Histone Deacetylase 3 Interacts with Runx2 to Repress the Osteocalcin Promoter and Regulate Osteoblast Differentiation*

The Runt domain transcription factor Runx2 (AML-3, and Cbfa1) is essential for osteoblast development, differentiation, and bone formation. Runx2 positively or negatively regulates osteoblast gene expression by in-teracting with a variety of transcription cofactor complexes. In this study, we identified a trichostatin A-sen-sitive autonomous repression domain in the amino terminus of Runx2. Using a candidate approach, we found that histone deacetylase (HDAC) 3 interacts with the amino terminus of Runx2. In transient transfection assays, HDAC3 repressed Runx2-mediated activation of the osteocalcin promoter. HDAC inhibitors and HDAC3-specific short hairpin RNAs reversed this repression. In vivo , Runx2 and HDAC3 associated with the osteocalcin promoter. These data indicate that HDAC3 regulates Runx2-mediated transcription of osteoblast genes. Suppression of HDAC3 in MC3T3 preosteoblasts by RNA interference accelerated the expression of Runx2 target genes, osteocalcin, osteopontin, and bone sialoprotein but did not significantly alter Runx2 levels. Matrix mineralization also occurred earlier in HDAC3-suppressed cells, but alkaline GST-Runx2-(1– GST-Runx2-(1–227), In con-trast, HDAC3 did not interact with GST, GST-Runx2-(94–327), or GST-Runx2-(321–513). These indicate that the extreme amino terminus (residues 1–117) of Runx2 is necessary and sufficient for the interaction with HDAC3. The runt domain (RuntD) of Runx2 is 93% identical to the RuntD of Runx1.

Runx2 is one of three mammalian homologues of the Drosophila protein Runt. Runx2 is required for osteoblast formation and chondrocyte differentiation and thus is an essential regulator of intramembranous and endochondral bone formation (1)(2)(3). Runx2 haploinsufficiency causes the autosomal dominant bone disorder cleidocranial dysplasia (4,5), but elevated Runx2 levels cause osteopenia (6) and contribute to murine T cell lymphomogenesis (7). Furthermore, Runx2 polymorphisms are associated with altered bone mineral density and susceptibility to osteoporotic fractures (8). The sensitivity of osteoblasts to changes in wild-type Runx2 levels provides biological proof that its activity must be tightly controlled. At the molecular level, Runx2 is regulated by transcriptional and translational mechanisms, post-translational modifications, including phosphorylation, and by interactions with other transcription factors and chromatin-modifying enzymes (9 -20). Because the Runx DNA-binding element is necessary but not sufficient to activate tissue-specific gene expression (21,22), it was hypothesized over a decade ago that Runx proteins are promoter organizers (23). It is now apparent that Runx factors indeed interact with other transcription factors as well as recruit both co-activator and co-repressor complexes to promoters. Known co-activators of Runx2 include YAP, HES, and the histone acetyltransferases, p300, CBP, MOZ, and MORF (24 -27). The list of Runx2-associating co-repressors includes just TLE, mSin3a, and histone deacetylase (HDAC) 1 6 (18, 19, 28, 29); however, among all known transcriptional co-repressors, only a small fraction have been tested for interactions with Runx2. Because Runx2 is crucial for bone formation, we are interested in understanding how its transcriptional activity is affected by co-repressor complexes in osseous cells.
HDACs are enzymatic components of multiprotein complexes that are recruited by transcription factors to specific DNA regulatory sequences. They remove acetyl residues from nucleosomal histones and other substrates leading to chromatin condensation and gene repression (30 -34). Eleven mammalian HDACs are classified into two groups on the basis of sequence similarities with yeast proteins. Class I HDACs (HDAC1-3, -8, and -11) are homologues of yeast RPD3 and are widely expressed in the nuclei of mammalian cells (35)(36)(37). Class II HDACs (HDAC4 -7, -9, and -10) are similar to yeast Hda1/2 and appear to shuttle between nuclear and cytoplasmic compartments in specific cell types (18, 38 -45). Major functions of HDACs are to regulate gene expression, transport ubiquitinated protein aggregates, and deacetylate tubulin (46 -48). Global suppression of HDAC activity by chemical inhibitors commonly causes cell cycle arrest and may promote cell differentiation (49,50).
HDAC3 is one of the better-studied HDACs. It associates with several transcription factors, including GATA1, peroxisome proliferator-activated receptor-␥, retinoblastma protein, thyroid hormone receptor, TEL, and TFII-I (51)(52)(53)(54)(55). HDAC3 also interacts with the co-repressors N-CoR and SMRT (53,56), which link HDAC3 to transcription factors and other co-repressors. HDAC3 is also the crucial catalytic component of several class II HDAC complexes (57,58). Thus, HDAC3 appears to be a very important factor in multiple co-repressor complexes that regulate gene expression. HDAC3-deficient animals have not been described; however, HDAC3 suppression in HeLa cells by RNA interference blocked cell proliferation and transcriptional repression by thyroid hormone receptor (59,60).
In this study, we identified HDAC3 as a transcriptional co-repressor of Runx2 using a candidate gene approach. HDAC3 interacted with the amino terminus of Runx2, which acts as an autonomous repression domain on a heterologous promoter. HDAC3 repressed Runx2-induced activation of the osteocalcin promoter and was found associated with the osteocalcin promoter in osteoblasts. Stable suppression of HDAC3 by RNA interference accelerated osteoblast mineralization and the expression of osteoblast differentiation genes. These data identify HDAC3 as a regulator of Runx2 transcriptional activity in osseous cells and suggest that its expression regulates bone formation.
Plasmids-The GAL-TK-luciferase (Luc) reporter plasmid and the GAL-Runx2 fusion proteins were previously described (18). Human and mouse HDAC3 mammalian expression vectors were kindly provided by Dr. Edward Seto (Lee Moffit Cancer Center, Tampa, FL) (61). The pSHAG and pSHAG-Firefly luciferase (Ffl) vectors were obtained from Dr. Gregory Hannon (Cold Spring Harbor Laboratories) (62). The pS-HAG-HDAC3 short-hairpin RNA vectors were constructed by anneal-FIG. 1. The amino terminus of Runx2 contains a repression domain that is sensitive to the HDAC inhibitor, TSA. A, schematic of the GAL-TK-Luc reporter plasmid that was used in these experiments. B, left: a diagram of GAL-Runx2 fusion proteins that we constructed for these studies. Right: immunoblot showing expression of the GAL-Runx2 fusion proteins. COS cells were transfected with the GAL-Runx2 fusion protein expression plasmids. Lysates were collected 40 h post-transfection and immunoblotted with an anti-Gal monoclonal antibody. RuntD, Runt domain; AD, activation domain. C, the amino terminus of Runx2 contains autonomous repression domains. NIH3T3 cells were transfected in triplicate with GAL-TK-Luc (1 g) and CMV-SEAP (1 g) reporter plasmids and expression plasmids for GAL (1 g) or GAL-Runx2 fusion proteins (1 g). Luciferase activity was measured 40 h post-transfection. The fold repression is relative to GAL alone, and the transfection efficiency is normalized using SEAP activity. D, the histone deacetylase inhibitor, TSA, reduces Runx2-dependent repression. The transfections and luciferase assays were performed as described in C except Me 2 SO (gray bars) or 300 nM TSA (black bars) was added to the samples 24 h post-transfection.
The self-inactivating (SIN) retrovirus was created from a murine stem cell virus (MSCV) expression plasmid, by removing the XbaI/NheI fragment from the 3Ј long terminal repeat of the vector, pMSCV-Puro-HpaI-Gateway (a gift from Dr. Gregory Hannon (63)). Short-hairpin RNAs recognizing Ffl or HDAC3 were transferred from pSHAG vectors into the pSIN-MSCV-puro-HpaI-Gateway vector with the Gateway cloning system (Invitrogen). The resulting plasmids are referred to as pSIN-MSCV-Ffl shRNA and pSIN-MSCV-HDAC3 shRNA.
Creation of Stable Cell Lines-Retroviruses were produced as previously described (64). Briefly, 293T cells were co-transfected with 20 g of pSIN-MSCV-shRNA vector and 20 g of the helper plasmid pCL2 by calcium phosphate precipitation. At 24 and 48 h, supernatants were collected, passed through a 0.45-m filter and stored at Ϫ80°C. MC3T3 cells were transduced with 1 ml of viral supernatant in the presence of 8 g/ml polybrene on two consecutive days. Transduced cells were selected by culturing in the presence of 10 g/ml puromycin for 3 days.
Antibodies-The polyclonal antibodies recognizing HDAC3 or HDAC4 were purchased from Cell Signaling Technologies (Beverly, MA). The polyclonal anti-Runx2 antibodies, S19 (N terminus) and C19 (C terminus) were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). The polyclonal anti-AML3 antibody was provided by Dr. Scott Hiebert. The monoclonal anti-HA and polyclonal anti-FLAG antibodies were purchased from Sigma-Aldrich Co. (St. Louis, MO). The monoclonal anti-␤-catenin antiserum was purchased from BD Biosciences Transduction Laboratories. The FHOD1 antiserum was previously described (64). Transcription Assays-COS and C3H10T1/2 cells were transiently transfected with the murine osteocalcin gene 2 (mOG2)-Luc reporter plasmid, pCMV-secreted alkaline phosphatase (SEAP), pCMV-Runx2 and pCMV-FLAG-HDAC3, and pSHAG-shRNA using DEAE-dextran or LipofectAMINE (Invitrogen), respectively. NIH3T3 cells were transfected with the GAL-TK-Luc reporter plasmid and GAL-Runx2 plasmids with Superfect (Qiagen, Valencia, CA). Luciferase activity was measured 40 h post-transfection with a luciferase assay system (Promega, Madison, WI). SEAP activity was measured as previously described (18,65). Luciferase values were normalized with SEAP values to control for transfection efficiency. The fold changes in activation or repression were calculated relative to samples transfected with mOG2-Luc or GAL-TK-Luc alone. Trichostatin A (TSA, Sigma) and cyclic hydroxamic acid-containing peptide 31 (CHAP31, kindly provided by Minoru Yoshida of The University of Tokyo) were added 24 h post-transfection.
GST Pull-down Assays-GST and GST-Runx1 or Runx2 fusion proteins were produced in DH5␣ Escherichia coli and extracted with a lysis buffer containing 1ϫ phosphate-buffered saline, 0.1% Nonidet P-40, and protease inhibitors (Mini Complete protease inhibitor mixture tablets, Roche Applied Science). Lysates were sonicated on ice and cleared by centrifugation. GST proteins were purified by incubating E. coli lysates with glutathione-Sepharose beads (Amersham Biosciences) for 1 h at 4°C. HDAC3 proteins were in vitro transcribed and translated in rabbit reticulocyte lysates (Promega) with 35 S-labeled amino acids (Tran 35 S-label, ICN Pharmaceuticals, Inc., Irvine, CA) from pBluescript-HDAC3 according to the manufacturer's instructions. GST proteins and radiolabeled murine HDAC3 were mixed and incubated for 1 h at 4°C, washed three times with lysis buffer, resolved by SDS-12% PAGE, and visualized by autoradiography.
Quantitative Real-time PCR-cDNA was prepared as described above. Real-time PCR was carried out in a Lightcycler (Roche Diagnostics) using the following primer sets: HDAC3 (5Ј-CGGGTGCTCTACA-TTGATATCGACATCCAC-3Ј and 5Ј-TACCAGGAGAGGGATATTGAA-ACTCTTGAC-3Ј), osteocalcin (5Ј-CTCTGTCTCTCTGACCTCACAG-3Ј and 5Ј-GGAGCTGCTGTGACATCCATAC-3Ј), osteopontin (5Ј-TGACC-CATCTCAGAAGCAG-3Ј and 5Ј-GCTGACTTGACTCATGGCT-3Ј), bone sialoprotein (5Ј-GAAACGGTTTCCAGTCCAG-3Ј and 5Ј-CTGCATCTC-CAGCCTTCTT-3Ј), and actin (5Ј-AAGGAAGGCTGGAAAAGAGC-3Ј and 5Ј-GCTACAGCTTCACCACCACA-3Ј). HDAC3, osteopontin, bone sialoprotein, and actin were amplified by adding 1 l of cDNA diluted 1:10 to the PCR mixture (19 l) containing each of the primers (0.5 M), 1.25 mM MgCl 2 , and 10 l of the SYBR Green Taq ReadyMix (Sigma-Aldrich Co.). The enzyme was activated at 95°C for 5 min prior to the start of cycling. The PCR program contained 40 cycles of denaturation at 94°C for 10 s, annealing of the primers at 56°C for 10 s, and elongation at 72°C for 10 s. Osteocalcin and actin were amplified by adding 2 l of cDNA diluted 1:10 to the PCR mixture (18 l) containing each of the primers (0.5 M), 0.05 mg/ml bovine serum albumin, and 10 l of the Platinum SYBR Green qPCR Supermix uracil glycosylase (Invitrogen). Prior to cycling, the reactions were preincubated at 50°C for 2 min for decontamination of dU-containing DNA by UDG, then at 95°C to inactivate UDG and activate Taq. The PCR program was the FIG. 4. HDAC3 represses Runx2-mediated activation of the osteocalcin promoter. A, HDAC3 represses Runx2transcriptional activation. COS cells were transfected with the reporter plasmids, mOG2-Luc (0.2 g) and CMV-SEAP (0.5 g), and mammalian expression vectors for Runx2 (MRIPV isoform) (1 g) and HDAC3 (at indicated concentrations). Luciferase activity was determined 40 h after transfection. The fold activation is relative to mOG2-Luc alone, and transfection efficiency is normalized using SEAP activity. B, HDAC3 repression of Runx2 is sensitive to HDAC inhibitors, TSA and CHAP31. The transfection and luciferase assays were performed in C3H10T1/2 cells as described in A except the mammalian expression vector for Runx2 (MASNS isoform) was used, 1 g of HDAC3 was transfected and TSA or CHAP31 were added at the indicated concentrations 24 h post-transfection.
Runx2 Interacts with HDAC3 same as described above. For all reactions, the temperature change rate was 20 C°/s for denaturation, annealing, and elongation. Fluorescence was measured at channel F1 at the end of each elongation cycle. Relative quantification of gene expression was determined by using the 2 -⌬⌬CT method where fold change in gene expression is relative to Day 0 Ffl shRNA samples (67). All samples were normalized to actin.
Differentiation Assays-To assess alkaline phosphatase production, differentiating MC3T3-E1 cells were stained at the indicated times in a buffer containing 100 mM Tris, 100 mM NaCl, 50 mM MgCl 2 , 337.5 mg/ml nitroblue tetrazolium, and 175 mg/ml 5-bromo-4-chloro-3-indolyl phosphate tolidium salt. Ca 2ϩ accumulation in the matrix was detected by staining the differentiating cell cultures with 40 mM Alizarin Red for 10 min (LabChem Inc., Pittsburgh, PA), washing with distilled water and developing in phosphate-buffered saline.

RESULTS
The Amino Terminus of Runx2 Acts as an Autonomous Repression Domain That Is Sensitive to the HDAC Inhibitor, TSA-We previously showed that Runx2 contains multiple repression domains (18). That report focused on a potent repression domain in the Runx2 carboxyl terminus (encompassing amino acids 383-513 of the MRIPV isoform), which interacts with TLE and HDAC6. We also showed that a truncated Runx2 protein (residues 1-383) lacking these repression domains suppressed transcription, albeit at lower levels. To identify the repression domain(s) within the first 383 residues of Runx2, we fused various Runx2 sequences to the GAL4 DNA binding domain (GAL) (Fig. 1B). NIH3T3 cells were transiently transfected with the GAL-Runx2 constructs and a reporter, GAL-TK-Luc, containing four GAL binding sites (Fig. 1A). Similar to our previously published results, GAL-Runx2-(1-513) repressed the promoter 6.2-fold, and GAL-Runx2-(1-383) repressed it 13.7-fold (Fig. 1C). Three Runx2 proteins ((1-327), , and (1-227)) lacking the activation domain also repressed transcription by 10.6-, 3.8-, and 3.3-fold, respectively. A GAL-Runx2 protein (321-383) containing just the Runx2 transactivation domain did not repress transcription (data not shown). These data demonstrate that Runx2 contains an autonomous repression domain within its amino terminus.
To determine if HDAC activity is involved in the repression by the amino terminus of Runx2, the HDAC inhibitor, trichostatin A (TSA), was added to the transcription assays (Fig.  1D). TSA partially inhibited the repression of Runx2-(1-383) and Runx2-(1-327), but completely reversed repression by Runx2-(1-250). These results suggest that an HDAC complex interacts with the first 250 residues of the Runx2 and contributes to Runx2-mediated transcriptional repression; however, HDAC-independent mechanisms also contribute to Runx2-mediated repression.
Runx2 Interacts with HDAC3-We used a candidate approach to identify HDACs that may interact with the amino terminus of Runx2. COS cells were co-transfected with pCMV5-GAL-Runx2-(1-227) and pCMV-FLAG-HDAC 1, 2, 3, 4, 5, or 6. GAL-Runx2-(1-227) was used because it repressed as well as GAL-Runx2-(1-250) in repression assays (Fig. 1C) and was more stable than GAL-Runx2-(1-250) (Fig. 1B). Whole cell extracts were immunoprecipitated with anti-FLAG-conjugated agarose beads. The immunoprecipitates and a fraction of the whole cell extracts were separated by SDS-PAGE and immunoblotted with anti-GAL and anti-FLAG antibodies. GAL-Runx2-(1-227) was only present in the HDAC3 and HDAC4 immune complexes (Fig. 2). Although more Runx2 appears to interact with HDAC3 complexes than with HDAC4 complexes, it is important to note that there is significantly less GAL-Runx2-(1-227) protein in the lysates of cells expressing HDAC4. In fact, all the HDACs tested, with the exception of HDAC3, blocked the expression of the GAL-Runx2 fusion protein. This likely occurs because the HDACs repress transcrip-tion from the CMV intermediate early promoter, 2 , and therefore in this experiment FLAG-HDACs are suppressing their own ectopic expression as well as that of GAL-Runx2. These results are informative and crucial for the interpretation of functional studies of HDACs described later in this report (see Fig. 4). These data show that HDAC3 does not suppress Runx2 activity by decreasing Runx2 levels. HDAC4, on the other hand, does suppress ectopic Runx2 expression. This limited our exploration of the HDAC4 interaction, because our functional assays utilize a CMV promoter to drive Runx2 expression. To gain a better understanding of the roles that HDAC3 and HDAC4 may play in regulating Runx2 in osteoblasts, we performed immunoblots on protein extracts from preosteoblastic MC3T3 cells, the myogenic/osteoblast progenitor cell line, C2C12, and two rat osteosarcoma cells lines, ROS17/2.8 and UMR-106. HDAC3 was detected in each of the cell lines, but HDAC4 was detectable only in UMR-106 osteosarcoma cells (Fig. 3A). All of these cells lines express Runx2 (data not shown). Osteocalcin mRNA was detectable in at low levels in MC3T3 (see Fig. 7E) and at higher levels in the ROS17/2.6 cells, but not in the UMR-106 or C2C12 cells (data not shown). Because HDAC3 was more broadly expressed than HDAC4 in this panel of cells, we decided to focus on its role in regulating Runx2 more closely for the rest of this study. Using immunoprecipitates from the rat osteosarcoma cell line ROS17/2.8, we ascertained that full-length Runx2 interacts with HDAC3 in vivo (Fig. 3B). HDAC3 was present in three Runx2 immune complexes that were collected with antibodies recognizing distinct epitopes of Runx2 (Fig. 3B), as well as in the HDAC3 immune complexes. HDAC3 was not present in the control immunoprecipitates. Together these data demonstrate that HDAC3 interacts with Runx2 in vivo.
HDAC3 Represses Runx2-dependent Activation of the Osteocalcin Promoter-The functional significance of the Runx2-HDAC3 interaction was determined by testing the effects of HDAC3 on Runx2 transcriptional activity. The osteocalcin promoter is a well characterized target of Runx proteins and contains three binding sites for Runx2 (70,71). COS cells were transiently transfected with an osteocalcin reporter plasmid (mOG2-luc) (72) and mammalian vectors expressing Runx2 and HDAC3. Cell lysates were collected, and luciferase assays were performed. Consistent with previous results from our laboratory and others, Runx2 (MRIPV isoform) activated the bone-specific osteocalcin promoter by 7.5-fold (Fig. 4A) (1, 73,  74). Similarly, the longer Runx2 isoform (MASNSL) activated the osteocalcin promoter by 11.5-fold (Fig. 4B, lane 6). HDAC3 did not effect the basal activity of the osteocalcin promoter but inhibited Runx2-dependent activation of this promoter in a concentration-dependent manner (Fig. 4, A and B). The repression was sensitive to the HDAC inhibitors, TSA and CHAP31 (Fig. 4B). These inhibitors modestly increased the basal activity of the promoter (lanes 1-5) and Runx2-dependent activation (lanes 6 -8). Importantly, both inhibitors completely reversed HDAC3-mediated repression of Runx2 activity (lanes 12-15); although, high concentrations of CHAP31 (100 nM) were slightly toxic. These data demonstrate that HDAC activity is required for HDAC3-mediated repression of Runx2.
To confirm that HDAC3 is required for the repression, sh-RNAs that suppress HDAC3 expression by RNA interference were added to the transcription assays. We created four sh-RNAs for the murine HDAC3 cDNA and subcloned them into the pSHAG vector. In transient expression assays, only one shRNA (denoted here as #1) suppressed HDAC3 levels in COS7 cells (Fig. 5A). A second shRNA (#2) that did not affect HDAC3 levels was used as a control for these experiments. In the absence of any HDAC3 shRNAs, Runx2 (MRIPV isoform) activated the osteocalcin promoter 8-fold, and HDAC3 blocked this activation as previously shown (Fig. 5B, top). Addition of HDAC3 shRNA #1 prevented HDAC3-mediated repression of Runx2 (Fig. 5B, middle). In contrast, HDAC3 shRNA #2, which does not suppress HDAC3 expression, did not reverse HDAC3mediated repression (Fig. 5B, bottom). Together these data demonstrate that HDAC3 represses Runx2-dependent activation of the osteocalcin promoter.
HDAC3 Associates with the Osteocalcin Promoter-Having determined that HDAC3 is expressed in osteoblasts, binds to Runx2, and represses Runx2-dependent activation of the osteocalcin promoter, we next asked if HDAC3 is associated with the osteocalcin promoter in vivo. To examine this, chromatin immunoprecipitation (ChIP) experiments were performed in ROS17/2.8 cells. Cross-linked and fragmented DNA-protein complexes were immunoprecipitated with either no antibody, anti-HA antibodies, Runx2 antibodies (S19 or C19), or anti-HDAC3 antibodies. Nested PCR analysis on purified DNA was performed with primers that span the proximal Runx2 binding site in the osteocalcin promoter. As shown in Fig. 6, osteocalcin promoter DNA was amplified in samples collected with Runx2 or HDAC3 antibodies (Fig. 6). There was little or no amplification in the mock sample (no chromatin input) and samples collected with beads alone (no antibody) or the anti-HA antibodies (nonspecific control). These data demonstrate that HDAC3 can interact with the region of the osteocalcin promoter containing a Runx2 binding site in ROS17/2.8 cells. Together with our data showing that HDAC3 interacts with Runx2 and represses Runx2-dependent transcription of the osteocalcin promoter, these results suggest that HDAC3 may modulate osteocalcin expression.
HDAC3 Suppression Accelerates Osteoblast Differentiation-To determine the functional significance of HDAC3 on osteoblast differentiation, HDAC3 expression was suppressed in MC3T3 cells using RNA interference. The HDAC3 shRNAs FIG. 6. HDAC3 associates with the osteocalcin promoter. ChIP assays demonstrate that HDAC3 interacts with the osteocalcin promoter. ROS17/2.8 cells were cross-linked with formaldehyde, washed, lysed, and immunoprecipitated with no antibody, nonspecific (anti-HA), Runx2 (S19 and C19), or HDAC3 antibodies. Two rounds of PCR were performed using nested primers that encompass the proximal Runx2 binding site within the osteocalcin promoter. Lanes 1 and 2 are PCR controls, lanes 3-5 are negative controls for ChIP, and lanes 6 -8 are test samples.

Runx2 Interacts with HDAC3
were placed into a pSIN-MSCV-Puro vector and stably introduced into MC3T3 cells by retroviral transduction. HDAC3 protein levels in MC3T3 cells expressing shRNA #1 were less than 50% of those in cells expressing Ffl shRNA (Fig. 7A). HDAC3 mRNA levels were down-regulated by ϳ5-fold in undifferentiated MC3T3 cells (Fig. 7, B and E) and remained low in differentiating populations. Runx2 mRNA levels were unchanged (Fig. 7B). We did not observe any drastic changes in cell proliferation or growth as a result of this level of HDAC3 suppression (data not shown). We also did not observe any changes in the levels of a double strand RNA response gene (oligoadenylate synthetase 1) as a result of shRNA production (data not shown). When the MC3T3 cells expressing shRNAs for HDAC3 or Ffl were placed in an in vitro differentiation assay, no gross changes in the rate of alkaline phosphatase production, which is a relatively early marker of osteoblast differentiation, were seen (Fig. 7C). In contrast, matrix mineralization and nodule formation occurred 2-3 days earlier in HDAC3-suppressed cells as compared with control cells (Fig.  7D). To determine whether genes that are both regulated by Runx2 and involved in mineralization were also expressed earlier in HDAC3-suppressed cells, we collected RNA at time points up to the onset of mineralization (day 6) and performed quantitative PCR. HDAC3 expression in MC3T3 cells containing shRNA #1 was down-regulated at day 0 and remained suppressed at days 4 and 6. Consistent with our ChIP and functional assays, osteocalcin mRNA was increased at day 4 in HDAC3-suppressed cells relative to MC3T3 cells expressing Ffl shRNA (Fig. 7E). Low levels of osteocalcin transcripts were detected in undifferentiated cells. By day 6, osteocalcin transcripts were expressed at equal levels in the control and HDAC3-suppressed cells. Relatively similar results were seen with osteopontin and bone sialoprotein, although bone sialoprotein transcripts were not detected in undifferentiated cells (75,76). Thus, HDAC3 suppression accelerates osteoblast mineralization and the expression of various genes that are upregulated during osteoblast differentiation. These data suggest that HDAC3 plays a role in regulating the timing of osteoblast differentiation, perhaps by directly regulating Runx2 activity. DISCUSSION In this report, we demonstrate that HDAC3 regulates Runx2 trans-activity and contributes to in vitro osteoblast mineralization. Our studies show that HDAC3 interacts with the amino terminus of Runx2 to regulate osteoblast-specific gene expression. A physical interaction between Runx2 and HDAC3 was demonstrated in vivo using co-immunoprecipitation. Functional interactions between these proteins were also observed in in vitro transcription assays. Additional studies indicated that Runx2 and HDAC3 co-localize to similar subnuclear structures (data not shown). We mapped the interaction to the amino-terminal 117 residues of Runx2, which contains an autonomous repression domain. To date, HDAC3 is the first protein shown to interact with the amino terminus of Runx2. Our biochemical studies were performed with the Runx2 isoform beginning with amino acids MRIPV; however, functional stud-ies demonstrated that HDAC3 represses this isoform as well as the alternative isoform starting with residues MASNSL (Fig.  4B). Thus HDAC3 can regulate both major Runx2 isoforms. The region of the Runx2 amino terminus that is shared by both isoforms contains a glutamine-alanine (QA) repeat region (residues . It is possible that the QA domain contributes to interactions with HDAC3. However, Runx1 also interacts with HDAC3 but does not contain a QA-rich region (68). Further structural and functional examinations of this domain will be important, because it is altered in a small percentage of cleidocranial dysplasia patients and in individuals with varying bone densities (5,8).
HDAC3 repressed Runx2-dependent activation of the osteocalcin promoter in transiently transfected cells. This repression was sensitive to the HDAC inhibitors, TSA and CHAP31. In addition, the activity was specific to HDAC3, because shRNAs directed against HDAC3 eliminated the repression. Previous studies implicated histone acetyltransferases in activation of the osteocalcin promoter (26) and described the amino terminus, including the QA-rich region, as an activation domain in the context of the entire protein but not autonomously (77). This is the first study to identify a HDAC as a co-repressor of the osteocalcin promoter. Our results may indicate that the activity of this region depends on interactions with specific cofactors. Moreover, these studies support the hypothesis that chromatin remodeling is essential for the regulation of the osteocalcin promoter (78,79). The one or more signals that switch Runx2 from an activator to a repressor remain to be identified. An interesting possibility is that cellular signaling cascades targeting Runx2 (80) will affect its interactions with co-repressors. Runx2 is a phosphoprotein (80), and phosphorylation of specific residues may dictate interactions with cofactor complexes. Phosphorylation of JNK and c-Jun relieve interactions between c-Jun-and HDAC3-containing repression complexes (81). Because the AP-1 complex cooperates with Runx2 to activate osteoblast genes (20,82), it is also a strong candidate to indirectly regulate HDAC3 interactions with Runx2. Of interest, Imai and colleagues (83) recently demonstrated that interactions between mSin3A and Runx1 (AML1) are relieved by Runx1 phosphorylation. The serine residues that are modified to control mSin3a interactions are conserved in Runx2. Current studies in our laboratory are focusing on their roles in regulating the interactions between Runx2 and co-repressors, including HDAC3.
Although mammalian Runt domain proteins are best known for their ability to activate gene expression, their Drosophila homologue Runt was originally described as a transcriptional repressor (84). Within the last 3 years, the underappreciated repression capabilities of mammalian Runt domain proteins have been identified and numerous mechanisms of repression reported. Runt domain proteins interact with a variety of transcriptional co-repressors to regulate gene expression (18,28). Each family member interacts with the Groucho/TLE and mSin3A proteins (28). Runx1 interacts with at least six HDACs (68). We have shown that Runx2 interacts with HDAC3, Fig. 7. HDAC3 accelerates later stages of osteoblast differentiation. A, stable suppression of HDAC3 by RNA interference in MC3T3 cells. Whole cell lysates (75 g) were prepared from undifferentiated MC3T3 cells stably expressing Ffl or HDAC3 shRNAs and immunoblotted with HDAC3 and ␤-catenin antibodies. B, HDAC3 suppression does not alter Runx2 expression in stable MC3T3 cell lines. Reverse transcription-PCR was performed on RNA collected from undifferentiated MC3T3 cells stably expressing Ffl or HDAC3 shRNAs. PCR was performed using primers recognizing the indicated cDNA. C, HDAC3 suppression does not alter alkaline phosphatase expression. MC3T3 cells stably expressing Ffl and HDAC3 shRNAs were differentiated in the presence of ascorbic acid and ␤-glycerol phosphate. Alkaline phosphatase staining was performed on the indicated days. D, HDAC3 suppression accelerates Ca 2ϩ accumulation in the matrix. Alizarin red staining was performed on differentiating MC3T3 cells on the indicated days. E, HDAC3 suppression accelerates osteocalcin, osteopontin, and bone sialoprotein expression. Real-time PCR was performed on cDNA from differentiated MC3T3 cells stably expressing either Ffl or shRNA #1 on the indicated days. Gray bars represent Ffl shRNA-expressing cells, and black bars represent HDAC3 shRNA-expressing cells, the asterisk indicates a p value of Ͻ0.05, between Ffl and HDAC3 shRNA cells at that time point, as determined by unpaired t testing.
HDAC6, and possibly HDAC4 (this report and Ref. 18). It remains to be conclusively determined if these co-repressor proteins interact with Runx factors as part of unique or shared co-repressor complexes. It is appreciated that HDACs are the enzymatic components of large multiprotein complexes that interact with many transcription factors. HDAC3 complexes contain nuclear receptor co-repressors, SMRT and N-CoR (53,56). It was also reported that HDAC3 is the required enzymatic component of complexes containing Class II HDACs, HDAC4, -5, and -7 (57,58). Interactions between HDAC3 and Class II HDACs (e.g. HDAC6) were not addressed in these studies. Our work indicates that HDAC3 and HDAC6 bind to distinct regions of Runx2. Our data also indicate that their expression patterns are different in that HDAC3 is strictly nuclear and expressed throughout osteoblast differentiation (data not shown), whereas HDAC6 shuttles from cytoplasm to the nucleus and is up-regulated during late stages of differentiation (18). Therefore, we currently favor the hypothesis that HDAC3 and HDAC6 regulate Runx2 activities in different contexts. Although it is almost certain that Runx2 repressor complexes contain many co-repressors, more work needs to be done to define the spatial and temporal organization of these complex components during osteoblast differentiation.
The physical and functional relationships between Runx2 and HDAC3 may be important for regulating bone formation. HDAC3 is expressed in C2C12 cells, which have osteogenic potential upon BMP2 stimulation, MC3T3 preosteoblasts, and two osteosarcoma cell lines. Suppression of HDAC3 by RNA interference in MC3T3 cells caused early expression of osteocalcin, osteopontin, and bone sialoprotein, and consequently accelerated a late stage event in osteoblast differentiation, namely calcium deposition in the mineralizing matrix. An earlier marker of osteoblast maturation, alkaline phosphatase production, was not affected by HDAC3 suppression. Interestingly, the dissociation of HDAC3 from peroxisome proliferatoractivated receptor-␥-retinoblastma protein complexes by retinoblastma protein phosphorylation in adipocytes stimulated their differentiation (52). Thus, HDAC3 may be a broad regulator of cellular maturation and terminal differentiation. Of note, general suppression of HDAC activity in tumor cells facilitates their differentiation, and several small molecule HDAC inhibitors are in phase I or II clinical trials (49,50). Our data showing that HDAC3 binds the osteocalcin promoter and directly regulates the activity of Runx2 suggests that HDAC3 may play a crucial role in regulating osteoblast differentiation. These results suggest that suppressing HDAC3 activity may be a mechanism of increasing bone formation.