DNA hypomethylation in rheumatoid arthritis synovial fibroblasts




Rheumatoid arthritis synovial fibroblasts (RASFs) are phenotypically activated and aggressive. We undertook this study to investigate whether the intrinsic activation of RASFs is due to global genomic hypomethylation, an epigenetic modification.


Global genomic hypomethylation was assessed by immunohistochemistry, flow cytometry, and L1 promoter bisulfite sequencing. The levels of Dnmt1 were determined in synovial tissue and cultured SFs by Western blotting before and after treatment with cytokines and growth factors. Normal SFs were treated for 3 months with a nontoxic dose of the DNA hypomethylation drug 5-azacytidine (5-azaC), and changes in gene expression were revealed using complementary DNA arrays. The phenotypic changes were confirmed by flow cytometry.


In situ and in vitro, RASF DNA had fewer 5-methylcytosine and methylated CG sites upstream of an L1 open-reading frame than did DNA of osteoarthritis SFs, and proliferating RASFs were deficient in Dnmt1. Using 5-azaC, we reproduced the activated phenotype of RASFs in normal SFs. One hundred eighty-six genes were up-regulated >2-fold by hypomethylation, with enhanced protein expression. These included growth factors and receptors, extracellular matrix proteins, adhesion molecules, and matrix-degrading enzymes. The hypomethylating milieu induced irreversible phenotypic changes in normal SFs, which resembled those of the activated phenotype of RASFs.


DNA hypomethylation contributes to the chronicity of RA and could be responsible for the limitation of current therapies.

Rheumatoid arthritis (RA), which affects ∼1% of the population, is a chronic autoimmune disease involving progressive destruction of the affected joints. Proinflammatory cytokines, such as interleukin-1β (IL-1β), IL-6, and tumor necrosis factor α (TNFα), are involved in its pathogenesis (1–3). Anti-TNFα therapies in particular have been shown to provide substantial benefit to patients not only through the reduction of signs and symptoms of the disease, but also by the inhibition of joint destruction (1). However, treatments with current biologics, including anti-TNFα, anti–T cell, and anti–B cell therapies, are only successful in at best 60% of the treated patients and are still unable to cure the disease. A cytokine-independent pathway appears responsible for the ongoing joint destruction mediated by synovial fibroblasts (SFs) (2). Since implantation of RASFs with human cartilage into SCID mice causes invasion into the cartilage without the support of the cells of the human immune system, it has been proposed that the activated phenotype is an “intrinsic” property of these cells (3).

SFs, more than other types of fibroblasts, acquire phenotypic characteristics commonly associated with transformed cells (4). RASFs show “spontaneous” activities associated with aggressive behavior, and they differ from the SFs of patients with osteoarthritis (OASFs) or normal SFs. For example, RASFs up-regulate protooncogenes (5), specific matrix-degrading enzymes (6), adhesion molecules (7), and cytokines (8). These observations of an intrinsically activated cellular phenotype prompted us to search for epigenetic modifications.

In somatic cells, Dnmt1 is the predominant DNA methyltransferase (9, 10). Reduction of Dnmt1 levels leads to hypomethylation, genomic instability, and tumorigenesis. Direct interaction between Dnmt1 and proliferating cell nuclear antigen (PCNA) ensures that patterns of methylation are faithfully preserved in DNA synthesis (10). Moreover, repetitive sequences such as L1, Alu, and satellite alpha repeats are silenced by methylation in normal cells and can be used as markers of global hypomethylation (11). Our group and others (12, 13) reported a reactivation of the endogenous retroviral element L1 in the RA synovial lining and at sites of invasion. These reports suggest that global genomic hypomethylation plays a role in the pathogenesis of RA, and that genes normally silenced by methylation might contribute to the activated phenotype of RASFs (12).

Here we show that DNA demethylation of normal SFs induces a cellular phenotype resembling that of activated RASFs. Genomic hypomethylation is a characteristic of RASFs and is involved in the pathogenesis of RA.


Cell cultures.

RASFs and OASFs were isolated from synovial tissue obtained during joint replacement surgery. Selected OA tissue showed no signs of inflammation or hyperplasia. Normal SFs were isolated from a small joint biopsy sample from a trauma patient. The cells were cultured in Dulbecco's modified Eagle's medium (DMEM) including 10% fetal calf serum (FCS) and used between passages 5 and 6 (reagents were from Life Technologies, Basel, Switzerland). The procedure was approved by the Ethics Committee of the University Hospital, Zurich, Switzerland. The characteristics of the study patients are shown in Table 1.

Table 1. Characteristics of the study patients*
 RA patients (n = 13)OA patients (n = 13)
  • *

    RA = rheumatoid arthritis; OA = osteoarthritis; NSAIDs = nonsteroidal antiinflammatory drugs; DMARDs = disease-modifying antirheumatic drugs; anti-TNF = anti–tumor necrosis factor; NA = not assessed; CRP = C-reactive protein.

  • Rheumatoid factor (RF) positivity was defined as >20 IU/ml.

Age, mean (range) years64 (48–79)66 (54–88)
No. of women/no. of men9/47/6
Medications, no. taking/no. assessed  
  Plus steroid4/130/13
  Plus anti-TNF2/130/13
RF, no. positive/no. assessed6/13NA
CRP, mean (range) mg/liter35.9 (8.2–60.8)2.4 (0.2–8.3)

Immunohistochemistry for 5-methylcytosine.

Formalin-fixed, paraffin-embedded sections of synovial tissue were deparaffinized and treated at 80°C for 30 minutes with citrate buffer (pH 3.4). The tissue slides were incubated with 2N HCl for 2 hours at 37°C. After acid treatment, the slides were washed well with phosphate buffered saline–0.05% Tween 20 (PBST). To determine the methylation in the synovial tissue, mouse monoclonal antibodies against 5-methylcytosine (Imgenex, San Diego, CA) were used. Mouse IgG isotype (Dako, Glostrup, Denmark) served as a negative control. Double staining with vimentin (using murine anti-human monoclonal antibodies, clone V9; Dako) was used to stain SFs. The antibodies were incubated overnight at 4°C. Bound antibodies were detected by incubation with biotinylated goat anti-mouse IgG (The Jackson Laboratory, Bar Harbor, ME) for 1 hour. The slides were incubated with avidin-conjugated horseradish peroxidase (HRP) (from Vectastain Universal Elite ABC kit; Vector, Burlingame, CA). The methylcytosine modification and vimentin were visualized using 3,3′-diaminobenzidine (Vector) and HistoGreen (Histoprime; Linaris, Wertheim-Bettingen, Germany), respectively.

ImageJ software (NIH Image, National Institutes of Health, Bethesda, MD; online at: http://rsbweb.nih.gov/ij/) was used to analyze the mean intensity of nuclei in the patient tissue. We acquired 3 different microscopic field images per patient tissue sample. The color images were converted into 8-bit gray/white images. The threshold was adjusted for each image in order to specify only the nuclei. The means of gray intensity and white intensity of all nuclei in the image were obtained and compared. The numerical values given by software were converted to a mean intensity unit using the following equation: mean intensity = (1/mean of gray intensity/white intensity) × 100.

Flow cytometry (fluorescence-activated cell sorting [FACS]) for 5-methylcytosine.

The cells were fixed with 0.25% paraformaldehyde for 10 minutes at 37°C and kept on ice for 10 minutes before addition of 88% methanol/12% PBS for 10 minutes at –20°C. The nuclei were washed twice with PBST/bovine serum albumin (BSA) and treated with 1N HCl for 40 minutes at 37°C. Neutralization of the acidic solution was performed by 1 wash step with 0.1M borate buffer (pH 8.5) and 2 wash steps with PBST/BSA. The nuclei were incubated for 20 minutes at 37°C with a blocking solution that contained PBST/BSA supplemented with 10% FCS. The nuclei were incubated with anti–5-methylcytosine antibodies for 1 hour at 37°C, washed twice with PBS, and incubated with fluorescein isothiocyanate (FITC)–conjugated anti-mouse antibodies for 30 minutes at 37°C (BD Biosciences, Heidelberg, Germany). Finally, the samples were stained with propidium iodide (PI) before being analyzed using FACS (FACSCalibur; BD Biosciences).

Bisulfite sequencing for L1 promoter.

Genomic DNA was prepared from RASFs and OASFs using the QiAmp DNA blood Mini kit (Qiagen, Hombrechtikon, Switzerland). The DNA (1 μg) was bisulfite modified using the EpiTect bisulfite kit (Qiagen). The modified DNA was eluted in 20 μl of Tris buffer (pH 8.5) and stored at –20°C. Polymerase chain reaction (PCR) amplification of bisulfite-modified DNA (2 μl) was performed using Hot Start PCR and AmpliTaq Gold polymerase (Applied Biosystems, Rotkreuz, Switzerland). Primers were designed for the CpG area upstream of the L1 promoter (X58075, –420 bp to –49 bp), as follows: 5′-TTT-ATT-AGG-GAG-TGT-TAG-ATA-GTG-GG-3′ (forward) and 5′-AAA-CCC-TCT-AAA-CCA-AAT-ATA-AAA-TAT-AAT-3′ (reverse). The online MethPrimer software was used (http://www.urogene.org/methprimer/). The PCR-purified fragment was cloned using the TOPO TA cloning kit according to the instructions of the manufacturer (Invitrogen, Carlsbad, CA). The positive clones were sequenced (Microsynth, Balgach, Switzerland). The data analysis was performed using BiQ analyzer software (Max Planck Institute, Munich, Germany).

Immunohistochemistry for L1 proteins.

L1 open-reading frame 1p (ORF1p, also called the L1 p40 protein) and L1 ORF2p (p150 protein) were detected by immunohistochemistry on paraffin-embedded sections of synovial tissue. Rabbit polyclonal antibodies against L1 ORF1p and chicken polyclonal antibodies against L1 ORF2p (obtained from G. G. Schumann, Paul Ehrlich Institute, Section PR2/Retroelements, Langen, Germany) and biotinylated anti-rabbit or anti-chicken antibodies and alkaline phosphatase–conjugated streptavidin (all from The Jackson Laboratory) were used, and fast red substrate (Vector) revealed the staining. Nonimmune rabbit or chicken sera were used as a control for primary antibodies.

Western blotting for Dnmt1 and PCNA.

Tissue or cells were prepared by lyses in radioimmunoprecipitation assay buffer (50 mM Tris HCl [pH 8], 150 mM NaCl, 1% Nonidet P40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate [SDS]; all from Sigma-Aldrich, Buchs, Switzerland). Proteins were separated in 10% SDS–polyacrylamide gels and transferred to nitrocellulose membranes. Membranes were blocked for 1 hour in 5% nonfat dry milk with 0.05% Tween 20 in Tris buffered saline (TBS) (pH 7.4) and were incubated overnight with antibodies against human Dnmt1 (Abcam, Cambridge, UK) or PCNA (PC10; Imgenex). After incubation with HRP-conjugated goat anti-mouse IgG secondary antibodies (The Jackson Laboratory) in 5% nonfat dry milk with 0.05% Tween 20 in TBS (pH 7.4), bound antibodies were visualized using enhanced chemiluminescence (Amersham, Buckinghamshire, UK). The intensity of the bands was evaluated by densitometry (Alphaimager 2200; Witec, Littau, Switzerland).

Treatment with TNFα, IL-1β, or platelet-derived growth factor (PDGF).

RASFs and OASFs were passaged for 48 hours and serum-starved with 0.5% FCS for 24 hours before adding 10 ng/ml recombinant human TNFα, 1 ng/ml IL-1β, 10 ng/ml PDGF (all from R&D Systems, Minneapolis, MN), or medium alone to the cell culture. The cells were kept in DMEM containing 10% FCS and collected after 24 or 48 hours.

Microarray of 5-azacytidine (5-azaC)–induced gene expression.

Normal SFs (n = 1 culture) were treated with a low dose of 5-azaC (0.1 μM/ml for 3 months, with medium changed every 3 days over 2 passages) (Sigma-Aldrich) or left untreated (control group). The long incubation was used to have enough cell divisions and to mimic the chronic state of the disease. Total RNA was isolated with the RNeasy MiniPrep Kit (Qiagen) including treatment with RNase-free DNase. Double-stranded complementary DNA (cDNA) was synthesized from 5 μg total RNA using the GeneChip One-Cycle cDNA Synthesis Kit (Invitrogen), including labeling with the One-Cycle Target Labeling Assay (Invitrogen). The labeled cDNA was hybridized with the probe sets present on the Human Genome U133 Plus 2.0 Gene Expression Array (Affymetrix, Santa Clara, CA), using the Fluidics Station 450, according to standard protocols. The hybridization picture was scanned with a GeneChip Scanner 3000 (Affymetrix) and further analyzed using the GCOS software (Affymetrix). Data were normalized (measurements <0.01 were set to 0.01, and normalization per chip was set to the 50th percentile) and analyzed with GeneSpring Microarray Analysis Software (Silicon Genetics, Redwood City, CA). Filters were set on 2-fold regulation, expression levels had to be >2.0, and flags had to be present or marginal in at least 1 of 2 compared samples.

FACS for phenotyping of RASFs.

RASFs and OASFs were treated with a low dose of 5-azaC (0.1 μM/ml for 2 weeks, with medium changed every 3 days). The cells were detached using Accutase (Omnilabo, Breda, The Netherlands) and incubated for 1 hour at 4°C in DMEM including 10% FCS with the following primary antibodies or isotype controls: anti-CD10, anti-CD26, anti-CD29, anti-CD36, anti-CD46, and anti-CD130 (murine monoclonal antibodies; BD PharMingen, San Diego, CA), anti–matrix metalloproteinase 14 (anti–MMP-14) hinge region (rabbit polyclonal antibodies; Chemicon, Zug, Switzerland), and anti–transforming growth factor β receptor type II (anti-TGFβRII) and anti–cathepsin K (anti-CTK) (murine polyclonal antibodies; Abnova, Heidelberg, Germany). CTK was measured in permeabilized cells using IntraStain (DakoCytomation, Carpinteria, CA). The cells were further incubated with FITC-conjugated secondary antibodies (goat anti-mouse IgG/IgM or rat anti-rabbit IgG; BD PharMingen). The mean fluorescence intensity was determined using a FACSCalibur.

Gene analysis.

The following online programs were used: Ensembl genome browser (http://www.ensembl.org/index.html), CpGplot (http://www.ebi.ac.uk/emboss/cpgplot/), Transcription Element Search System (http://www.cbil.upenn.edu/cgi-bin/tess/tess), and TFBS list (http://lgsun.grc.nia.nih.gov/geneindex/mm6/TFBS/list.html).

Statistical analysis.

Differences between patient groups were evaluated using the Mann-Whitney U test, while changes in the same cell culture were evaluated using the Wilcoxon signed rank test. Analysis of frequency was performed using chi-square tests.


Global genomic hypomethylation in RA synovial tissue.

To visualize global genomic methylation, paraffin-embedded synovial tissue sections from RA and OA patients were stained with anti–5-methylcytosine monoclonal antibodies (Figure 1). In RA, the vimentin-positive synovial fibroblasts showed decreased stainings of cell nuclei in both the lining and the sublining, reflecting a generalized genomic hypomethylation (Figure 1A). Detailed image analysis of nuclei in the synovial lining and sublining showed that RASF nuclei were stained less densely than OASF nuclei (Figure 1B). Thus, synovial cell nuclei were 13% less methylated in RA (range 10–20%) than in OA (P < 0.05) (n = 6 samples each) (Figure 1C).

Figure 1.

Immunohistochemistry shows global genomic hypomethylation in rheumatoid arthritis (RA) synovial tissue. A, Osteoarthritis (OA) and RA synovial tissue was stained with monoclonal antibodies against 5-methylcytosine (brown) and vimentin (green). B, RA synovial fibroblasts showed lower intensity of brown staining (5-methylcytosine; arrows) than OA synovial fibroblasts. Representative intensity histograms were obtained with ImageJ software (NIH Image, National Institutes of Health, Bethesda, MD; online at: http://rsbweb.nih.gov/ij/). C, RA synovial tissue was significantly less methylated than OA synovial tissue, as reflected by the lower nuclear staining of 5-methylcytosine (n = 6 samples each, determined by ImageJ software analysis). Horizontal bars indicate the mean. (Original magnification × 200 in A and B.)

Global genomic hypomethylation in RASFs and effects of cytokines and growth factors.

To determine whether RASFs are still hypomethylated in vitro, SFs were isolated from tissue and cultured for 5–6 passages. In addition, we investigated the effect on DNA methylation of exposure to physiologic concentrations of proinflammatory cytokines and growth factors. Cells were stimulated or left untreated, and then harvested. Cell nuclei were stained with anti–5-methylcytosine monoclonal antibodies (Figure 2A) and PI (Figure 2B) and analyzed by flow cytometry. The cell nuclei of untreated RASFs, compared with those of untreated OASFs, showed significantly less 5-methylcytosine staining (P < 0.05) (n = 6 samples each) (Figures 2A and C). In RASFs, 5-methylcytosine remained significantly reduced even in the presence of the proinflammatory cytokines TNFα or IL-1 (P < 0.05) (n = 6 samples each). TNFα significantly increased the 5-methylcytosine content (P < 0.05 for both OASFs and RASFs), while IL-1 and PDGF had no effect. However, the relative deficiency of 5-methylcytosine remained in RASFs, when compared with OASFs (P < 0.05) (n = 6 samples each), except in the presence of PDGF.

Figure 2.

Flow cytometric analysis of cell nuclei confirms genomic hypomethylation of rheumatoid arthritis synovial fibroblasts (RASFs) in vitro; effects of proinflammatory cytokines and growth factors. A, Nuclei of RASFs and osteoarthritis (OA) SFs labeled with fluorescein isothiocyanate (FITC)–conjugated anti–5-methylcytosine (anti–5-MeC) antibodies and evaluated by flow cytometry. Shown is a representative example of reduced 5-methylcytosine in the nuclei of RASFs compared with that in the nuclei of OASFs (gray lines represent isotype control [IC]). B, Example of cell cycle analysis using propidium iodide (PI) staining, showing distinct G1, S, and G2/M phases in unstimulated SFs or in SFs exposed for 24 hours to tumor necrosis factor α (TNFα). C, Histograms of mean fluorescence intensity (MFI) of stainings with 5-methylcytosine. Unstimulated RASFs showed a significant reduction in 5-methylcytosine. The cells were treated with TNFα, interleukin-1β (IL-1β), or platelet-derived growth factor (PDGF). In all conditions tested except for that with PDGF, 5-methylcytosine remained significantly deficient in RASFs compared with OASFs (n = 6 samples each). D, Percentage of total cells in the G2/M phase of the cell cycle in unstimulated and stimulated SFs (same experiment as in C). As expected, the proliferation increased upon stimulation, but no difference could be detected between OASFs and RASFs. NS = not significant. Values in C and D are the mean and SD. ∗ = P < 0.05 versus unstimulated control cells.

Cell cycle analysis.

We determined the percentage of OASFs and RASFs in the G2/M phase in the different conditions (i.e., with or without the addition of cytokines or growth factors) (Figures 2B and D). The strongest increase occurred with TNFα within 24 hours. However, no significant difference was detected between OASFs and RASFs in the different conditions (P = 0.8–0.9) (n = 6 samples each).

Expression of Dnmt1 and PCNA and effects of cytokines and growth factors.

We performed Western blotting with specific monoclonal antibodies to search for a defect in the methylation pathway. Tissue and cell lysates were analyzed for the expression of Dnmt1 and PCNA. The Dnmt1:PCNA ratio was significantly lower in RA synovial tissue than in OA synovial tissue (P < 0.05) (n = 5 samples each) (Figure 3A). In RA, an increased rate of cell proliferation was associated with high levels of PCNA. In proliferating cells, one would expect to find an increased expression of Dnmt1; this was not the case in RA synovial tissue. The levels of Dnmt1 were even lower in RA synovial tissue than in OA synovial tissue, which showed low expression of PCNA and a low rate of proliferation. This observation suggested a deficient production or a decreased half-life of Dnmt1 in the RA synovial tissue.

Figure 3.

Relative deficiency of Dnmt1 in proliferating cells. A, The level of Dnmt1 is very low in RA synovial tissue compared with OA synovial tissue, in spite of an increased expression of proliferating cell nuclear antigen (PCNA). Western blots show representative examples of Dnmt1 and PCNA expression in synovial tissue from 2 OA patients and 2 RA patients. The PCNA:Dnmt1 ratio was significantly reduced in RA (n = 5 samples each). B, Representative Western blots show that the expression of Dnmt1 was also decreased in RASFs in vitro compared with that in OASFs. The histogram shows that the PCNA:Dnmt1 ratio was significantly lower in RASFs than in OASFs (n = 6 samples each). C, Representative Western blots show that stimulation of cells with TNFα for 24 hours increased the expression of PCNA, but the levels of Dnmt1 in RASFs remained very low. Importantly, in all conditions tested (including those with IL-1β and PDGF), the PCNA:Dnmt1 ratio remained significantly reduced in RASFs (n = 6 samples each). Values are the mean and SD. ∗ = P < 0.05 versus unstimulated control cells (shown in B). See Figure 2 for other definitions.

In vitro, the Dnmt1:PCNA ratio remained significantly lower in RASFs than in OASFs (P < 0.05) (n = 6 samples each) (Figure 3B). We tested the effect of proinflammatory cytokines and growth factors on the expression of Dnmt1 (Figure 3C). In OASFs, the Dnmt1:PCNA ratio was reduced upon stimulation with TNFα (P < 0.05) and increased upon stimulation with IL-1 (P < 0.05); this difference was in great part due to a strong increase of PCNA after TNFα stimulation. Important to note is that in all conditions tested, the Dnmt1:PCNA ratio was significantly reduced in RASFs (P < 0.05) (n = 6 samples each).

Hypomethylated L1 promoter in RASFs.

Because L1 proteins (L1 ORF1p and L1 ORF2p) are expressed in RA synovial tissue (Figure 4A), we sought to determine whether their expression is due to genomic hypomethylation. Eighteen CG sites of the L1 promoter 5′-untranslated region (i.e., GenBank accession no. X58075, 372 bp between nucleotides –49 and –420) were analyzed for changes in methylation by bisulfite sequencing. Genomic DNA derived from RASFs revealed significantly fewer methylated CpG sites upstream of the L1 ORF1 in comparison with DNA derived from OASFs (mean ± SD 78 ± 2% CpG methylation versus 85 ± 3% CpG methylation; P < 0.05) (n = 7 patients with RA and 7 patients with OA, and 25 clones analyzed for each) (Figure 4B). Furthermore, we found that the methylation of L1 promoter in the isolated synovial fibroblasts correlated significantly with the expression of L1 ORF1p protein in the corresponding synovial tissue (Figure 4C). Fibroblasts derived from RA tissue with increased L1 ORF1p expression showed a higher percentage of unmethylated and missing CpG than control fibroblasts derived from an OA tissue.

Figure 4.

L1 proteins and promoter hypomethylation in RASFs.A, L1 proteins (L1 open-reading frame 1p [ORF1p]/p40 in fibroblasts at sites of cartilage destruction and L1 ORF2p/p150 in round-shaped cells at sites of bone destruction) are expressed in RA synovial tissue (original magnification × 400). B, The L1 promoter (5′-untranslated region, 372 bp containing 18 CpG sites) is hypomethylated in RASFs compared with OASFs. Bisulfite sequencing was performed in 25 clones from 7 RASF cultures and 7 OASF cultures. The results were analyzed using BiQ analyzer software and expressed as the mean and SD. CpG sites at positions 1 and 9 were significantly hypomethylated, in addition to a significant overall hypomethylation. P values were obtained by Mann-Whitney U test. C, L1 promoter methylation in isolated SFs is correlated with L1 ORF1p protein expression in corresponding RA and OA synovial tissue. SFs derived from RA tissue have a higher percentage of unmethylated and missing CpG (non CpG) than SFs derived from OA tissue. The table shows the percentage of unmethylated CpG among the 18 CpG and the percentage of missing CpG in the sequence of the patient (non CpG). The total includes both unmethylated and missing CpG. This has been correlated with the expression of L1 ORF1p (formerly called L1 p40 protein) using a semiquantitative evaluation of the immunohistochemical staining (from + = low expression to +++ = high expression). See Figure 2 for other definitions.

Phenotype of hypomethylated normal SFs.

To explore whether a hypomethylating milieu is responsible for the activated phenotype of RASFs, we continuously treated normal SFs for 3 months with a nontoxic dose of the DNA hypomethylator 5-azaC, and we analyzed the modification in gene expression using microarrays. One hundred eighty-six genes were up-regulated >2-fold during this condition. Many of these genes are implicated in RA, including interleukins, growth factors and their receptors, extracellular matrix proteins and enzymes, matrix-degrading enzymes and inhibitors, adhesion molecules (see Supplementary Table 1, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/), protein kinases, transcription factors, components of Wnt, Ras, and Rho signaling pathways, and apoptosis-related proteins (see Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/). The induced genes were categorized according to the occurrence of CpG islands in their respective promoters, in exon 1, or in both. Fifty-two of these genes had no CpG islands. One hundred thirty-four of them had CpG islands in their promoters and/or exons 1 (P < 0.001). Transcription factors (18 of 22; P < 0.005) and adhesion molecules (10 of 11; P < 0.01) showed more CpG islands in their promoters and/or exons 1. We chose 9 genes that were up-regulated >5-fold, and we measured their expression by flow cytometry in the presence and absence of 5-azaC.

Figure 5.

Protein expression as determined by flow cytometric analysis of gene products up-regulated in a hypomethylating milieu in OASFs and RASFs. AC, Confirmation of the data obtained by Affymetrix cDNA arrays for normal SFs continuously treated with 5-azacytidine (5-azaC), inducing genomic hypomethylation. Data are presented as box plots of mean fluorescence intensity, where the boxes represent the 25th to 75th percentiles, the lines within the boxes represent the median, and the lines outside the boxes represent the 10th and 90th percentiles. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus OASFs, by Mann-Whitney U test (n = 6 samples per group). + = P < 0.05; ++ = P < 0.01, versus untreated control, by Wilcoxon signed rank test. D, Mean and SD percentages of increase from baseline in gene expression in OASFs and RASFs. Significant differences in gene expression (n = 6 patients of each diagnosis) between OASFs and RASFs in reaction to 5-azaC treatment were observed for CD10, CD36, and CD46. MMP-14 = matrix metalloproteinase 14; TGFβR2 = transforming growth factor β receptor type II; CTK = cathepsin K (see Figure 2 for other definitions).

Up-regulation of genes with CpG islands in their promoters.

To confirm the findings on the 5-azaC expression microarray, OASF and RASF cultures (n = 6 samples each) were treated for 2 weeks with a nontoxic dose of 5-azaC, and the protein expression was analyzed by FACS. Examples of genes with CpG islands in their promoters were CD10, CD29, and CD130. These gene products were significantly increased after 2 weeks of treatment with 5-azaC (Figure 5A). Moreover, the expression of these proteins was significantly greater on the cell surface of untreated RASFs than on that of untreated OASFs.

Up-regulation of genes with CpG islands in exon 1.

Several genes that showed a >5-fold up-regulation of messenger RNA (mRNA) had no clear CpG island in their promoters. However, some of them had a CpG island in exon 1. Examples were CD26, MMP-14, and TGFβRII (Figure 5B). The baseline levels of CD26 and MMP-14 mRNA were the same in OASFs and RASFs. In the presence of 5-azaC, however, CD26 and MMP-14 mRNA levels increased significantly over baseline in both OASFs and RASFs. TGFβRII had a CpG island in exon 1, and its promoter showed an accumulation of CpG. In OASFs, 5-azaC increased the levels of expression significantly. However, in RASFs, TGFβRII expression on the cell surface was maximal in the absence and presence of 5-azaC.

Up-regulation of genes without CpG islands in their promoters.

Several genes without CpG or increased frequency of CpG were up-regulated >5-fold in the presence of 5-azaC, including CD36, CD46, and CTK (Figure 5C). The baseline expression of CD36 and CD46 on the cell surface was maximal in RASFs, and these values were significantly higher than those in OASFs. Both genes were up-regulated in OASFs by 5-azaC. The baseline levels of CTK were the same in OASFs and RASFs, but its expression was up-regulated upon treatment with 5-azaC.

Differences between OASFs and RASFs in the response to hypomethylation.

In OASFs, CD10, CD36, and CD46 showed significantly more increases in expression upon treatment with 5-azaC (Figure 5D). This was in large part because their expression was already maximal in RASFs.


Here we report that genomic hypomethylation developed in situ is conserved in RASFs in vitro even after >5 passages, and we confirm that the expression of L1 proteins in RASFs is associated with a partially hypomethylated L1 promoter. The degree of CG hypomethylation (78% methylation in RASFs versus 85% in OASFs) in the L1 promoter is similar to the degree of hypomethylation in tumor cells (9).

It is essential to understand more about the induction and maintenance of this global genomic hypomethylation. Proinflammatory cytokines such as TNFα, IL-1β, and IL-6 have multiple influences on the pathogenesis of RA. IL-1β (14) and IL-6 (15) can affect genomic methylation. TNFα, however, had not previously been associated with epigenetic changes. We observed that a physiologic dose of TNFα accelerated the cell cycle within 24 hours of exposure even more than IL-1β or PDGF. In OASFs, stimulation of cell proliferation was accompanied by increased DNA methylation. This was not the case in RASFs, and therefore the relative degree of DNA hypomethylation remained in RASFs treated with TNFα or IL-1β. As a consequence, it can be hypothesized that RASFs become progressively more hypomethylated during inflammation. This results in further activation of genes in RASFs.

Previous studies have shown that deficiency of Dnmt1 is associated with genomic hypomethylation (10, 16, 17). Dnmt1 interacts with PCNA at the DNA replication fork, and this system is responsible for the correct transmission of methylation marks to daughter cells. In RASFs, however, the expression of Dnmt1 appeared deficient, either in unstimulated cells or after exposure to proinflammatory cytokines. Thus, in RASFs, a relative deficiency of Dnmt1 during cell proliferation could result in the observed genomic hypomethylation.

Our work raises the question of whether the global genomic hypomethylation is accompanied, or followed by, specific promoter hypermethylation, since this is the case in various tumors (9). At least 1 example has been reported in the literature (i.e., by silencing death receptor 3 [18]), which could, at least in part, explain the relative resistance to apoptosis reported for RASFs in certain patients (2).

In SFs, the loss of methylation marks in daughter cells could cause an irreversible differentiation into an aggressive phenotype. Based on our observations, we hypothesized that normal SFs continuously treated with the Dnmt1 inhibitor 5-azaC will resemble RASFs. Indeed, a large number of gene transcripts (73 of 186, 39%) found to be up-regulated upon stimulation with 5-azaC and detected in cDNA microarrays were previously described to be involved in the pathogenesis of RA. It is known that DNA methylation silences genes with CpG island promoters. Therefore, from the list provided by the microarrays, we chose 3 genes that showed a >5-fold up-regulation of mRNA and the presence of CpG islands in their promoters, namely, CD10, CD29, and CD130. We confirmed that they were expressed on the surfaces of RASFs more than on the surfaces of OASFs and that their expression was increased within 2 weeks of treatment with a low dose of 5-azaC. It is well established that RASFs attach to cartilage through adhesion molecules, including CD29 and CD61 (β1 and β3 integrins) (7, 19). Invasion of RASFs into cartilage requires the availability of these 2 integrins (19, 20). CD10, a neutral endopeptidase, is highly expressed on RASFs (21) and presumably plays a critical role in the local regulation of peptide levels in the joint. IL-6 signaling involves both a specific IL-6 receptor (IL-6Rα) and a ubiquitous signal-transducing protein, CD130 (gp130), which is also used by oncostatin M. Both IL-6 and oncostatin M are involved in the pathogenesis of RA (8, 22, 23).

We also evaluated the expression of genes that have no CpG island in their promoters but that do have a CpG island in exon 1, namely, CD26, MMP-14, and TGFβRII. CD26 (dipeptidylpeptidase 4) was found to be highly expressed in RA synovial tissue (24) and in proliferating RASFs (25). The destruction of cartilage and bone in RA is in large part mediated by MMPs (20). MMP-14 has a central role because it cleaves other proMMPs and converts them into active forms. Inhibition of MMP-1 and/or MMP-14 results in a significant reduction of cartilage invasion by RASFs (26). Our results are in line with previously reported data in pancreatic cancer cells showing that 5-azaC also up-regulates the expression of MMP-14 and MMP-1 (27). Expression of TGFβRII was already maximal on RASFs, as proposed earlier (28).

Approximately one-fourth of genes (51 of 186, 27%) (see Supplementary Tables 1 and 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/) that were up-regulated upon treatment with 5-azaC in normal SFs contained no CpG island (e.g., CD36, CD46, CTK, caspase 1, and IL-1RI). For example, CD36, abundantly expressed in RA synovial tissue, binds proinflammatory oxidized low-density lipoproteins (29) and thrombospondin 1 (30). CD46 is a C3b binding protein, which could be involved in tissue damage in RA (31), whereas CTK, a key enzyme in bone resorption, is highly expressed in RA synovial tissue, not only by osteoclasts, but also by RASFs (32). Caspase 1 (IL-1β–converting enzyme), which activates IL-1β in RA synovial tissue (33), is also up-regulated in normal SFs by 5-azaC. IL-1β stimulates both the synthesis and the activity of MMPs involved in cartilage destruction (34).

Of the gene products that were up-regulated in normal SFs upon 5-azaC treatment, a particularly high proportion (52 of 73, 71%) are involved in intercellular processes and interactions with the extracellular matrix and have been described in RA synovial tissue and/or RASFs (see Supplementary Table 1, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/). These include 22 interleukins, growth factors, and their receptors, 13 extracellular matrix proteins and related enzymes, 10 matrix-degrading enzymes and their inhibitors, and 4 adhesion molecules. Most importantly, among them, cathepsins, MMPs, mannosidase α1, carbonic anhydrases, and ADAM-12 are involved in joint destruction, and lysyl oxidase has been shown to increase the crosslinking of mature collagen, an early step in cartilage destruction.

Many of the gene products also up-regulated in normal SFs upon 5-azaC treatment are involved in intracellular processes and play a role in RA (20 of 94, 21%) (see Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/). They include 4 protein kinases, 10 transcription factors, 2 proteins in the Wnt pathway, 2 proteins involved in the regulation of actin filaments and Rho signaling, and 2 regulators of apoptosis. Furthermore, transcription factors whose expression increased upon 5-azaC treatment (18 of 22, 82%) (see Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/) had CpG islands in their promoters and/or exon 1 more often than other genes revealed by the cDNA arrays. Many of them may play a role in RA, including Ets-related transcription factor, activating transcription factor 2 (ATF-2; which binds to activator protein 1), CCAAT/enhancer binding protein δ, nuclear factor of activated T cells 5, CREB/ATF, hypoxia-inducible factor 2α, and STAT-1. The sustained up-regulation of multiple signaling and transcription pathways in a hypomethylating milieu clearly could be responsible for the intrinsically activated phenotype of RASFs. The cDNA arrays identified proteins that are implicated in the normal or pathologic function of SFs, including 1 recently reported as a potential autoantigen in RA, human cartilage gp-39 (35).

CD10, CD36, and CD46 were also more up-regulated in OASFs than in RASFs upon 5-azaC treatment. The expression of all other genes tested was up-regulated to the same extent in OASFs and RASFs. Bisulfite sequencing of the CD10 CpG island, however, showed that it is hypomethylated even in normal SFs (data not shown). Therefore, the gene is regulated indirectly or by a methylation-independent mechanism. Other investigators have reported similar effects of 5-azaC on myeloid leukemia genes (36). CD46 and CD36 do not have a CpG island in their gene promoter; most likely they are regulated by indirect mechanisms. Other 5-azaC microarray studies have shown up-regulation of genes in the absence of CpG islands in their promoters (37). Thus, 5-azaC can apparently influence the expression of certain genes by different mechanisms. For example, it can affect histone modifications and up-regulate transcription factors, transcriptional repressors, and/or the expression of microRNA (38, 39). We have provided a list of transcription factors that could be candidates for future investigations (see Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/).

In summary, we report reduced 5-methylcytosine DNA in RA synovial tissue and in cultured RASFs. Specifically, the promoter of an L1 element was partially demethylated, confirming the global genomic hypomethylation in RASFs. Moreover, our observations suggested a progressive loss of methylation marks. It can be hypothesized 1) that the loss of methylation marks could be responsible for the intrinsically activated and aggressive phenotype of RASFs and 2) that tissue-specific transcription factors, which are not normally expressed in synovial tissue, are up-regulated in the disease and can be responsible for the activation of many genes involved in the pathogenesis of RA. Moreover, genomic hypomethylation could explain the increased expression of multiple receptors, adhesion molecules, and matrix-degrading enzymes, which play a role in RA and explain the enhanced response of RASFs to proinflammatory cytokines, leading all together to joint destruction. Thus, the epigenetic modifications of RASFs may be responsible, at least in part, for the fact that current therapies do not work in all patients and do not yet cure the disease.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Neidhart had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design.Karouzakis, R. E. Gay, Michel, S. Gay, Neidhart.

Acquisition of data.Karouzakis, Neidhart.

Analysis and interpretation of data.Karouzakis, S. Gay, Neidhart.


We thank Maria Commazzi for her technical assistance in immunohistochemistry and Dr. Caroline Ospelt for her assistance in microarray analysis.