Regulatory Factor for X-box Family Proteins Differentially Interact with Histone Deacetylases to Repress Collagen α2(I) Gene (COL1A2) Expression*

Our studies indicate that the regulatory factor for X-box (RFX) family proteins repress collagen α2(I) gene (COL1A2) expression (Xu, Y., Wang, L., Buttice, G., Sengupta, P. K., and Smith, B. D. (2003) J. Biol. Chem. 278, 49134–49144; Xu, Y., Wang, L., Buttice, G., Sengupta, P. K., and Smith, B. D. (2004) J. Biol. Chem. 279, 41319–41332). In this study, we examined the mechanism(s) underlying the repression of collagen gene by RFX proteins. Two members of the RFX family, RFX1 and RFX5, associate with distinct sets of co-repressors on the collagen transcription start site in vitro. RFX5 specifically interacts with histone deacetylase 2 (HDAC2) and the mammalian transcriptional repressor (mSin3B), whereas RFX1 preferably interacts with HDAC1 and mSin3A. HDAC2 cooperates with RFX5 to down-regulate collagen promoter activity, whereas HDAC1 enhances inhibition of collagen promoter activity by RFX1. Interferon-γ promotes the recruitment of RFX5/HDAC2/mSin3B to the collagen transcription start site but decreases the occupancy by RFX1/mSin3A as manifested by chromatin immunoprecipitation assay. RFX1 binds to the methylated collagen sequence with much higher affinity than unmethylated sequence, recruiting more HDAC1 and mSin3A. The DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine, which inhibits DNA methylation, reduces RFX1/HDAC1 binding to the collagen transcription start site in chromatin immunoprecipitation assays. Finally, both RFX1 and RFX5 are acetylated in vivo. Trichostatin A stimulates the acetylation of RFX proteins and activates the collagen promoter activity. Collectively, our data strongly indicate two separate pathways for RFX proteins to repress collagen gene expression as follows: one for RFX5/HDAC2 in interferon-γ-mediated repression, and the other for RFX1/HDAC1 in methylation-mediated collagen silencing.

The major focus of research at this laboratory has been toward understanding the transcriptional events occurring at the transcription start site of type I collagen genes. Earlier investigations have led to the discovery of a binding site for the regulatory factor for X-box (RFX) at the transcription start site of both ␣1(I) and ␣2(I) genes, COL1A1 and COL1A2 (38,39). Binding of two members of RFX proteins, RFX1 and RFX5, to the start sites of the type I collagen genes represses their expression. RFX1 is able to form dimers with itself as well as with two other RFX members, RFX2 and RFX3, with which RFX1 shares significant homology, including the dimerization domain that mediates the complex formation. On the other hand, RFX5, lacking the dimerization domain, is less homologous to other family members and forms a trimeric complex with two other proteins, RFXB and RFXAP (1). RFX5 is also responsible for recruiting class II transactivator (CIITA), the master regulator for major histocompatibility II (MHC II) expression, to the collagen transcription start site during IFN-␥ response (2). Although both RFX1 and RFX5 down-regulate collagen expression, their distinct association with other transcription factors suggests that they are involved in different physiological and pathophysiological events leading to the repression of collagen synthesis. Our results presented here demonstrate that RFX1 and RFX5 differentially interact with class I HDACs, underlying the different pathways when repressing collagen synthesis.
In other studies, HT1080 cells were plated at a density of 5 ϫ 10 5 cells per p35 tissue culture dish and treated with 5-aza-2Ј-deoxycytidine (aza-dC) (500 nM and 1 M) for 3 days in DMEM adding fresh aza-dC each day. In some studies HT1080 cells were treated with TSA (300 nM) for 24 h.
DNA Affinity Pull-down Assay-The collagen sequence (COL1A2 Ϫ25/ϩ30, GenBank TM accession number AF004877) with a HindIII overhang was synthesized as complementary strands and annealed as described previously (40). Double-stranded collagen DNA was biotinlabeled by incubating with Klenow fragment (New England Biolabs, Beverly, MA) and biotin-14-dATP (Invitrogen) supplemented with regular dCTP, dTTP, and dGTP at room temperature for 30 min. The reaction mixture was phenol/chloroform-extracted and alcohol-precipitated to remove unincorporated biotin.
Nuclear protein extracts from IMR-90 cells were obtained as described previously using 450 mM sodium chloride (1,41). The streptavidin beads (Promega, Madison, WI) were washed three times with ice-cold phosphate-buffered saline supplemented with 1 mM PMSF. Nuclear proteins (100 -200 g) were pre-cleared by incubating with the washed beads for 30 min at 4°C on a shaking platform as described previously (2). Pre-cleared nuclear proteins were prepared by capturing the beads on a magnetic stand and removing the supernatant. The supernatant was then incubated with biotin-labeled collagen DNA probe (Ϫ25/ϩ30) for 1 h at room temperature in binding buffer (60 mM NaCl, 20 mM HEPES, pH 7.9, 0.1 mM EDTA, 4% glycerol, 2 mM dithiothreitol) supplemented with bovine serum albumin, poly(dI-dC), and sonicated salmon sperm DNA to remove nonspecific binding. DNAprotein complex formed was captured by the magnetic beads and washed extensively with binding buffer supplemented with 0.01% Triton X and 100 mM KCl. The bound proteins were eluted with 1ϫ electrophoresis sample buffer by incubating at 90°C for 10 min and analyzed by SDS-polyacrylamide gels.
Plasmids, Transfections, and Luciferase Assays-The COL1A2-luciferase construct (pH20) (42) contains sequences from Ϫ221 to ϩ54 bp of mouse COL1A2 promoter fused to the luciferase reporter gene. Fulllength class I HDAC expression constructs (HDAC1, HDAC2, and HDAC3) in pcDNA3 and silencing HDACs (43) were kindly provided by Dr. Edward Seto. Full-length FLAG-RFX5 were kindly provided by Dr. Jenny Ting. RFX1 cDNA was excised from pHISB-RFX-1 plasmid, respectively, and inserted into the pIRES-hrGFP-2〈 construct (Stratagene) that has green fluorescent protein-coding sequence. RFX1 plasmid was digested with NotI and XhoI to produce the 5-kb vector along with 3-kb fragment. The RFX1 fragment was inserted into the EcoRI/ XhoI site of the pIRES-hrGFP-2〈 plasmid. This bicistronic plasmid can encode the protein along with the green fluorescent protein when expressed in mammalian cell lines.
Cells were plated at the density of 3 ϫ 10 5 cells/well in 6-well tissue culture dishes (for IMR-90 cells) or 5 ϫ 10 6 cells per p100 tissue culture dish (for 293FT cells). Transfections were performed with Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's protocol. Cells were harvested 48 h post-transfection, and luciferase assays were performed with a luciferase reporter assay system (Promega).
Immunoprecipitations-To investigate whether factors interact in vivo, co-immunoprecipitations were performed. Whole cell lysates (IMR-90 or 293FT with transfected constructs as indicated where applicable) were obtained by resuspending cell pellets in RIPA buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100) with freshly added protease inhibitor (Roche Applied Science) and PMSF (100 g/ml RIPA). Anti-RFX5 (194, Rockland) or anti-RFX1 (I-19; Santa Cruz Biotechnology) antibody was added to and incubated with IMR-90 cell lysate overnight before being absorbed by protein A/G PLUS-agarose beads (Santa Cruz Biotechnology). Precipitated immune complex was released by boiling with 1ϫ SDS electrophoresis sample buffer. Alternatively, FLAG-conjugated beads (M2; Sigma) were added to and incubated with 293FT cell lysate overnight. Precipitated immune complex was eluted with 3ϫ FLAG peptide (Sigma).
Westerns-Proteins were separated by 8 or 10% polyacrylamide gel electrophoresis with pre-stained markers (Bio-Rad) for estimating molecular weight and efficiency of transfer to blots. Proteins were transferred to nitrocellulose membranes (Bio-Rad) in a Mini-Trans-Blot Cell (Bio-Rad). The membranes were blocked with 5% milk powder in Tris-buffered saline buffer (TBST: 0.05% Tween 20, 150 mM NaCl, 100 mM Tris-HCl, pH 7. After three washes with TBST, the membranes were incubated with appropriate secondary antibodies, either anti-goat IgG (Sigma), anti-mouse IgG, or anti-rabbit IgG (Amersham Biosciences) conjugated to horseradish peroxidase, for another 1 h at room temperature. Then protein blots were visualized using ECL reagent (PerkinElmer Life Sciences) on a Kodak image station (PerkinElmer Life Sciences).
RNA Isolation and Real Time PCR-Cells were harvested, and RNA was extracted using an RNeasy RNA isolation kit (Qiagen, Valencia, CA) according to the manufacturer's protocol. Reverse transcriptase reactions were performed using a SuperScript first-strand synthesis system (Invitrogen) according to the manufacturer's protocol. Real time PCRs were performed on a ABI Prism 7700 sequence detection PCR machine (Applied Biosystems, Foster City, CA) according to manufacturer's protocol. The oligonucleotide forward and reverse PCR primers and fluorescent probes are described in Table 1.

RFX1 and RFX5 Differentially Interact with Class I HDACs on
COL1A2 Transcription Start Site-Previously, we reported that two members of the RFX family, RFX1 and RFX5, bind to the transcription start site of the COL1A2 gene and repress its expression (39). During IFN-␥ treatment, when RFX5 occupies the COL1A2 transcription start site, there is a coordinate decrease in acetylation of histones (1). Because transcriptional repression is usually associated with the recruitment of HDACs, we examined whether RFX1 and/or RFX5 is responsible for recruiting the HDACs to the collagen site. To this end, DNA affinity pull-down experiments were performed, as described previously (2), with nuclear proteins extracted from human IMR-90 cells and a biotinlabeled double-stranded DNA probe that spans from Ϫ25 to ϩ30 of the collagen promoter containing the RFX-binding site. Different DNA oligonucleotides were also used as competitors for RFX binding to test the specificity of the interactions. Streptavidin-conjugated magnetic beads were used to sequester the biotinylated probe with bound nuclear proteins.
After extensive washing, bound proteins were eluted with SDS electrophoresis buffer, separated on SDS-10% PAGE, and transferred to membranes for Western analysis using specific antibodies for RFX and HDAC family members. No proteins were present in the eluates without a DNA probe (Fig. 1, A, lane 2, and B, lane 3). In the presence of the DNA probe, RFX1 is detected in the eluates along with HDAC1 ( Fig. 1,  A, lane 3, and B, lane 4). When a methylated sequence, which acts as a competitor specifically eliminating the binding of RFX1 (mpBR), is added as demonstrated previously (40), HDAC1 binding is also lost ( Fig.  1, A, lane 4, and B, lane 5). This indicates that RFX1 selectively interacts with HDAC1 on the collagen start site.
In contrast, RFX5 is able to bind to DNA in the absence of RFX1 along with HDAC2 ( Fig. 1, A, lane 4, and B, lane 5), suggesting a selective interaction between RFX5 and HDAC2. The MHC II X-box sequence abolishes the binding of RFX1 (Fig. 1A, lane 5) and RFX5 (Fig. 1B, lane 6), as well as the two HDACs, whereas neither Sp1 nor Ets consensus sequence alters the binding of RFX1, HDAC1, RFX5, or HDAC2, implying that the interactions are specific. A third class I HDAC, HDAC3, is not detectable under any of these circumstances, probably because it is not recruited to the collagen transcription start site through RFX proteins.
IFN-␥ Enhances the Interaction between RFX5 and Co-repressors-Earlier studies indicated that IFN-␥ increased RFX5 protein levels, nuclear localization, and occupation on the collagen start site in IMR-90 cells (1). Because there was an interaction between RFX5 and HDACs, RFX5 may recruit more HDACs and IFN-␥ could enhance the association between RFX5 and HDACs at the collagen transcription start site. Increased HDACs could inhibit histone acetylation and ultimately repress collagen transcription. To test this hypothesis, DNA affinity pull-down experiments were performed first with nuclear proteins extracted from IMR-90 cells treated with or without IFN-␥ (100 units) for 24 h.
Both RFX1 and HDAC1 bind to the collagen sequence with similar affinity either with or without IFN-␥ treatment (Fig. 1B, compare lane 4 and lane 10). With the addition of methylated mpBR, binding of RFX1 as well as HDAC1 is eliminated, whereas RFX5, HDAC2, and mSin3B associate with the probe (Fig. 1B, lane 5). Binding of RFX5, HDAC2, and mSin3B are all enhanced with IFN-␥ treatment (Fig. 1B, compare lane 5 to lane 11). The specificity of the interactions was maintained in the presence of IFN-␥ (Fig. 1B, lanes 12-14), suggesting that more co-repressors are recruited to the collagen transcription start site by RFX5.

Interactions between RFX Proteins and HDACs Are DNA-independent-
Next, co-immunoprecipitation experiments were performed between RFX proteins and HDACs to determine whether interactions depend on DNA. The first immunoprecipitation experiments were performed using epitope-tagged RFX5 and HDAC-expressed proteins in 293FT cells because they are fast-growing human kidney cells that stably express the large T antigen of SV40 allowing increased amounts of overexpressed protein. FLAG-conjugated beads were used to precipitate protein extracts from 293FT cells transfected with either FLAG-tagged RFX5 or HDAC2. Eluates were separated by SDS-polyacrylamide gels and examined for endogenous proteins that might co-precipitate with FLAG-tagged proteins. As depicted in Fig. 2A, FLAG-RFX5 co-precipitates with endogenous HDAC2 (left panel) and vice versa (right panel), indicative of a reciprocal interaction. In both cases, mSin3B is present in the immune complex.
Next, nuclear proteins from IMR-90 cells treated with or without IFN-␥ were precipitated using an anti-RFX5 or anti-HDAC2 antibody to examine endogenous interactions. Eluates were separated by SDSpolyacrylamide gels and examined for HDAC2 ( Fig. 2A, left panel) or RFX5 and HDAC2 (right panel). HDAC2 is immunoprecipitated with the RFX5 antibody but not with a pre-immune IgG (Fig. 2B, right panel). More HDAC2 is co-immunoprecipitated with RFX5 antibody from IFN-␥ treated cells. HDAC2 antibody also co-immunoprecipitated increased amounts of RFX5 after IFN-␥ treatment (Fig. 2B, right panel) suggesting increased interactions between these proteins during IFN-␥ treatment.
Finally, to examine endogenous interactions, RFX1 was immunoprecipitated from IMR-90 nuclear extracts with an anti-RFX1 antibody. Proteins in the supernatant and precipitate were separated by SDSpolyacrylamide gels and examined for endogenous proteins that remain in the supernatant or co-precipitate. A considerable fraction of HDAC1 co-precipitates with RFX1 (Fig. 2C). In contrast to RFX5, RFX1 did not co-precipitate with HDAC2 (Fig. 2C). These data suggest that RFX proteins probably associate with HDAC before recruiting them to the collagen transcription start site. were precipitated using FLAG-conjugated beads (40 l) (M2) or pre-immune IgG-conjugated beads (40 l) (IgG) as described under "Materials and Methods." Eluates were separated by 10% SDS gel and Westerns performed with anti-FLAG, RFX5, HDAC2, or mSin3B antibody as indicated. 10% of the original lysate was also loaded as input. B, RFX5 interacts with HDAC2 in IMR-90 cells. Whole cell extracts (500 g) prepared from IMR-90 cells with (ϩ) or without (Ϫ) IFN-␥ treatment were precipitated using an anti-RFX5 antibody (5 g) (left panel), an anti-HDAC2 antibody (5 g) (right panel), or anti-pre-immune IgG (5 g) as indicated. Eluates were separated by 10% SDS gel and Westerns performed with an anti-HDAC2 antibody (left panel) or both anti-HDAC and anti-RFX5 antibody. The original lysate proteins (10%) was also loaded and labeled as input. IP, immunoprecipitation. C, RFX1 interacts with HDAC1 but not HDAC2 in IMR-90 cells. Nuclear extracts (300 g) prepared from IMR-90 cells were precipitated using an anti-RFX1 antibody (5 g). Eluates were separated by 10% SDS gel and Westerns performed with anti-HDAC1 or HDAC2 antibody as indicated. 10% of the original lysate was also loaded as input. Sup, supernatant.
Class I HDACs Display Distinct Functions Regulating COL1A2 Promoter Activity-Because there were interactions between RFX proteins and HDACs, we examined whether these interactions bear any functional significance. Transient transfections were performed in IMR-90 cells with different HDAC expression constructs with or without RFX expression plasmids. As shown in Fig. 3, both HDAC1 and HDAC2 repress the collagen promoter activity, but HDAC3 does not significantly alter COL1A2 promoter activity, which is in accordance with our binding data (Fig. 1). RFX5 represses better with HDAC2 (Fig. 3A). Similarly, RFX1 represses better with HDAC1 (Fig. 3B).
Next, siHDACs (43) were transfected with or without RFX expression constructs (Fig. 3, C and D). Silencing either HDAC1 or HDAC2, but not HDAC3, activated collagen promoter activity. Surprisingly, without these HDAC enzymes, RFX5 also further activated the collagen promoter (Fig. 3C). RFX1 remained a repressor with siHDAC2 and siHDAC3 but lost its ability to repress collagen promoter without HDAC1. This further demonstrates the different functional interactions between the RFX family members.
RFX5/HDAC2/mSin3B Are Involved in IFN-␥-mediated Collagen Transcriptional Repression-Because there is differential association between RFX proteins and HDACs, both physically and functionally, we hypothesized that RFX5/HDAC2 and RFX1/HDAC1 might be involved in different mechanisms responsible for collagen repression. Our previous reports suggest that RFX5 complex might be responsible for mediating IFN-␥ repression of collagen transcription (1, 2). Because certain co-repressor proteins were associated with RFX5 on the collagen start site in vitro in DNA affinity pull-down assays (Fig. 1) and the association could be increased by IFN-␥ (Fig. 1B), it was hypothesized that these repressors might be recruited to the collagen site by RFX5 during IFN-␥ response in vivo to repress collagen expression. Chromatin immunoprecipitation (ChIP) assays were performed with anti-RFX5, RFX1, HDAC1, HDAC2, HDAC3, mSin3A as well as mSin3B antibodies in IMR-90 cells treated with IFN-␥ for 0, 8, 16, or 24 h. Similar amounts of genomic DNA surrounding the COL1A2 transcription start site is precipitated by either RFX5 or RFX1 antibody (Fig.  4A). During IFN-␥ treatment, however, binding of RFX5 to the collagen sequence is greatly enhanced, but RFX1 binding is gradually and slightly decreased, suggesting that RFX5, but not RFX1, is involved in IFN-␥mediated collagen repression.
More DNAs are precipitated by mSin3A binding to the collagen start site than mSin3B before IFN-␥ is added (Fig. 4B, compare 0 h mSin3A to  0 h mSin3B). Although binding of mSin3B increases with time during IFN-␥ treatment, mSin3A occupancy is continuously decreased. By 24 h of IFN-␥ treatment, mSin3A binding is decreased to the level of mSin3B binding before IFN-␥ treatment, whereas binding of mSin3B is stimulated to the level of mSin3A binding at the beginning of IFN-␥ treatment. In other words, it appears as though IFN-␥ induces an exchange of mSin3A for mSin3B on the collagen transcription start site.
Meanwhile, when the occupancy of HDACs was examined, it was discovered that HDAC2 was the predominant form of histone deacety- A, HDAC2 enhances repression of COL1A2 promoter by RFX5. A COL1A2 promoter construct (pH20, 0.5 g) was co-transfected with different class I HDAC constructs (0.5 g) in the presence (ϩ) or absence (Ϫ) of RFX5 (0.5 g), along with GFP (0.1 g) for normalization, into IMR-90 cells in duplicate wells as described under "Materials and Methods." Average luciferase activities were normalized by both protein concentration and GFP fluorescence and were expressed as relative percentage activity compared with the control group in which an empty vector was transfected. This representative experiment was repeated at least three times. B, HDAC1 enhances repression of COL1A2 promoter by RFX1. A COL1A2 promoter construct (pH20, 0.5 g) was co-transfected with different class I HDAC constructs (0.5 g) in the presence (ϩ) or absence (Ϫ) of RFX5 (0.5 g), along with GFP (0.1 g) for normalization, into IMR-90 cells in duplicate wells as described under "Materials and Methods." Average luciferase activities were normalized by both protein concentration and GFP fluorescence and were expressed relative to percentage activity compared with the control group in which an empty vector was transfected. This representative experiment was repeated at least three times. C and D, silencing HDAC constructs activate COL1A2 promoter activity. A COL1A2 promoter construct (pH20, 0.5 g) was co-transfected with different class I siHDAC constructs (43) in the presence (ϩ) or absence (Ϫ) of RFX5 (C, 0.5 g) or RFX1 (D, 0.5 g), along with GFP (0.1 g) for normalization, into IMR-90 cells in triplicate wells as described under "Materials and Methods." Average luciferase activities were normalized by both protein concentration and GFP fluorescence and were expressed as relative luciferase activity per g of protein per ng of GFP. This representative experiment was repeated at least twice.
lase present on the collagen transcription start site before IFN-␥ treatment. Binding of HDAC1 and HDAC3 was minimal compared with the binding of HDAC2, although the amount of DNA precipitated by either anti-HDAC1 or anti-HDAC3 antibody was significantly above control levels (Fig. 4C).
RFX1/HDAC1/mSin3A Are Involved in Methylation-mediated Collagen Transcriptional Repression-Our recent studies indicate that collagen expression is greatly decreased when the gene is methylated (44) and that RFX1 binding to the collagen site is methylationsensitive (38,40), raising the possibility that RFX1 might be involved in methylation-mediated collagen repression by recruiting co-repressors to the collagen transcription start site. To examine the valid-ity of this hypothesis, DNA affinity pull-down experiments were performed using either a regular biotinylated COL1A2 DNA probe (Ϫ25 to ϩ30) or the same probe that was methylated at the ϩ7 site that enhances the binding of RFX1 to the collagen transcription start site (40). As depicted in Fig. 5A, binding of RFX1 was much more robust on the methylated probe than on the unmethylated probe as expected. Interestingly, both HDAC1 and mSin3A bind to the methylated probe more strongly just like RFX1, supporting the notion that during methylation of collagen DNA more RFX1 is able to bind to the transcription start site and recruit more co-repressor complexes to repress collagen transcription. No RFX5 binding is detectable under these conditions. The inhibitor, aza-dC, a compound that inhibits DNA methyltransferases, dramatically (40-fold) increases collagen gene expression in a fibrosarcoma cell line, HT1080, in which 50% of collagen genomic DNA is methylated (38,44). Therefore, we examined whether the effect of aza-dC is mediated through diminished binding of RFX1 as well corepressors to the collagen transcription start site because of de-methylation. ChIP assays were performed with HT1080 cells treated with different concentrations of aza-dC for 72 h using anti-RFX1, HDAC1, as well as HDAC2 antibodies. RFX1 binding is decreased with aza-dC treatment in a dose-response manner by up to 60% (Fig. 5B). Interestingly, HDAC1 binds to the partially methylated collagen transcription start site in vivo much more strongly than HDAC2 (Fig. 5C). Binding of HDAC1 is also greatly decreased by aza-dC treatment similar to RFX1, further confirming that RFX1/HDAC1 might be involved in DNA methylation-mediated collagen transcriptional repression.

TSA Has Differential Cell-specific Effects on Steady State mRNA Levels-
Because HDACs are clearly involved in the repression of collagen transcription, we postulated that inhibition of overall HDAC activity would increase collagen transcription. TSA is a general inhibitor for class I and II HDACs that alters expression of many genes presumably through histone deacetylation. First, HT1080 cells were treated with several doses of TSA (50, 100, 500, and 1000 nM) for 24 h. There was no change in collagen mRNA levels at low doses. At higher doses of TSA there was a 5-(500 nM) or 10-fold (1000 nM) increase in collagen mRNA (Fig. 6A).
On the other hand, when IMR-90 cells were treated with TSA (500 nM), there was no significant increase in collagen mRNA levels (Fig. 7B). IFN-␥ repressed collagen mRNA levels by 50% in 24 h. TSA partially reversed the repression of collagen mRNA expression.
The mRNA steady state levels are a combination of transcription and degradation or processing of mRNA. In order to examine early tran- FIGURE 5. Methylation of the collagen gene (COL1A2) enhances the interaction between RFX1 and co-repressors. A, methylation enhances the binding of RFX1, HDAC1, as well as mSin3A to COL1A2 sequence in vitro. DNA affinity pull-down assays were performed with IMR-90 nuclear extract using either unmethylated probe (U) or methylated probe (M) as described under "Materials and Methods." Eluates were separated by 10% SDS gel and Westerns performed with anti-RFX1, HDAC1, RFX5, or mSin3A antibody as indicated. 10% of the original lysate was also loaded as input (I). B and C, Aza-dC treatment of HT1080 cells leads to decreased occupancy of RFX1 and HDAC1 on COL1A2 transcription start site. Chromatin immunoprecipitation assays were performed with HT1080 cells treated with Aza-dC at indicated concentrations for 72 consecutive hours using anti-RFX1 (B), anti-HDAC1, or anti-HDAC2 (C) antibodies as described under "Materials and Methods." This experiment was repeated at least twice and a representative graph was shown. Data are expressed as picograms of DNA precipitated by indicated antibody per ng of total genomic DNA. FIGURE 6. TSA stimulates COL1A2 mRNA in a cellspecific manner. A, HT1080 cells were untreated or treated with 300 nM of TSA for 24 h before harvesting as described under "Methods and Materials." Total RNAs were prepared and transcribed to form cDNA, and real time PCRs were performed with the cDNA samples using primers to detect COL1A2 mRNA. B and C, IMR-90 cells were untreated or treated with IFN-␥ (100 units/ml) and/or TSA (500 nM) for 24 h before harvesting as described under "Materials and Methods." Total RNAs were prepared and transcribed to form cDNA, and real time PCRs were performed with the cDNA samples using primers to detect COL1A2 mRNA (B) or COL1A2 heterogeneous nuclear RNA (hnRNA) (C) (see Table 1 for primers). Each experiment was repeated at least three times in duplicate wells. Data are expressed as relative RNA levels compared with control levels, normalized to 18 S RNA, and presented as average Ϯ S.D. A, paired sample t test, two-tailed; B and C, one-way ANOVA was performed to evaluate the statistical significance; ***, p Ͻ 0.01. scription, primers within the second intron and second exon (Table 1) were used to measure heterogeneous RNA transcripts. Total RNA was DNase-treated, and no reverse transcriptase was used as controls to be sure that RNA, not DNA, was measured by these primers. TSA treatment increased transcription of heterogeneous collagen mRNA more than steady state mRNA in the IMR-90 cells (Fig. 7C), suggesting that TSA increases transcription and degradation or processing of mRNA. IFN-␥ repressed collagen heterogeneous RNA to the same extent as steady state mRNA.
TSA Stimulates Collagen Promoter Activity and Partially Blocks Repression by RFX Proteins-Even though TSA effects on collagen mRNA were low, TSA activated collagen promoter-luciferase activity in IMR-90 cells (Fig. 8A). Overexpression of RFX1 and RFX5 repressed collagen synthesis, and TSA treatment was able to partially block this repression (Fig. 8B). RFX5 seemed to be more sensitive to TSA than RFX1 because TSA (500 nM) almost completely blocked repression by RFX5 overexpression.

DISCUSSION
Previously, we demonstrated that there is a binding site for the RFX family of transcription factors within the first exon (Ϫ1 to ϩ20) of the COL1A2 gene (40). Two members of the RFX family, RFX5 and RFX1, bind to this site and repress collagen expression (39). RFX1 homodimers and RFX1/RFX2 heterodimers bind to the collagen gene with higher affinity especially when it is methylated on the coding strand (38 -40, 44, 45). In this study, it is clear that RFX1 interacts on the collagen start site with HDAC1 by DNA affinity precipitation (Fig. 1). This interaction of RFX1 and HDAC1 with collagen DNA is increased when the CpG site within the RFX consensus sequence is methylated (Fig. 5A). In normal fibroblasts, there is a small but measurable amount of RFX1 binding within chromatin as judged by ChIP assays (Fig. 4A). Collagen type I genes are methylated in DNA from certain cancer lines and in colorectal tumors (44). The collagen gene in HT1080 cells is 50% methylated at the CpG within the RFX1-binding site (44), and aza-dC increases collagen expression as much as 40-fold in this fibrosarcoma cell line (44,46). RFX1 occupies the site 2-fold more in HT1080 cells than in IMR-90 cells (data not shown). In HCT116 cells that have higher methylation status of the collagen gene, RFX1 occupancy is further increased (data not FIGURE 7. TSA stimulates COL1A2 promoter activity in a dose-response manner. A and B, COL1A2 promoter construct (pH20, 0.5 g) was co-transfected with or without RFX expression constructs (0.5 g), along with GFP (0.1 g) for normalization, into IMR-90 cells in duplicate wells as described under "Materials and Methods." 24 h after transfections, cells were treated with TSA at indicated concentrations for additional 24 h before harvesting. Luciferase activities were normalized by both protein concentration and GFP fluorescence and were expressed relative percentage activity compared with the control group in which an empty vector was transfected. This experiment was repeated three times and data were shown as average Ϯ S.D. Oneway ANOVA was used to evaluate the statistical significance. IP, immunoprecipitation; IB, immunoblot. **, p Ͻ 0.05 and ***, p Ͻ 0.01. C and D, TSA enhances the acetylation of both RFX5 and RFX1. Whole cell extracts prepared from IMR-90 cells with TSA (500 nM) treatment for the indicated period of time were precipitated using an anti-RFX5 antibody or an anti-RFX1 as indicated. Eluates were separated by 10% SDS gel and Westerns performed with anti-acetyl-lysine, RFX5, or RFX1 antibody as indicated.
shown), suggesting that RFX1 does interact more with the collagen start site when the gene is methylated. Aza-dC, which decreases methylation by inhibiting DNA methyltransferases, also decreases the occupancy of RFX1 as well as HDAC1 on the collagen start site (Fig. 5). These data again suggest that RFX1 and HDAC1 interact at the DNA within chromatin. In fact, RFX1 also co-immunoprecipitates with HDAC1, but not HDAC2 or HDAC3, suggesting that these protein-protein interactions occur independent of RFX1-DNA interactions.
On the other hand, RFX5 interacts with two other RFX proteins, RFXB and RFXAP, as well as CIITA to repress collagen transcription during IFN-␥ stimulation (1,2). This study demonstrates that RFX5 occupies the collagen gene, and RFX5 occupancy increases with IFN-␥ treatment as RFX1 occupancy decreases (Fig. 4A). HDAC2, but not HDAC1 or HDAC3, also increases on the collagen gene with time of IFN-␥ treatment (Fig. 4B) suggesting that HDAC2 is responsible for the deacetylation of histones surrounding the collagen start site observed earlier (1). Therefore, although both RFX1 and RFX5 are repressors for the COL1A2 gene, the underlying mechanisms might differ through different co-repressor interactions that could account for altered cellular responses impacting collagen expression. Two collagen-related events, namely IFN-␥-regulated collagen repression during pulmonary fibrosis or inflammation and methylation-mediated collagen down-regulation in carcinogenesis, have led to an association of these two events with the RFX family. RFX1 might be responsible for methylation, whereas RFX5 may be responsible for IFN-␥-mediated collagen down-regulation.
CIITA is also recruited to the collagen transcription start site by RFX5 with time of IFN-␥ treatment (2). Earlier studies demonstrate that CIITA interacts with HDAC1 and mSin3A to terminate MHC II activation (47,48). However, this investigation suggests that HDAC2 and Sin3B occupy the collagen transcription start site during IFN-␥ treatment (Fig. 4B). Our preliminary results suggest that CIITA can complex with HDAC2 and Sin3B (data not shown). Although our effort so far has been concentrated on RFX5 and CIITA, recent evidence has been published that the other two members of the RFX5 complex, RFXB and RFXAP, are both capable of interacting with certain histone-modifying factors (49,50), suggesting a universal role for RFX5 proteins in transcriptional control. RFXAP, for example, complexes with Brahma related gene 1 (BRG1), an ATPase-dependent chromatin remodeling molecule, to regulate MHC II expression (49). RFXANK is a binding partner for HDAC4, although the biological significance of this interaction remains unclear (50). Therefore, there might be extensive interactions between histone-modifying proteins and RFX-associated transcription factors. This may explain why silencing of any one of the class I HDACs still caused an increase in collagen promoter activity. Certainly other classes of HDACs, such as HDAC4 discussed above, could be involved in collagen repression through the RFX complex. Co-repressor complexes have been isolated containing several HDACs with HDAC1 and HDAC2 usually isolated together with Sin3A (36,37). However, recently interactions of HDACs with individual transcription factors have been noted especially in the Sp1/kruppel zinc finger family (51)(52)(53)(54). HDACs interact with Sp1 protein without activation of the enzyme activity by blocking access of the Sp1 protein to the DNA (53). Collagen is an Sp1-activated promoter, and therefore, the recruitment of HDAC to the collagen gene may block assess to Sp1 sites in the promoter. In this case, HDAC represses transcription without active catalytic activity.
TSA inhibits multiple histone deacetylases in class I and class II causing wide ranging changes in gene expression. The effects of TSA on collagen gene expression are dependent on cell type. In our studies, TSA activated gene expression in the cancer cell line HT1080. Gene expression of collagen in human hepatoma cells is induced by TSA accompanied by changes in histone acetylation (55). However, one of the cell lines, PLC/PRF, had no detectable collagen expression. Our investigations on this cell line indicate that the collagen gene is very highly methylated, and therefore, TSA cannot activate a gene silenced by complete methylation. On the other hand, TSA can induce collagen gene expression in a variety of cancer cell lines with partially methylated collagen genes such as breast cancer cell line, MCF7, or colorectal cancer cell line, HCT116. 3 Our studies also indicate that in fibroblasts TSA produces no significant changes in steady state mRNA levels. However, this seems to be caused by a combination of increased transcription with accompanying increases in degradation and/or processing. Others have demonstrated that TSA suppresses collagen synthesis in dermal fibroblasts and hepatic stellate cells (56 -58). Although steady state mRNA and protein levels of collagen decrease after TSA treatment, nuclear run-on studies indicated that transcription of type I collagen actually was activated thus supporting our results with heterogeneous nuclear RNA (58). The decrease in collagen found by these investigators is most likely due to indirect mechanisms involving increased degradation or processing of collagen mRNA and protein because of multiple changes in gene expression from TSA inhibition of HDACs.
The actual increased acetylation of RFX1 and RFX5 proteins with TSA treatment suggests that the mechanism for repression of collagen through the RFX family may require active deacetylation activity of HDAC. The role of acetylation of RFX family needs to be further analyzed. Protein acetylation of transcription factors has been implicated in multiple cellular processes, including differentiation, metabolism, proliferation, and survival against stress (59 -62). NF-B, c-Myc, Sp3, p53, FOXO, GATA, and SMAD7 have all been identified as targets for acetylation (61,(63)(64)(65)(66). For example, acetylation of SMAD7 is involved with stabilizing the protein (67). On the other hand, acetylation of p53 is required for efficient recruitment of co-activators to promoter regions as well as activation of target genes (68 -70). Transcriptional output mediated by FOXO is controlled via a two-tiered mechanism of phosphorylation and acetylation (71). FOXO and RFX family have similar winged helix DNA binding domains (72) so they could be regulated in a similar manner by acetylation. The acetylation may alter the interaction of RFX1 with co-repressors or with DNA. The overall consequence of acetylation is not well understood and seems to vary from one factor to another and to depend on specific circumstances.
The roles of HDACs in several diseases, especially in lung inflammation (73,74) and in cancer (75), have been recently outlined. HDAC2 specifically may decrease in chronic obstructive pulmonary disease patients causing corticosteroid resistance (74,76). HDAC1 has been associated with certain types of breast cancer (77). These studies point out the biological significance of acetylation and deacetylation and the need to investigate the mechanism of action on individual genes associated with these diseases.
In summary, the RFX family represses collagen transcription at the transcription start site through either methylation-specific binding by RFX1-3 family members or through IFN-␥-stimulated binding during inflammation by the RFX5-CIITA complex (Fig. 8). In each case, a specific co-repressor complex most likely contains RFX family members. RFX1 interacts best with HDAC1 and mSin3A, whereas RFX5/CIITA interacts with HDAC2 and mSin3B on the collagen transcription start site. During IFN-␥ treatment, RFX5 synthesis, translocation to the nucleus, and complex formation are increased (1). CIITA is expressed and recruited to the collagen start site with RFX5 (2). Most likely RFX5 becomes deacetylated by HDAC2, which increases complex interactions and repressor activity of RFX5. If the collagen gene is methylated, RFX1 interacts with the collagen gene and represses transcription through HDAC1 deacetylation of chromatin. This study reveals that different members of the same family of transcription factors repress collagen transcription through similar but distinct mechanisms.