Association Analysis of Single Nucleotide Polymorphisms in Cartilage-Specific Collagen Genes With Knee and Hip Osteoarthritis in the Japanese Population


  • The authors have no conflict of interest.


Osteoarthritis (OA) is one of the most common diseases in the elderly. Although its pathophysiology is complex and its molecular basis remains to be determined, much evidence suggests that OA has strong genetic determinants. To search for susceptibility loci of OA, we selected seven candidate genes encoding cartilage-specific collagens (type II, IX, X, and XI collagens) and performed association analysis for OA using single nucleotide polymorphisms (SNPs) in the coding region of these genes. Four hundred seventeen OA samples and 280 control samples were collected from the Japanese population, and 12 SNPs were genotyped. Our studies have identified two susceptibility loci of OA: COL2A1 and COL9A3. An SNP in COL9A3 showed significant association with knee OA (p = 0.002, odds ratio [OR] = 1.48). Haplotype analysis showed significant association between a specific haplotype of COL2A1 and hip OA (p = 0.024; OR = 1.30). Further analysis of these two genes will shed light on the molecular mechanisms of OA.


OSTEOARTHRITIS (OA) is a debilitating disease that is defined as the degeneration of various synovial joints such as joints of the hand, spine, hip, and knee. OA may result from autoimmune diseases including rheumatoid arthritis and other polyarthritis, trauma, and infection; however, in most cases, OA is idiopathic. Idiopathic OA is a common disease, affecting >5% of adults.(1)

The etiology of idiopathic OA is unknown. Its pathophysiology is complex, with numerous changes in bone and cartilage metabolism; the only common findings of OA are articular cartilage loss and subchondral bone reaction with age. However, evidence has accumulated showing that OA is not merely an inevitable result of the aging process but instead has a genetic background.(2,3) Identifying susceptibility genes is a promising approach to understanding the molecular basis and pathogenesis of OA.

A number of candidate genes have been proposed as potential susceptibility loci for OA.(4–6) These genes predominantly encode structural proteins of the extracellular matrix. Particular emphasis has been placed on the gene for type II collagen (COL2A1), which is the major cartilage collagen. COL2A1 mutations have been found in various types of chondrodysplasias, including spondyloepiphyseal dysplasia(7,8) and familial OA.(9) These chondrodysplasias have been characterized as monogenic model diseases of idiopathic OA because they show early degenerative changes in multiple joints. Several studies have suggested an association of polymorphisms in COL2A1 with OA.(10–12)

Mutations in genes encoding additional structural proteins of the cartilage extracellular matrix have been found in other types of chondrodysplasias. These include two fibrillar collagens (types IX and XI collagens) and one short collagen (type X collagen). These genes, which encode so-called cartilage-specific collagens, also are candidate genes for OA. Type IX collagen is a fibrillar collagen located on the surface of cartilage collagen fibrils cross-linked with type II collagen. It is a heterotrimeric protein composed of α1-α3(IX)-chains, which are encoded by different genes (COL9A1, COL9A2, and COL9A3).(13) A COL9A1-null mouse shows early degenerative changes in multiple joints,(14) and mutations in COL9A2 and COL9A3 have been found in patients with multiple epiphyseal dysplasia (MED; MIM 600204, 600969). MED is a relatively common chondrodysplasia characterized by mild short stature, flattened irregular epiphyses, and early onset generalized OA that predominantly affects the hip and knee joints.

Type X collagen is a short-chain homotrimeric collagen expressed by hypertrophic chondrocytes in the enchondral growth plates.(15) Although its precise function remains to be determined, it also is reactivated in cartilage of OA patients.(16) Mutations in COL10A1 have been found in Schmid metaphyseal chondrodysplasia (SMCD; MIM156500), one of the most common types of chondrodysplasias. SMCD is characterized clinically by short-limbed short stature and skeletal deformities such as coxa vara and genu varum. Although early degenerative changes of multiple joints usually are not shown in SMCD, its characteristic skeletal deformities are similar to patterns typical of idiopathic OA in hip and knee joints.

Type XI collagen is a fibrillar collagen found in the center of cartilage collagen fibrils associated with type II collagen. It is a heterotrimer composed of α1-3(XI)-chains.(17) The α3(XI)-chain is a post-translational variant of α1(II). α1(XI)- and α2(XI)-chains are encoded by COL11A1 and COL11A2.(18–20) Mutations in COL11A1 and COL11A2 have been found in the Stickler syndrome (MIM 604841,184840), which is characterized by early onset OA of multiple joints with or without ocular involvement.

Given their critical and unique roles in cartilage metabolism, as well as their mutant phenotypes, these genes are strong candidate susceptibility loci for OA. Although a linkage study has suggested COL9A1 as a susceptibility locus for hip OA,(21) no association study has examined relationships between single nucleotide polymorphisms (SNPs) in these genes and OA. We have performed association analysis of SNPs in COL2A1, COL9A1, COL9A2, COL9A3, COL10A1, COL11A1, and COL11A2 with knee and hip OA in the Japanese population.


Selection of SNPs in candidate genes

SNP information for each candidate gene was obtained from published literature,(22–27) the public IMS-JST cSNP database (http//, and our own data obtained through mutation analysis of chondrodysplasias (T. Ikeda and A. Mabuchi, unpublished data, 2002). SNPs used for association analysis (Table 1) were selected from those existing in coding regions (cSNPs), with minor allele frequencies of >10%. A total of 12 cSNPs were selected using these criteria; seven were synonymous and five were nonsynonymous. Three cSNPs resulted in nonconservative amino acid substitutions.

Table Table 1.. Association Analysis of 12 cSNP Loci With OA
original image

Sample collection

Given the heterogeneity of OA, we elected to study two OA phenotypes, hip and knee OA. Diagnosis of hip and knee OA was made by experienced orthopedists based on physical examination followed by roentgenogram. Four hundred seventeen OA patients and 280 controls were selected from the population participating in the “Genetic Study Program of Bone and Joint Disease” at the Tokyo University Hospital and related institutions. All hip and knee OA patients were symptomatic and were treated in these hospitals on a regular basis. The control population was drawn from volunteers undergoing treatment for injury or other orthopedic diseases in the hospitals. Blood samples were obtained from all participants, and genomic DNA was prepared from peripheral leukocytes according to standard protocols. To avoid population admixture, all individuals selected for this study were Japanese living in and around Tokyo. The study protocol was approved by the ethical committees of the participating institutions, and written informed consent was obtained from each participant.

Assessment of OA

After the patients agreed to join the study, they were referred to the specialists of hip and knee diseases and received further detailed clinical and radiographic examination. For each patient with knee OA, standard three-direction knee radiographs (i.e., anteroposterior, lateral, and skyline view) were taken and assessed by a single expert observer. For each patient with hip OA, anteroposterior radiographs were taken and assessed by a second observer. Diagnosis of OA was confirmed when a change of grade 2 or higher was observed using Kellgren's scale,(28) with definite joint space narrowing. For all OA patients, parameters that have been reported to confound the result(2,3,6,12) including family history, body mass index (BMI), and complication of clinical hand OA (Heberden's node) were examined.

Rheumatoid arthritis and polyarthritis associated with autoimmune diseases were excluded, as were posttraumatic OA and infection-induced OA. Patients who had clinical and radiographic findings suggestive of skeletal dysplasias including overt short stature, multiple symmetric involvement of epiphyses, and a positive Mendelian family history were excluded from the study. Control subjects also received detailed clinical and radiographic examination by expert orthopedists.


SNPs were detected by TaqMan assay(29) in 96-well plates after optimization for each primer set (ABI, Foster, CA, USA). The reaction was performed using an ABI GeneAmp polymerase chain reaction (PCR) system 9700 (50°C for 2 minutes, 95°C for 10 minutes, 95°C for 15 s, and 62°C for 1 minute, for 40 cycles) and read using an ABI Prism 7700 sequence detector according to the manufacture's instruction.

Statistical analysis

For each SNP, a χ2 test was performed between the OA and control groups on both genotypic and allelic frequencies. The odds ratio (OR) and 95% CI were calculated with respect to the minor allele compared with the major allele. A likelihood ratio test of linkage disequilibrium was performed for each candidate genotyped using two or more SNPs.(30) Haplotype frequencies also were estimated by the maximum-likelihood method on both OA and control groups, and the difference was calculated by the exact test.(31) Analyses were performed using “Arlequin” software (


Sample population

The OA sample population (n = 417) consisted of 44 men and 373 women, with a mean age of 62.2 ± 14.8 years. Two hundred twenty-eight patients had knee OA, and 189 patients had hip OA. Although 20 of the hip OA patients also had knee OA, it was not their major problem. No knee OA patient had clinically significant hip OA. The knee OA group (71.4 ± 7.1 years) was older than the hip OA group (50.9 ± 13.8 years). Although the control group (61.9 ± 12.1 years) was younger than the knee OA group, the mean age of the control group was comparable with the mean age of onset in the knee OA group (63.1 ± 10.0 years). BMI was slightly higher in the OA group (23.7 ± 3.7) compared with the control (22.7 ± 3.3) but was within a normal range in both populations. Forty-five percent of OA patients had at least one OA patient in their first-degree relatives. Thirty-nine percent of OA patients also had clinical hand OA (Heberden's nodes). The mean Kellgren's radiographic scale was 2.9.

Association analysis

We analyzed 12 cSNPs in seven candidate genes and found a significant association between a cSNP, c.1740C > T of COL9A3, and OA (p = 0.015; OR = 1.31; Table 1). This association was more significant in the knee OA population (p = 0.002; OR = 1.48; Table 2). A nonsynonymous cSNP in exon 51 of COL2A1 showed a marginally significant association (p = 0.093). Other SNPs showed no evidence of association with OA. Stratification by other clinical and radiographic features including family history, complication of hand OA, and Kellgren's radiographic scale didn't improve the result.

Table Table 2.. Association of the c.1740C > T SNP in COL9A3 With Knee OA
original image

Linkage disequilibrium and haplotype analysis

All SNPs located within a single gene were in strong linkage disequilibrium with each other (Table 3). The maximum physical distance of linkage disequilibrium was 23.5 kb in the COL2A1 gene. Haplotype analysis of COL2A1 showed a significant total population difference between hip OA and control groups (Table 4). The “122” haplotype, a combination of the major allele of the exon 5 SNP and the minor alleles of the SNPs in exons 32 and 51 were significantly overrepresented in the OA population (p = 0.024; OR = 1.30), especially in hip OA. The “222” haplotype also was overrepresented in the OA population, but it was not statistically significant.

Table Table 3.. Pairwise Linkage Disequilibrium of COL2A1, COL10A1, and COL9A3
original image
Table Table 4.. Haplotype Analysis of COL2A1 in OA
original image


In this study, we selected seven genes encoding cartilage-specific collagen and performed association analysis of SNPs in these genes with OA. This is the first association analysis of idiopathic OA with genes encoding so-called minor cartilage collagen (COL9A1-3, COL10A1, and COL11A1,2). Our results suggest that COL9A3 and COL2A1 are susceptibility loci for OA in the Japanese population.

Association between a minor haplotype of COL2A1 and idiopathic OA has been reported previously in a white population.(11) In this study, we also have found a risk haplotype for OA in COL2A1. The 122 haplotype was overrepresented in the OA population, specifically in hip OA. The haplotype frequency was 40% in the OA population, much higher than the previously reported risk haplotype frequency (4%). This large difference may reflect a population difference or a difference of the polymorphic loci genotyped in the two studies. The haplotype of the previous study(11) was constructed from the exon 5 SNP, the intron 33 SNP (HindIII RFLP), and the 3′ variable number of tandem repeat (VNTR),(32) and the haplotype of this study was constructed from three cSNPs (exon 5, 32, and 51). Only the exon 5 SNP is common to the two studies. However, the difference between the OA and control groups in this study (6%) is comparable with the difference reported previously (3%). The observation that no single SNP of COL2A1 showed significant association with OA indicates that a true disease-causing polymorphism linked to this risk haplotype is likely to exist. This disease-associated polymorphism may be common to both Japanese and white populations. In the three SNPs genotyped in COL2A1, the exon 51 SNP showed a marginal result (p = 0.093), and the 22 haplotype of the exon 32 and 51 SNPs was associated mainly with OA. This fact implies that the disease-causing variant may exist around exon 51. Although the 3′ VNTR was not associated with OA in previous studies,(11,33) further analysis of polymorphisms around exon 51, including the 3′ VNTR and other coding and intronic SNPs, may reveal the true disease-causing polymorphism.

The exon 30 SNP in COL9A3 also showed significant association with OA, specifically with knee OA. This SNP is synonymous, and no splicing variants are predicted by flanking sequences on either allele; we found no splicing variations in lymphoblast cDNA from five individuals (data not shown). Therefore, this association also would reflect the linkage disequilibrium between the exon 30 polymorphism in COL9A3 and a true disease-causing polymorphism that may exist near exon 30. However, we could not find any other informative cSNPs in COL9A3; the A250E polymorphism reported by Finn(34) was rare in Japanese. Further investigation using the intronic and regulatory SNPs in COL9A3 will be needed to detect the true disease-causing polymorphism and to confirm the importance of COL9A3 on the onset and development of OA.

Although a previous linkage study has identified COL9A1 as a susceptibility locus of OA,(21) we found no association with the COL9A1 locus in this study. Although an association analysis appears to be more efficient than linkage analysis in identifying major susceptibility genes for common diseases(35) and our sample size is comparable with that of the previous linkage study, this conflicting result may be derived from the difference of the statistical methods used in the two studies, that is, linkage and association. It also is possible that our negative result may reflect the different population examined in this study. Alternatively, there may be an additional susceptibility polymorphism not linked to the cSNP examined in this study. However, genotype and haplotype analyses of another intron 1 SNP in COL9A1 also have failed to reveal any association (data not shown). This result indicates that COL9A1 has, at the most, only a minor contributing effect on OA in the Japanese population.

We used only cSNPs for this association analysis because their information was relatively easily obtained from published literature, a public database, and our previous works. However, most of these cSNPs were synonymous or exist in propeptide coding regions, and, hence, they are thought to have no effect on mature protein products. Regulatory and intronic SNPs near exon-intron junctions also may affect gene function through modification of expression levels or creating alternative splice variants. Such SNPs should be included to avoid false negative results. The continued accumulation of data in SNP databases will enable these studies. However, the wide range of linkage disequilibrium in the Japanese population that is apparent in this study may justify association analysis using only cSNPs as markers.

Collecting cases with strong genetic backgrounds is important for raising statistical power in case-control association studies.(36) However, most previous association studies of OA have collected cases from the general population. The prevalence of symptomatic knee OA is only 9.5% of radiographically defined knee OA,(1) and the majority of radiographically detected cases from general population samples are asymptomatic. On the other hand, our study recruited patients undergoing treatment in the hospital, which resulted in a greater percentage of symptomatic and severe cases being included. Therefore, our study subjects are expected to have stronger genetic backgrounds for OA and hence greater statistical power than population-based samples.

Some SNPs showed highly different allelic frequencies compared with a previously reported study in a white population. For example, in the nonsynonymous exon 19 SNP in COL9A2, the threonine allele, which is rare and associated with lumbar disc disease (LDD) in the Finnish population,(25) was common in the Japanese population (0.13 in the OA group and 0.17 in the control group). Because the prevalence of LDD is not predicted to differ greatly between these two populations, it is probable that this SNP is in linkage disequilibrium with a nearby susceptibility locus or the threonine allele may play on a minor role in the general population. On the other hand, estimated haplotype frequencies of COL11A2 shared a similar distribution with previously reported haplotypes in the populations collected from Kagoshima and Aomori,(26,27) which are geographically separate from Tokyo.

In this study, we have used SNP data to identify specific associations between cartilage-specific collagen genes and knee and hip OA, providing additional evidence to support the hereditary basis of OA. We have found that COL9A3 and COL2A1 are susceptibility loci for OA in the Japanese population. Continued genetic analysis of COL2A1 and COL9A3 will help define the molecular mechanisms of OA, leading to improved methods for screening, diagnosis, and treatment.


We thank the patients and doctors who cooperated in this study. We also thank Aya Narita, Masako Ogawa, and Hiromi Kakoi for excellent technical assistance.