MBD3 and HDAC1, Two Components of the NuRD Complex, Are Localized at Aurora-A-positive Centrosomes in M Phase*

MBD3, a component of the histone deacetylase NuRD complex, contains the methyl-CpG-binding domain (MBD), yet does not possess appreciable mCpG-specific binding activity. The functional significance of MBD3 in the NuRD complex remains enigmatic, partly because of the limited availability of biochemical approaches, such as immunoprecipitation, to analyze MBD3. In this study, we stably expressed the FLAG-tagged version of MBD3 in HeLa cells. We found that MBD3-FLAG was incorporated into the NuRD complex, and the MBD3-FLAG-containing NuRD complex was efficiently immunoprecipitated by anti-FLAG antibodies. By exploiting this system, we found that MBD3 is phosphorylated in vivo in the late G2and early M phases. Moreover, we found that Aurora-A, a serine/threonine kinase active specifically in the late G2and early M phases, phosphorylates MBD3 in vitro, physically associates with MBD3 in vivo, and co-localizes with MBD3 at the centrosomes in the early M phase. Interestingly, HDAC1 is distributed at the centrosomes in a manner similar to MBD3. These results suggest the highly dynamic nature of the temporal and spatial distributions, as well as the biochemical modification, of the NuRD complex in M phase, probably through an interaction with kinases, including Aurora-A. These observations will contribute significantly to the elucidation of the yet-uncharacterized cell cycle-controlled functions of the NuRD complex.

Core histones are subjected to a variety of post-translational modifications, including acetylation and methylation (1). These modifications affect the chromatin structure and/or the accessibility of transcriptional regulatory factors to chromatin and thereby influence gene expression. Acetylation has been the most extensively investigated among them. Generally, hyperacetylated histones are correlated with the transcriptionally active state, whereas hypoacetylated histones are correlated with the silent state (reviewed in Ref. 2). The relative level of histone acetylation is determined by the equilibrium between two opposite enzymatic activities, histone acetyltransferases (HATs) 1 and histone deacetylases (HDACs). Each of the HAT and HDAC groups comprises a protein family consisting of many member proteins. These proteins do not function by themselves. Instead, they form large multiprotein complexes together with other components, such as DNA/chromatin-binding proteins and co-activator/co-repressor, as well as other rather ill-defined proteins. Among the members of the HDAC family, HDAC1, HDAC2, and HDAC3 comprise one subgroup (Class I HDACs) because of their homology to yeast HDAC Rpd3. HDAC1 and HDAC2 are present as two major complexes, namely, NuRD and SIN3 (reviewed in Ref. 3). These two HDAC complexes share some components, including HDAC1 and HDAC2, and contain specific components. NuRD is unique in that it possesses another catalytic activity, the ATP-dependent chromatin-remodeling activity performed by Mi2.
The acetylation of histones is not a stationary process. Instead, it is known that the general levels of acetylated histones and the relative activities of HATs and HDACs vary in a regulated manner during cell cycle progression (4,5). These cell cycle-dependent changes of the histone acetylation levels are believed to participate in controlling the progression of cell cycle by regulating the expression of genes required for cell cycle progression and directly modifying the structures and functions of chromosomes and their functional components, such as kinetochores. Little is known, however, regarding how the cell cycle progression and the cell-cycle-regulated HAT and HDAC activities are linked.
Gene activities are also modified by DNA methylation, the covalent modification of cytosine in the context of the CpG nucleotide. DNA methylation is generally coupled with gene repression. Importantly, DNA methylation and histone deacetylation are cooperatively involved in transcriptional repression (6). Several lines of evidence indicate that methyl-CpG (mCpG)-binding proteins play a major role in mCpG-dependent gene repression (7). To date, five mCpG-binding proteins, MBD1-4 and MeCP2, have been identified in mammals and are collectively called MBD family proteins, because they share the methyl CpG-binding domain (MBD). Among them, MBD1, MBD2, and MeCP2 have been well characterized for their gene-repressing activities. These proteins bind to mCpG through MBD and recruit HDAC through another domain called the transcriptional repression domain. MBD4 is a DNA repair enzyme specific to T/G mismatched base pairs (8,9). In contrast to these well-characterized MBD proteins, the function of MBD3 is not fully understood. Although MBD2 and MBD3 are highly homologous in and outside MBD, experiments performed in vitro and in vivo failed to demonstrate the mCpG-binding activity of recombinant mouse and human MBD3 (10,11). Controversially, Xenopus MBD3 was demonstrated to possess mCpG-binding activity (12). Recent studies have pointed out that amino acids in MBDs critical for binding to mCpG are present in Xenopus MBD3 but not in mammalian MBD3 (13,14). These amino acid substitutions are responsible for the inability of mammalian MBD3 to bind to mCpG (11). It was reported that the NuRD complex contains MBD3 as a component (15,16). However, the role of MBD3 in the NuRD complex remains unclear. Other components of the NuRD complex include HDAC1, HDAC2, Mi2, RbAp46, RbAp48, and MTA-2 (12,16,17). It was shown that the human NuRD complex does not bind to mCpG but is recruited to mCpG by interacting with the authentic mCpG-binding protein MBD2 (16,18).
In this study, we characterized the biochemical and cytological behaviors of MBD3. We found that MBD3 physically binds with and is phosphorylated in vitro by Aurora-A. Aurora-A is a recently identified serine/threonine kinase, which is active in the late G 2 and early M phases and localizes to the centrosome and spindle (reviewed in Refs. 19 and 20). The inactivation or lack of Aurora-A leads to the failure of centrosome separation and the formation of monopolar spindles (21)(22)(23). Therefore, Aurora-A has been implicated in mitotic regulation. Moreover, we found that a fraction of MBD3 and HDAC1 are co-localized at the Aurora-A-positive centrosomes in early M phase. These results suggest that the NuRD complex is spatially and biochemically regulated in M phase.

EXPERIMENTAL PROCEDURES
Plasmids-The cDNA of MBD3 was isolated and the mammalian expression vector of HA-tagged MBD3 was constructed as described previously (24). pMX-puro, which is a puromycin-resistant derivative of pMX-neo, was used to construct a retroviral expression vector (25). The cDNA coding MBD3 with a FLAG epitope was inserted into the pMXpuro (referred to as pMXpuro-MBD3-FLAG). For the preparation of a bacterial expression vector, the cDNA encoding MBD3 with an HA epitope was inserted into pGEX5X-1 (Amersham Biosciences, referred to as pGEX5X-1-HA-MBD3) or pET32c (Novagen). The former plasmid was digested by EcoRI and XhoI to construct another expression vector for only the HA epitope fused with GST. The resulting 3Ј-terminal ends were filled in by Klenow fragment (Takara) and ligated using a DNA ligation kit (Takara). Plasmids for point mutants of MBD3 were generated with a GeneEditor in vitro site-directed mutagenesis system (Promega). Products were confirmed by DNA sequencing. Mammalian and bacterial expression vectors for Aurora-A have been described elsewhere (26).
Construction of a Cell Line Stably Expressing MBD3-A cell line stably expressing MBD3 was constructed with a retrovirus-mediated system (25). In brief, transfection of NX packaging cells with pMXpuro or pMXpuro-MBD3-FLAG was performed with LipofectAMINE reagent (Invitrogen). Retroviruses in the culture supernatants were used to infect HeLa cells. Transfected cells were selected in the presence of 1.0 g/ml puromycin (referred to as HeLa-puro and HeLa-MBD3-FLAG, respectively).
Immunoprecipitation and Immunoblot Analysis-For analysis of the interaction between ectopic FLAG-tagged MBD3 and other components of the NuRD complex, HeLa-MBD3-FLAG cells were lysed using modified radioimmune precipitation assay buffer (50 mM Tris-HCl, pH 8.0, 0.15 M NaCl, 1 mM dithiothreitol, 1ϫ Complete (Roche Molecular Biochemicals), 0.1% SDS, 0.5% deoxycholate, 1% Nonidet P-40, 25 mM sodium fluoride, and 25 mM ␤-glycerophosphate). Protein extracts were pretreated with normal mouse IgG (NMI) followed by a mixture of protein G-Sepharose and protein A-Sepharose (Amersham Biosciences). The precleared extracts were then treated with anti-FLAG antibodies or NMI. The immunocomplexes were collected with mixture of protein G-Sepharose and protein A-Sepharose and subjected to immunoblot analysis. In brief, samples were resolved by 6 or 10% SDS-PAGE and transferred to Immobilon (Millipore) sheets. The blots were incubated in Block-Ace (Dai Nippon Pharmaceuticals) solution and then probed with primary antibodies. After rinsing with in TNT buffer (20 mM Tris, 0.14 M NaCl, and 0.05% Tween 20), the blots were incubated with appropriate horseradish peroxidase-conjugated goat secondary antibodies (Amersham Biosciences). After a second rinse in TNT buffer, immunocomplexes were visualized using ECL chemiluminescence (Amersham Biosciences).
Cell Cycle Synchrony and Metabolic Labeling-HeLa-MBD3-FLAG cells were synchronized at G 1 /S boundary according to the thymidine/ aphidicolin double-block protocol and released for various lengths of time as described previously (27). To obtain samples in M phase, one sample was released from the block and cultured for 6.5 h, followed by an additional culture for 4 h in the presence of 0.5 g/ml nocodazole. In 32 P-metabolic labeling experiments, cells were incubated in phosphatefree Dulbecco's modified Eagle's medium supplemented with [ 32 P]orthophosphate to a final concentration of 0.5 mCi/ml for 90 min prior to harvesting. Cells were lysed using the modified radioimmune precipitation assay buffer. Protein extracts were subjected to immunoprecipitation experiments with NMI or anti-FLAG antibodies, and immunocomplexes were resolved by parallel 10% SDS-PAGE. One gel was dried and analyzed by BAS2000 imaging analyzer (Fuji Film). Another gel was subjected to immunoblot analysis as above, except that visualization was performed using a POD Immunostain kit (Wako).
Indirect Immunofluorescence-Cells were cultured on coverslips and fixed with 2% paraformaldehyde for 10 min followed by permeabilization with 0.1% Triton X-100 for 10 min. Cells were subsequently incubated with primary antibodies for 1 h at room temperature. After extensive washing with 0.2% Tween 20, these antibodies were detected by appropriate secondary antibodies as follows: Alexia 488-conjugated goat anti-rabbit Ig antibodies (Molecular Probes), Alexia 488-conjugated donkey anti-goat Ig antibodies (Molecular Probes), Cy3-conjugated goat anti-mouse Ig antibodies (Amersham Biosciences), and Cy5conjugated donkey anti-mouse Ig antibodies (Amersham Biosciences). DNA was stained with propidium iodide (Nacalai Tesque) or TOTO3 (Molecular Probes). Images of fluorescent staining were observed with a confocal laser scanning microscope (Zeiss).
Transient Transfection-For analysis of the interaction between MBD3 and Aurora-A, COS cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (JRH Biosciences). These cells were transfected with the plasmid for HA-tagged MBD3 and Glu-tagged Aurora-A, independently or simultaneously using the DEAE-dextran method.
For transient overexpression of MBD3 wild-type or mutant, HeLa cells were maintained in the same medium as above. These cells were transfected with the plasmid for HA-tagged MBD3 wild-type or mutant using LipofectAMINE plus (Invitrogen).
Phosphoamino Acid Analysis-In vitro labeled sample was resolved by 10% SDS-PAGE. After staining with CBB, phosphorylated MBD3 was excised and incubated in 6 N HCl. Acid hydrolysis was performed for 90 min at 110°C. After lyophilization, the hydrolysates were dissolved in a marker mixture containing phosphoserine, phosphothreonine, and phosphotyrosine (Sigma). The sample was subjected to electrophoresis on a silica gel thin-layer plate at 1 kV for 60 min in the electrophoresis buffer (acetic acid:pyridine:H 2 O, 50:5:945, v/v). The markers were stained with 0.25% ninhydrin, and phosphoamino acids were detected with the BAS2000 imaging analyzer.

MBD3-FLAG Is a Component of the NuRD Complex and
Immunoprecipitated by Anti-FLAG Antibodies-It was previously reported that anti-MBD3 antibodies did not immunoprecipitate the NuRD complex, presumably because MBD3 is buried in the complex and is not accessible to antibodies (16). We also experienced the inability of anti-MBD3 antibodies raised in our laboratory to immunoprecipitate MBD3. We reasoned that tagged versions of MBD3 might be recognized by anti-tag antibodies presumably because the tag might be extended from and exposed to the outside of the complex. We therefore constructed a subline of HeLa cells (HeLa-MBD3-FLAG) that stably express the C-terminally FLAG-tagged human MBD3 protein. Two versions of MBD3, MBD3a and MBD3b, are known to be produced by alternative splicing (16). In this study, MBD3a, the longer version, is used throughout and will be called simply MBD3 hereafter. In an immunoblot experiment (Fig. 1A), anti-MBD3 antibodies recognized a slower migrating band (closed arrowhead) in HeLa-MBD3-FLAG cell extracts but not in the mock-transfected cell (HeLa-puro) extracts, in addition to a faster migrating band (open arrowhead) that was also found in HeLa-puro extracts. The slower migrating band was reactive with anti-FLAG antibodies, indicating that it represents the C-terminally FLAG-tagged MBD3 recombinant protein. Accordingly, the faster migrating band is the endogenous MBD3 protein.
It has been shown that MBD3 is a component of the NuRD complex (15,16). We examined whether MBD3-FLAG func-tions as a component in the NuRD complex or not. The protein complex containing MBD3-FLAG was immunoprecipitated from HeLa-MBD3-FLAG extracts using anti-FLAG antibodies.
In immunoblot experiments using specific antibodies, we recognized the presence of Mi2, MTA2, HDAC1, and HDAC2 (Fig.  1B) and RbAp48 and RbAp46 (data not shown) in the immunoprecipitates. These results indicate that the MBD3-FLAG protein constitutes the NuRD complex and is successfully immunoprecipitated by anti-FLAG antibodies.
MBD3 Phosphorylation in G 2 /M Phase-The abundance of native MBD3 does not change significantly during the cell cycle (data not shown). To investigate the possibility that the MBD3 function is controlled by post-translational mechanisms, we analyzed whether MBD3 was phosphorylated during the cell cycle. HeLa-MBD3-FLAG cells were synchronized at the G 1 /S boundary using the thymidine/aphidicolin double-block protocol. The cells were released and harvested at various intervals. Prior to each harvest, the cells were cultured for 90 min in the presence of [ 32 P]orthophosphate. A parallel culture was used for determining the stage in cell cycle by FACScan and fluorescence microscopy. MBD3-FLAG was immunoprecipitated using anti-FLAG antibodies from cell extracts, fractionated by SDS-PAGE, and examined by autoradiography and immunoblot experiments (Fig. 2, A and B). The control lanes for the samples of 12-h post-release and nocodazole-arrested cells showed a high background. These samples were derived from metaphase-rich populations (see below), and it is likely that the high background was due to the relatively higher total kinase activities in metaphase. Specific 32 P-labeling efficiency of MBD3-FLAG was calculated as (signals in autoradiography)/ (signals in immunoblot). The results showed that MBD3-FLAG was significantly phosphorylated at 8, 10, 12, and 14 h and peaked at 10 h, after the release of the block (Fig. 2C). It was also highly phosphorylated in nocodazole-arrested cells. FAC-Scan analysis (Fig. 2D) and DNA staining (Fig. 2E) revealed that the cells at 8 -12 h consisted of the G 2 and M phase cells. Because the cells at 12 h, which consisted of the largest number of metaphase cells (cells in metaphase ϭ 38.8%, n ϭ 1000), showed a lower labeling efficiency than cells at 10 h (cells in metaphase ϭ 0.4%, n ϭ 1000), it was suggested that the phosphorylation predominantly occurred at the late G 2 or early M phase. It should be noted, however, that a significant level of MBD3 phosphorylation was observed at 6-h post-release. It is  A and B, an aliquot of the G 1 /S-arrested cells without release (Release(Ϫ)), exponentially growing asynchronized cells (Expo), and nocodazole-treated metaphase-arrested cells (Noco) were labeled for 90 min with [ 32 P]orthophosphate, followed by harvesting. The G 1 /S-arrested cells were released and harvested at the indicated times, preceded by [ 32 P]orthophosphate labeling for 90 min. Cell extracts were subjected to immunoprecipitation with NMI (NMI) or anti-FLAG (␣-FLAG). Immunoprecipitates were resolved by SDS-PAGE in duplicate. One gel was dried and analyzed by autoradiography (A), and the other was subjected to immunoblot analysis with anti-MBD3 antibodies (B). C, the signal intensity of each band in A and B was quantitated. Specific labeling efficiency was calculated as (signals in autoradiography)/(signals in immunoblot), normalized such that the value for the exponentially growing cells is 1, and plotted. D and E, cells cultured in parallel with the labeling experiments were stained with PI. The DNA contents were analyzed by FACScan (D), and the cell morphology was analyzed using a confocal laser scanning microscope (E). Scale bar, 10 m. The percentages of metaphase cells, obtained by examining 1000 cells in each case, are shown at the bottom.
possible that the MBD3 phosphorylation is not tightly regulated by the cell cycle or, alternatively, that both cell-cycle-dependent and -independent kinases are involved in phosphorylating MBD3. We also found that MTA2, which was immunoprecipitated together with MBD3-FLAG by anti-FLAG antibodies, was phosphorylated in similar kinetics to that of MBD3-FLAG (data not shown).
MBD3 Is Co-localized with Aurora-A at Centrosomes in M Phase-The results described above suggested that MBD3 might be phosphorylated by protein kinases active in the early M phase. To know the potential consequence of this mitotic phosphorylation, we first examined the mitotic localization of endogenous MBD3 using specific antibodies in HeLa cells (which did not produce the MBD3-FLAG protein). In the interphase, MBD3 distributed as numerous fine dots in the nucleus (Fig. 3A). These dot-like structures did not overlap with the centrosomes revealed by anti-␥-tubulin antibodies. However, when cells entered M phase, a fraction of MBD3 signals overlapped with the ␥-tubulin signals at the centrosomes. MBD3 at the centrosomes was not evident when the two centrosomes had not migrated to opposite poles in prophase (Fig. 3B) but became evident when the centrosomes occupied the opposite poles in prometaphase (Fig. 3C). This centrosome-associated MBD3 was observed in metaphase (Fig. 3D) and early anaphase (Fig. 3E). However, it was noted that the MBD3 signals at the centrosomes in anaphase were much weaker than those in the earlier M phase, and MBD3 accumulated at the mid zone of spindles in this stage (Fig. 3E).
The above results indicate that the kinetics of MBD3 phosphorylation was well correlated with that of the appearance of centrosome-associated MBD3. Accordingly, kinases present at mitotic centrosomes may be responsible for the phosphoryla- tion of MBD3. Aurora-A is one such protein kinase. Aurora-A is expressed, activated, and localized at the centrosomes from the late G 2 phase to anaphase (19,29,30). We therefore examined the potential role of Aurora-A in MBD3 phosphorylation. First, localizations of endogenous MBD3 and Aurora-A in HeLa cells were examined using specific antibodies (Fig. 4). In the late G 2 phase and prophase, the earliest time point when Aurora-A localization at the centromeres was manifested, MBD3 was not yet localized at the centrosomes (Fig. 4, A and B). However, a fraction of MBD3 was co-localized with Aurora-A in prometaphase and metaphase cells in which two centrosomes were present at opposite poles (Fig. 4C for prometaphase; Fig. 4D for metaphase). MBD3 remained associated with Aurora-A at the centrosomes in the early anaphase (Fig. 4E). These results suggest that a population of MBD3 is transiently localized at Aurora-A-positive centrosomes from prometaphase to early an-aphase. This association appears to occur only after Aurora-A is already present at the centrosomes and the two centrosomes have moved to opposite poles (Figs. 3 and 4).
MBD3 is a component of the NuRD complex. Given the transient localization of MBD3 at the centrosomes in the early M phase, we were interested in whether other components of NuRD may show similar distributions. IF studies using anti-HDAC1 antibodies revealed that HDAC1 shows essentially the same distribution as MBD3 during M phase. The results of the double staining of HDAC1 and Aurora-A are shown in Fig. 5, and similar results were obtained in the double staining of HDAC1 and ␥-tubulin (data not shown). HDAC1 was co-localized at two oppositely positioned centrosomes with Aurora-A from prometaphase to early anaphase. Interestingly, we did not observe any localization of HDAC1 at the mid zone in late anaphase, although future studies are necessary to confirm these different distributions of MBD3 and HDAC1 in the late M phase. We observed MTA2 similarly distribute at the Aurora-A-positive centrosomes in early M phase (data not shown). Taken together, it was suggested that the NuRD complex transiently interacts with Aurora-A-positive centrosomes in the early M phase.
MBD3 Physically Associates with Aurora-A in Vivo-To obtain biochemical evidence showing the association between MBD3 and Aurora-A, we performed an immunoprecipitation experiment. Genes encoding HA-tagged MBD3 and Glu-tagged Aurora-A were transiently transfected into COS cells individually or simultaneously. The cells were lysed, and protein extracts were subjected to immunoprecipitation with anti-Glu antibodies. Immunoprecipitates were resolved by 10% SDS-PAGE. Immunoblot analysis revealed the presence of HA-MBD3 in the immunoprecipitated complex only when both HA-MBD3 and Glu-Aurora-A were overexpressed (Fig. 6, lane  3). Therefore, it was concluded that MBD3 and Aurora-A associate in vivo either directly or indirectly.
Aurora-A Phosphorylates MBD3-Because MBD3 is phosphorylated in the late G 2 to early M phase (Fig. 2) and Aurora-A is active during these cell cycle stages, it is possible that Aurora-A phosphorylates MBD3. To examine this possibility, we prepared bacterially expressed HA-tagged MBD3 fused with GST (GST-HA-MBD3), HA-tagged GST (GST-HA), and Aurora-A. Aurora-A and either GST-HA-MBD3 or GST-HA were incubated in the presence of [␥-32 P]ATP. The amounts of proteins measured by Coomassie Brilliant Blue (CBB) staining are indicated in Fig. 7B. We found that GST-HA-MBD3, GST-HA, and Aurora-A were efficiently labeled in this in vitro kinase assay (Fig. 7A). One explanation for these results was that Aurora-A phosphorylates the GST-HA sequence present in both MBD3 and control peptides. To examine whether Aurora-A specifically phosphorylates MBD3, we next analyzed the phosphorylated amino acids on MBD3. It was reported that human Aurora-A phosphorylates serine 10 (RKS) of histone H3 in vitro (31) and that Xenopus Aurora-A/Eg2 phosphorylates serine 174 (RLDS) of the cytoplasmic polyadenylation elementbinding factor (32). We assumed that RX(S/T) or RXX(S/T) was a putative targeting sequence by Aurora-A. MBD3 has two serine residues and one threonine residue existing within these contexts (RR-S24, RYD-S85, and RQ-T104, respectively). We prepared HA-and 6xHis-double-tagged wild-type MBD3 (wildtype) in E. coli. Similarly, double-tagged mutant proteins, in which Ser-24, Ser-85, or Thr-104 was replaced with alanine, were also prepared (S24A, S85A, and T104A, respectively). These recombinant MBD3 proteins were subjected to in vitro kinase assay using recombinant Aurora-A. The labeled MBD3 proteins were further digested with trypsin and subjected to two-dimensional phosphopeptide mapping (Fig. 8). Numerous labeled peptide spots (Fig. 8A), schematically shown in Fig. 8B, were observed for wild-type MBD3. Spots 1 and X were the most strongly labeled (Fig. 8B). When the spots were compared between the wild-type and mutant MBD3 proteins, spots 1 and 2, spots 3 and 4, and spot 5 disappeared in S24A, S85A, and T104A, respectively. Spot X remained in all mutants. Because spot 1 is one of the major labeled peptides, we concluded that Ser-24 is one of the major amino acids phosphorylated by Aurora-A in vitro. Although the origin of peptide X is unknown at present, it is likely derived from linker sequences. Finally, we determined the phosphoamino acids in wild-type MBD3 (Fig. 9). The most significantly phosphorylated amino acid was serine, an observation consistent with the results obtained in Fig. 8.

DISCUSSION
In this study, we have shown that MBD3 and HDAC1, and possibly the NuRD complex, exhibit dynamic changes in their spatial distributions during M phase, which have not been described before. The kinetics of these spatial movements is similar to that of MBD3 phosphorylation that transiently occurs during the late G 2 and early M phases, another novel finding described here. These results suggest that the biochemical changes of MBD3 may be intimately related to the targeting of MBD3 to centrosomes. We also found that Aurora-A is a candidate kinase for the phosphorylation of MBD3.
Essential Roles of HAT/HDAC System in M Phase-Although the dynamic changes in the spatial distributions of MBD3 and HDAC1 in M phase have not been described before, they are not necessarily unexpected. A fair number of studies revealed that histone acetylation is involved in cell cycle progression. In normal human fibroblasts, HDAC inhibitors induce G 2 /M arrest, possibly by activating a checkpoint mechanism (33). In transformed cells that are presumably defective in this checkpoint, HDAC inhibitors result in a catastrophic mitosis. A similar observation was also reported in yeast. The deletion of ESA1, a gene encoding the H4-specific HAT belonging to the MYST family, is lethal in budding yeast. Interestingly, in these cases, cells are arrested at G 2 /M, indicating defects in the progression of mitosis, and this arrest depends on the Rad9 checkpoint pathway (34,35). The checkpoint is generally believed to monitor DNA damage, chromosomal dysfunction, and inappropriate progression of the cell cycle. Therefore, it is likely that these G 2 /M arrests caused by the imbalance in HAT/HDAC equilibrium are due not to failure in controlling the expression of specific genes, but to more global defects in chromatin structures and/or chromosomal functions. Given this important role of HATs and HDACs at the entry of and during the progression of M phase, one would expect that HATs and HDACs change their intracellular distribution and catalytic activity dynamically at this stage. Indeed, a recent study has shown that HATs and HDACs are displaced from condensing chromosomes and unable to modify chromatin-bound histones in M phase (36). As an extension of this notion, we have shown in this study that at least a fraction of HDAC1 and MBD3 is targeted to the centrosomes during the early M phase. This novel finding suggests that HDAC1 is not passively sequestered from chromatin but positively plays a yet unrecognized role at the centrosomes.
It should be noted that the essential role of the HAT/HDAC system in M phase does not necessarily indicate that histone proteins are the substrates of the system. Because we have demonstrated that HDAC1 is transiently localized at M phase centrosomes, with which histone is presumably not associated, the possibility that non-histone proteins are modified by HDAC1 at the centrosomes should be considered.
Roles of HDAC1 at Centrosomes-A growing number of nonhistone proteins have been identified as a substrate for HAT. Acetylation regulates many aspects of protein function, including DNA recognition, protein-protein interaction, and protein stability (reviewed in Ref. 37). Similarly, an increasing number of acetylated proteins are deacetylated by HDACs. Among them, p53 is of special interest in the context of this study. The p300/CBP HAT acetylates and activates p53 as a transcription factor by increasing its DNA-binding activity (38,39). p53 is deacetylated and inactivated in both TSA-sensitive and -insensitive pathways. HDAC1 in the NuRD complex is responsible for the TSA-sensitive deacetylation pathway (40), whereas SirT1 is responsible for the TSA-insensitive pathway (41,42). Interestingly, it was found that MTA2 directly binds to p53, and accordingly MTA2 was proposed to recruit p53 to the NuRD complex. It is known that p53 is localized at the centrosomes and spindles in M phase (43). Because we have found that HDAC1, MBD3, and MTA2 are co-localized at the centrosomes in M phase, it is possible that the TSA-sensitive deacetylation of p53 occurs at the mitotic centrosomes.
Phosphorylation of MBD3-We also demonstrated that MBD3 is phosphorylated during the late G 2 and early M phases. Aurora-A is a key regulator of structure and function of the mitotic spindle in yeast, fly, worm, and frog (21,22,44,45). Mutants defective in Aurora-A show such phenotypes as the failure in centrosome separation and the formation of monopolar spindles. Moreover, human Aurora-A is amplified in several human cancers, and overexpression of Aurora-A causes malignant transformation in cultured cells (27, 46 -49). Importantly, the expression, enzymatic activity, and subcellular localiza- tion are tightly regulated in a cell cycle-dependent manner. The expression of Aurora-A peaks at the G 2 /M transition, and the enzymatic activities also peak in G 2 and in prophase. Studies on the subcellular localization of Aurora-A revealed its association with mitotic structures: Aurora-A is highly localized at the centrosomes and spindle of mitotic cells. These spatio-temporal profiles of Aurora-A are similar to those of MBD3. Indeed, we found that 1) MBD3 accumulates on Aurora-A-positive centrosomes, 2) Aurora-A binds to MBD3 both in vivo and in vitro, and 3) MBD3 is phosphorylated by Aurora-A in vitro. From these observations, we suggest that Aurora-A is a candidate kinase that phosphorylates and accordingly regulates MBD3 in vivo. It should be noted, however, that the peak phosphorylation of MBD3 (G 2 -early M) appeared relatively earlier than the co-localization of MBD3 with Aurora-A at centrosomes (prometaphase). Moreover, phosphorylation reactions in vitro do not always reflect reactions in vivo. We therefore do not exclude the possibility that kinases other than Aurora-A is responsible for the regulation of MBD3. Several reports have shown that HDAC1 and HDAC2 are phosphorylated and their catalytic activities and complex formations were regulated by this post-translational modification (50 -52). However, the physiological significance of HDAC phosphorylation remains to be determined. It would be interesting to examine whether HDAC1 and HDAC2 also interact with and are phosphorylated by Aurora-A or not.
Although the precise function of Aurora-A in the mitotic NuRD complex in vivo needs to be determined in the future, the interaction between MBD3 and Aurora-A is intriguing in the light of the potential link between p53 and the NuRD discussed above. p53 controls cell cycle progression, not only through its action on cell cycle regulators, but also through regulation of the function of centrosomes. The absence of intact p53 leads to abnormal centrosome duplications (53). It is therefore attractive to envision that p53 may be controlled by the Aurora-Aregulated NuRD complex at the centrosomes. FIG. 9. Serine residues in MBD3 are phosphorylated by Aurora-A in vitro. Wild-type MBD3 was phosphorylated by purified recombinant Aurora-A. The protein was separated by SDS-PAGE and excised from the gel. Amino acids obtained after hydrochloric acid hydrolysis were separated by high voltage electrophoresis and analyzed by autoradiography. The identities of the spots are shown.

FIG. 7. MBD3 is phosphorylated by
Aurora-A in vitro. HA-tagged MBD3 or only the HA epitope was fused to GST. Bacterially expressed GST-HA-MBD3 or GST-HA fusion proteins were subjected to in vitro kinase reactions with purified Aurora-A. The reaction samples were resolved by 10% SDS-PAGE and analyzed by autoradiography (A) or stained by CBB (B).

FIG. 8. MBD3 Ser-24 is significantly phosphorylated by Aurora-A in vitro.
6xHis-HA-tagged wild-type (A) and mutant MBD3 (S24A in C, S85A in D, and T104A in E) were phosphorylated by purified recombinant Aurora-A and digested with trypsin. The peptides were separated by high voltage electrophoresis followed by ascendant chromatography. Phosphopeptide maps were obtained by autoradiography. Schematic presentation of spots for wild-type MBD3 is shown in B.