Large-scale gene expression profiling reveals major pathogenetic pathways of cartilage degeneration in osteoarthritis

Authors


Abstract

Objective

Despite many research efforts in recent decades, the major pathogenetic mechanisms of osteoarthritis (OA), including gene alterations occurring during OA cartilage degeneration, are poorly understood, and there is no disease-modifying treatment approach. The present study was therefore initiated in order to identify differentially expressed disease-related genes and potential therapeutic targets.

Methods

This investigation consisted of a large gene expression profiling study performed based on 78 normal and disease samples, using a custom-made complementary DNA array covering >4,000 genes.

Results

Many differentially expressed genes were identified, including the expected up-regulation of anabolic and catabolic matrix genes. In particular, the down-regulation of important oxidative defense genes, i.e., the genes for superoxide dismutases 2 and 3 and glutathione peroxidase 3, was prominent. This indicates that continuous oxidative stress to the cells and the matrix is one major underlying pathogenetic mechanism in OA. Also, genes that are involved in the phenotypic stability of cells, a feature that is greatly reduced in OA cartilage, appeared to be suppressed.

Conclusion

Our findings provide a reference data set on gene alterations in OA cartilage and, importantly, indicate major mechanisms underlying central cell biologic alterations that occur during the OA disease process. These results identify molecular targets that can be further investigated in the search for therapeutic interventions.

Osteoarthritis (OA) is one of the most common disabling conditions, affecting many parts of the joint, including bone, synovium, ligaments, and articular cartilage. Although OA is mainly characterized by functional loss of the articular cartilage matrix covering the joint surfaces, it is obvious that cells are active players during the disease process. In many laboratories, small-scale expression analyses have been performed on normal and OA cartilage specimens. These analyses have revealed activation (1, 2) as well as phenotypic instability (3–5) of articular chondrocytes. A gene expression profile of ∼1,200 different genes and some 20 samples provided an initial set of interesting data on differentially expressed genes, but was limited due to the commercial array used, which focused on cancer-relevant genes (6). Since then, a few more studies on gene expression profiles in articular cartilage have been reported (for review, see ref.7). In addition, gene association studies have identified numerous genes that might confer susceptibility to OA (for review, see ref.8).

However, knowledge about changes in OA cartilage remains limited. A broader gene expression profile of OA chondrocytes needs to be established using modern screening technologies, in order to better characterize the cellular events and regulatory pathways directly involved in cartilage destruction. In this study, we used extensive gene expression profiling to investigate molecular alterations that would provide clues regarding the major pathogenetic factors in the OA disease process.

MATERIALS AND METHODS

Clinical cases and RNA isolation.

For study of messenger RNA (mRNA) expression levels by complementary DNA (cDNA) array and quantitative polymerase chain reaction (PCR) techniques, cartilage from human femoral condyles was processed as described previously (6). Normal articular cartilage (18 specimens, from subjects ages 45–88 years) and cartilage with early degeneration (20 specimens, from subjects ages 43–91 years) were obtained at autopsy, within 48 hours of death. OA cartilage was obtained at the time of total knee replacement (21 samples with mild OA according to the Mankin scale [9], from patients ages 61–84 years; 19 samples with moderate or severe OA, from patients ages 61–84 years). Cartilage was considered to be normal if it showed no significant macroscopic softening or surface fibrillation and had a Mankin grade of <3. Early degenerative cartilage was defined as cartilage that showed moderate fibrillation and softening but no advanced erosion of the articular cartilage, corresponding to a Mankin grade of 3–6. Cartilage from patients with rheumatoid arthritis was excluded from the study. Only primary degenerated cartilage was used; regenerative cartilage (osteophytic tissue) was not studied. RNA from cartilage tissue was isolated as described previously (10).

Oligonucleotide fingerprinting.

Two cDNA libraries were constructed from 2 separate pools of mRNA, using chondrocytes that were originated from OA and normal cartilage (starting from 1 mg total RNA each) and transfected into Escherichiacoli. Bacterial colonies were plated, and 200,000 clones per library were placed into microtiter plates and subjected to oligonucleotide fingerprinting as described previously (11). Briefly, the inserts of all selected clones were PCR amplified and spotted onto nylon filters. One hundred ninety-five short oligonucleotides were radioactively labeled and hybridized to these filters. Hybridizations were analyzed with phosphorimagers and automated detection software and clustered by the k-means algorithm (12). Approximately 330,000 cDNA clones had sufficient hybridization information to be subjected to clustering, which resulted in 8,821 different clusters. The accuracy of the clustering was controlled by hybridization of randomly selected cDNA probes onto the same nylon filters.

Complementary DNA array production and expression analysis.

Complementary DNA arrays were produced with 7,808 complementary DNAs spotted using a custom-made needle printer. More than 700 identical arrays were produced for the gene expression studies. For probe synthesis (i.e., cDNA synthesis by reverse transcription), 1 μg of total RNA was used. Using a 2-split protocol, 500 ng total RNA each was labeled in 2 independent reactions using 33P and Superscript II (Invitrogen Life Technologies, Paisley, UK), and subsequently pooled and purified. The reaction mixtures were primed using random hexamers. Between 0.7 × 108 and 1 × 108 counts were hybridized onto 4 identical arrays using semiautomated hybridization machines. Hybridizations were carried out in plastic boxes in a volume of 60 ml formamide-based buffer, overnight at 50°C. After washing, filters were exposed for 5 days to phosphorimager screens and then scanned using a Fuji (Tokyo, Japan) BAS 5000 system. Images were then processed using Image Split software (GPC Biotech, Munich, Germany) to generate images of the individual filters. Using Consolen Batch software (GPC Biotech), grids were automatically set onto the images. Spot intensities were then determined, and data were automatically transferred into the database. Visual grid–based software (GPC Biotech) was used for image analysis to determine the raw spot intensities. Local background was determined and subtracted from the target spot intensities. Duplicate spots on an array that showed a large difference in expression were considered outliers and eliminated from further processing.

To consolidate replicate data, the mean and SEM for each target spot were calculated. By comparison of the mean of the target replicate values with the distribution of the background values, the probability of significant expression was estimated for each target. If individual filters in an experiment showed an uneven distribution of the signal or other abnormalities, the experiment was repeated. Correlation factors within and between runs were determined; the mean within-run and between-run correlations were 0.979 and 0.965, respectively. Various possible methods for data processing were compared, and those yielding the most robust outcome (i.e., differentially regulated genes) were chosen.

Complementary DNA array normalization and outlier detection.

Data were normalized by median absolute deviation (MAD) scale normalization (13) to fix the median and MAD (a measure of the variability within the data) of each sample to a common level. The overall expression value distributions of all samples were thereby rendered similar. The normalized data include negative values due to background correction of the original data. For further analysis, the general background level was estimated to be 0.01, and expression values <0.01 were set to 0.01.

The individual hybridizations were checked for data consistency using cluster analysis, principal components analysis, and analysis of data distributions. We found 5 of the original 83 samples to be outliers in at least 2 of the 3 types of analysis and removed them prior to further processing (for details, see supplementary text, available online at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/).

Detection of differential expression.

Within the given microarrays the number of spots per gene varies, and the experimental technique of spotting different clones for a given gene results in high variability of the spot intensities representing the same gene. This necessitates use of a robust combination method. Differentially expressed genes were detected according to the following procedure. First, the 2-sided Wilcoxon rank sum test was applied to calculate P values for spots. Spot P values were combined into gene P values by Stouffer's method (14), in which P values are transformed to Z scores that are combined under consideration of the direction of regulation, and finally the Z scores are transformed back to P values.

The gene P values are converted into q values by use of the R-library q value (15). The q value quantifies the false discovery rate, i.e., a q value of 0.01 indicates that when the subset of all genes having a q value of ≤0.01 is used to define the group of genes with significant differential expression, 1% of the selected genes have to be expected to be false-positive. The false discovery rate was used to rank differentially expressed genes. We applied cutoff criteria for false discovery rate and fold change to select genes for detailed analysis. To estimate the overall fold change for each gene, we determined individual spot fold changes (separately for each pair of samples from the groups to be compared); the overall gene fold change is estimated as the mean of the corresponding spot fold changes.

Clustering.

Two different clustering protocols were performed. For both, data were log2-transformed and subsequently transformed to Z scores so that, for all genes, the mean is 0 with a standard deviation of 1.

In the first protocol, clustering was done on neutrally preselected genes, i.e., without using our knowledge of sample-to-disease assignment (“unsupervised” clustering). Genes were selected by significant expression and significant variance between the analyzed samples. Clustering was done on this gene subset and the entire sample set by applying Spearman's rank correlation and average linkage.

In the second protocol, clustering was done on genes that were preselected by disease group (“supervised” clustering): the top 50 genes were selected on the basis of their fold change and the P value between groups (normal versus late OA and normal versus early OA). The genes preselected for significant differences between the normal and the late OA samples were clustered against all samples. The genes preselected for significant differences between the normal and the early OA samples were clustered against the normal and early OA samples. Clustering was done by applying Spearman's rank correlation and complete linkage.

Complementary DNA synthesis and real-time PCR.

First-strand cDNA was synthesized and real-time PCR (TaqMan) was performed for 10 selected genes (Agg, btg2, COL1A1, COL2A1, COL3A1, GAPDH, GPX3, SOD2,SOX9, and tob1) as described previously (16, 17) (for a more detailed description as well as sequence information on the primers and probes used, see supplementary Table 1 and text, available online at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/).

General screening strategy.

In order to achieve the largest amount of data possible for analysis (18), a large series of nonpooled samples was profiled (4 independent hybridization experiments each). For analysis of the expression data we used a 4-step strategy, as follows: 1) The primary data were normalized. 2) The expression levels of gene groups of known relevance to chondrocyte anabolism were examined, to validate the data obtained. 3) Differences in expression levels between the different sample groups of interest (normal versus early degenerative disease, early degenerative disease versus mild late-stage OA, mild versus moderate or severe late-stage OA) were analyzed. 4) Finally, a clustering analysis was performed for all samples based on their gene expression profiles. All data are available online at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/.

Statistical analysis.

For in vivo investigations, the significance of differences in expression levels was evaluated with the nonparametric Wilcoxon/Mann-Whitney test. This nonparametric test is more likely to be appropriate than, for example, the t-test, because it is not based on assumptions regarding the distribution of expression values (e.g., normal distribution). For the in vitro probes, the t-test for pairwise comparison was used, due to the limited number of comparisons being made. P values less than 0.05 were considered significant.

RESULTS

Findings of oligonucleotide fingerprinting and array generation.

To select OA-relevant cDNA clones, extensive screening based on OliCode technology (11) was performed (11). For this analysis 400,000 clones, derived from 2 cDNA libraries prepared from normal and OA tissue, were hybridized with 195 oligonucleotides and cDNA clones grouped into 8,821 clusters and 36,000 singletons. A total of 5,280 clusters with >3 hits were selected for expression profiling. Additionally, >1,000 genes of interest with regard to cartilage/chondrocyte biology were spotted on the arrays (for the full gene list, see http://www.bio.ifi.lmu.de/publications/OA_cDNA_AR2006/).

Validation and evaluation of expression analysis of marker genes of cellular differentiation and anabolic activity.

For examination of the chondrocyte gene expression profile we first focused on the main function of this cell type, preservation and turnover of the cartilage matrix. Genes involved in cartilage anabolism were first investigated. Types II and III collagen were strongly up-regulated in peripheral and central OA cartilage, which is consistent with the findings of previous studies (1, 4). Type I collagen (COL1A1, COL1A2) was also up-regulated (Figure 1a). This is not surprising and does not necessarily indicate a major shift in the cellular phenotype of chondrocytes (19), but reflects the general metabolic activation of OA chondrocytes, as has been shown by quantitative PCR (17). Collagen types VI (COL6A1, COL6A2, COL6A3), IX (COL9A2, COL9A3), and XI (COL11A1, COL11A2) were also found to be significantly up-regulated, but to a much lesser degree. Similar results with regard to COL6 have been reported previously (20). Interestingly, expression of other collagens (COL5A1, COL15A1), so far not known to be expressed in adult articular chondrocytes, was also detectable.

Figure 1.

Gene expression patterns of a, collagens and b, noncollagenous matrix proteins in cartilage samples with early degeneration (deg.) and with late-stage osteoarthritis (OA). Values are the mean fold difference from normal. ∗ = q (false discovery rate; see Materials and Methods) < 0.05; ∗∗ = q < 0.01; ∗∗∗ = q < 0.001.

In contrast to the collagens, noncollagenous matrix proteins were generally less up-regulated in OA chondrocytes, except for fibromodulin, cartilage intermediate-layer protein, fibronectin, tenascin, and osteonectin/secreted protein, acidic and rich in cysteine (Figure 1b). Another interesting phenomenon was that most of the noncollagenous proteins were much more strongly expressed in normal articular cartilage than were the collagens. This supports the notion of a rather high turnover of the noncollagenous cartilage matrix compared with the collagenous compartment. These observations are consistent with previous data showing almost no overall regulation of aggrecan and decorin (6, 17, 21), in contrast to the strong up-regulation of fibronectin, tenascin, and osteonectin (6, 22, 23), in OA chondrocytes.

Expression of SOX9, the major transcription factor known to be relevant to chondrocyte phenotypic stability (24, 25), was significantly down-regulated. This finding was also in accordance with previously described data (26).

Quantitative PCR analysis confirmed the up-regulation of the matrix proteins type I collagen (fold change observed by quantitative PCR [fold changeqPCR] 12.6; P < 0.005), type II collagen (fold changeqPCR 17.6; P < 0.0001), and type III collagen (fold change 36.4; P < 0.0001), as well as the down-regulation of SOX9 (fold changeqPCR 0.4; P < 0.01) in OA cartilage samples. Aggrecan was not significantly regulated in diseased tissue compared with normal tissue (fold changeqPCR 0.7).

Comparison of genes that were differentially expressed in normal and early degenerative cartilage.

Table 1 summarizes the genes that were significantly (q < 0.05) differentially regulated between normal and early degenerative cartilage. Overall, only 15 genes were significantly up- or down-regulated, even if less stringent criteria were applied (regulation >50%, q < 0.05); this indicated that normal and early degenerative cartilage lesions are similar in terms of general cell biology. The alternative interpretation that early lesions within the joint cartilage are mostly focal could largely be excluded because samples were obtained only from the lesional areas, in which the expression of the genes was altered by 2-fold at most. Thus, the difference between normal and early degenerative cartilage was only minor, in terms of both the number of regulated genes and the intensity of gene regulation.

Table 1. Genes up- or down-regulated in early degenerative cartilage lesions compared with normal samples*
RefSeqAnnotationNormal meanFold changeqP
  • *

    Only genes that were expressed at a level of >0.1 (assigned by the program used) in at least 1 sample group are shown. Up- or down-regulation is defined as a fold change of >1.5 or <0.66 in early degenerative versus normal cartilage, and a false discovery rate (q) of <0.05 (i.e., 5% of the selected genes have to be expected to be false-positive).

NM_002669Pleiotropic regulator 1 (PRL1 homolog, Arabidopsis) (PLRG1)0.101.920.0350.000
NM_005252v-Fos FBJ murine osteosarcoma viral oncogene homolog (FOS)0.121.880.0000.000
NM_002026Fibronectin 1 (FN1)1.691.580.0000.000
 
NM_003068Snail homolog 2 (Drosophila) (SNAI2)0.280.470.0210.000
NM_198057Delta sleep–inducing peptide, immunoreactor (DSIPI)0.570.550.0350.000
NM_006854KDEL endoplasmic reticulum protein retention receptor 20.270.550.0280.000
NM_003479Protein tyrosine phosphatase type IVA, member 2 (PTP4A2)2.740.550.0210.000
NM_001109A disintegrin and metalloproteinase domain 8 (ADAM8)1.700.550.0210.000
NM_004960Fusion, derived from t(12;16) malignant liposarcoma (FUS)9.400.550.0160.000
NM_004161RAB1A, member RAS oncogene family (RAB1A)3.870.570.0290.000
NM_006833COP9 subunit 6 (MOV34 homolog, 34 kd) (COPS6)12.460.570.0000.000
NM_153425Tumor necrosis factor receptor I–associated death domain (TRADD)1.620.580.0260.000
NM_015692α2-macroglobulin family protein VIP (VIP)0.240.590.0030.000
NM_016732RNA binding protein (hnRNP-associated with lethal yellow) (RALY)0.500.630.0030.000
NM_006423Rab acceptor 1 (prenylated) (RABAC1)0.110.650.0030.000

One of the up-regulated genes was fibronectin (fold change 1.58; q < 0.001), a gene that is well known to be up-regulated early in cartilage degeneration (27). However, the increase of fibronectin might be ambiguous since its degradation products have been shown to be detrimental to cartilage homeostasis (28, 29). Regardless, however, this finding further emphasizes the potential importance of fibronectin in maintenance and degradation of articular cartilage.

Comparison of genes that were differentially expressed in normal cartilage and severe late-stage OA cartilage.

In the next step, we analyzed the genes that were significantly differentially regulated between normal and late-stage OA chondrocytes. This revealed a significantly higher number of differentially expressed genes than in the other comparisons (Table 2 and supplementary Tables 2 and 3, available online at http://www.mrw.interscience.wiley.com/suppmat/0004-3591/suppmat/). An ontology analysis based on Geneontology (www.geneontology.org) revealed a broad spectrum of regulated genes (Figure 2) and emphasized the up-regulation of numerous genes involved in extracellular matrix formation, as indicated above. In contrast, many genes involved in oxidative damage defense, namely GPX3 (glutathione peroxidase 3) (fold change 0.12; q < 0.001), SOD2 (superoxide dismutase) (fold change 0.48; q < 0.01), SOD3 (fold change 0.62; q < 0.001), and TXNIP (thioredoxin-interacting protein) (fold change 0.29; q < 0.001), appeared to be down-regulated. Findings with regard to GPX3 and SOD2 expression were confirmed by quantitative PCR (GPX3 fold changeqPCR 0.08 [P < 0.0001]; SOD2 fold changeqPCR 0.08 [P < 0.06]).

Table 2. Genes up- or down-regulated in cartilage lesions with moderate/severe late-stage OA compared with normal samples*
RefSeqAnnotationNormal meanFold changeqP
  • *

    Only a selection of genes that were expressed at a level of >0.1 (assigned by the program used) in at least 1 sample group is shown. Up- or down-regulation is defined in this table as a fold change of >2 or <0.5 in cartilage with moderate/severe late-stage osteoarthritis (OA) versus normal cartilage, and a false discovery rate (q) of <0.05 (i.e., 5% of the selected genes have to be expected to be false-positive). For the complete list of genes with a fold change of >1.5 or <0.66, see supplementary Table 3.

NM_000088Collagen, type I, α1 (COL1A1)0.099.550.0000.000
NM_170746Selenoprotein H (SELH)0.109.270.0000.000
NM_002160Tenascin C (hexabrachion) (TNC)0.058.060.0000.000
NM_000090Collagen, type III, α1 (COL3A1)0.077.370.0000.000
NM_033150Collagen, type II, α1 (COL2A1)0.146.070.0000.000
NM_000093Collagen, type V, α1 (COL5A1)0.065.810.0000.000
NM_173343Interleukin-1 receptor, type II (IL1R2)0.085.340.0000.000
NM_002026Fibronectin 1 (FN1)1.694.690.0000.000
NM_002775Protease, serine, 11 (IGF binding) (PRSS11)0.094.170.0000.000
NM_000089Collagen, type I, α2 (COL1A2)0.063.410.0000.000
NM_003118Secreted protein, acidic, cysteine-rich (osteonectin)0.193.340.0000.000
NM_018058Cartilage acidic protein 1 (CRTAC1)0.343.010.0000.000
NM_058175Collagen, type VI, α2 (COL6A2)C20.282.960.0000.000
NM_003613Cartilage intermediate-layer protein (CILP)0.282.720.0000.000
NM_080630Collagen, type XI, α1 (COL11A1)0.042.660.0000.000
NM_001848Collagen, type VI, α1 (COL6A1)0.382.580.0000.000
NM_002023Fibromodulin (FMOD)1.492.560.0000.000
NM_032977Caspase 10, apoptosis-related cysteine0.372.530.0000.000
NM_004000Chitinase 3–like 2 (CHI3L2)0.102.110.0000.000
NM_003254Tissue inhibitor of metalloproteinases 1 (TIMP1)0.362.050.0000.000
 
NM_002084Glutathione peroxidase 3 (plasma) (GPX3)3.990.120.0000.000
NM_005195CCAAT/enhancer binding protein (C/EBP), delta0.570.150.0000.000
NM_006472Thioredoxin-interacting protein (TXNIP)0.690.290.0000.000
NM_002422Stromelysin 1 (MMP3)0.380.290.0000.000
NM_024906Stearoyl-CoA desaturase 4 (SCD4)0.770.310.0000.000
NM_005412Serine hydroxymethyltransferase 2 (mitochondrial)0.160.320.0070.001
NM_175617Metallothionein 1E (functional) (MT1E)0.570.330.0000.000
NM_004417Dual-specificity phosphatase 1 (DUSP1)0.790.330.0000.000
NM_018659Cytokine-like protein C17 (C17)0.770.350.0000.000
NM_006763BTG family, member 2 (BTG2)0.170.360.0070.001
NM_005950Metallothionein 1G (MT1G)0.200.380.0000.000
NM_002178Insulin-like growth factor binding protein 6 (IGFBP6)0.440.430.0040.000
NM_005952Metallothionein 1X (MT1X)0.630.430.0000.000
NM_005749Transducer of ERBB2, 1 (TOB1)0.210.450.0000.000
NM_002065Glutamate-ammonia ligase (glutamine synthase) (GLUL)0.380.480.0160.002
NM_006169Nicotinamide N-methyltransferase (NNMT)0.150.480.0000.000
NM_000636Superoxide dismutase 2, mitochondrial (SOD2)0.220.480.0050.001
Figure 2.

Functional role of genes that were significantly (P ≤ 0.01, log2 fold change >1) down-regulated (a) or up-regulated (b) in the comparison of normal versus late-stage osteoarthritis cartilage. The area in each chart assigned to a functional category corresponds to the negative logarithm of the P value determined for the category by hypergeometric distribution, i.e., the larger an area, the more significant the overrepresentation of the category.

Other findings of interest pertained to genes of the transducer of ERBB2 (tob)/B cell translocation gene (btg) family, namely tob1 (fold change 0.39; q < 0.001), btg1 (fold change 0.53; q < 0.001), and btg2 (fold change 0.36; q < 0.05). For tob1 and btg2, the findings were confirmed by quantitative PCR (tob1 fold changeqPCR 0.13 [P < 0.001]; btg2 fold changeqPCR 0.17 [P < 0.05]).

A third group of regulated genes included numerous genes for cytokines or genes involved in cytokine signaling. One interesting finding was that many genes related to the interleukin-1 (IL-1) pathway were, in contrast to our expectation, not up-regulated, but rather, down-regulated, in OA chondrocytes. This included IL1B (fold change 0.5; q < 0.001) itself as well as IL6, IL-8, and LIF (leukemia inhibitory factor). Additionally, functional antagonist/scavenger receptor type II was down-regulated (fold change 5.3; q < 0.001). However, the expression levels of relevant genes were very low in all samples, which warrants a caveat about data reliability despite statistical significance. Thus, additional investigations are necessary in order to evaluate the importance of the IL-1β pathway in OA joint disease.

Comparison of genes that were differentially expressed in mild late-stage and moderate or severe late-stage OA cartilage.

Only 14 genes fulfilled the (rather nonstringently defined) criteria of differential expression (regulation >50%; q < 0.05) when specimens of cartilage with mild versus moderate or severe late-stage OA were compared (Table 3). The results were in accordance with the cluster analysis findings discussed below, showing the close similarity of gene expression in mild and moderate/high late-stage OA cartilage.

Table 3. Genes up- or down-regulated in cartilage lesions with moderate/severe late-stage OA compared with mild late-stage OA*
RefSeqAnnotationLow-grade OA meanFold changeqP
  • *

    Only genes that were expressed at a level of >0.1 (assigned by the program used) in at least 1 sample group are shown. Up- or down-regulation is defined as a fold change of >1.5 or <0.66 in cartilage with moderate/severe late-stage osteoarthritis (OA) versus cartilage with mild late-stage OA, and a false discovery rate (q) of <0.05 (i.e., 5% of the selected genes have to be expected to be false-positive).

NM_000214Jagged 1 (Alagille syndrome) (JAG1)0.792.520.0040.000
NM_153498CamKI-like protein kinase (CKLiK)0.132.410.0220.000
NM_015913Endoplasmic reticulum thioredoxin superfamily member, 18 kd0.072.180.0220.000
NM_002026Fibronectin 1 (FN1)3.922.140.0000.000
NM_000088Collagen, type I α1 (COL1A1)0.672.130.0000.000
NM_002775Protease, serine, 11 (IGF binding) (PRSS11)0.231.830.0000.000
NM_00602526 serine protease (P11)0.141.820.0150.000
NM_002160Tenascin C (hexabrachion) (TNC)0.261.800.0000.000
NM_001823Creatine kinase, brain (CKB)0.071.780.0420.000
NM_018058Cartilage acidic protein 1 (CRTAC1)0.591.770.0000.000
NM_000093Collagen, type V α1 (COL5A1)0.231.670.0000.000
NM_000701ATPase, Na+/K+ transporting, α1 polypeptide (ATP1A1)0.101.610.0260.000
NM_005752C-type lectin, superfamily member 1 (cartilage-derived)0.420.610.0000.000
NM_018659Cytokine-like protein C17 (C17)0.410.610.0100.000

Clustering of sample probes.

In order to investigate differences and overlaps in the expression of genes within the different sample groups, clustering analysis was performed. This analysis revealed 2 primary clusters (Figure 3), with a clear separation between the normal and early degenerative cartilage samples versus the peripheral and central OA cartilage samples. The normal and early degenerative cartilage samples could not be clearly distinguished from one another, nor could the peripheral and central OA samples. This provides evidence in support of the hypothesis that differences between normal and early degenerative cartilage, as well as between peripheral and central OA cartilage, are relatively minor, but the combined normal and early degenerative cartilage group is clearly different from the combined peripheral and central OA group.

Figure 3.

Dendrogram and heat map of all samples and genes preselected for significant expression and variance. The displayed genes were preselected based on significant expression and significant variance between the analyzed samples. Clustering was done by applying Spearman's rank correlation and average linkage. A good separation between the normal (N) and early degenerative (E) versus the peripheral (P) and central (C) osteoarthritis (OA) cartilage samples is seen. Only 1 of the early degenerative cartilage samples is clustered closer to the peripheral and central OA samples than to the other normal and early degenerative cartilage samples. The normal and early degenerative cartilage samples cannot be distinguished from one another, as is also the case for the peripheral and central samples. This provides evidence in support of the hypothesis that normal and early degenerative cartilages are similar and peripheral and central OA cartilages are also similar, but the 2 combined groups are clearly different.

Figure 4 shows the biased cluster analysis of genes that were strongly preselected (the 50 most highly differentially expressed genes) in the analysis of normal versus late OA cartilage and in the analysis of normal versus early degenerative cartilage. These can be seen as marker genes to distinguish between the groups at the gene expression level. Figure 4a shows perfect separation of the normal samples versus the late OA samples based on the selected genes. Interestingly, Figure 4b shows that the samples from early degenerative cartilage again clustered, at least in part, with the normal samples despite the selection of maximally discriminating genes. This again supports the notion that the separation of normal versus early degenerative cartilage is less clear, even with highly preselected genes. However, there was a distinction between the groups at the gene expression level (see Table 1).

Figure 4.

Dendrogram and heat map of a, genes selected from the analysis of normal versus late-stage OA cartilage and b, genes selected from the analysis of normal versus early degenerative cartilage. For both analyses, the top 50 genes were selected based on their fold change and P value between the groups considered. Clustering was done by applying Spearman's rank correlation and complete linkage. Thus, the selected genes can be seen as marker genes to distinguish between the corresponding groups at the gene expression level. a, A perfect separation between the normal samples and the late-stage OA samples is seen. Interestingly, the early degenerative cartilage samples again clustered with the normal samples, even though they were not considered in the preselection of genes for this analysis. b, The separation between normal and early degenerative cartilage is not perfect, and several samples are misclassified. This provides further evidence in support of the notion of a lack of clear distinction between normal and early degenerative cartilage, even in an analysis using highly preselected genes. There is, however, a distinction between the groups at the gene expression level. See Figure 3 for definitions.

Comparative analysis of differential expression levels of susceptibility genes associated with incidence of knee OA.

A number of genes have been previously described to be linked to the development and/or progression of OA of the knee (30–34). Expression levels of these genes in articular cartilage varied and were increased in some cases and decreased in others; most were not significantly up- or down-regulated (Table 4). This is not surprising since the involvement of these genes in the development of OA might entail modification of developmental processes or other indirect effects, such as promotion of obesity and/or and increased pain (in)sensitivity. Also, the present data set is not very suitable for evaluating the importance of expression levels of these genes with regard to OA development, because a longitudinal database would be needed for assessment of predisposing factors. Thus, if, e.g., reduced expression of certain genes increases the risk for development of OA, then OA-susceptible normal samples should show reduced levels of these genes; however, expression levels of these genes would not necessarily be increased or reduced during the development of OA.

Table 4. Detected differential expression levels of candidate genes for knee osteoarthritis, which were suggested in the literature*
GeneRefSeqAnnotationRef.Normal meanLate OA vs. normal
Fold changeqP
  • *

    Other “hot candidate” molecules, such as asporin (57), LRCH1 (30), and FRZB (31), were not represented on the arrays.

AACTNM_001085Serine (or cysteine) proteinase inhibitor, clade A321.130.900.2850.141
ANKHNM_054027Ankylosis, progressive homolog550.161.590.0000.000
BLP2NM_025141BBP-like protein 2 (BLP-2), transcript variant 2340.090.970.6590.769
BMP2NM_001200Bone morphogenetic protein 2320.021.320.3140.170
CILPNM_003613Cartilage intermediate-layer protein320.282.700.0000.000
CIRBPNM_001280Cold-inducible RNA binding protein340.180.590.0000.000
COX2NM_000963Prostaglandin-endoperoxide synthase 2320.041.560.2740.129
DUSP1NM_004417Dual-specificity phosphatase 1340.790.300.0000.000
EIF4A1NM_001416Eukaryotic translation initiation factor 4A, isoform 1340.140.940.5040.440
ESR1NM_000125Estrogen receptor 1320.061.200.4900.414
GPRK6NM_002082G protein–coupled receptor kinase 6340.031.640.1220.037
H3F3BNM_005324H3 histone, family 3B341.200.430.0000.000
HIST2H2AANM_003516Histone 2, H2aa (HIST2H2AA)340.130.810.1630.056
IL1R1NM_000877Interleukin-1 receptor type I560.071.000.6180.679
NCOR2NM_006312Nuclear receptor corepressor 2320.150.940.5620.560
OPGNM_002546Osteoprotegerin320.051.070.5900.619
RHOBNM_004040Ras homolog gene family, member B (ARHB)340.770.290.0000.000
S100BNM_006272S100 calcium-binding protein, β (neural)340.270.970.6050.652
SUI1NM_005801Putative translation initiation factor340.380.610.0000.000
TNFAIP6NM_007115Tumor necrosis factor α–induced protein 6320.300.920.6910.858

DISCUSSION

This work provides the first extensive gene expression profile of normal, early degenerative, and mild and moderate/severe late-stage OA chondrocytes. The most striking feature in the data from all 4 groups is the high variability of gene expression levels between the different donors, for many genes. Whereas this might be expected for the disease samples for various reasons (e.g., different stages), the finding of a comparable variability among normal donors is very surprising. Since no rhythms (e.g., circadian) in articular cartilage tissue have been reported to date, this indicates that a wide range of gene expression levels is compatible with normal functioning of the tissue (as well as tissue dysfunction). Thus, networks of molecules, rather than single components, appear to determine the normal state, an issue that is particularly relevant with regard to development of therapeutic interventions.

The second important result of our study is that macroscopically less damaged or relatively normal-appearing cartilage from joints with late-stage OA cannot be considered normal, nor is it similar to early degenerative lesions. This is in accordance with previous studies showing that cartilage that is obtained from joints with advanced OA but has a lower Mankin grade is metabolically activated and shows matrix alterations similar to those in cartilage with moderate/high Mankin grades (1, 35–38). Our present analysis indicates that these rather normal-looking areas are severely changed in terms of their gene expression profile. Also, the similarity between the peripheral and central areas provides evidence against the notion that peripheral OA cartilage is a good model for early OA (39), for which it is often used, and it further suggests that there is no major shift between the 2 tissue and cell types, as recently suggested (40). Obviously, the most severely damaged areas with no or little cartilage remaining are difficult to assess with the methods as used in our study, and were therefore not investigated. However, these areas are of little clinical relevance since they represent a stage of the disease at which therapy will not be of benefit.

Certainly, great caution would be required if early degenerative cartilage from any donor were to be studied as a substitute for early OA cartilage (40). A significant portion of it might not represent progressive disease. Still, these early lesions were shown to be biochemically similar to early lesions of OA (41). Also, we cannot rule out the possibility that gene expression levels might be altered during the (short) time delay between death and autopsy, which is nearly unavoidable in studies of human samples (40). Obviously, this time has to be kept as short as possible, and further studies are needed in order to evaluate more directly its influence on changes in gene expression. However, extensive studies on mRNA expression levels in human brain tissue have shown no significant changes in the gene expression pattern even several days postmortem (42, 43). Also, a comparative analysis of changes in gene expression levels in relation to time since death did not reveal any evidence of significant changes occurring during the time frame investigated (up to 48 hours); in particular, no hypoxia-induced genes appeared to be up-regulated (Aigner T, et al: unpublished observations). Cartilage is, in this respect, a unique tissue in that chondrocytes remain viable for a long time (44).

The similarity of mild and moderate/severe late-stage OA specimens presumably reflects the fact that all cartilage areas in a joint are exposed to the same synovial factors, i.e., potent cytokines and growth factors. These are secreted by activated synoviocytes (45) and diffuse into the cartilage from the synovial joint space.

Another interesting finding is the identification of gene groups that are most differentially expressed between OA and normal chondrocytes. In this respect, the strong up-regulation of matrix constituents was to be expected (1, 2, 9), and thus represented an excellent internal positive control.

The strong down-regulation of major components of oxidative cellular defense is a new and surprising observation, given that oxidative stress is increased in OA chondrocytes (for review, see ref.46). The dramatic down-regulation of enzymes that are key in cellular oxidative defense is presumably one important reason for the increased accumulation of oxidatively damaged molecules within the cells. Thus, even a reduction of the gene dose of SOD2 of 50% in heterozygous knockout mice leads to significantly increased oxidative stress and cell damage (47). Also, the down-regulation of the extracellular isoform, SOD3, and of GPX3 might increase reactive oxygen species in the extracellular space and might enhance cartilage matrix breakdown in this way (48).

Also, the down-regulation of a third group of genes, for members of the tob/btg group of proteins, fits very well into the scenario of cell biologic changes occurring in OA cartilage (49, 50). This might represent an important clue for understanding of the cell biology of OA: the molecules of the btg/tob family are thought to be involved in phenotype stabilization of cells and inhibition of proliferative activity, both of which are reversed in OA chondrocytes.

A fourth group of proteins of interest found in our study comprised proteins from the IL-1 signaling pathway, which appear to be down-regulated rather than activated in OA cartilage. This is somewhat in contrast to many previous assumptions (for review, see ref.51), but is in accordance with recent in vitro and in vivo data. Thus, investigators at our laboratory have shown clearly reduced responsiveness of OA chondrocytes to stimulation with IL-1 (52). Also, activation of IL-6 and LIF expression in OA cartilage, as would be expected after stimulation with IL-1β, was not observed (53). Although attenuation of the IL-1β pathway might at first appear to be a potentially beneficial treatment avenue, providing protection of articular cartilage from catabolic stimulation, recent data suggest that IL-1 activity might be important for cartilage tissue homeostasis (54).

Overall, besides providing an extensive primary data set, the present study offers significant information on the types of samples important for studying normal, degenerative, or OA cartilage, which is clearly an area that holds great potential in terms of gene expression profiling analyses. At the same time, we have identified foci of gene alterations, confirming known alterations (matrix synthesis, proliferation, phenotypic instability of cells) as well as identifying innovative new ones (oxidative defense). Most importantly, the results provide not only general direction, but also details on specific targets, in terms of therapeutic intervention: the tob1, btg2, SOD2, SOD3, and GPX3 genes should be the subject of further research.

Acknowledgements

We would like to acknowledge Drs. G. Zeiler and W. Eger (Orthopedic Hospital, Rummelsberg, Germany) for providing OA cartilage samples, Dr. Stephan Söder (Erlangen, Germany) for help in acquisition of normal cartilage tissue, and Freya Boggasch, Anke Nehlen, and Brigitte Bau for expert technical assistance.

Ancillary