two-hybrid test, 3pK also interacted with a larger HPH2 Cterminal fragment of 298 aa and the putative full-length HPH2 (432 aa) (Fig. 1, A and B). The C-terminal 3pK interaction domain of HPH2 (Fig. 1A) overlaps with the ␣-helical HDII/ SPM/SAM domain involved in hetero-and homotypic interactions between several Polycomb proteins (4,5,35,36). Mutational analysis of the C-198 fragment was used to further investigate the role of the HDII domain in the HPH2/3pK interaction. The HDII domain contains at least five distinct ␣-helical (␣1-␣5) structures (37). C-terminal deletions removing these ␣-helices prevent HPH2/3pK interaction (Fig. 1A, 3pK Phosphorylates Polycomb Group Protein Bmi1 lower inset). Removal of the C-terminal-most helix (␣5) is sufficient to completely abolish interaction. The requirement of an intact ␣5 was confirmed by the introduction of selected point mutations in ␣5 (Fig. 1A, asterisks on lower panel), which are predicted to be present on the surface of the protein and therefore may be required for the interaction with 3pK. Removing the basic side chain through exchange of K422A prevented interaction between HPH2 C-198 and 3pK completely, whereas exchange of L417A led to a reduction of protein association. 3pK (K73M) displays high affinity for the shorter HPH2 fragments, in particular for the C-terminal 73 fragment (Fig. 1A). This may relate to binding-site masking in full-length HPH2. Interestingly, mutation of known phosphorylation sites on 3pK (e.g. T313E, T313A, T201E/T313E) does not alter binding affinity as compared with the wild type kinase (not shown), suggesting that HPH2 binding is independent of Thr 201 /Thr 313 phosphorylation. Thus, the HDII/ SPM/SAM domain plays an important role in HPH2/3pK interaction.
3pK Complexes with PcG Proteins in Vivo-Interaction between 3pK and the individual HPH2 fragments in mammalian cells was confirmed by co-immunoprecipitation experiments (data not shown) with results consistent with the quantitative yeast two-hybrid data (Fig. 1A). The finding that 3pK is a binding partner of HPH2 suggests that the kinase may be part of a PcG complex and may also be associated with other PcG proteins. Biochemical analysis indeed confirms that HA-tagged Bmi1, a direct binding partner of HPH1/2 (4, 5), is specifically co-immunoprecipitated with GST-3pK and vice versa (Fig. 1B). In vivo, GST-3pK co-precipitates with endogenous PcG proteins, such as (h)Rnf2 (Fig. 1C), which is part of the hPRC1 complex (38,39). Exposure of cells to a 3pK-activating agent, the stress-inducer arsenite (3), does not alter the interaction quantitatively in these cells (not shown), again suggesting that phosphorylation of 3pK does not affect PcG association. This is in line with the comparable interaction of 3pK phosphorylation-site mutants with HPH2 described above. In a reciprocal experiment, Bmi1 was immunoprecipitated with an anti-3pK antiserum in both U2OS cell extracts (Fig. 1C) and HeLa cell extracts with high Bmi1 levels (Fig. 1D). Finally, we showed that endogenous proteins co-precipitate (Fig. 1E); this interaction was specific, because two unrelated kinases, AKT and Gsk3␤, did not co-precipitate hPRC1 core members (Fig. 1E). These biochemical data demonstrated association of 3pK with PcG proteins in vivo, without the need for overexpressed binding partners.  2 and 4). C, endogenous PcG complexes from U2OS (osteosarcoma) cell extracts co-precipitate GST-3pK (left panel); Bmi1 is co-precipitated with 3pK (right panel); overexpression of GST-3pK (GST-3pK⅐IRES⅐GFP retroviral vector) is indicated (ϩ). D, 3pK associates with PcG proteins in HeLa (cervical carcinoma) cells overexpressing Bmi1-PY; Bmi1 precipitates with its binding partners hRnf2 and HPH1. E, co-immunoprecipitation of endogenous proteins, only, from HeLa cell extracts. Antisera used for immunoprecipitation (IP) are indicated. ns, normal serum. Two unrelated kinases, AKT and GSK3␤, were used in control immunoprecipitations. Immunoprecipitation was confirmed by reduced detection in the depleted extract. Antisera used in Western detection are as indicated (arrows): anti-GST (for GST-3pK) and anti-Bmi1 (F6; for either endogenous or overexpressed Bmi1).

3pK Phosphorylates Polycomb Group Protein Bmi1
Mitogen-and Stress-induced Phosphorylation of Bmi1 in Vitro and in Vivo-3pK is a kinase activated by both stress stimuli and mitogenic signals. Activation by mitogens, such as serum and TPA, occurs almost exclusively via the ERK pathway, whereas stress stimulation with arsenite recruits the MAPKAPK to the p38 MAPK cascade (3). Thus, we next examined whether PcG proteins may represent downstream targets of these MAP kinase cascades by applying the respective stimuli. We focused on Bmi1, because chromatin association of Bmi1 correlates with its phosphorylation status (26). ERK activation, via mitogenic stimulation of serum-starved cells, leads to a rapid and readily detectable phosphorylation of Bmi1 in vivo; activation of p38, via arsenite, induces a much stronger hyperphosphorylation of Bmi1 as compared with mitogenic stimulation (Fig. 2A). Importantly, serum starvation prior to activation in these experiments precludes cell cycle-dependent phosphorylation (26) from interfering with these assays. Phosphatase treatment of PcG proteins extracted from arseniteexposed cells establishes phosphorylation as the main posttranslational modification on PcG proteins in response to p38 activation (Fig. 2B). We then studied the effect of protein phosphorylation on PcG core complex/protein interaction. As indicated in Fig. 2C, interaction of known PcG core complex partners, such as HPH1 and HPc2 (14), is at least in part preserved, as they still co-immunoprecipitate with Bmi1. Stress (Fig. 2D) or mitogenic (data not shown) stimulation result in Bmi1 phosphorylation in HeLa and 293T cells and, importantly, also in primary human TIG3 fibroblasts, showing that the observation is not cell-type restricted or linked to a specific cellular phenotype. Ground-state phosphorylation comparison in the different cell types also reveals that Bmi1 is already phosphorylated in some serum-starved established cell lines (Fig. 2, A and D). This would be consistent with a constitutive activation of MAP kinase signaling pathways in some cancer cell lines. Arseniteinduced Bmi1 phosphorylation is reduced in the presence of the p38 inhibitor SB202190 (Fig. 2E), providing additional evidence for specific catalytic targeting of Bmi1 by the stressactivated phosphorylation cascade, which acts through p38/ SAPK. Arsenite also activates the JNK pathway upstream of 3pK (40), which may explain the incomplete reduction of Bmi1 phosphorylation upon treatment with the p38 inhibitor. The mitogen-activated protein kinase/extracellular signal-regulated kinase kinase inhibitor UO126 reduces mitogen-induced phosphorylation of Bmi1 (Fig. 2F) and of HPH1 and HPc2, two known PcG core complex factors. Finally, the degree of Bmi1 phosphorylation downstream of mitogen-or stress-activated kinase signaling (Fig. 2, A, E, and F), correlates well with the reported difference in 3pK activation by these respective stimuli (40).
Signaling-induced Phosphorylation Abolishes Polycomb/ Chromatin Association-Next, the effect of MAPK signaling on PcG proteins/chromatin association was examined by immunocytochemistry. Serum-starved/refed cells (ERK1/2-mediated signaling) show a clearly visible decrease of chromatin-associ- 3pK Phosphorylates Polycomb Group Protein Bmi1 ated nuclear Bmi1 domains (Fig. 3A) (26,41). Likewise, arsenite exposure (p38-mediated signaling) relieves Bmi1/chromatin association by a process that is partially inhibitable by blocking p38 activity (Fig. 3A, right panel). In addition to Bmi1, a number of other PcG proteins are released from chromatin. These include HPc2, hRing1, and hRnf2 (Fig. 3B). Diminished fluorescence can be explained in a number of ways. Proteolytic breakdown was excluded by comparing straight lysates in the presence or absence of proteasome inhibitors; detection was unaltered, independent of physiological condition or inhibition of proteolysis (data not shown). Epitope masking by phosphorylation is rather unlikely, because a number of PcG proteins were detected by multiple independent antisera (data not shown). Differential extraction of soluble and chromatin-bound PcG proteins confirms the above immunocytochemical findings. As in M phase control extracts, which mainly contain soluble phosphorylated Bmi1 (26), upon p38 activation (arsenite), a substantial amount of phosphorylated Bmi1 is retrieved in the soluble fraction (Fig. 3C). Interestingly, HPH1 appears to undergo a nuclear redistribution upon exposure of cells to arsenite; consistently, less HPH1 is released from chromatin as compared with other PcG core proteins (Figs. 3B, 4A, and 5E). Also, less phosphorylated HPH1 is extractable as soluble pro-tein from arsenite-stimulated cells, as compared with M phase cells (Fig. 3C). Despite this HPH1 "chromatin retention," a fraction of HPH1 interacts with Bmi1, even when phosphorylated (Fig. 2C). Multiple distinct interaction partners for HPH1 may indeed exist in various nuclear (sub)compartments; Drosophila polyhomeotic was found in multiple protein interactions, some of which were free from posterior sex combs and polycomb (42).
p38/SAPK activation may affect chromatin remodeling more directly through phosphorylation of histones (43). In analogy to the reported interference of p(phospho)Ser 10 with heterochromatin protein (HP1)/3m(trimethyl)Lys 9 interaction (44), it is possible that simultaneous histone H3 phosphorylation and PcG/chromatin association are mutually exclusive. We therefore studied the dynamics of PcG/chromatin dissociation versus histone H3 Ser 28/10 phosphorylation, as trimethylation at the adjacent lysine 27 residue is required for Polycomb complex/ histone H3 interaction (12,15,16). Serum-starved U2OS cells showed very low histone H3 phospho-Ser (pSer 28 ; Fig. 4A, and pSer 10 , not shown) levels ( Fig. 4A and not shown); arsenite exposure led to the accumulation of both pSer 10 and pSer 28 (Fig. 4). There was a good correlation between histone H3 Ser 28 phosphorylation and loss of PcG/chromatin association (Fig.   FIG. 3. Phosphorylation-dependent PcG/chromatin interaction. A, Bmi1 visibly dissociates from chromatin upon mitogenic or stress signaling. U2OS cells were serum-starved for 48 h prior to serum/TPA (15 min) or arsenite (45 min) treatment. PcG/chromatin release is partially inhibited by the p38 inhibitor SB202190 (far right; compare with Fig. 2E). B, antibodies against a number of different PcG proteins show PcG/chromatin association under serum-starved conditions and PcG/chromatin dissociation following exposure to arsenite. Cells were fixed and permeabilized in 100% methanol, which allows for the detection of chromatin-bound PcG proteins (26). C, differential extraction (see "Experimental Procedures") of soluble fractions shows extensive chromatin dissociation of Bmi1 upon arsenite stimulation (ars), whereas HPH1 release is substantially less. Synchronized G 1 and M phase control cell extracts reveal the predicted chromatin association/dissociation, respectively (26); loading was corrected for cell numbers. 4A). Global histone H3 Ser 28 phosphorylation varied significantly between cells, whereas individual PcG proteins dissociated or relocated uniformly (Fig. 4A). This suggests PcG/chromatin dissociation may occur relatively fast in response to direct phosphorylation by 3pK (or other arsenite-activated kinases) or in response to local histone H3 Ser 28 phosphorylation. Kinase assays on N-terminal peptides, however, suggest that H3 and H4 are not direct 3pK phosphorylation targets in vitro. 2 In contrast, pSer 10 and PcG/chromatin association are not mutually exclusive (Fig. 4B), consistent with the observation that histone H3 pSer 10 phosphorylation is thought to interfere with recognition and binding of the 3mLys 9 silencing mark by HP1 chromodomain proteins but not by HPc chromodomain proteins (44).
Identification of 3pK as a Candidate Bmi1 Kinase-We studied 3pK as a candidate Bmi1 kinase by in vitro kinase assays and by stimulation of cells with altered 3pK expression levels.
Using recombinant proteins, we showed that Bmi1 is a direct phosphorylation target for 3pK in vitro (Fig. 5A). For additional in vitro kinase assays, active 3pK was obtained by mitogen or arsenite stimulation of cells (3). Immunoprecipitated Bmi1 is strongly phosphorylated in the presence of the active kinase (Fig. 5B, lower panel), similar to levels obtained with the control substrate Hsp27 (Fig. 5B, lower panel). Preliminary data obtained with PcG complexes purified by immunoprecipitation suggest that other PcG complex members (HPH1, HPc2) are also directly phosphorylated in vitro by 3pK. 3 To provide additional evidence for phosphorylation of Bmi1 by 3pK in vivo, genetically matched cells with increased or diminished 3pK expression were used in the phosphorylation assays. The expression of a nucleus-retained active 3pK mutant results in more intensive Bmi1 phosphorylation after serum stimulation compared with wild type cells (Fig. 5C). Serum stimulation reveals subtle but clear differences in phosphorylation between control and 3pK-overexpressing cells (Fig. 5D); Bmi1 is phosphorylated more rapidly and more intensely in 3pK-overexpressing cells. We were able to relate these findings to specific biological effects of 3pK overexpression. 4 Conversely, in cells in which 3pK is knocked down by RNA interference (25) using two independent RNA interfering sequences (Fig. 5E, lanes 1 and  3), decreased phosphorylation of Bmi1 correlates nicely with decreased 3pK levels. These findings connect 3pK expression levels and Bmi1 phosphorylation in vivo. Cells knocked down for 3pK expression retain Bmi1 and HPH1/chromatin association upon arsenite stimulation (Fig. 5F), whereas cells overexpressing 3pK show a more rapid Bmi1 release (data not shown), again directly implicating 3pK in this process. Com- FIG. 4. Post-translational histone H3-tail modifications during stress signaling. A, histone H3 Ser 28 phosphorylation correlates with loss of PcG/chromatin association in U2OS cells. Arsenite-exposed cells show loss of Ring1B binding and the characteristic relocation of HPH1 (see also Fig. 3B). Phosphorylation of Histone H3 Ser 28 is not detectable in serum-starved cells (control), whereas it clearly is upon arsenite exposure. B, histone H3 Ser 10 phosphorylation in cells does not correlate with PcG/chromatin dissociation in arsenite-exposed cells. Shown is an example of cells in which histone H3 Ser 10 phosphorylation in response to arsenite, as well as the typical punctate nuclear PcG staining, is clearly visible. Histone H3 Ser 10 phosphorylation in control cells (serum starved) is non-detectable (not shown). Arrows indicate cells positive for both pSer 10 and chromatin-bound PcG complexes; cells were exposed to arsenite on glass slides and fixed in formaldehyde/acetone. DAPI, 4Ј,6-diamidino-2-phenylindole.

3pK Phosphorylates Polycomb Group Protein Bmi1
bined, these findings strongly suggest that PcG proteins are phosphorylated downstream of activated MAPK (ERK/p38) pathways and identify 3pK as a Bmi1 kinase.
Overexpression of 3pK Reduces PcG Complex/Chromatin Association-To qualify as a PcG complex-associated Bmi-1 kinase, 3pK would be expected to associate with chromatin. Indeed, differential nuclear extraction of cells overexpressing human 3pK constructs shows, in line with the immunoprecipitation data presented above, that a fraction of the wild type 3pK protein is associated with chromatin (Fig. 6A). Neither constitutively active nor kinase-inactive mutants of 3pK associate with chromatin to the extent wild type 3pK does (Fig. 6A), in concordance with observations that such mutants are mainly found in the cytoplasm (45)(46)(47)(48). Combined with the immunoprecipitation data, this demonstrates that at least a fraction of cellular MAPKAP kinase 3 is associated with PcG complexes at the chromatin level.
Because repressive PcG complexes are intimately associated with chromatin and MAPK-induced PcG phosphorylation results in chromatin dissociation, it would follow that 3pK acti-vation or overexpression causes PcG/chromatin dissociation, which would then result in de-repression of PcG target genes. To begin to understand the biological relevance of PcG complex phosphorylation by 3pK, we studied the cellular function of 3pK. 3pK overexpression induces a profound effect on cell proliferation; cell cycling slows down significantly or arrests, depending on the cell type studied. 4 We therefore asked whether overexpression of 3pK induces de-repression of the Cdkn2a/ INK4A locus, a well established Bmi-1 target (49, 50) that is transcriptionally silent in U2OS cells (51). 3pK-overexpressing cells showed reduced (or loss of) nuclear PcG staining (Fig. 6C,  middle panel). In agreement with this, in U2OS cells overexpressing 3pK, less Bmi1 is present in the chromatin-bound fraction as compared with control cells (Fig. 6B), which supports the idea that 3pK overexpression results in chromatin release of Bmi1. Cells that have lost PcG/chromatin association show re-expression of p14 ARF , one of the gene products of the Cdkn2a/INK4A locus (Fig. 6C). The above findings are consistent with the notion that PcG repression needs to be lost to de-repress a target gene. FIG. 5. Identification of 3pK as a candidate Bmi1 kinase. A, 3pK directly phosphorylates Bmi1. Recombinant GST-3pK (T201E/T313E) and recombinant His-tagged Bmi1 protein were used in in vitro kinase assays. B, in vitro kinase assays show that immunoprecipitated Bmi1 is phosphorylated by 3pK. Active GST-3pK was obtained from HEK293 cells (arsenite-or serum/TPA-stimulated for 60 min). HA-Bmi1 substrate and kinase were immunoprecipitated separately and then combined in kinase assays. Hsp27, a known 3pK substrate, was used in control reactions (lower panel). C, increased in vivo phosphorylation of Bmi1 by a nucleus-retained (NR) 3pK mutant, which lacks the C-terminal (aa 308 -383) regulatory region. Lane c, serum-starved cells; control cells or cell expressing NR-3pK were serum-stimulated (10% FCS), following which Bmi1 was immunoprecipitated from cell extracts. P 32 incorporation was determined by autoradiography. D, exposure to serum (5%) reveals differences in the kinetics of Bmi1 phosphorylation between control and 3pK overexpressing U2OS cells; time (min) between start of stimulation and harvest is indicated. E, reduced Bmi1 phosphorylation in response to hsRNA-mediated 3pK reduction. Various stably expressed hsRNAs (lanes 1-3; see "Experimental Procedures") reduce 3pK protein levels (upper panel); diminished Bmi1 phosphorylation upon mitogenic stimulation correlates well with lowered 3pK expression levels (lower panel). Lane 0 is an empty vector control. U2OS cells were serum-starved prior to activation (lower panel, far left lane c). F, reduced chromatin dissociation of PcG proteins Bmi1 and HPH1 upon arsenite stimulation of cells correlates with 3pK expression in U2OS cells. Cells were treated as described in the legend to Fig. 3A. RNA interfering sequence used: 1 (see Fig. 5C and "Experimental Procedures").
3pK Phosphorylates Polycomb Group Protein Bmi1 DISCUSSION We here identify 3pK as a component of PcG complexes. Initially identified in a yeast two-hybrid screen as a binding partner of HPH2, 3pK was subsequently shown to associate with chromatin and PcG complexes by biochemical analysis. The region within HPH2 responsible for 3pK binding is the HDII/SPM/SAM domain, which also mediates homo-and heterotypic interactions with other PcG proteins, such as SCM and Bmi1 (4,5,35,36). We established an intact ␣5 motif as a crucial structure for HPH2/3pK interaction within the HDII domain. MAPKAP kinase 2, a close homolog of 3pK, was recently found to bind HPH2 in a yeast two-hybrid screen (52). The interaction region was mapped to the C-terminally located HDII domain as well. In addition, full-length MPh1 (mouse Polyhomeotic) interacts with 3pK in direct interaction assays. 3 This combined information supports the notion that the HDII/ SPM/SAM domain plays an important role in HPH/3pK interaction and suggests that complex composition may vary with cell type or physiology, or alternatively, that there may be ternary complex formation between HPH1/2, Bmi1, and 3pK or other binding partners in vivo.
3pK phosphorylates Bmi1 and possibly other PcG proteins in vitro, and 3pK overexpression or knockdown in genetically matched cell lines results in enhanced and reduced Bmi1 phosphorylation in vivo, respectively. In addition, we showed that the typical punctate nuclear PcG staining pattern is lost upon ERK or p38 activation. This implicates a connection between MAP kinase signaling and chromatin structure remodeling through modulation of PcG complex function. A simple mechanistic model proposes that PcG-associated 3pK is an integral regulatory element within PcG complexes; 3pK phosphorylation and activation by upstream nuclear shuttle kinases (ERK, p38 and possibly JNK) results in phosphorylation and subsequent chromatin dissociation of PcG complexes from known chromosomal binding sites (26). Although chromatin dissociation is expected to result in reactivation of repressed genes, it is currently unknown what the molecular effect of protein phosphorylation is on PcG members. Phosphorylation is thought to affect subcellular localization or homodimerization of some PcG proteins (53,54). We find some relocation of PcG proteins to the cytoplasmic fraction. 5 However, the fate of dissociated PcG complexes, as well as that of the activated MAPKAP kinase, is unclear, as studies with proteasome inhibitors do not show marked proteolytic breakdown. Given our observation that at least part of the mammalian PcG protein core complex (14) remains intact upon phosphorylation, it is unlikely that Bmi1 phosphorylation directly causes chromatin dissociation. A more likely scenario is that PcG/chromatin binding is mediated by critical amino acid residues on a (limited number of) PcG complex member(s) in a phosphorylation-dependent manner. Potentially interesting candidates are the mammalian orthologs for Pc (HPc1/2/3), which may directly bind histone H3 3m(trimethyl)Lys 27 -silencing imprints through its chromodomain. It is possible that, besides chromatin association, protein/protein interaction and/or (associated) catalytic activity of individual PcG proteins is altered upon phosphorylation. In this light, the redistribution of HPH1 is interesting. Although the exact im-5 J. W. Voncken, unpublished data. (3) were differentially extracted. Shown are the chromatin-bound protein fractions. In concordance with published data (48), only wild type 3pK resides in the nucleus. B, 3pK overexpression reduces chromatin association of Bmi1 (right panel). Shown is the chromatin-bound fraction of differential nuclear extracts. Cells were kept under high selection pressure (puromycin) for two passages. C, reduced PcG/chromatin association (middle panel) and increased p14 ARF expression (right panel) in 3pK-overexpressing U2OS cells. U2OS cells were infected on glass slides with GST-3pK⅐IRES⅐GFP retrovirus; infection was confirmed by fluorescence. Under these conditions, 16 -20% (n Ͼ 150) of the cells have lost the typical nuclear PcG domains (arrowhead). The majority of these cells is positive for p14 ARF expression. Control cells, infected with empty vector, showed neither PcG dissociation nor p14 ARF expression (n Ͼ 250). D, connection between cell signaling and epigenetic regulation of gene expression via chromatin remodeling factors. MAPK phosphorylation cascades are known to target transcription factors (TF, right) and chromatin (nucleosomes, left) directly. We propose that activated MAPK cascades (schematically depicted by circles on arrow) also target chromatin (beads) through modulation of chromatin remodeling complexes (gray structures), such as PcG complexes. Activation of 3pK leads to phosphorylation of PcG complexes; subsequent chromatin dissociation results in de-repression of silenced loci.

3pK Phosphorylates Polycomb Group Protein Bmi1
plications of this HPH1-redistribution are currently not clear, it may well serve to stall DNA replication and or gene expression upon activation of the stress kinase pathway through its interaction with DNA replication inhibitors (55,56). Likewise, recently identified enzymatic activities of PcG proteins (i.e. histone methyltransferase activity of EZH2) (12), small ubiquitin-like modifier E3 ligase activity of HPc orthologs (57), and ubiquitin E3 ligase activity for the Ring finger proteins (such as Ring1B (58) and possibly Bmi1) are likely to be affected by phosphorylation. Clearly, identification of relevant cellular substrates of these enzymes and mutation analysis of relevant phosphorylation sites are essential to increasing our understanding of the exact role of phosphorylation in PcG-mediated silencing.
ERK and/or p38 and JNK MAP kinases and downstream MAPK effector kinases (one of which is 3pK) are implicated in a variety of fundamental cellular processes, including proliferation, differentiation, and cell cycle regulation (1). The involvement of PcG and trxG proteins in development and cancer is well documented (reviewed in Ref. 7). Similar to 3pK, we have also observed in vitro HPH2 binding (as previously reported (52)), Bmi1 phosphorylation by MAPKAP kinase 2, and reduced chromatin dissociation upon knockdown of MAPKAP kinase 2 protein levels. 6 These observations suggest that additional kinases may target PcG proteins downstream of active MAPK cascades. Interestingly, other kinases within the MAP-KAP kinase family act on histones (59), further supporting the idea that such kinases, besides directly activating transcription factors, also target chromatin components and chromatin remodeling factors (Fig. 6D). Phosphorylation of residues (e.g. histone H3 pSer 10 or pSer 28 ) adjacent to silencing marks (e.g. histone H3 3mLys 9 or 3mLys 27 ) could act as a (re-)activating event (44) and serve to release repressive (HP1 or PcG, respectively) effector complexes and effectively de-repress target genes. In accordance with this "binary switch" model (44), our data suggest that PcG/chromatin dissociation is a relatively early event and may be directly connected to local phosphorylation of histone H3 Ser 28 ; i.e. histone H3 Ser 28 phosphorylation at target loci is sufficient to release PcG complexes from chromatin. The co-existence of pSer 10 and Polycomb complexes in arsenite-stimulated cells, however, seems to rule out a major role for histone H3 pSer 10 in PcG/chromatin dissociation, in line with a specific role for H3 3mLys 9 in HP1-mediated heterochromatin formation (60). However, to formally exclude the possibility that these respective phosphorylation events are independent, detailed examination of the histone H3 Ser 28 status at target genes in relation to PcG/chromatin association is required.
Bmi1 and its direct binding partners, Rnf2, MPh1 and M33, through transcriptional control of the Cdnk2a/INK4A locus, are involved in proliferative control during embryogenesis and postnatal life and in stem cell renewal in hematopoiesis and neurogenesis (61)(62)(63)(64)(65). We have studied the effect of 3pK on cell cycle progression and its genetic interaction with PcG proteins in detail. 4 Relevantly, overexpression of 3pK de-represses the Cdnk2a/INK4A locus, further supporting the interaction of MAPK signaling and PcG-dependent chromatin remodeling. Both mitogenic as well as stress activation lead to PcG/chromatin dissociation. Clearly, different target genes are predicted to be (de)activated in response to these different physiological stimuli. Recent findings indicate that 3pK activation and translocation kinetics are dependent on the pathway through which it is activated (3,48); this is expected to influence the type of substrates and genes targeted by activated kinases downstream of mitogens or cellular stressors.
In summary, we have described a novel link between signaling through MAP kinase cascades and an epigenetic transcription regulatory system. We propose this as a molecular mechanism by which cells modulate gene expression profiles to adequately respond to changes in their microenvironment; consequentially, PcG-mediated silencing may be subject to dynamic modulation. Chromatin association of both trxG⅐ and PcG⅐protein complexes has been reported to depend on their phosphorylation status (26,34). We propose that MAPK signaling impinges on gene expression through functional regulation of chromatin-modifying complexes via PcG and trxG kinases and/or phosphatases (Fig. 6D). Unraveling the nature of kinase/PcG and PcG/chromatin association and the exact biological significance of phosphorylation for PcG function presents an exciting and relevant challenge for future research. It will be of particular interest to examine whether and how this molecular connection relates to the proposed role of MAPK pathways as switching mechanisms that modulate gene activity (2). PcG, trxG proteins, and MAPK signaling pathways are connected to tumorigenesis; the hereindescribed findings will improve our understanding of how cells modulate gene expression at the epigenetic level in response to extracellular cues and have the potential to yield therapeutic strategies to fight cancer.