Hypoxia Promotes the Differentiated Human Articular Chondrocyte Phenotype through SOX9-dependent and -independent Pathways*

The chondrocyte is solely responsible for synthesis and maintenance of the resilient articular cartilage matrix that gives this load-bearing tissue its mechanical integrity. When the differentiated cell phenotype is lost, the matrix becomes compromised and cartilage function begins to fail. We have recently shown that hypoxia promotes the differentiated phenotype through hypoxia-inducible factor 2α (HIF-2α)-mediated SOX9 induction of the main matrix genes. However, to date, only a few genes have been shown to be SOX9 targets, while little is known about SOX9-independent regulators. We therefore performed a detailed microarray study to address these issues. Analysis involved 35 arrays on chondrocytes obtained from seven healthy, non-elderly human cartilage samples. Genes were selected that were down-regulated with serial passage in culture (as this causes loss of the differentiated phenotype) and subsequently up-regulated in hypoxia. The importance of key findings was further probed using the technique of RNA interference on these human articular chondrocytes. Our results show that hypoxia has a broader beneficial effect on the chondrocyte phenotype than has been previously described. Of especial note, we report new hypoxia-inducible and SOX9-regulated genes, Gdf10 and Chm-I. In addition, Mig6 and InhbA were induced by hypoxia, predominantly via HIF-2α, but were not regulated by SOX9. Therefore, hypoxia, and more specifically HIF-2α, promotes both SOX9-dependent and -independent factors important for cartilage homeostasis. HIF-2α may therefore represent a new and promising therapeutic target for cartilage repair.

The chondrocyte is solely responsible for synthesis and maintenance of the resilient articular cartilage matrix that gives this load-bearing tissue its mechanical integrity. When the differentiated cell phenotype is lost, the matrix becomes compromised and cartilage function begins to fail. We have recently shown that hypoxia promotes the differentiated phenotype through hypoxia-inducible factor 2␣ (HIF-2␣)-mediated SOX9 induction of the main matrix genes. However, to date, only a few genes have been shown to be SOX9 targets, while little is known about SOX9-independent regulators. We therefore performed a detailed microarray study to address these issues. Analysis involved 35 arrays on chondrocytes obtained from seven healthy, non-elderly human cartilage samples. Genes were selected that were down-regulated with serial passage in culture (as this causes loss of the differentiated phenotype) and subsequently up-regulated in hypoxia. The importance of key findings was further probed using the technique of RNA interference on these human articular chondrocytes. Our results show that hypoxia has a broader beneficial effect on the chondrocyte phenotype than has been previously described. Of especial note, we report new hypoxia-inducible and SOX9-regulated genes, Gdf10 and Chm-I. In addition, Mig6 and InhbA were induced by hypoxia, predominantly via HIF-2␣, but were not regulated by SOX9. Therefore, hypoxia, and more specifically HIF-2␣, promotes both SOX9-dependent and -independent factors important for cartilage homeostasis. HIF-2␣ may therefore represent a new and promising therapeutic target for cartilage repair.
The main function of articular cartilage is to protect the underlying bone by withstanding loading of the joint. Because it is the only cell type in the tissue, the chondrocyte is solely responsible for synthesis and maintenance of the resilient extracellular matrix that gives cartilage its mechanical integrity. The matrix predominantly consists of type II collagen fibers and aggrecan. The former enables resistance to tensional forces, whereas the latter draws water into cartilage and therefore gives the tissue its ability to withstand compressive forces (1,2). However, when the differentiated cell phenotype is lost, production of these proteins decreases greatly and the matrix becomes compromised, and therefore cartilage function begins to fail. An altered phenotype is found in diseased states such as arthritis, where the cartilage matrix wears away, eventually exposing the underlying bone (3). In addition, current cellbased therapies aimed at cartilage repair such as autologous chondrocyte implantation rely on subculture of chondrocytes ex vivo. However, the differentiated phenotype is also lost in such standard cell culture conditions (4 -6).
Cartilage is an avascular tissue and therefore exists in a low oxygen environment (7,8). We have previously identified hypoxia as a promoter of the differentiated phenotype (9,10) and shown this occurs through HIF-2␣ 2 -mediated induction of master chondrocyte regulator, transcription factor SOX9 (11). Sox9 is essential for cartilage development in mice (12), and in humans, heterozygous mutation of the Sox9 gene causes camptomelic dysplasia, a severe skeletal disease (13)(14)(15). However, to date, only a few genes have been shown to be SOX9 targets (16 -18), while little is known about SOX9-independent regulators. Therefore, the aim of the present study was to identify novel regulators of the chondrocyte phenotype. Such findings could be used to improve current cell-based therapies or identify new treatments for restoration and maintenance of cartilage homeostasis in diseased or damaged tissue.
Potential chondrocyte regulators and markers were identified as genes that were down-regulated with serial passage in culture (as this causes loss of the differentiated phenotype) (4 -6) and subsequently up-regulated in hypoxia (a condition shown to promote the differentiated phenotype, (9,10,19). An extensive microarray analysis was thus performed: 35 arrays from seven individual patient samples, with two time points, and primary chondrocytes from each patient acting as a positive control. The SOX9 dependence of the most promising candidates was subsequently assessed using the technique of RNA interference. In this manner, we identified novel SOX9-dependent and -independent genes relevant to expression of the differentiated human chondrocyte phenotype and hence to cartilage repair.

EXPERIMENTAL PROCEDURES
Human Articular Chondrocyte Isolation and Expansion-Healthy articular cartilage was obtained from the femoral con-* This work was supported by Grant BBSB0126X from the UK Biotechnology and Biological Sciences Research Council and by the UK Arthritis Research Campaign. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 To whom correspondence should be addressed. dyle and tibial plateau of seven patients amputated for osteosarcoma or soft tissue sarcoma that did not involve the joint (six male, one female, from 8 to 45 years old). Samples were obtained after informed patient consent and following local ethics committee guidelines. The cartilage specimens were dissected on the same day as surgery and cut into small pieces (1-2 mm 2 ) before overnight digestion in 1.5 mg/ml collagenase 2 (Worthington) in Dulbecco's modified Eagle's medium containing 10% serum (Biosera), with incubation on a shaker (ϳ50 rpm) at 37°C. Freshly isolated cells were seeded at a density of 8ϫ10 3 /cm 2 and with medium (Dulbecco's modified Eagle's medium containing 10% serum) changed twice a week. Cultures were passaged upon reaching confluency (ϳ7 days). For subsequent subculture, cells were seeded at 5 ϫ 10 3 /cm 2 . Expansion of cells was performed in 20% oxygen (37°C in humidified atmosphere of 5% carbon dioxide). After two rounds of subculture (second passage cells) the cells had undergone an ϳ12-16-fold increase in number.
Hypoxia Experiments-Dedifferentiated human articular chondrocytes (i.e. second passage cells) were seeded at high density (1 ϫ 10 5 /cm 2 ) in either 20 or 1% oxygen, with primary chondrocytes (in 20% oxygen) acting as a positive control. The high density seeding prevented proliferation and cell spreading and therefore inhibited further dedifferentiation of the cells. Medium was pre-equilibrated at the appropriate oxygen tension before addition to cultures. Primary cultures were obtained by seeding freshly isolated chondrocytes as above and leaving for 1 week in 20% oxygen before RNA extraction and analysis. Second passage cells were lysed after 1 or 4 days of exposure to hypoxia (or control normoxia conditions).
Microarray Sample Preparation-RNA was isolated using the RNeasy kit (Qiagen). Total RNA quality was verified by measuring A 260 /A 280 ratio with a spectrophotometer (Nanodrop ND-1000), and an aliquot was analyzed with the Agilent Bioanalyser 2100 system. 100 ng of total RNA was used for preparing cDNA. The reaction was carried out with T7oligo(dT) primer and Superscript II (Affymetrix). Second strand cDNA synthesis was done at 16°C by adding DNA polymerase I and RNase H to the reaction. cDNA was then used for a first in vitro transcription of cRNA using the MEGAscript T7 kit (Ambion). cRNA was treated with the Genechip Sample Cleanup Module (Qiagen), and quality was analyzed with the Agilent Bioanalyser 2100 system. 600 ng of cRNA was used for a second amplification cycle. For this round, double strand cDNA synthesis was followed by incubation with T4 DNA polymerase (Affymetrix). After cleaning up cDNA with the Genechip Sample Cleanup Module (Qiagen), in vitro transcription of cRNA was done in the presence of biotinylated nucleotides using the Genechip IVT labeling kit (Affymetrix). cRNA was purified again with the Genechip Sample Cleanup Module (Qiagen kit) and quantified (Nanodrop ND-1000). Quality of cRNA was analyzed with the Agilent Bioanalyser 2100 system. cRNA was further tested by hybridizing to a Test 3.0 Array (Affymetrix). Once samples passed these control measures, 20 g of biotinylated cRNA were hybridized to Human Genome U133 Plus 2.0 Array (56921 transcripts representing 45500 genes; Affymetrix).
Microarray Data Analysis-Gene expression analysis was performed using Resolver software 7.0 (Rosetta). Data were obtained on human articular chondrocyte cultures from a total of seven patients. The following cultures were prepared for microarray analysis: primary chondrocytes (1 o ) in 20% oxygen (n ϭ 7) and corresponding second passage (P2) chondrocytes after exposure to 20 and 1% oxygen (n ϭ 7 each). These latter P2 comparisons were done after both 1 and 4 days of treatment, giving a total data set of (5 ϫ 7) 35 arrays.
Real-time PCR-This was performed on a real-time PCR Corbett Research thermocycler as previously described (11). In all cases, pre-developed primer/probe sets were obtained from Applied Biosystems. cDNA was generated from 1 g of RNA (Qiagen) using the Promega kit with random primers (Promega). Expression was normalized using ribosomal protein (RPLP0), as previous experiments had shown this gene was not regulated by hypoxia.
siRNA Transfection-Human articular chondrocytes were transfected as previously described (11). Briefly, 5 ϫ 10 4 cells were seeded in a 3.5-cm dish. siRNA was transfected at a final concentration of 10 nM using Lipofectamine 2000 (Invitrogen) for 4 h in serum-free OptiMEM I. siRNAs against HIF-1␣, HIF-2␣, and SOX9 were used (11). siRNA against luciferase (Dharmacon) was transfected as a non-targeting control. Efficiency of transfection was also assessed using a fluorescently labeled siRNA (siGlo) from Dharmacon. Four to six hours after trans- fection, medium was changed with pre-equilibrated Dulbecco's modified Eagle's medium (in 20 or 1% oxygen) containing 10% fetal calf serum and cells were incubated in 20 or 1% oxygen for 3 days. Protein Extraction and Western Blotting-Cells were lysed in urea sample buffer (8 M urea, 10% glycerol, 1% SDS, 5 mM dithi-othreitol, 10 mM Tris-HCl) for HIF analysis or otherwise in radioimmune precipitation assay buffer. A protease inhibitor mixture (Sigma) was added just prior to cell lysis. Protein (20 g) was separated by SDS-PAGE, transferred onto polyvinylidene difluoride membranes for Western blotting, and finally visualized using the ECL method. Primary antibodies Statistical Analysis-Unless otherwise stated, data were compared using one-way analysis of variance with Bonferroni's post test using Prism 4 software (GraphPad). Results are mean Ϯ S.E. from four to seven independent experiments (i.e. using chondrocytes from four to seven individual patients). p Ͻ 0.05 was considered statistically significant.

RESULTS
The Main Cartilage Matrix Genes Are Hypoxia-inducible-An extensive microarray analysis was performed using 35 genome-wide arrays on human articular chondrocytes obtained from seven patients (showing no signs of cartilage disease). Genes potentially relevant to the differentiated phenotype were identified as those whose expression was decreased with serial passage of chondrocytes in normoxia (dedifferentiation) and subsequently up-regulated in hypoxia (redifferentiation) (Fig. 1A). A final list of 101 transcripts was obtained for which their expression met these criteria at both 1-and 4-day time points (n ϭ 7, p Ͻ 0.01) (Fig. 1B). A threshold for hypoxic induction Ն1.5-fold was also applied. Results have been classified according to known function in Table 1, which also shows validation of microarray data by real-time PCR (for 23 genes). All of the established cartilage extracellular matrix genes (aggrecan, Col2a1, Col9a1, COl11a2), in addition to master chondrocyte regulator transcription factor Sox9 were identified as hypoxia-inducible genes (Table 1). In addition, the extracellular matrix protein Prelp was both highly expressed in pri- mary human chondrocytes and hypoxia-inducible (Fig. 2D).
Of the matrix proteins, higher expression in primary chondrocytes was only observed for aggrecan and Col2a1 (Fig. 2, A and B, respectively, and Table 1).
Applying an unsupervised genome-wide hierarchical cluster analysis, the cartilage-associated genes Hapln1 (link protein), Fgfr3, Ctgf, and the recently discovered cartilage collagen Col27a1 (20) were shown to cluster tightly with the previously identified aggrecan, Col2a1, Col11a2, and Col9a2 (Fig. 3). Because this analysis was performed on the whole genome for the complete data set of 35 arrays, the cluster containing these genes indicates a very similar pattern of expression through the dedifferentiation/redifferentiation process, albeit with some variation between patient samples.
Hypoxia-inducible Genes Gdf10 and Chm-I Are SOX9dependent-Chm-I and Gdf10 were identified as hypoxia-inducible genes relevant to the differentiated chondrocyte phenotype (Fig. 2, G and H, respectively). To more directly assess the potential importance of these genes to cartilage function, we investigated their dependence on the master cartilage transcription factor SOX9 by depleting the latter using RNA interference. Fig. 4A shows that virtually all the cells were successfully transfected using Lipofectamine to deliver the siRNA oligonucleotides, with no detrimental effects on cell viability. SOX9 depletion was comprehensive as shown by Western blotting (Fig. 4B). The expression of both Chm-I and Gdf10 was significantly reduced in both 20 and 1% oxygen in SOX9-depleted cells. Specifically, hypoxic induction of Chm-I was completely abolished by depletion of SOX9 (Fig. 4C), while that of Gdf10 was greatly reduced (Fig. 4D).
INHBA and MIG6 Are Potential Chondrocyte Regulators Whose Expression Is Independent of SOX9-The highly expressed Mig6 (Fig. 2C) and InhbA (Fig. 2E) were both identified through microarray analysis as potential regulators of the chondrocyte phenotype. In addition, InhbA clustered with Col2a1 and aggrecan, among other cartilage-specific proteins, FIGURE 3. Dendrogram showing genes clustering with Col2a1. An unbiased, genome-wide, hierarchical clustering analysis was performed on the entire data set of 35 arrays from human articular chondrocyte samples from a total of seven patients. All of the main established cartilage matrix genes clustered with Col2a1, in addition to other cartilage-associated genes and, more unexpectedly, muscle transcription factor Mef2c. 1 o , primary human chondropcytes. Note, the color coding refers to Z-scores rather than expression intensities; Z-score normalization was applied as it helps to remove the multiplicative error effect of the intensity measurements.

DISCUSSION
The aim of this study was to identify novel regulators and markers of the differentiated human chondrocyte phenotype. This is important because repair of damaged and diseased cartilage relies on appropriate matrix synthesis from the resident chondrocytes. Using a microarray-based approach we identified a tightly regulated group of potentially important chondrocyte genes, whose regulation was further probed using the technique of RNA interference. We thus identified matrix protein Chm-I and growth factor Gdf10 as new SOX9-regulated genes and Mig6 and InhbA as hypoxia-inducible (predomi-nantly through HIF-2␣), but SOX-9-independent, genes defining the differentiated human chondrocyte phenotype.
This study was performed using chondrocytes from healthy human cartilage, i.e. non-diseased and nonelderly tissue (mean age of 22 years). Typically, human cartilage research is done on the only tissue readily available: elderly, arthritic cartilage obtained following joint replacement. However, our samples were obtained from amputations due to sarcomas that did not involve the joint space. Thus this tissue represents a rarely obtained and highly therapeutically relevant research tool. With normal cartilage obtained from a total of seven different patients, we identified potential chondrocyte regulators by analyzing their changing expression in three different conditions: primary chondrocytes indicating the normal, differentiated phenotype; dedifferentiated cells obtained through serial subculture; and redifferentiating chondrocytes (by exposure of dedifferentiated cells to hypoxia). It has long been known that serial subculture of chondrocytes leads to loss of phenotype (4 -6). In addition, recent evidence has shown that hypoxia promotes the differentiated phenotype (9,10,19) and that the master chondrocyte regulator SOX9 is instrumental in this (11). We therefore applied this doublecomparison analysis (genes down-regulated with passage and subsequently up-regulated in hypoxia) to the present study to ensure that genes thus identified were not simply hypoxiainducible but were also potentially relevant to the differentiated cell phenotype. This is in contrast to the single-comparison analyses that are normally made when adopting a microarray approach, e.g. diseased versus"normal" (21,22) or differentiated versus dedifferentiated chondrocytes (23). The inclusion of two time points (1 and 4 days) and the use of cells from seven different patients further strengthened the analysis and gave a comprehensive data set based on 35 genome-wide arrays. The established and cartilage-specific matrix genes were thus identified as hypoxia-inducible (i.e. aggrecan, Col2a1, Col9a1, Col11a2), in addition to transcription factor SOX9, which controls expression of these genes (11,16,18). The cluster analysis did reveal variations in the degree of this response between samples, which most likely reflects the fact that each sample represents cells from each of seven individual patients rather than, for example, a  single cell line with replicates. Therefore these findings confirmed the validity of our approach of using hypoxia (1% oxygen) to restore the differentiated human articular chondrocyte phenotype. Because cartilage is avascular, the chondrocytes are exposed to low oxygen levels in vivo. A gradation of oxygen tension from 2 to 7% has been measured across the epiphyseal plate in rats and rabbits (7), whereas lower tensions would be expected in (thicker) human cartilage. Our results highlight the importance of studying chondrocyte function in such physiological levels of oxygen (i.e. hypoxia) compared with standard in vitro conditions (20% oxygen).
Using RNA interference, we provide the first evidence that Chm-I and Gdf10 are SOX9-regulated genes. In fact, their hypoxic induction was abolished by SOX9 depletion. Hypoxic induction of matrix protein Prelp was also reduced in SOX9-depeleted cells, although not to a statistically significantly degree (data not shown). ChM-I is a relatively cartilage-specific secreted protein that has been shown to have anti-angiogenic properties (24). Therefore, our current finding that it is hypoxia-inducible and that this is highly SOX9-dependent offers an entirely plausible mechanism for its regulation because the permanent articular cartilage, with its primary function of withstanding high mechanical loads, cannot afford a delicate blood supply and therefore must constantly inhibit vascular invasion throughout life.
Although various growth factors (including transforming growth factor ␤1 (TGF-␤1), TGF-␤3, bone morphogenetic protein, and fibroblast growth factor family members) have been used in different human chondrocyte studies (25)(26)(27), there is little evidence that these molecules are actually endogenously produced in significant amounts by adult human articular cartilage. In the current study we identified two TGF-␤ superfamily members, GDF10 and INHBA. The latter dimerizes to form activin-A. In a recent proteomic study, colleagues in our institute have identified activin-A as a major secreted factor from human cartilage explants (28). They further demonstrated that activin-A induces tissue inhibitor of matrix metalloproteinases-1 (TIMP-1) in human chondrocytes. Activin-A may therefore act as an anti-catabolic factor in cartilage, because TIMP-1 inhibits the matrix-degrading enzymes matrix metalloproteinases (MMPs). Indeed, Chang et al. (29) recently reported that activin-A inhibits MMP-3 in human chondrosarcoma cells and conclude that activin-A may have therapeutic potential as a chondroprotective agent. The present study provides the first evidence that this endogenously produced factor is hypoxiainducible and furthermore that this induction is independent of SOX9.
MIG6 was also identified as a potential chondrocyte regulator and was found to be very highly expressed in primary human chondrocytes. Although its mechanism of action in cartilage is not well defined, its importance to maintenance of the tissue is shown by the fact that deletion of the gene in mice leads to early onset joint disease with marked cartilage  Fig. 4B. C, Western blot showing MIG6 protein was strongly hypoxia-inducible and this induction was greatly reduced specifically by depletion of HIF-2␣, but not HIF-1␣ or SOX9. D, hypoxic induction of Mig6 mRNA was similarly specifically HIF-2␣-dependent. Data from n ϭ four cultures (cells from four patients). E, hypoxic induction of InhbA was HIF-1␣-and HIF-2␣-dependent, but SOX9-independent (n ϭ 4). For gene expression data, transcripts were analyzed by real-time PCR, normalizing expression levels to endogenous control RPLP0. Values are means Ϯ S.E. ***, p Ͻ 0.001 versus siLuc in 1% oxygen.
degradation (30). Studies in different cell lines have shown MIG6 can be induced by stress stimuli such as mechanical/ osmotic stress (31) and hypoxia (32). The present study is the first to show its hypoxic induction in chondrocytes. RNA interference experiments further established this occurred specifically through HIF-2␣, not HIF-1␣. Hypoxic induction of COL2A1, the main cartilage matrix protein, specifically involves HIF-2␣, which increases COL2A1 levels via SOX9 (11). In contrast, HIF-2␣-mediated hypoxic induction of MIG6 does not require SOX9. In addition, because MIG6 depletion did not affect SOX9 levels (data not shown), it appears hypoxic induction of MIG6, like activin-A, represents a SOX9-independent pathway in chondrocytes important for cartilage homeostasis. In addition to Sox9, Mef2c was unexpectedly identified by our analysis. Interestingly, similarly to Sox9, in a range of mouse tissues we found Mef2c predominantly expressed in both heart muscle and cartilage (data not shown). Furthermore, Mef2c clustered with cartilage-specific genes Col2a1 and aggrecan in our microarray analysis. Although well established as a key transcription factor involved in muscle differentiation, MEF2C was recently shown to be required for craniofacial development (33) and to play a role in chondrocyte hypertrophy (34). This raises the intriguing possibility that this classic muscle transcription factor plays a role in articular cartilage, and further work in this area is warranted.
We have recently demonstrated that hypoxic induction of SOX9 specifically requires HIF-2␣ (11). In the present study, in addition to uncovering new SOX9-regulated genes (Gdf10 and Chm-I), we identify Mig6 and InhbA as HIF-2␣-regulated, but SOX9-independent, genes (summarized in Fig. 6). Therefore, although SOX9 is crucial, hypoxia, and more specifically HIF-2␣, also promotes SOX9-independent factors important for cartilage function. This raises the exciting possibility that manipulation of HIF-2␣, rather that SOX9, may better promote the chondrocyte phenotype and, hence, cartilage repair. Future studies should address this important and clinically relevant issue.