Association between the abnormal expression of matrix-degrading enzymes by human osteoarthritic chondrocytes and demethylation of specific CpG sites in the promoter regions




To investigate whether the abnormal expression of matrix metalloproteinases (MMPs) 3, 9, and 13 and ADAMTS-4 by human osteoarthritic (OA) chondrocytes is associated with epigenetic “unsilencing.”


Cartilage was obtained from the femoral heads of 16 patients with OA and 10 control patients with femoral neck fracture. Chondrocytes with abnormal enzyme expression were immunolocalized. DNA was extracted, and the methylation status of the promoter regions of MMPs 3, 9, and 13 and ADAMTS-4 was analyzed with methylation-sensitive restriction enzymes, followed by polymerase chain reaction amplification.


Very few chondrocytes from control cartilage expressed the degrading enzymes, whereas all clonal chondrocytes from late-stage OA cartilage were immunopositive. The overall percentage of nonmethylated sites was increased in OA patients (48.6%) compared with controls (20.1%): 20% versus 4% for MMP-13, 81% versus 47% for MMP-9, 57% versus 30% for MMP-3, and 48% versus 0% for ADAMTS-4. Not all CpG sites were equally susceptible to loss of methylation. Some sites were uniformly methylated, whereas in others, methylation was generally absent. For each enzyme, there was 1 specific CpG site where the demethylation in OA patients was significantly higher than that in controls: at −110 for MMP-13, −36 for MMP-9, −635 for MMP-3, and −753 for ADAMTS-4.


This study provides the first evidence that altered synthesis of cartilage-degrading enzymes by late-stage OA chondrocytes may have resulted from epigenetic changes in the methylation status of CpG sites in the promoter regions of these enzymes. These changes, which are clonally transmitted to daughter cells, may contribute to the development of OA.

Osteoarthritis (OA) is characterized by the progressive failure of the extracellular cartilage matrix, which leads to the destruction of articular cartilage (1, 2). Primary OA is a late-onset complex disease with genetic, mechanical, and environmental components. Concordance of the disease in monozygotic twins is 40–60%, and the overall contribution of genetic factors is estimated to be ∼50% (3). Given that more than 60% of unrelated adults over the age of 60 years are affected, other factors must also play a role in the development of this disease. Age, obesity, abnormal joint loading, and sports injuries are all risk factors, but OA is more than just the result of “wear and tear” (4, 5). Articular chondrocytes are increasingly being suspected of playing major roles in the initiation and progression of the disease (1). This warrants closer examination of the cellular changes that occur in OA.

The main extracellular matrix components of articular cartilage are collagens (principally, types II, IX, and XI) and proteoglycans (mainly, aggrecan). The major enzymes that mediate the destructive processes are aggrecanases 1 and 2 (ADAMTS-4 and ADAMTS-5) and matrix metalloproteinases (MMPs) 2 (gelatinase A), 3 (stromelysin 1), 9 (gelatinase B), and 13 (collagenase 3) (6–10). One source of these proteinases may be the synovium, with the presumption that the enzymes diffuse into the cartilage matrix. Some MMPs tend to be elevated in the synovial fluid of OA patients (11, 12), although not as much as in the inflammatory synovium of rheumatoid arthritis patients (11). The second most important source of degradative enzymes is the chondrocytes of OA cartilage (13, 14). Evidence is accumulating to suggest that chondrocyte-derived degradative enzymes are important in the pathology of OA. For example, it has been shown that postnatal expression of MMP-13 in the articular chondrocytes of mice leads to an OA-like degradation of articular cartilage (15). In addition, deletion of ADAMTS-5 was shown to prevent cartilage degradation in a murine model of OA (16), although it is not yet known whether this also applies to OA in humans.

Based on advances in quantitative polymerase chain reaction (PCR) and complementary DNA (cDNA) microarray technology (17, 18), there is general agreement that 1) chondrocytes from severely degraded OA cartilage repress normal cartilage genes, such as aggrecan and type II collagen; 2) other genes that are not normally expressed by chondrocytes are activated (e.g., cytokines and MMPs); and 3) the changed chondrocytes undergo proliferation as well as increased apoptosis (1, 19–24). These changes in gene expression suggest that OA chondrocytes possess a changed or modulated phenotype (25), although the mechanistic basis of this change remains obscure.

We (26) and other investigators (14, 19) have observed and reported that the abnormal synthesis of MMPs does not occur in all OA chondrocytes, but is notably present in chondrocyte clusters. If, as is generally assumed, these clusters are clones of a single chondrocyte, then the daughter cells would appear to have inherited the changed pattern of gene expression from a single abnormal chondrocyte. Heritable changes in gene expression are either genetic (i.e., mutations) or epigenetic (i.e., changes in the regulation of gene expression without corresponding changes in the genetic sequence). We hypothesize that the expression of abnormal genes by clonal OA chondrocytes is due to epigenetic changes in the methylation status of the promoter region of the genomic DNA. In differentiated mammalian somatic cells, nonexpressed genes are silenced largely by DNA methylation in the promoter regions (27, 28), where the cytosine residues in the sequence 5′-CG-3′ (so-called CpG motifs) are methylated. Methyl-CpG binding proteins attach to this region (29), forming complexes with histone deacetylases and corepressors. This, in turn, leads to histone deacetylation and to changes in chromatin structure and prevents access of the transcriptional machinery to the promoter site (30).

The CpG methylation pattern within the genome of differentiated somatic cells is relatively stable and clonally inherited (31). However, pathologic alterations in the methylation status do occur in tumor cells (32, 33) and in other cells as they age (34). Many studies have linked hypermethylation of CpG island promoters of tumor suppressor genes to cancers (for review, see ref. 35), but few studies have investigated the linkage of hypomethylation with gene expression in disease.

Very few studies have examined the DNA methylation status of chondrocytic genes. An early study by Fernandez et al (36) compared the DNA methylation status of the promoter region of the proα1(II) collagen gene in chick embryo chondrocytes with that in chick fibroblasts and found less methylation in the chondrocytes than in the fibroblasts. Poschl et al (37) investigated whether the loss of aggrecan expression in OA was linked to methylation changes in the promoter, but were unable to demonstrate an association, concluding that DNA hypermethylation was unlikely to be involved in the silencing of this chondrocytic gene.

To our knowledge, no studies have investigated whether the abnormal gene expression in OA is due to epigenetic “unsilencing” (i.e., a demethylation of CpG sites). We hypothesized that the abnormal gene expression of clonal OA chondrocytes is associated with heritable epigenetic alterations in the DNA methylation pattern. In particular, we propose that demethylations have taken place in the promoter region of the genes that are expressed de novo in OA chondrocytes. To test the hypothesis, we selected MMPs 3, 9, and 13 as well as ADAMTS-4 as representative degradative enzymes. These enzymes are not normally synthesized by articular chondrocytes, but are produced by clonal OA chondrocytes, with protein synthesis on the whole correlating with messenger RNA (mRNA) expression (9, 14, 38, 39). We then compared the methylation status of the promoter regions of DNA obtained from patients with OA with that of DNA obtained from non-OA patients. We found that increased hypomethylation was present in the DNA promoter regions of enzymes that are abnormally expressed in OA.


Selection of materials.

Femoral heads were obtained at the time of joint replacement surgery for OA (16 patients) or following a fracture of the neck of the femur (10 patients). Tissues were obtained with the patients' consent and with the approval of the local ethics committee. Patients with femoral neck fracture generally have osteoporosis and, because of an inverse relationship between OA and osteoporosis (40), the articular cartilage from these patients can serve as controls (non-OA cartilage). In these patients, the fracture had typically occurred 1–3 days before surgery, and it was possible that the femoral head had undergone avascular necrosis as a consequence. The viability of the articular chondrocytes from these patients was therefore assessed by using live/dead markers, which confirmed that avascular necrosis had not seemed to have affected the chondrocytes, presumably because the cells were nourished by the synovial fluid. Histologic analysis also confirmed that the cartilage from patients with femoral neck fractures had near-normal morphology (26).

Additional non-OA cartilage samples were obtained, in collaboration with Prof. David Wilson and Dr. Neil Hanley (Human Genetics, Southampton, UK), from the cartilaginous femurs of 8–10-week-old fetuses following termination of pregnancies. Tissues were obtained according to guidelines issued by the Polkinghome Report (in the UK) and with approval of the local ethics committee. Tissues from 6 fetal femurs were combined to supply sufficient DNA for the analyses.

Histologic assessment.

Cutting with a scalpel to the subchondral bone provided slices of articular cartilage measuring ∼4 × 10 mm. To label live/dead cells, some samples were incubated for 2–3 hours prior to embedding with the fluorescent live/dead markers CellTracker green and ethidium homodimer 1 (25 mM; Molecular Probes, Eugene, OR). Samples were then fixed overnight in freshly prepared 4% paraformaldehyde and then processed into paraffin wax. Sections measuring 5–7 μm were stained with Safranin O to identify proteoglycans and with hematoxylin/Alcian blue/Sirius red to identify regions of aggrecan loss. We had previously shown (26) that Sirius red, which stains types I and II collagen, does not normally stain cartilage because proteoglycans present in the tissue prevent the staining. Only when the proteoglycans are lost, as in OA cartilage, will Sirius red have access to and stain the type II collagen in the matrix of articular cartilage.

For immunocytochemistry, the following antibodies were used: anti-human MMP-3 (MAB3312; Chemicon, Southampton, UK), anti-human MMP-9 (MAB3309; Chemicon), rabbit anti-human MMP-13 (AHP751; Serotec, Oxford, UK), and anti-human ADAMTS-4 (AHP821; Serotec). After inhibition of endogenous peroxidase with hydrogen peroxide, the sections were incubated overnight with the relevant antibodies. Binding of the primary antibodies was visualized with the aid of the appropriate biotinylated secondary antibody, followed by treatment with avidin–peroxidase and 3-amino-9-ethyl-carbazole. This yielded a brown reaction product. The sections were counterstained with 1% Alcian blue, viewed with a Zeiss Universal light microscope (Zeiss, Welwyn Garden City, UK), and images were captured with a digital camera.

Isolation of DNA.

To demonstrate that changes in the methylation status of genomic DNA had occurred in vivo, it was essential to select the sampling site with extreme care (see Results) and to isolate the DNA directly from human articular cartilage, rather than from isolated or cultured chondrocytes, since these procedures themselves might alter the methylation status. Thin slices of articular cartilage were cut with a scalpel, stored at −80°C, then ground in a liquid nitrogen–cooled freezer mill (Spex CertiPrep 6750; Spex, Stanmore, UK). DNA was extracted from the milled cartilage powder using Qiagen DNeasy Maxi kits for plants (Qiagen, Crawley, UK). This kit contained an extra step for the removal of proteins and polysaccharides, a step that also effectively removes the large amounts of proteoglycans that are present in cartilage.

Analysis of the promoter regions.

Using the published sequences of the 5′-flanking regions, which contained the promoters, the locations of the CpG sites were identified. To determine which methyl-sensitive restriction enzymes cut near the CpG sites, we used RestrictionMapper, an online restriction-mapping tool (available at The 53 commercially available enzymes, for which cleavage is blocked at all sites by methylation, were selected from the list of all restriction enzymes, and the specific enzymes and cleavage sites were identified by the online program.

Assessment of DNA methylation status.

Judicious selection of the sampling area (see Results) ensured that the sample contained predominantly chondrocytes of the altered phenotype. This meant, however, that only a small proportion of the available articular cartilage could be sampled. Typically, ∼1–2 μg of DNA per OA patient and ∼2–5 μg of DNA per control patient were extracted. There are 2 methods of analyzing methylation status: bisulfite modification (41) and cleavage with specific methylation-sensitive restriction enzymes (42, 43). The bisulfite modification method is essential for examination of methylation in CpG islands, because information can be obtained on the methylation status of all CpG sites within the region of interest. However, the method is laborious and prone to a number of reaction artifacts (44). In addition, the modified DNA must be cloned or sequenced before data can be obtained. The second method provides a sensitive scanning technique that is suitable for detecting methylation if only small amounts of DNA are available and if the promoters contain sparse CpG sites. The method utilizes the methylation sensitivity of restriction enzymes: If the cytosines are methylated, PCR amplification produces a band equivalent to that of untreated control samples. Conversely, if restriction enzyme cleavage at unmethylated sites induces DNA strand breaks, no band will be detected (45).

Since the principal aim of the present studies was to provide “proof-of-concept” data on whether abnormal synthesis of degrading enzymes was associated with loss of methylation (rather than a complete mapping of the methylation status of all CpG sites), the simpler methylation-sensitive restriction enzyme method was used. Appropriate enzymes were identified, and primers were designed to bracket the regions containing CpG sites (detailed in Table 1). The following enzymes were used: Aci I, Ava I, Bst BI, Hha I, Hpy 99I, Hpy CH4IV (all from New England Biolabs, Beverly, MA), and Hpa II (Promega, Southampton, UK).

Table 1. Primer pairs and enzymes used for the promoter regions of matrix metalloproteinases (MMPs) 9, 3, and 13 and ADAMTS-4
PromoterPrimer namePrimer sequenceLocation relative to transcription start siteLength of segmentSegment cut by
Distal MMP-9M9a forward5′-CCCGAGGTCCTGAAGGAAGAG-3′−714 to −694356 bpAci I at −624 and −562
 M9a reverse5′-GCCAAGGGAAAGTGATGGAAG-3′−379 to −359  
Distal MMP-9M9d forward5′-ACCTCGGCCTCCCAAAGTGCTAAG-3′−962 to −939406 bpAva I at −712
 M9d reverse5′-GGGGCGGAAGGAATGGGCTCTGCTA-3′−581 to −557  
Proximal MMP-9M9b forward5′-TGGGCCAGGGGGATCATTAGTTTCAG-3′−421 to −396486 bpAci I at −185, −223, and −233; Hha I at −36
 M9b reverse5′-CAGCCCAGCACCAGGAGCACCAGGAC-3′+40 to +65  
Far distal MMP-3M3b forward5′-AAAATAGAGTAGCAGAGGCAGGTA-3′−2,015 to −1,992381 bpHpy 99I at −1,923
 M3b reverse5′-AGAGTGGTGGCAGTGATGTGAA-3′−1,656 to −1,635  
Distal MMP-3M3a forward5′-GGTGGCAGGTGGCAAGAG-3′−1,026 to −1,009448 bpHpa II at −686; Hpy CH4IV at −635
 M3a reverse5′-ATGGGCAGAATAGAACAAAGAGG-3′−599 to −577  
Proximal MMP-3M3c forward5′-AAAAATAAATCACCCCTTCTAAAT-3′−507 to −484528 bpAci I at −12; Hha I at +27; Ava I at +33
 M3c reverse5′-CTGTCTTGCCTGCCTCCTT-3′+3 to +21  
Distal MMP-13M13a forward5′-CAACCATGGGGCTCAATCCT-3′−564 to −543460 bpAci I at −544; Bst BI at −343; Hpy CH4IV at −323
 M13a reverse5′-CTTACGTGGCGACTTTTTCTTTTC-3′−127 to −105  
Proximal MMP-13M13b forward M13b reverse5′-GCCGTTTATTTTGCCAGATGGGTTTTGA-3′ 5′-CCCGCGAGATTTGTAGGATGGTAGTATGAT-3′−228 to −201 +173 to +202430 bpAva I at −136; Hpy CH4IV at −110; Aci I at −14; Hha I at +70
Distal ADAMTS-4ADAMa forward5′-GCCCCAGGGAGAATTAAAAAGAAA-3′−1,135 to −1,112547 bpAci I at −829, −722, and −672; Hpy CH4IV at −753
 ADAMa reverse5′-TGGGGGAGGCAAGGAGACAGT-3′−609 to −589  
ADAMTS-4ADAMb forward5′-TGCCTCCCCCAATGTTCTGC-3′−599 to −580545 bpHpa II at −323
 ADAMb reverse5′-GGTGAGCCTCTGCGATTGATTGG-3′−77 to −55  
Proximal ADAMTS-4ADAMc forward5′-CCTCTCCTGGGGCTCTGTCC-3′−294 to −275400 bpHha I at −37; Aci I at −35, +57, and +90
 ADAMc reverse5′-ACTCTGTCTGGGCCGCTTCTGTG-3′+84 to +106  

To optimize the conditions of enzymatic digestion and to ensure completeness of digestion, a source of unmethylated DNA was required. We generated PCR products from all primers and digested the purified cDNA (using a PCR purification kit from Qiagen) with the relevant enzymes, followed by agarose electrophoresis. From this, we established that digestion was complete when 0.1 μg of DNA was incubated for 15 hours at 37°C (65°C for Bst BI) in a 10-μl reaction mixture containing 4 enzyme units in the appropriate buffer (with bovine serum albumin, if required). Heat inactivation was performed at 65°C for all enzymes, except Ava I, which was performed at 80°C. Bst BI–digested DNA was maintained at 4°C, a temperature at which the enzyme was inactive.

The promoter regions of digested and undigested DNA samples were amplified by hot-start PCR, using Platinum Supermix (Invitrogen, Paisley, UK). Following activation of the Taq polymerase (2 minutes at 94°C), 35 cycles were run (94°C for 30 seconds, 55°C for 30 seconds, and 72°C for 1 minute), followed by a final extension at 72°C for 5 minutes. To verify that overnight incubation and heat activation per se did not damage the DNA, samples were also incubated without the relevant enzymes. No loss of PCR amplification was observed in these no-enzyme controls. In addition, we included a sample of universally methylated DNA (Chemicon) for each primer/enzyme combination as a positive control.

Statistical analysis.

The incidence of demethylation was compared for each individual site between non-OA and OA samples using Fisher's exact test. In addition, the overall percentages of demethylation were calculated for each patient, and the differences between the means of OA and non-OA samples were compared using Student's t-test, where the populations showed a normal distribution; otherwise, the nonparametric Mann-Whitney U test was used.


Localization of enzyme-producing chondrocytes.

Immunocytochemical analysis showed that the vast majority of articular chondrocytes from patients with femoral neck fracture (Mankin score 1–2) did not synthesize MMP-3 (Figures 1A and C), nor did they synthesize MMP-9, MMP-13, or ADAMTS-4 (results not shown). An exception was the superficial zone, where some flattened cells were immunopositive for the enzymes (shown for MMP-3 in Figure 1B and for MMP-9 in Figure 1G). Occasionally, isolated cells in the intermediate and deep zones were also immunopositive, especially for MMP-3.

Figure 1.

Localizations of matrix-degrading enzymes in human articular cartilage. AC, In cartilage from patients with femoral neck fracture (Mankin score 0–2), only a few cells in the superficial layer are immunopositive for matrix metalloproteinase 3 (MMP-3) (B), whereas the bulk of chondrocytes are still negative (C). These findings also apply to the other enzymes examined. DF, In cartilage from patients with early-stage osteoarthritis (OA) (Mankin score 4–6), the flattened superficial cells have disappeared, and the enzyme-positive chondrocytes are present in the upper approximately one-fourth of the cartilage (illustrated for MMP-13). GK, Increase in the number of MMP-9–positive chondrocytes with increasing Mankin score. In cartilage with a Mankin score of 1, only a few single cells are immunopositive (G). These are present in doublets in cartilage with a Mankin score of 4 (H) and in quadruplets in cartilage with a Mankin score of 7 (J). In severe OA (Mankin score 14), all cells of the typical clones are immunopositive for MMP-9 (K). LP, All cells of clones are also immunopositive for MMP-3 (L), MMP-13 (M), and ADAMTS-4 (N). In addition, the clonal cells are highly metabolically active, as indicated by the intense fluorescence after incubation with CellTracker green (P). Bars in A and D = 100 μm; bars in all other images = 10 μm.

In cartilage with low-grade OA (Mankin score 4–6), which was obtained from the non–weight-bearing regions of the femoral heads of patients with OA, the overall thickness of the cartilage was still similar to that of cartilage from patients with femoral neck fracture, but the flattened cells were absent (Figures 1D and E). This suggested that the very superficial zone had been degraded. The region in which enzyme-positive chondrocytes (now rounded cells) were found had increased to approximately one-fourth of the total cartilage thickness (shown for MMP-13 in Figures 1D and E), but chondrocytes in the remaining three-fourths of the cartilage remained negative for MMP-13 (Figure 1F). Again, the same localization was found for chondrocytes positive for MMP-3, MMP-9, and ADAMTS-4. With increasing Mankin scores, the thickness of the cartilage decreased as the cartilage was degraded. At the same time, the depth increased at which enzyme-producing chondrocytes were found. These chondrocytes, which were present as single cells in cartilage with a low Mankin score (Figure 1G), divided to produce 2 cells (Figure 1H), then 4 cells (Figure 1J), and finally gave rise to the typical clones of OA cartilage, where all cells were immunopositive for all 4 enzymes (Figures 1K–N). It should be noted that the location of cells producing MMPs 3, 9, and 13 and ADAMTS-4 always overlapped (shown for the clonal cells in Figures 1K–N). In other words, the same cell seemed to produce all 4 enzymes.

Live/dead labeling demonstrated intense green fluorescence in clonal chondrocytes (Figure 1P). This contrasted with a barely visible green fluorescence of nonclonal chondrocytes (results not shown). Because the green fluorescence is due to metabolic activity, these observations suggest that metabolic activity was greatly increased in the clonal, enzyme-expressing chondrocytes as compared with the nonclonal chondrocytes. To distinguish between as-yet-unchanged chondrocytes and chondrocytes that produced degrading enzymes, we refer to the latter as “degradative” chondrocytes.

Selection of sampling areas for DNA analyses.

The preceding findings made it evident that the sampling site is critical for comparison of the methylation status of DNA from normal articular chondrocytes with that of DNA from degradative chondrocytes. From the histology results, it was apparent that a reasonably homogeneous population of typical degradative chondrocytes was found only in the thin cartilage of late OA or in the superficial zones of early OA, close to the weight-bearing area. Hence, only the upper cartilage regions were sampled for DNA extraction. In non-OA patients, the superficial zone often contained some cells that expressed matrix-degrading enzymes. Therefore, in cartilage from these patients, the superficial zone was removed before the cartilage was sampled for DNA extraction.

Features of the promoter regions.

The promoter regions of the 4 degradative enzymes are illustrated schematically in Figure 2. Notably, all 4 promoter regions can be classified as “sparse” CpG promoters, since none contain a concentration of CpG sites that would constitute a CpG “island.” For MMP-3 (GenBank accession no. U43511 [46]), the total number of CpG sites that are present in the 2,000-bp region upstream of the transcription start site is just 7. For MMP-9 (GenBank accession no. D10051 [47]), MMP-13 (GenBank accession no. U52692 [48]), and ADAMTS-4 (GenBank accession no. AY044847), the numbers of CpG sites are 10 in 1,000 bp, 6 in 600 bp, and 13 in 900 bp, respectively. Methylation-sensitive restriction enzymes were commercially available for all CpG sites, except for 2 sites in MMP-3 and MMP-9 and 6 sites in ADAMTS-4. Figure 2 shows where the enzymes cut and the primer pairs selected. In most cases, a particular primer/enzyme combination identified a unique CpG site, but for MMP-9 and ADAMTS-4, the Aci I sites were too close together, so that 2–3 Aci I sites were assessed simultaneously. Details of the primer sequences are given in Table 1.

Figure 2.

Diagrammatic representation of the 5′-flanking regions, which contain the promoters, of the 4 degradative enzymes examined. CpG sites are indicated by short vertical lines, and the methylation-sensitive restriction enzymes are mapped. Where no enzyme is assigned to a CpG site, none was commercially available. The locations of the polymerase chain reaction primers are also indicated. Transcription start sites are indicated by the boldfaced arrows. MMP = matrix metalloproteinase.

Completeness of enzyme cutting and absence of nonspecific DNA degradation.

The enzyme data sheets state that 1 unit of enzyme activity in a 50-μl assay buffer will cleave 1 μg of λ DNA in 1 hour. When we tested 2 units of enzyme in a 10-μl assay buffer containing ∼0.1 μg of DNA, digestion was not complete, because a weak band remained at the site of the original product in some cases. We therefore doubled the enzyme concentration and obtained complete digestion. Figure 3A illustrates typical results for each enzyme and shows that digestion was complete in all cases, because bands of the expected sizes for the breakdown products appeared instead of bands for the no-enzyme controls.

Figure 3.

Digestion of unmethylated and methylated DNA. A, Verification of complete digestion of unmethylated DNA. To demonstrate completeness of digestion, polymerase chain reaction (PCR) products were combined from 6 PCR analyses. After purification, the products (which are unmethylated) were digested with the relevant enzymes. Undigested (C) and digested products were electrophoresed in parallel. The bands obtained after digestion corresponded to the expected size of the breakdown products. In all cases, digestion had reached completion, as indicated by the absence of a band at the level of the corresponding control. B, To check for the absence of nonspecific digestion of methylated DNA, universally methylated DNA was digested by the relevant enzymes and amplified by PCR with the indicated primers. The labeling is the same as in A. The ladders show 200-bp intervals.

Overnight digestion in the absence of the enzyme and heat inactivation had no effect on the intensity of the appropriate PCR band (results not shown). Similarly, when universally methylated DNA was digested with the enzymes, followed by PCR amplification, the bands obtained were equivalent to those for the untreated samples (Figure 3B). This confirmed that nonspecific DNA degradation did not take place under the experimental conditions used.

Methylation status of the promoter regions of degradative enzymes.

The absence of PCR amplification indicated the absence of methylation. Care was taken to avoid false-negative results, since a PCR band will be absent if the sample contains insufficient DNA. Hence, the relevant no-enzyme control sample was always amplified with the same primers in parallel with the digested samples, and both were subjected to electrophoresis in adjacent lanes. This provided a positive control for DNA loading and for the PCR reaction. Absence of the band in enzyme-digested samples was registered only if a band for the no-enzyme control sample was clearly visible. Because of our stringent selection criteria for OA cartilage, we did not have sufficient DNA to produce a band in the no-enzyme control samples for about one-third of the OA samples. However, we were still able to obtain results if 1 μl of the PCR product was used for a second PCR reaction in both control and enzyme-digested samples.

It was more difficult to avoid false-positive results because the PCR reaction is known to amplify even low concentrations of product. For example, if the OA sample contained, say, 95% degradative chondrocytes and 5% non–enzyme-producing chondrocytes, the latter cells might produce a faint PCR band. Again, we used comparison of the intensity with the PCR band from the corresponding no-enzyme control as our criterion. Very faint bands were registered as the absence of methylation, but bands with >50% of the intensity of the no-enzyme controls were registered as the presence of methylation. Consequently, we may have underestimated the amount of demethylation in some samples.

Figure 4 shows typical results for all 4 enzymes. As expected, fetal DNA was generally methylated, with just 3 exceptions: the Ava I site of MMP-9, the Hpy 99I site of MMP-3, and the Hpy CH4IV site of ADAMTS-4. All 3 sites are at some distance from the transcription start site (−712, −1,923, and −753, respectively). Similarly, the DNA from patients with a femoral neck fracture was generally methylated, again with some exceptions: the same Ava I site of MMP-9, as well as the Aci I sites at −624 and/or −562. However, the largest loss of methylation was present in the DNA from OA patients. For MMP-13, 5 CpG sites remained fully methylated, but loss of methylation was detected at the Ava I site at −134 and the Hpy CH4IV site at −110. At the latter site, a faint, but extremely weak, band was just visible as compared with the corresponding no-enzyme control sample. However, because PCR can amplify even very small quantities of methylated DNA, which may be present if the sample contains a few nondegradative chondrocytes, the presence of a faint band is still consistent with the absence of methylation in the majority of chondrocytes. For MMP-9, all sites showed the absence of methylation, whereas for MMP-3, the Aci I site at −12 was still methylated. All other CpG sites were demethylated. All enzyme digestions abolished the PCR bands in the ADAMTS-4 promoter, although one cannot attribute the Aci I digestions to a single site because the enzyme potentially cut at 3 sites.

Figure 4.

Representative results of polymerase chain reaction (PCR) amplifications of the promoter regions with and without digestion by methylation-sensitive restriction enzymes, as indicated. A typical pattern of reactions is shown for DNA extracted from fetal cartilage, from the deep zone of cartilage from a patient with femoral neck fracture (# NOF), and from the surface zone around the weight-bearing area of cartilage from a patient with osteoarthritis (OA). The ladders show 200-bp intervals. MMP-13 = matrix metalloproteinase 13; Hpy4 = Hpy CH4IV; Hpy99 = Hpy 99I.

While Figure 4 shows the typical results, there was considerable variation between patients. This is illustrated in Table 2, which lists the methylation status for each patient. The absence of a band or the presence of a faint band (provided that the band for the corresponding no-enzyme control was strong) is shown as the absence of methylation (or, −), whereas the presence of a band of >50% intensity compared with the no-enzyme control is shown as the presence of methylation (or, +). For 3 patients, we also determined the methylation status of the deep zone of OA samples taken from the thick cartilage of the non–weight-bearing region, which was usually avoided because it did not contain degradative chondrocytes. In these samples, the same pattern of methylation was observed as for the non-OA control samples from patients with femoral neck fracture, confirming that the absence of expression correlated with promoter methylation (Table 2).

Table 2. Presence and absence of methylation at specific CpG sites for universally methylated DNA as well as DNA derived from fetal cartilage, the deep zone of articular cartilage from the femoral heads of patients with femoral neck fracture, and the deep and superficial zones of articular cartilage from the femoral heads of OA patients*
Sample, sex/ageMMP-13MMP-9MMP-3ADAMTS-4
−544−343−323−136−110−14+70−712−562 and −624−185, −223, and −233−36−1,923−686−635−12+27+33−829, −722, and −672−753−323−37−35, +57, and +90
  • *

    Cartilaginous femurs from 6 fetuses (8–10 weeks old) were combined to supply sufficient DNA for the analyses. OA = osteoarthritis; MMP-13 = matrix metalloproteinase 13; + = methylation;– = absence of methylation; ND = not determined; NS = not significant.

  • P < 0.005.

  • P value approached significance (just over 0.05).

  • §

    P < 0.001.

Methylated DNA++++++++++++++++++++++
Fetal cartilage+++++++++++++++++++
Non-OA cartilage deep zone                      
OA cartilage deep zone                      
OA cartilage superficial zone                      
% of cartilage unmethylated                      
 Femoral neck fracture00030000896730204001020506000000

Based on the data shown in Table 2, the following conclusions may be drawn. First, the overall degree of demethylation varied considerably between the 4 degradative enzymes. MMP-13 is the most heavily methylated, whereas MMP-9 shows considerable hypomethylation even in femoral neck fracture samples.

Second, not all CpG sites were equally susceptible to demethylation: Some CpG sites, such as those at −544, −343, and +70 in the MMP-13 promoter, remained uniformly methylated in all samples. In other CpG sites, such as those at −323 and −14 of MMP-13 and at −12 of MMP-3, loss of methylation was observed in only a small number of samples. Other CpG sites, such as those at −712 of MMP-9 and +33 in MMP-3, showed a high level of demethylation in both femoral neck fracture patients and OA patients. In other sites, there was an increase in demethylation, which failed to reach significance, with P values just over 0.05. Nevertheless, for each enzyme, there was 1 crucial CpG site, where a statistically significantly higher percentage of demethylation was found in OA patients as compared with femoral neck fracture patients (P < 0.005). These sites were −110 for MMP-13, −36 for MMP-9, −635 for MMP-3, and −753 for ADAMTS-4.

Third, it is also of interest to record the percentage of demethylation for individual patients (Table 3). For each enzyme, the overall percentage of demethylation was increased in OA samples compared with controls: 20.2% versus 4.2% for MMP-13, 81.2% versus 47.5% for MMP-9, 59.3% versus 30.1% for MMP-3, and 50% versus 0% for ADAMTS-4. All increases were highly statistically significant.

Table 3. Demethylated CpG sites, expressed as a percentage of all CpG sites examined for each patient*
Sample, sex/ageMMP-13MMP-9MMP-3ADAMTS-4Overall %
  • *

    Using the data presented in Table 2, the overall methylation status was calculated for each patient, each enzyme, and across all enzymes. Differences between patients with femoral neck fracture and those with osteoarthritis were compared using the Mann-Whitney U test, which showed a statistically significant difference for all comparisons. MMP-13 = matrix metalloproteinase 13; ND = not determined.

  • The data for ADAMTS-4 could not be analyzed statistically because the standard deviation was 0.

Femoral neck fracture     
 Mean ± SEM4.2 ± 2.147.5 ± 10.830.1 ± 6.7020.1 ± 3.6
 95% confidence interval0–923–6915–45012–28
 Mean ± SEM20.2 ± 4.181.2 ± 5.859.3 ± 6.250 ± 6.548.6 ± 3.0
 95% confidence interval11–2972–9346–7242–53
P (femoral neck fracture group vs. osteoarthritis group)0.0120.0150.0076<0.0001

Fourth, for individual patients, the overall percentage of demethylation also varied. In femoral neck fracture patients, the mean was 20.1% (95% confidence interval 12–28), whereas in OA patients, the mean was 48.6% (95% confidence interval 42–53) (P < 0.0001). The degree of demethylation was unrelated to the age or sex of the patient. One could speculate that the amount of demethylation was related to the stage of OA, i.e., the time since disease onset and/or the rate of progression. However, this information is not available.


The difficulties in identifying susceptibility genes for OA (3, 49, 50) led us to consider the possibility that epigenetic changes, possibly in combination with genetic factors, also play a major role in the etiology of OA. To investigate this possibility, we sought to determine whether epigenetic changes in DNA methylation occur in genes that are abnormally expressed by OA chondrocytes. The present study provides the first direct evidence of an association between loss of DNA methylation in the promoters and abnormal expression of MMP-3, MMP-9, MMP-13, and ADAMTS-4 by OA chondrocytes. The enzyme-expressing cells, which we have called “degradative” chondrocytes, were located within regions of proteoglycan loss, either in the superficial regions of low-grade OA cartilage or throughout the thin fibrillated cartilage of late-stage OA, especially in the typical chondrocyte clusters. In contrast, chondrocytes from the intermediate and deep zones of non–weight-bearing regions of cartilage from OA patients were normal, at least as far as the absence of expression of the degradative enzymes was concerned. OA is a progressive disease with focal erosion of cartilage, starting at weight-bearing areas, but progressing to peripheral regions. The location of degradative chondrocytes seemed to precede matrix degradation, consistent with a role of these enzymes in the cartilage erosion.

Our findings confirm those of earlier studies on immunocytochemical localization of the MMPs in OA cartilage (13, 19, 51–53). The present study is the first, however, to have localized ADAMTS-4 to clonal OA chondrocytes.

Available data on mRNA expression are, on the whole, consistent with the findings of our immunolocalization studies. Using microarray or quantitative reverse transcription–PCR (RT-PCR), increased expression of MMP-9, MMP-13, and ADAMTS-4 has been found in OA cartilage compared with normal cartilage, although expression levels were low (17, 18). Since in those studies, RNA was extracted from the whole OA cartilage, rather than only the superficial zones, the low levels of expression are not unexpected. The data for mRNA expression of MMP-3 are more contradictory. Using Northern blotting, Nguyen et al (54) found abundant expression of MMP-3 mRNA in “normal” cartilage and loss of message in late-stage OA cartilage. By in situ hybridization, the message in “normal” cartilage was localized to chondrocytes of the superficial zone, similar to our immunolocalization in low-grade OA. More recently, using quantitative RT-PCR of knee cartilage from normal subjects and patients with late-stage OA (defined as those requiring knee replacement), Bau et al (17) found a marked decrease in MMP-3 expression levels in late-stage OA. When in situ hybridization was used to localize mRNA in OA cartilage, MMP-9 and MMP-3 were preferentially expressed in clustered chondrocytes, with little expression in single cells (14, 38, 55), findings that are consistent with the present immunocytochemical observations. These different results may be related to the type of cartilage selected for study (femoral head versus tibial plateau), and clearly, further work needs to be done on MMP-3 to resolve the discrepancies between findings of immunocytochemical/in situ hybridization analyses and the RT-PCR data.

It is noteworthy that abnormal enzyme expression seemed to start in a few cells in the superficial zone, which was found in both OA and non-OA cartilage. However, subsequent cell divisions and transmission of abnormal enzyme expression to daughter cells was found only in OA cartilage. What might underlie this heritable abnormal gene expression? We hypothesized that demethylation or loss of methylation of certain CpG sites in the promoter regions may “unsilence” genes that are normally silenced by methylation. The promoters of all 4 enzymes examined contain relatively few CpG sites, a fact that may favor a pathologic loss of methylation. This contrasts with the situation of “island” promoters, which contain CpG motifs at high frequency, and where an association between hypermethylation and gene silencing is well established (56). However, we know of no examples of demethylation of CpG island promoters that correlate with unsilencing.

Examination of methylation status in promoters with sparse CpG sites has been neglected, largely because it was thought that methylation of many CpG sites was needed to repress gene transcription. However, silencing of genes with sparse promoters is feasible. Although the methyl-binding protein MeCP1 (MBD4) requires multiple, closely spaced CpG sites, another protein, MeCP2, can bind to single CpG sites, where it promotes chromatin condensation into an inactive conformation, thereby affecting gene expression at a distance from the methylated region (57). However, if a sparse CpG promoter also contains a strong enhancer element, gene transcription may take place even though the CpG sites are methylated (56). At present, it is not known whether the promoters of the degradative enzymes contain strong enhancer elements.

Initially, we had expected that all promoter CpG sites would be methylated in control patients with femoral neck fracture and all would be unmethylated in OA patients. This was not the case. Instead, sites differed in their susceptibility to demethylation, ranging from a CpG site with no methylation in both control and OA samples (at the −712 site in the MMP-9 promoter) to sites with ubiquitous methylation in all patients (e.g., 3 of the CpG sites in the MMP-13 promoter). Nevertheless, the overall percentage of nonmethylated CpG sites had more than doubled in the degradative chondrocytes from OA patients. In addition, it proved possible to identify 1 CpG site for each enzyme where the most frequent loss of methylation had occurred in samples from OA patients. These sites may well be crucial for the epigenetic regulation of gene transcription, along with overall levels of methylated CpG sites.

It is important to note that a causal relationship between loss of methylation and abnormal gene expression can, as yet, only be postulated. Other studies, however, have shown that promoter methylation is important in the regulation of MMP-9 expression (58, 59). In mouse lymphoma cell lines, constitutive MMP-9 expression was found to vary inversely with levels of methylation. Moreover, as reported here, specific sites varied in their degree of methylation. Interestingly, when the mouse lymphoma cells, which did not express MMP-9, were demethylated with the aid of 5-aza-deoxycytidine, MMP-9 expression was induced (58). Similarly, the expression of several MMPs in pancreatic adenocarcinoma cells was found to be associated with DNA demethylations (60). The CpG sites in the promoters of MMPs 2, 7, or 9 in cells that did not express these enzymes were partially or completely methylated. Treatment of the cells with 5-aza-deoxycytidine caused several MMPs, including MMP-3 and MMP-9, to be expressed (60). The findings of these studies give support to the concept that the methylation status of the promoter regions is directly related to MMP expression. No reports, however, have been published so far that relate demethylation to the expression of ADAMTS-4.

The present studies suggest that complete demethylation of every CpG site in a promoter is not needed to activate gene expression, inasmuch as the methylation status of a single CpG site may be sufficient for altered gene expression. In the X chromosome–linked phosphoglycerate kinase 1 gene, an Hpa II site at −21 bp was the only Hpa II site in the promoter whose methylation pattern was found to correlate fully with altered expression (43). Similarly, 5-azacytidine–induced transcriptional activation of the Epstein-Barr virus latency C promoter was found to be the result of demethylation at a single CpG site (61). In the hepatitis B virus core gene, methylation of a single Hpa II site at −280 bp regulated the expression of the core gene (62). In the p53 promoter, a single-site methylation at −450 bp reduced gene expression in reporter gene constructs (42). Taken together, these findings confirm that a change in the methylation status of a single specific CpG may be sufficient for the activation of gene transcription.

Loss of DNA methylation may favor gene expression, but only in the presence of the appropriate transcription factors that activate the promoters. MMPs 3, 9, and 13 as well as ADAMTS-4 are induced in cultured chondrocytes by interleukin-1β (IL-1β) (19, 20, 39, 63), either via the MAP kinase or the NF-κB pathways. Inflammatory cytokines, such as IL-1β, are produced in OA by activated synoviocytes, mononuclear cells, and OA chondrocytes (20). These observations suggest that transcription factors are available to induce transcription once CpG sites have been demethylated.

With increasing age, the degree of methylation may increase or decrease, with the actual change depending on the tissue and the gene in question. For example, island promoters may become hypermethylated, while other genes/tissues may at the same time lose methylation (34). Aging may contribute to the loss of methylation in OA in a stochastic manner, but is unlikely to be the only cause of site-specific loss of methylation in OA, since the femoral neck fracture patients were generally older than the OA patients, yet showed less loss of methylation. The mechanisms of pathologic demethylations are as yet not known. It is possible to directly demethylate double-stranded DNA in vitro (64). However, this reaction necessitates disruption of the C[BOND]C and O[BOND]H bonds and therefore requires large activation energy. In the pathologic situation, demethylations may be replication-coupled. In that case, interference with methyltransferase activity would result in hemi-methylated and unmethylated DNA (65). Experimental demethylations induced by 5-aza-deoxycytidine result from inhibiting the methyltransferases during cell division (66). However, normal aged articular chondrocytes no longer undergo division, even though cell division is stimulated in the degradative chondrocytes. This process ultimately forms the clones that are typical of severe OA. The possibility that division of cells in the superficial zone is reinitiated by some unknown stimulatory factor prior to the appearance of the first degradative chondrocytes cannot be excluded.

The proposal that OA may be causally related to epigenetic changes is new, but such changes have been described in other diseases, especially in cancers, where DNA hypomethylation has played a role in the reexpression of previously silenced genes (67). In hepatoblastoma, for example, activation of the insulin-like growth factor 2 gene has been correlated with demethylation of the P3 promoter (68). Demethylation has also contributed to the aberrant expression of BCSG1, the breast cancer–specific gene (69). In leukemia, hypomethylation of the promoter has been shown to accompany the reactivation of the homeobox HOX11 protooncogene (70). It has been suggested (71, 72) that epigenetic misregulation in combination with genetic variations in DNA sequence gives rise to the pathology of other complex, non-Mendelian diseases, such as Alzheimer's disease, Parkinsonism, multiple sclerosis, or hyperparathyroidism. Atherosclerosis has also been linked to methylation changes (73).

Primary OA is a late-onset, complex disease. Based on the evidence from the present study, epigenetic changes may play an important role in the pathology of OA. If so, this would constitute good news for OA patients, because DNA methylation, although heritable, is potentially reversible. Reversibility, in turn, could lead to novel avenues for therapeutic intervention, especially in early disease.


Our thanks go to the orthopedic surgeons of Southampton General Hospital for supplying us with the femoral heads obtained during joint replacement surgery. We are also grateful to Prof. David Wilson and Dr. Neil Hanley for providing the fetal femurs. The critical reading of the manuscript by Dr. Fred Anthony was much appreciated.