Candidate Gene Association Study of Magnetic Resonance Imaging-based Hip Osteoarthritis (OA): Evidence for COL9A2 Gene as a Common Predisposing Factor for Hip OA and Lumbar Disc Degeneration

MATERIALS AND METHODS

The study sample was the part of the population-based Twin Spine Study material⁴ that has been investigated by MRI. It consists of 588 Finnish monozygotic (MZ) and dizygotic (DZ) male twins aged 35–70 years. The studied phenotype was MRI-based calcified structure changes also visible in radiographs. Imaging of the hips was done using Siemens Vision 1.5 T scanning, both hips at the same time, using a body matrix coil. Analysis of the scans was done using Siemens MRI workstation at the Puijon Magneetti Oy company, Kuopio, Finland. The grading scale was based on Li, et al⁵ and the Kellgren-Lawrence radiological assessment of OA⁶: Individuals with 2 normal MRI images were considered as healthy, and individuals with at least small local osteophytes in caput or acetabulum in 1 hip were considered affected, yielding a phenotype allowing for early-onset OA (Table 1).

Polychoric correlations for MZ and DZ twins and heritability using quantitative genetic models allowing for additive genetic, common, and unique environmental variance in the liability were estimated using Mx software (M. Neale, Virginia Commonwealth University; available from: http://www.vcu.edu/mx/). Lifestyle factors (body mass index, mean age, smoking pack-years, and exercise frequency) were compared between the healthy and affected subjects (Mann-Whitney U test).

Genotyping and ethical approval is described in Videman, et al⁴. One individual from each MZ twin pair was used in the SNP association analysis (n = 345, 216 unrelated). Evidence for association of the studied SNP was evaluated using the Pseudomarker program that can use various pedigree structures including combined data of individuals and sibling pairs. The program is able to separate the linkage evidence from evidence for association. The test statistic for association linkage was used with dominant and recessive modes⁷. Association was also calculated for unrelated individuals using Fisher’s exact test and the size of genetic effect for unrelated individuals was analyzed. The Haploview program⁸ was used to calculate the linkage disequilibrium (LD) between the markers.

To define variation at the codon level, we sequenced the flanking region of SNP rs7533552 in all samples (ABI3730xl DNA Analyzer, BigDye Terminator v3.1 Cycle Sequencing Kit, Life Technologies Corp., Carlsbad, CA, USA). Genotype call rate for rs7533552 was 97.5% and Hardy-Weinberg equilibrium was 0.826.

To correct for multiple testing, the SNP spectral decomposition (SpD) method⁹ with modifications by Li and Ji¹⁰ was used to calculate the p value threshold for 5% significance (0.00073; Genetic Epidemiology Laboratory, Queensland Institute of Medical Research; available from: http://genepi.qimr.edu.au/general/daleN/SNPSpD).

RESULTS

The differences in lifestyle factors were nonsignificant between the healthy and affected subjects (Table 1). The polychoric correlation for hip OA was 0.71 (SE 0.11) for MZ twins and 0.40 (SE 0.13) for DZ twins. A quantitative genetic model allowing for additive genetic and unique environmental variance estimated the heritability of hip OA to be 0.72 (95% CI 0.49–0.87).

One variant in the COL9A2 gene (rs7533552; location chr1:40545736, hg18, second base in the codon for Gln³²⁶), which has previously shown association with disc bulging in the lumbar spine (L1 to L4)⁴, also showed association with hip OA (p = 0.0025; Table 2). The G allele was the predisposing allele in both diseases, with a frequency of 30.8% in the hip OA cases and 17.5% in the healthy subjects. The G allele changes a polar negatively charged glutamine (Gln³²⁶) to a polar positively charged arginine (Arg). The same amino acid will be further changed to a nonpolar hydrophobic tryptophan (Trp), if the rare SNP in the first position of the same codon is a T allele (rs12077871; location chr1:40545737, hg18; Figure 1). Since the genotyping method used was only designed to monitor variation at rs7533552, we further sequenced the locus to define the genotypes of both the variants in order to reveal the carried amino acid for each individual. Eleven individuals carried a Trp allele as a combination of C/T heterozygote for SNP rs12077871 (first base of the codon) together with A/G heterozygote or G/G homozygote for SNP rs7533552 (second base of the codon; Table 3). This indicated that the minor alleles T and G of the 2 studied SNP were inherited together (D’ LD = 1.00, r² LD = 0.04, calculated using the Haploview program)⁸.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Single-nucleotide polymorphisms in the amino acid Gln³²⁶.

The size of genetic effect for unrelated individuals was analyzed by comparing amino acid counts of the wild-type Gln allele with the predisposing Arg or Trp alleles between hip OA cases (45 predisposing alleles, 101 wild-type alleles) and controls (50 predisposing alleles, 236 wild-type alleles). The disease odds ratio for carrying either of the predisposing variant alleles Arg or Trp versus the wild-type allele Gln was 2.10 (95% CI 1.66–2.67; p = 0.0021, Fisher’s exact test).

Also, 3 SNP in LD with each other (r² > 0.63, D’ > 0.83) in the COL10A1 gene showed association with hip OA in the present study (p = 0.0015) but were not associated with any of the lumbar disc degeneration traits in our previous study (Table 2).

DISCUSSION

We monitored for association between hip OA and 99 SNP that were previously studied with MRI-based lumbar disc degeneration phenotypes⁴. One SNP (rs7533552, also known as rs2228564) in the COL9A2 gene was found to be associated with hip OA (p = 0.0025) and also with greater disc bulging in the L1-L4 lumbar region (p = 0.036; p value for the within-family component of multivariate regression analysis of disc bulging in L1-L4 discs using the quantitative transmission disequilibrium test program version 2.4.3, calculated with 1000 Monte Carlo permutations for correction of the skewed phenotype)⁴. It is possible that either one of the observed SNP associations is affected by the other trait, or that the SNP independently affects both traits.

Hip OA has previously been shown to associate with growth differentiation factor 5 (GDF5) with genome-wide significance (p = 1.8 × 10⁻¹³)^11,12, while variants in SMAD family member 3¹³ (SMAD3), acidic (leucine-rich) nuclear phosphoprotein 32 family member A¹⁴ (ANP32A), deiodinase, iodothyronine type II¹⁵ (DIO2), and frizzled-related protein^16,17 (FRZP) genes have shown association in replicated studies. Our finding is, to our knowledge, the first to suggest that the change from Gln³²⁶ to Arg or Trp in COL9A2 gene predisposes to hip OA. Neither of the 2 SNP in the codon nor any SNP in high LD with them in European populations were represented on the Illumina HumanHap 610 genome-wide chip used in the genome-wide association studies. The Gln³²⁶ to Arg change has previously been reported to be more frequent in patients with severe and chronic back pain surgically treated for intervertebral disc herniation, although the association did not reach statistical significance¹⁸. The Gln³²⁶ to Trp change (“Trp2 allele”) has been shown to be associated with unilateral back and radiating leg pain¹⁹ and with degenerative lumbar spinal stenosis²⁰.

COL9A2 encodes one of the 3 alpha chains of type IX collagen, having an important role in load-bearing cartilages stabilizing the collagen fibrils, usually found in tissues containing type II collagen. Lack of type IX collagen can increase the susceptibility of the matrix to mechanical damage²¹. According to the PolyPhen protein structure prediction program, the change from glutamine to arginine is predicted not to cause major structural damage for the protein, but the change from glutamine to tryptophan is predicted as possibly damaging (Division of Genetics, Department of Medicine, Brigham and Women’s Hospital – Harvard Medical School, Cambridge, MA, USA; available from: http://genetics.bwh.harvard.edu/pph/).

Association was also observed with the COL10A1 gene that encodes 3 alpha 1 chains in type X collagen homotrimer. It is expressed by hypertrophic chondrocytes in a correlation with endochondral ossification and thus can be considered as a marker for new bone formation in articular cartilage²². Additionally, Boos, et al²³ found type X collagen in osteoarthritic cartilage of the hip when absent from normal adult cartilage.

We focused on a strictly narrowed measurable phenotype instead of on symptoms such as joint pain potentially caused by various biological mechanisms. The phenotype for bone changes was selected because of its known reliability/validity in OA grading on the basis of radiographic studies, but the lack of information on measurement reliability is a limitation. The high MZ correlation suggests that reliability must be high. The age range used also allows cases of early-onset hip OA to be included in the group of affected individuals. However, including the younger, seemingly healthy individuals, who may later develop the disease, in the group of unaffected individuals could decrease the power of detecting a true association. Also, the associations did not exceed the 5% significance level (p = 0.00073, according to the SNP SpD method)⁹ when multiple testing was taken into account, and thus the results need to be interpreted carefully. Analysis in larger materials and functional studies is needed to validate the finding by excluding the possibility of a false-positive finding.

Ninety-nine SNP were analyzed in a population-based male twin cohort to study if there were common variants affecting hip OA and lumbar disc degeneration. One SNP (rs7533552) changing Gln³²⁶ to Arg or Trp in the COL9A2 gene predisposed to hip OA and lumbar disc degeneration, making type IX collagen an interesting target for future studies.