The RFX family interacts at the collagen (COL1A2) start site and represses transcription.

The transcription start site of the collagen alpha2(1) gene (COL1A2) has a sequence-specific binding site for a DNA methylation-responsive binding protein called regulatory factor for X-box 1 (RFX1) (Sengupta, P. K., Erhlich, M., and Smith, B. D. (1999) J. Biol. Chem. 274, 36649-36655). In this report, we demonstrate that RFX1 forms homodimers as well as heterodimers with RFX2 spanning the collagen transcription start site. Methylation at +7 on the coding strand increases RFX1 complex formation in gel shift assays. Methylation on the template strand, however, does not increase RFX1 complex formation. DNA from human fibroblasts contains minimal methylation on the coding strand (<4%) with variable methylation on the template strand. RFX1 acts as a repressor of collagen transcription as judged by in vitro transcription and co-transfection assays with an unmethylated collagen promoter-reporter construct. In addition, an RFX5 complex present in human fibroblasts interacts with the collagen RFX site, which is not sensitive to methylation. This is the first demonstration of RFX5 complex formation on a gene other than major histocompatibility complex (MHC) promoters. Also, RFX5 represses transcription of a collagen promoter-reporter construct in rat fibroblasts that have no detectable RFX5 complex formation or protein. RFX5 complex activates MHC II transcription by interacting with an interferon-gamma (IFN-gamma)-inducible protein, major histocompatibility class II trans-activator (CIITA). Collagen transcription is repressed by IFN-gamma in a dose-dependent manner in human but not in rat fibroblasts. IFN-gamma enhances RFX5 binding activity, and CIITA is present in the RFX5 complex of IFN-gamma-treated human fibroblasts. CIITA repressed collagen gene transcription more effectively in human fibroblasts than in rat fibroblasts, suggesting that the RFX5 complex may, in part, recruit CIITA protein to the collagen transcription start site. Thus the RFX family may be important repressors of collagen gene transcription through a RFX binding site spanning the transcription start site.

The transcription start site of the collagen ␣2(1) gene (COL1A2) has a sequence-specific binding site for a DNA methylation-responsive binding protein called regulatory factor for X-box 1 (RFX1) (Sengupta, P. K., Erhlich, M., and Smith, B. D. (1999) J. Biol. Chem. 274, 36649 -36655). In this report, we demonstrate that RFX1 forms homodimers as well as heterodimers with RFX2 spanning the collagen transcription start site. Methylation at ؉7 on the coding strand increases RFX1 complex formation in gel shift assays. Methylation on the template strand, however, does not increase RFX1 complex formation. DNA from human fibroblasts contains minimal methylation on the coding strand (<4%) with variable methylation on the template strand. RFX1 acts as a repressor of collagen transcription as judged by in vitro transcription and co-transfection assays with an unmethylated collagen promoter-reporter construct. In addition, an RFX5 complex present in human fibroblasts interacts with the collagen RFX site, which is not sensitive to methylation. This is the first demonstration of RFX5 complex formation on a gene other than major histocompatibility complex (MHC) promoters. Also, RFX5 represses transcription of a collagen promoterreporter construct in rat fibroblasts that have no detectable RFX5 complex formation or protein. RFX5 complex activates MHC II transcription by interacting with an interferon-␥ (IFN-␥)-inducible protein, major histocompatibility class II trans-activator (CIITA). Collagen transcription is repressed by IFN-␥ in a dose-dependent manner in human but not in rat fibroblasts. IFN-␥ enhances RFX5 binding activity, and CIITA is present in the RFX5 complex of IFN-␥-treated human fibroblasts. CIITA repressed collagen gene transcription more effectively in human fibroblasts than in rat fibroblasts, suggesting that the RFX5 complex may, in part, recruit CIITA protein to the collagen transcription start site. Thus the RFX family may be important repressors of collagen gene transcription through a RFX binding site spanning the transcription start site.
Extracellular matrix plays a critical role in morphogenesis and growth. Remodeling of the extracellular matrix is a complex and tightly regulated process occurring during embryo-genesis, angiogenesis, and wound repair. In the majority of normal adult tissue, there is limited turnover of extracellular matrix until injury occurs. In contrast, the balance between synthesis and degradation of extracellular matrix in many pathological conditions is disrupted leading to abnormal matrix remodeling. This disruption can lead to excessive accumulation of matrix proteins in fibrotic diseases such as scleroderma (1), liver cirrhosis (2), and lung fibrosis (3). Alternatively, excessive breakdown exists in conditions such as rheumatoid arthritis (4), osteoarthritis (5), periodontitis (6), tumor invasion, and metastasis (7). Cytokines have been implicated in the development of pathological conditions (6). Among these factors interleukin 1 (IL-1), 1 tumor necrosis factor-␣ (TNF-␣), interferon-␥ (IFN-␥), and transforming growth factor-␤ (TGF-␤) (8) have been detected and are considered essential in inflammatory sites, healing wounds, and during tissue remodeling. These growth factors can positively (TGF-␤) or negatively (IL-1, TNF-␣, and IFN-␥) regulate the expression of extracellular matrix genes, in particular collagen genes. For example, TNF-␣ and IFN-␥ work synergistically to oppose activation of collagen transcription by TGF-␤ (9).
Type I collagen is a predominant protein in fibrotic lesions and consists of two ␣1(1) (COL1A1) chains and one ␣2(1) chain (COL1A2). Studies in diverse systems have identified various transcription factors that regulate collagen transcription. These include ubiquitously expressed transcription factors such as Sp1 and Sp3 and nuclear factor for Y box or CAAT binding factor (NF-Y/CBF) that are essential for constitutive expression (10 -13). Little is understood how these factors function on collagen genes through coordination with co-activators. It is known that on other genes NF-Y/CBP interacts with GCN5 and PCAF (14) and Sp1 interacts with c-AMP response element protein (CREB) binding protein (CBP) or p300 (CBP/ p300) on other promoters (15).
In addition, transfection studies have located responsive elements in the collagen promoters that can be modulated by growth factors (16 -21). Several transcription factors bind cooperatively to COL1A2 promoter to mediate activation by TGF-␤. For examples, Smad3 interacts cooperatively with the co-activator CBP/p300 and Sp1 (22)(23)(24) during TGF-␤ activation whereas NF-B and C/EBPs appear to be involved in TNF-␣ inhibition of COL1A2 transcription (25,26). Although response elements have been demonstrated for IFN-␥ in the proximal promoter (20,21,27), the transcription factors that bind directly to the collagen promoter have not been characterized. In addition, IFN-␥ antagonizes TGF-␤ possibly through competition for p300/CBP and indirectly through interferon regulatory factor-1 and signal transducers and activators of transcription 1␣ (12,21).
Our previous studies indicate that the collagen ␣2(1) transcription start site contains a sequence-specific, methylationresponsive binding site for regulatory factor for X-box 1 (RFX1, also referred to as methylated DNA binding protein) (28). This protein belongs to a family with conserved DNA binding domains (29). Three of the family members (RFX1, RFX2, and RFX3) form heterodimers as well as homodimers and contain both activation and repression domains (30 -32). RFX4 was identified as part of a variant estrogen receptor expressed in breast cancer (29) and has now been characterized as a testesspecific family member that can dimerize with RFX2 and RFX3 (33).
The fifth member of the family, RFX5, has been well characterized as a major protein involved in IFN-␥-activated and constitutive transcription of major histocompatibility complex type II (MHC II) proteins (34,35). This protein forms a trimeric complex with two proteins, RFXB/RFXANK and RFXAP (36,37). However, an additional protein, class II transactivator, CIITA, is necessary for transactivation of MHC II transcription. Mutations in the proteins CIITA, RFX5, RFXB/RFXANK, or RFXAP cause bare lymphocyte disorder, a severe immune deficiency that eliminates MHC II from cell surfaces. RFX5 complex formation on MHC II promoter is required to recruit CIITA. The CIITA protein interacts cooperatively with several other proteins on the promoter such as NF-Y/CBF (38), TFIIB (39), TAF II (40,41), and CBP/p300 (42). CIITA is considered a master regulator of MHC II and is induced in many cell types by IFN-␥ (34, 43) through signal transducers and activators of transcription and interferon regulatory factor-1 binding sites in its fourth promoter (44).
In certain instances, CIITA can repress as well as activate transcription of genes (45)(46)(47). For example, CIITA suppresses transcription of several thyroid-specific genes while at the same time it activates MHC II (45). Synthesis of two other Th2 cell proteins, FAS ligand and IL-4, are also repressed by CIITA (46,48). CIITA inhibits IL-4 gene transcription by competing with a transcription factor that interacts with the co-activator CBP/p300 (48). Most importantly, CIITA down-regulates collagen gene transcription through an interaction with CBP/p300 (49).
This report establishes that RFX1 homodimers, RFX1⅐RFX2 heterodimers, and RFX5 extracted from human fibroblast nuclei can form complexes on the normal unmethylated collagen transcription start site as well as on methylated sites. RFX5 complex binds when RFX1 and RFX2 are removed. Both RFX1 and RFX5 repress collagen gene expression in transfection and in vitro transcription assays. IFN-␥ suppresses collagen gene expression by increasing the binding activity of RFX5 complex and CIITA at the collagen promoter. IFN-␥ does not suppress collagen gene activity in cells with low amounts of RFX5 binding activity on the collagen site, suggesting that RFX5 binding recruits the CIITA protein.
Bisulfite Modification of DNA and Methylation-sensitive Singlenucleotide Primer Extension-Bisulfite modification followed by MS-SNuPE was used to analyze methylation of collagen at the ϩ7 CpG TAAAAACACCCTAAACCAAAAACTTTT a The PCR and MS-SNuPE primers are the predicted bisulfite-converted collagen sequences from COL1A2 sequences in the GenBank™. The PCR primers have been chosen from regions that do not contain any methylation sites (CpGs). site as previously described (50,51). The bisulfite modification causes unmethylated cytosines (Cs) to be converted to uracil. Methylated cytosine is resistant to deamination. Briefly, genomic DNA (ϳ2 g/50 l) isolated from human fibroblasts was incubated with 5.5 l of 2 N NaOH at 37°C for 15 min. Hydroquinone (0.5 mM) and sodium metabisulfite (2.6 M at pH 5) were added to DNA, and the DNA was incubated at 50°C for 20 h. The DNA was purified and extracted in water then desulfonated by incubation for 5 min in 0.3 M NaOH at room temperature. The DNA was reprecipitated and suspended in 20 l of Tris-HCl (10 mM)/ EDTA (1 mM) buffer, pH 8. The horizontal line in the diagram represents the COL1A2 gene surrounding the transcription start site. The vertical arrows pointing upwards indicate the location of the ϩ7 site (ϩ7 primer) within the RFX1 binding site and the control site Ϫ43 (control primer) analyzed by the MS-SNuPE assay. The coding and template strand was amplified by PCR with primers diagramed schematically on the map of the COL1A2 gene as forward primer (FP) and reverse primer (RP). Bisulfite-modified sequences for all primers on both strands are given in Table I. Each tick mark above the horizontal line represents an individual CpG dinucleotide in rat COL1A2. The tick marks below the horizontal line represent an individual CpG dinucleotide in human COL1A2. The curved arrow represents the start site, and the circle with RFX within shows the RFX consensus binding site. Boxes represent the TATA and reverse CAAT boxes, sites of TBP, and NF-Y⅐CBP binding, respectively. B, the COL1A2 gene in human fibroblasts is unmethylated on the coding strand and partially methylated on the template strand. Methylation at a control site (Ϫ43) and at the ϩ7 site was analyzed by primer extension with singleradiolabeled nucleotides on PCR-amplified bisulfite-modified templates. The products were resolved on a 15% denaturing polyacrylamide gel visualized by exposure to autoradiography and quantified by a flatbed radioactive counter (Packard Instant Imager). The presence of a band in the first panel indicates primer extension at a control site, Ϫ43. M refers to a methylated C, while U refers to unmethylated bisulfite converted C. The pair of MS-SNuPE primers on the left measures the coding strand methylation, and the pair of MS-SNuPE primers on the right measures the template strand methylation. The percentage of methylation beneath the bands is cpm measured on a flat-bed counter in the M lane divided by the total counts in both lanes times 100. The bar graph below the third panel represents the percentage of methylation calculated from five separate bisulfite experiments shown with a standard error bar. C, RFX1 has greater affinity for the hemi-methylated coding strand than template strand, and RFX5 binds independently of methylation status of the collagen site. Five micrograms of nuclear proteins was incubated in binding buffer containing 1 g of poly(dI⅐dC)⅐poly(dI⅐dC) with collagen probe (Ϫ1 to ϩ20). The probe methylation is indicated as follows: unmethylated (u/u), methylated on both strands (m/m), hemi-methylated on the template strand (u/m), or hemi-methylated on the coding strand (m/u). The incubations for 30 min at room temperature were performed with (ϩ) or without (Ϫ) the competitor pB1 at 50-fold molar excess (sequence of pB1 given in Table II). Protein and DNA were separated in native polyacrylamide gel (5%). In all electrophoretic mobility shift assay figures, the arrows 1 and 2 indicate the RFX1-3 complexes and arrow 3 indicates the RFX5 complex. FP is the free probe.
PCR was used to amplify the modified DNA replacing uracil residues with thymine. Primers were designed using a converted sequence (Cs to Ts) in a region that did not contain any possible methylation sites. Separate primers were designed to amplify the coding strand (Ϫ65 to ϩ151) or the template strand (Ϫ71 to ϩ144) (see Table I). The PCR product was separated on 2% low melting agarose gel and purified using a Qiagen gel extraction kit (Qiagen).
For determining methylation at individual sites, we used methylation-sensitive single-nucleotide primer extension (MS-SNuPE). This method established whether the nucleotide at ϩ7 remained a C (methylated) residue or had been converted to T (unmethylated). A control primer was used to establish whether conversion of C to T was completed. The MS-SNuPE primers are given in Table I, and their relationship to the collagen gene is diagramed in Fig. 1A. The reaction mixture contained gel-purified PCR fragments, primers, and 1 Ci of radiolabeled [ 32 P]dCTP or [ 32 P]dTTP for coding strand and [ 32 P]dGTP or [ 32 P]dATP for template strand in 1X Taq polymerase buffer (Promega, Madison, WI). After denaturation at 95°C for 3 min, the primers were annealed at 40°C for 2 min and extended at 72°C for 1 min with Taq polymerase (1 unit) (Promega). The reaction was stopped by addition of 10 l of stop buffer. The reaction products were heated at 95°C for 2 min before loading onto a 15% acrylamide/7 M urea gel. The radioactivity of the specific bands were quantified by a Packard Instant Imager (PerkinElmer Life Sciences) and pictured by autoradiography. The percentage of methylation was calculated by the amount of radiolabeled dCTP or dGTP incorporated into the primer divided by the total radioactivity (Fig. 1B).
Electrophoretic Mobility Gel Shift Assay-Nuclear extracts were prepared according to Dignam et al. (52) with some modifications. Extractions of protein from isolated nuclei were performed at higher salt conditions than normal using 500 mM NaCl or 420 mM NaCl rather than 350 mM NaCl in buffer C. All buffers contained the phosphatase inhibitor orthovanadate (1 mM) and the protease inhibitors leupeptin (40 g/ml), aprotinin (200 g/ml), pepstatin A (40 g/ml), and phenylmethylsulfonyl fluoride (0.5 mM). Protein concentration of the extracts was FIG. 2. A sequence-specific protein complex from human fibroblast nuclear extracts forming at the start site of collagen ␣2(1) gene co-migrates with a complex on the X-box of MHC II promoter. Five micrograms of nuclear proteins were incubated in binding buffer containing 1 g of poly(dI⅐dC) with collagen probe (Ϫ1 to ϩ20) with (ϩ) or without (Ϫ) 50-fold molar excess of competitors for 30 min at room temperature. Protein and DNA were separated in native polyacrylamide gel (5%). NS is a nonspecific complex. A, electrophoretic mobility shift assay using human fibroblast nuclear extract was performed using the unmethylated (U col, lanes 1 and 2) and methylated (M col, lanes 3 and 4) collagen start site sequence (Ϫ1 to 20) compared with the well-characterized Xbox sequence (X-box, lanes 5 and 6) in the MHC II promoter. Lanes 2, 4, and 6 have the methylated pBR322 competitor (pB1) (for sequences of probes see Table II and Ref. 28). B, electrophoretic mobility shift assay comparing human fibroblasts nuclear extracts (lanes 1-6) to rat fibroblast nuclear extracts (lanes 7-9) incubated with a 50-fold molar excess of competitors. Five micrograms of nuclear proteins was incubated in binding buffer containing 1 g of poly(dI⅐dC) with unmethylated collagen probe (Ϫ1 to ϩ20) and with or without additional competitors for 30 min at room temperature. Protein and DNA were separated in native polyacrylamide gel (5%). The left panel shows the specific RFX5 complex formation in human fibroblasts (arrow 3), and the right panel indicates the lack of the same complex in rat fibroblasts. Lanes 1 and 7 contain no competitor. Different competitors (col, COL1A2 sequence Ϫ1 to ϩ20, lane 2 and lane 8; pB1, methylated pBR322, lane 3 and lane 9; X, X-box from MHC II promoter, lane 4; pB1 ϩX, methylated pBR322 and X-box, lane 5; TAE, unrelated sequence in distal site of COL1A1, lane 6) were added at a 50-fold molar excess (for sequences of competitors see Table II and Ref. 28). C, sequences of the transcription start sites of both human collagen type I genes. The top sequence is ␣2(I) (COL1A2); the bottom sequence is ␣1(I) (COL1A1). Underlined sequences are consensus sequences for RFX1 binding on each collagen gene. Sequences above and below the collagen sequences are sequences from the MHC II X-box. The capital letters represent the contact sites for RFX5 and RFXB/RFXANK as described by Boss (37). determined by the Bradford reagent using bovine serum albumin as a standard.
Oligonucleotides containing collagen sequences (Ϫ1 to ϩ20) with and without methylation or methylated DNA binding protein/RFX consensus sequences with HindIII overhangs were synthesized (Oligo Etc., Gulford, CT and Integrated DNA Technology, Coralville, IA). The oligonucleotide sequences used for gel shift assay and competition assays are given in Table II. Complementary strands were annealed to make double-stranded oligonucleotides. Hemi-methylated probes were produced by annealing a methylated coding strand oligonucleotide or a methylated template strand oligonucleotide with the corresponding complementary unmethylated sequences. Annealing was performed at 95°C for 7 min in presence of 200 mM NaCl and cooled to room temperature overnight. The double-stranded oligonucleotides were radiolabeled using the [␣-32 P]dATP and the Klenow fragment to fill in the HindIII overhangs.
For DNA mobility shift assay, the binding reaction was performed for 30 min at room temperature in 20 l of binding buffer containing ϳ20,000 cpm/20 fmol of labeled probe, 1 g of poly(dI⅐dC)⅐poly(dI⅐dC) and nuclear extract containing 5.0 g of protein. Double-stranded oligonucleotides were used as competitors as described previously. Separation of free radiolabeled DNA from DNA-protein complexes was carried out on a 4 -5% non-denaturing polyacrylamide gel with a standard Tris borate electrophoresis buffer (TBE) at 300 V at a cold temperature (4°C). Autoradiography was performed by overnight exposure to Kodak Biomax film (Eastman Kodak Co.). The intensities of the bands were quantified using Packard Instant Imager (PerkinElmer Life Sciences).
Nuclear extracts and antibodies were preincubated for 20 min at room temperature before the radiolabeled probe was added followed by another 20-min incubation with the probe. RFXB/RFXANK, CIITA, and RFX1-RFX3 antibodies were purchased from Santa Cruz Biotechnology, Santa Cruz, CA. The anti-RFX5 antibodies 191 (C terminus) and 194 (amino acids 320 -494) were purchased from Rockland, Gilbertsville, PA.
Western Blots-Human fibroblasts or rat fibroblasts were treated with IFN-␥ for 24 h before harvesting. Proteins were extracted using radioimmune precipitation assay buffer (1ϫ phosphate-buffered saline, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM sodium orthovanadate) with freshly added phenylmethylsulfonyl fluoride (100 g/ml) and aprotinin (5-10 trypsin inhibitory units/ml). The amount of protein was determined by the Bradford protein assay (Bio-Rad). Pro-teins in extracts were separated by 10% polyacrylamide gel electrophoresis with prestained markers (Bio-Rad) used for estimating molecular weight and the 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 at room temperature for 3 h and hybridized for 1 h to several antibodies (polyclonal antibody against RFX5 (194) (Rockland). CIITA (N-20), RFX1 (I- 19), RFX2 (C-15), or RFX3 (T-17) (Santa Cruz Biotechnology)). After washing with buffer for three times, the membranes were incubated with appropriate secondary antibodies, either anti-goat IgG (Sigma) or anti-rabbit IgG (Amersham Biosciences), 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).
Luciferase assays were performed under standard conditions (Luciferase kit, Promega, Madison, WI). Briefly, the cells were washed twice with PBS buffer and scraped with lysis reagent. The cells and solution were centrifuged at 12,000 ϫ g to pellet the debris. The cell extract was mixed with the luciferase assay reagent, and light emission was measured in a luminometer. The luciferase activity was assayed in duplicate within the linear range of the instrument. Values were normalized to fluorescence of the green fluorescent protein and to total protein as measured by the Bradford reagent using bovine serum albumin as a standard.
In Vitro Transcription Assay-The reaction mixture for in vitro transcription contained 50 -90 g of nuclear extract, 1 g of super-coiled template collagen promoter (pH20, purified on by two CsCl gradients), 20 mM HEPES, pH 7.9, 4 mM MgCl 2 , 60 mM KCl, 2 mM EDTA, 0.5 mM dithiothreitol, 12% glycerol, 600 M of each rNTP in a final volume of 25 l. Double-stranded oligonucleotides containing binding sites for RFX1 protein, methylated pBR322 (pB1), and X-box (28) were used as competitors in some reactions. The oligonucleotides used for these experiments are given in Table II. The reactions were carried out at 30°C for 1 h and terminated by the addition of 175 l of stop solution, which contained 0.3 M sodium acetate, 0.5% SDS, 3 g/ml tRNA, pH 5.2. After extraction of protein with phenol/chloroform, the RNA was precipitated by ethanol. To detect the newly synthesized transcript, an antisense oligonucleotide primer corresponding to a sequence in the luciferase gene was generated as previously described (28). The primer sequence is given in Table II. The primer was end-labeled with polynucleotide kinase and [␥-32 P]ATP, hybridized to in vitro transcription products and extended using Moloney murine leukemia virus reverse transcriptase. The primer-extended products were analyzed by 5% polyacrylamide gel electrophoresis containing 7 M urea. Transcription reaction and primer extension reactions always included an RNase inhibitor protein (cloned human pancreatic RNase A lytic enzyme inhibitor, Ambion, Inc., Austin, TX). Gels were dried and autoradiographed at Ϫ80°C with an intensifying screen.

RESULTS
The COL1A2 transcription start site contains a sequencespecific binding site (Ϫ1 to ϩ20) for RFX1 protein. In our previous studies (28,56), using nuclear proteins extracted from rat fibroblasts, we demonstrated that RFX1 has a higher affinity binding to the collagen site if the ϩ7 C is either methylated or mutated from C to T. An important consideration is whether this site is methylated in vivo or in cells. We have already demonstrated that this region of the promoter is methylated in a chemically transformed cell line that does not express COL1A2 (57). However, the method used did not detect methylation status within the RFX1 consensus site.
To examine the methylation status at the ϩ7 CpG site in the COL1A2 gene, DNA was extracted from human fibroblasts, IMR-90, modified with bisulfite treatment, and analyzed by methylation-sensitive, single-nucleotide primer extension (MS-SNuPE) (51). Bisulfite treatment converts cytosines to uracil in single-stranded DNA under conditions that do not alter 5-methylcytosine. After bisulfite modification, the collagen promoter and first exon region were amplified by PCR using primers for each strand schematically represented in Fig. 1A with primer sequences in Table I. To analyze methylation within the collagen RFX binding site, specific primers (ϩ7 primers, Fig.  1A and Table I) were annealed to a sequence adjacent to the ϩ7 CpG site followed by single-nucleotide primer extension with radiolabeled nucleotides. A control primer (Fig. 1A and Table I) at a site that cannot be methylated within the collagen promoter was designed to test the efficiency of bisulfite conversion. The control primer single-nucleotide extension indicated that less than 2% of the Cs in the promoter was left unconverted. The ϩ7 CpG site was only 4% methylated in the human fibroblasts on the coding strand. Unlike the coding strand, the template strand showed variable methylation at the ϩ7 site. The graph in Fig. 1B represents the average of five separate experiments with standard error bars.
We next investigated whether proteins present in human fibroblast nuclear extracts have differential affinity for a fully methylated or a hemi-methylated collagen site (Ϫ1 to ϩ20). As demonstrated in our earlier reports (28,56), RFX1 interacts with the unmethylated collagen sequence in a sequence-specific manner as judged by interaction with recombinant protein and competition with consensus and mutated sequences. RFX1 complex formation is three times stronger with methylated probes than with unmethylated probe (Fig. 1C, compare lane 1  to lane 2). When the template strand is methylated, RFX1 complex formation is similar to unmethylated complex formation (Fig. 1C, compare lane 1 to lane 3). Interestingly, when the coding strand alone is methylated, RFX1 complex formation is similar to the fully methylated complex formation (Fig. 1C,  compare lane 2 to lane 4). This suggests that the increased RFX1 affinity for methylation is mediated through the coding strand but not the template strand.
RFX1 is one of five proteins in the RFX family that can bind to an X-box site in the MHC II promoter, which has no possible methylation sites. Several investigators (37,55,58) use the methylated sequence from pBR322 (pB1) to deplete RFX1 from nuclear extracts to detect RFX5 complex formation at the X-box site in MHC II complex. This sequence interacts well with RFX1-3 proteins (59), but it is not a consensus binding site for RFX5. When pB1 was added as a competitor, a clearly different faster migrating complex (complex 3) formed on the collagen site. Complex 3 was not as responsive to methylation as RFX1 (Fig. 1C, lanes 5-8). The radioactivity in each complex 3 was within 20% of the unmethylated sequence when using probes with the same specific activity. We hypothesized that complex 3 is RFX5 complex, because RFX5 is not considered to be methylation-responsive (29).
Next, the collagen sequences with and without methylation were compared with the X-box consensus sequence from MHC II promoter using human fibroblast nuclear extracts ( Fig. 2A). In Fig. 2A (lanes 2, 4, and 6) and Fig. 2B (lane 3), there is an increase in binding of complex 3 in the presence of pB1. Complex 3 migrates similarly to RFX5 on the MHC II consensus X-box sequence as previously described (37,55,58). Similar complexes were formed with all probes suggesting that the collagen site binds to the same proteins as the well-characterized MHC II X-box site. Complex 3 formation on unmethylated sequences without pB1 varies between experiments (see Fig.  2A, lanes 1-4 and Fig. 2B, lane 1) depending on the amount of RFX1 extracted from the nucleus. The difference in the amount of complex 3 formation in the presence of pB1 in Fig. 2B is probably due to lower specific activity of the unmethylated probe.
The complexes formed on the unmethylated collagen transcription start site are sequence-specific (Fig. 2B). Several RFX1 binding sites were used in competition gel shift assays to determine the specificity of protein-DNA binding using human fibroblast nuclear proteins (Fig. 2B, left panel) and rat fibroblast nuclear proteins (Fig. 2B, right panel). Collagen binding sequence (Fig. 2B, lane 2 and lane 8) and the MHC II RFX binding sequence, X-box (Fig. 2B, lane 4), competed for the slowest migrating complexes, complexes 1 and 2. However, the methylated pBR322 (pB1) consensus sequence allowed the formation of a unique faster migrating complex, complex 3 (Fig.  2B, lane 3), when human fibroblast nuclear extracts were used. Interestingly, complex 3 did not form when rat fibroblast nuclear extracts were used (Fig. 2B, lane 9) (56,60). When methylated pB1 sequence was present with the X-box consensus sequence as a competitor, complex 3 was not visible (Fig. 2B,  lane 5). TAE, a nonspecific DNA sequence from the ␣1(I) promoter (17), did not compete with (data not shown) or without pB1 (Fig. 2B, lane 6). Several nuclear extracts from human skin fibroblasts also contained complex 3 in the presence of pB1 (data not shown).
RFX5 has different binding site preferences than RFX1-3 (37). The position and orientation of the RFX5 complex on the X-box binding sequence (37) has been determined. The se- FIG. 5. Both RFX1 and RFX5 repress collagen promoter activity in rat fibroblast (FR) as judged by cotransfection assay. The collagen promoter luciferase construct (0.5 g) (pH20) was co-transfected using LipofectAMINE into rat fibroblasts with the indicated amounts of RFX1 (A) or RFX5 (B) expression vectors. Cells were also co-transfected with a GFP construct (0.1 g) driven by CMV promoter to normalize the transfection efficiency. Each plate contained the same amount of DNA by adding appropriate pcDNA empty construct (Vector). Luciferase activity was measured after 24 h of transfection with 20 g of protein. The activities were normalized for numbers of cells containing green fluorescent proteins. Each bar indicates the relative luciferase activity as percent control with empty vector. The results indicate the average of three independent sets of transfections in duplicate. A, RFX1 cotransfection in rat fibroblasts. B, RFX5 co-transfection in rat fibroblasts. quence required for binding at the X-box is homologous with the collagen sequence surrounding the transcription start site of both collagen type I genes (Fig. 2C). This further suggests that RFX5 complex forms on the collagen site.
Next, commercially available antibodies were used to confirm the identity of the complexes binding to methylated (Fig.  3, left panel) and unmethylated (Fig. 3, right panel) collagen sequences. RFX1 antibody clearly supershifted complexes 1 and 2 (Fig. 3, lane 2 and 7). The RFX2 antibody blocked binding of only complex 2 (Fig. 3, lanes 3 and 8). RFX3 antibody and non-immune serum did not supershift or block binding of either complex. In our earlier report (28), recombinant RFX1 formed complex 1. These data are consistent with the idea that complex 1 is a homodimer of RFX1, whereas complex 2 is a heterodimer of RFX1 and RFX2.
Next, RFX1 binding was depleted using methylated pB1 to determine if complex 3 was indeed RFX5 complex containing RFX5 and RFXB/RFXANK. All RFX5 and RFXB/RFXANK antibodies reduced protein binding to COL1A2 transcription start site DNA (Fig. 4A, lanes 1-4), whereas a non-immune serum did not alter complex 3 formation (Fig. 4A, lane 5). These commercially available antibodies disrupted complex 3 binding. Others have demonstrated that all three proteins are necessary for binding to DNA and can be co-immunoprecipitated (36). Therefore, an RFX5 complex forms on the RFX site spanning the collagen transcription start site.
In human fibroblasts, IFN-␥ induces CIITA (61), which is considered a master regulator for IFN-␥ induction of MHC II genes (62). RFX5 recruits CIITA to MHC II promoter to activate transcription (63,64). On the other hand, IFN-␥ represses collagen expression (21). Because the collagen gene contains an RFX5 site, it was of interest to examine the role of CIITA and its interaction with RFX5 in human fibroblasts treated with IFN-␥. Nuclear extracts from control and IFN-␥-treated cells were used in gel shift assays. A CIITA antibody reduced complex formation only in the IFN-␥-treated nuclear extracts (Fig.  4B, lane 6 compared with lane 3 without IFN). Most impor-tantly, there is an increase in complex formation when cells are treated with IFN-␥ (Fig. 4B, compare lanes 1 and 4) indicating the possibility that RFX5 may recruit CIITA.
The presence of CIITA and RFX family members in both rat and human fibroblasts was examined by Western analysis. All RFX5 and CIITA antibodies reacted with the appropriate molecular weight bands when compared with recombinant proteins expressed in COS cells (data not shown). CIITA protein in the human fibroblast extracts could only be observed when high amounts of protein (120 g) were separated on 10% SDSgel electrophoresis. However, CIITA extracted from rat fibroblasts was visible in less total protein (30 g) (Fig. 4C). On the other hand, RFX5 was clearly visible in human fibroblast extracts (30 g). Unlike CIITA, RFX5 was not detected in rat fibroblast extracts (120 g) (Fig. 4C). This confirms the gel shift analysis and suggests that RFX5 binding proteins are less abundant in the rat fibroblast cells compared with human fibroblasts.
RFX1 has both an activation and repression domain. Because increased binding of RFX1 to methylated or ϩ7 mutated collagen cDNA (C to T) repressed collagen transcription (28), it was hypothesized that RFX1 was a collagen transcription repressor on the normal collagen promoter. To further investigate the role of RFX1 on collagen gene expression, co-transfection assays were performed. Overexpression of RFX1 repressed collagen promoter activity (40 -50%) in rat fibroblasts (Fig. 5A) and in human fibroblasts (data not shown). Therefore, RFX1 can repress the normal unmethylated collagen promoter. Because RFX5 complex also can bind to the collagen transcription start site, the function of RFX5 was also determined by cotransfection studies. RFX5 was transfected into rat fibroblasts that contained undetectable RFX5 binding activity (Fig. 2B, lane 9) and RFX5 protein (Fig. 4C). Increasing amounts of RFX5 suppressed collagen promoter activity by 50% in a dosedependent manner (Fig. 5B). There was little effect of RFX5 overexpression in human fibroblasts (data not shown), possibly because these cells have high endogenous RFX5 binding activity ( Fig. 2A, lanes 2 and 4, and Fig. 2B, lane 3) and RFX5 protein levels (Fig. 3C).
In vitro transcription assays were performed to further characterize the regulation of collagen by RFX5 complexes. Our earlier reports (28,56) demonstrated that, when RFX1 binding is enhanced using methylated or mutated templates, transcription is repressed in an in vitro transcription assay. In this study, RFX5 binding was enhanced using competitors. The control template had activity that was not altered by the addition of X-box (Fig. 6, lane 4) or several other RFX1 consensus sequences (data not shown). However, the addition of methylated pB1 to nuclear extracts inhibited collagen transcription (Fig. 6, lane 3). The pB1 oligonucleotides were added under conditions expected to decrease RFX1 binding while enhancing RFX5 complex formation. These data suggest that RFX5 complex formation represses transcription. Therefore, we propose that under favorable conditions RFX5 can bind to the collagen transcription start site and inhibit collagen basal transcription.
CIITA is essential for constitutive and IFN-␥ activation of MHC II protein transcription but can also coordinately repress other proteins (45)(46)(47). IFN-␥ treatment of human fibroblasts decreases collagen mRNA levels (data not shown) and inhibits collagen promoter constructs in a dose-dependent manner (Fig.  7A). Rat fibroblast collagen expression was not down-regulated by IFN-␥ treatment, and IFN-␥ did not down-regulate the collagen promoter in a dose-dependent manner (data not shown). Therefore, IFN-␥ repressed collagen promoter activity in cells that contain RFX5 complex (Fig. 7A) but did not repress collagen in cells without RFX5 complex formation. Finally, FIG. 6. RFX1 as well as RFX5 are the repressors of collagen transcription as judged by in vitro transcription assay. An in vitro transcription assay was performed with 50 g of human fibroblasts nuclear proteins using supercoiled pH20 as template (see "Materials and Methods" for details). The arrow indicates the correctly initiated transcript. Lane 1 is the standard (Std), which is the x174 DNA digested by Hinf1. Lanes 2-4, the control template; lane 3, methylated pBR322 competitor sequence at a 50-fold molar excess to the template; lane 4, X-box sequence oligonucleotide at a 50-fold molar excess to pH20. CIITA was transfected into human fibroblasts. Increasing doses of CIITA suppressed collagen transcription as judged by the lower luciferase activity (Fig. 7B) in human fibroblasts and to a lesser extent in rat fibroblasts (Fig. 7C). This suggests that RFX5 and CIITA are necessary for collagen repression induced by IFN-␥. DISCUSSION Collagen transcription in most tissues undergoes multiple alterations during the life of an organism, alternating between activation and repression. Transcription is active during development and growth, but transcription is repressed during adult life except when tissue is injured and requires remodeling. Therefore, the gene most likely requires both activators and repressors. Even though the gene is predominantly repressed, little is understood about the mechanism of collagen transcriptional repression.
Previous results (28,56) demonstrate that RFX1 interacts with the COL1A2 gene, spanning the transcription start site. This protein binds to the collagen start site with higher affinity if the cytosine at ϩ7 is either methylated or mutated (C to T). These mutations or methylation at ϩ7 reduce transcription in transfection and in vitro transcription assays. Most likely, RFX1 is acting at a repressive element located at the transcription start site. In this report, we demonstrate that the coding strand methylation is more important than template strand methylation for increased RFX1 affinity (Fig. 1C). Furthermore, human fibroblasts in culture have low methylation at the ϩ7 site on the coding strand with more variable methylation on the template strand (Fig. 1B). The variable template strand methylation may be related to age in culture (65)(66)(67)(68), proliferation (69), cell cycle (70), or confluence (66), because methylation status does change under these culture conditions. In addition, the ϩ7 site is methylated on the coding strand or both strands in several cancer cell lines, and increased methylation correlates inversely with collagen gene activity. 2 The fact that methylation and gene silencing is prevalent in the coding strand of COL1A2 in cancer cells could be correlated to the important role of RFX1 in collagen methylation and in the consequent down-regulation. We hypothesize that methylation increases the binding of RFX1 at the transcription start site as part of the mechanism for methylation-induced transcriptional repression of collagen. In mammalian systems, a repressive element at the transcription start site is not a common finding, although such elements have been reported in the viral genes (71). However, RFX1 sites are present near the start site of several viral and mammalian genes (72,73). Therefore, this protein may be involved with repression at start sites in other genes as well as collagen. Because the protein does bind to the unmethylated collagen sequence, we concentrated in this report on defining the role of RFX1 on the normal collagen gene.
RFX1 is a member of a protein family with highly conserved DNA binding domains. This report demonstrates that RFX1 homodimers and RFX1⅐RFX2 heterodimers can bind to the collagen start site (Fig. 3). Co-transfection experiments (Fig.  5A) demonstrate that RFX1 protein represses rather than activates collagen transcription on unmethylated DNA. Overexpressed wild-type RFX1 often has no inducing or repressing activity even though RFX1 has both activation and repression domains (74). However, co-transfection data clearly demonstrate inhibition of the collagen gene by RFX1 overexpression. Mutational analysis of the human proliferating cell nuclear antigen promoter was necessary to demonstrate that RFX1 acts as a repressor of human proliferating cell nuclear antigen (75), possibly through the interaction with p107 (76). RFX1 activates a promoter with multiple RFX1 binding sites from an interleukin-5␣ receptor in a lineage-specific manner, suggesting that there is cooperation with other tissue-and lineagespecific cofactors involved in RFX1 function (32). RFX1 associates with an Myc intron binding factor (MIBP1) to activate Myc expression (77) and with a B-cell-specific activity protein (BSAP/Pax5) that regulates B-cell specificity of Epstein-Barr virus growth-transforming function (78). RFX1 also interacts with and activates c-Abl kinase, a non-receptor tyrosine kinase activated in the nucleus during S phase (79). In addition, c-Abl binds to a viral consensus RFX site when it is phosphorylated (80,81). Our unpublished results suggest that c-Abl can interact with the RFX1 collagen complex. Most likely, RFX1 represses the unmethylated active collagen gene under certain conditions even though the repression element in the transcription start site is a low affinity binding site. A low affinity binding site may allow easily reversible repression at the start site.
RFX5 complex formation is an essential component in the activation of MHC II proteins (34). This complex has been thoroughly studied only on MHC I or II promoters. Unlike the other members of the RFX family, RFX5 does not form dimers but rather forms a complex with two other proteins called RFXB/RFXANK and RFXAP. Formation of the trimer is essen-tial for RFX5 binding. RFX5 interaction with the collagen start site was investigated based on conditions established for the MHC II X-box using a methylated competitor to remove other family members (37,55,58). RFX5 complex in human fibroblast nuclear extracts interacts with the collagen transcription start site as judged by gel shift migration patterns and blocking of binding by RFX5 and RFXB/RFXANK antibodies similar to X-box (Fig. 2). This is the first demonstration that RFX5 complex formation occurs at a site other than in MHC II promoters. Interestingly, there is little RFX5 complex in a rat fibroblast cell line originally examined in our earlier studies (28). Further Western analysis revealed that there is no detectable RFX5 protein in rat fibroblasts whole cell extracts, whereas RFX5 can be easily detected in human fibroblasts (Fig. 4C).
Next, the function of RFX5 was tested using co-transfection assays. RFX5 repressed collagen transcription in rat fibroblasts in a dose-dependent manner (Fig. 5B), whereas it had no effect on collagen promoter in human fibroblasts (data not shown). RFX5 is pre-assembled in a complex with RFXB/ RFXANK and RFXAP before binding to DNA (82). Individual proteins do not bind well to DNA. It is possible that in the rat fibroblasts there is a small excess of RFXB/RFXANK and RFXAP proteins to form a complex with the overexpressed RFX5. The human fibroblasts have more RFX5 complex binding activity as judged by the gel shift experiments (Figs. 2 and  3). Therefore, additional overexpressed RFX5 in these cells might not be able to function as a repressor.
CIITA, considered a master regulator for MHC II transcription, is essential for activation of MHC II promoters (34) and antigen presentation (54). Certain mutations in CIITA inactivate MHC II transcription causing bare lymphocyte syndrome, characterized by a lack of MHC II proteins on lymphocyte membranes. In these cases, RFX5 complex forms on the MHC II promoter, but transcription is not activated. Mutations in the three proteins present in the RFX5 complex also cause bare lymphocyte syndrome and alter the complex formation on The theoretical activated promoter is at the top, and the inactive promoter is at the bottom. The lines with boxes represent the first 350 bases of the promoter with its multiple cis-acting sites shown as boxes (Sp1, Smad, AP1, NFkB, Ets, IFN, CCAAT, and TATA). The various DNA binding proteins that interact at these sites are depicted as circles. The co-activators that may be involved with collagen transcription, CBP/p300, GCN5/PCAF, and CIITA, are illustrated as ovals along with the polymerase II and TAFs.
DNA. Occupancy at the RFX site on the MHC II promoter by RFX5 complex is essential for recruitment of CIITA in certain instances. Because collagen contains an RFX5 site, CIITA was examined for its ability to repress or activate collagen synthesis. Co-transfection with CIITA expression constructs in human fibroblasts reduced the activity of the collagen promoter down to 25% of the original activity in a dose-dependent manner (Fig. 7B). The same CIITA construct in rat fibroblasts repressed activity but to a lesser extent (Fig. 7C), possibly because these fibroblasts have less RFX5 complex and/or more CIITA protein (Fig. 4C). CIITA is also a repressor of FAS ligand, IL-4 synthesis, and thyroid-specific genes (45)(46)(47). In the case of the IL-4 gene, the IL-4 promoter is regulated by multiple protein-protein interactions between CIITA, NF-AT, and coactivator CBP/p300 (48). It is not clear whether there is an RFX site in these promoters.
Importantly, CIITA transfection represses endogenous collagen synthesis as well as promoter activity using a construct containing 350 bases of a collagen promoter (49). This construct contains several potential binding sites for factors that could interact with CBP/p300, including a Smad site, an Ets site, and numerous Sp1 sites. Our expression studies with CIITA corroborate the conclusion that CIITA represses collagen transcription. It should be noted that our construct in human fibroblasts is more sensitive to overexpression of CIITA and IFN-␥, possibly because it is a smaller collagen promoter construct (225 bases of promoter) without activator sites. In addition, these investigators (49) suggest that the repression occurs through an interaction with CBP/p300, because an N-terminal 36-amino acid sequence in CIITA interacts with CBP/p300 and is essential for collagen down-regulation. Most importantly, IFN-␥ does not repress collagen transcription in a cell line (G1A) that does not induce CIITA. CIITA expression rescues the IFN-␥ response suggesting that CIITA is indeed an active protein involved in collagen suppression. CIITA is often considered a scaffold protein interacting with several proteins at multiple promoter sites (38). Therefore, it is possible that it represses collagen through interactions with RFX5 proteins and CBP/p300. Because collagen has an RFX site, this protein complex may help recruit CIITA to the gene. CIITA transcription is regulated by several promoters and cytokines. INF-␥ induces and TGF-␤ down-regulates CIITA transcription through its fourth promoter in many cell types, including fibroblasts (61). Most importantly, IFN-␥ represses collagen type I and opposes TGF-␤ activation in many cell types. In the two cell lines used in this study there was a difference in sensitivity to IFN-␥. Collagen transcription was highly sensitive to the cytokine in the human fibroblasts (Fig.  7A), yet the rat fibroblasts required 150 units/ml to see any reduction in collagen promoter activity, and that decrease was not dose-responsive. This difference in activity could be due to the fact that there are reduced amounts of RFX5 and higher amounts of CIITA in the rat fibroblasts (Fig. 4C).
Because an RFX5 site was detected in the collagen transcription start site, the binding of CIITA with and without IFN-␥ was investigated in the human fibroblasts responsive to IFN-␥. IFN-␥ increased the RFX5 binding to the collagen site and antibodies to CIITA blocked binding only in the treated nuclear extracts. This suggested that CIITA might be involved with stabilizing the interaction of RFX5 to the collagen DNA.
RFX5 complex forms a stable interaction with NF-Y/CBF complex on MHC II promoters (82)(83)(84). Collagen has a wellcharacterized NF-Y/CBF complex that forms on the collagen promoter (85,86). It is possible that RFX5 proteins form cooperative interactions with NF-Y/CBF on the collagen promoter. In addition, others have detected IFN-␥ response elements in the collagen promoter (20,21) but the proteins involved have not been characterized. The interaction between CIITA and CBP/p300 has been implicated in IFN-␥-mediated suppression of collagen transcription (49,87). CIITA could form a repression scaffold on the collagen promoter interacting with RFX5, NF-Y/CBF, and protein binding to the IFN-␥-responsive element. Thus, this CIITA complex may replace CBP/p300, preventing activation of transcription as shown in our hypothetical model (Fig. 8). In this model, IFN-␥ inactivates RFX1 and activates CIITA transcription. Once CIITA is produced it may be recruited, in part, to the collagen promoter through its interactions with RFX5 complex on the collagen transcription start site. Our data suggest that RFX5 complex binding is increased in the presence of IFN-␥ treatment and CIITA. The CIITA most likely forms cooperative interactions with other proteins, including CBP, to repress collagen transcription.