Functional characterisation of the osteoarthritis susceptibility locus at chromosome 6q14.1 marked by the polymorphism rs9350591

The Broad Institute [10] online software was used to conduct a search of SNPs in high linkage disequilibrium (LD; r ²  ≥ 0.8) with the association SNP, rs9350591. The RegulomeDB [11] online database was used to explore the functionality of the polymorphisms.

Ethical approval was granted by the Newcastle and North Tyneside research ethics committee (REC reference number 09/H0906/72) to collect tissue from two groups of donors: those with primary OA who had undergone elective total joint replacement of the hip (total hip replacement; THR) or knee (total knee replacement; TKR), and those that had undergone a THR a result of a neck-of-femur (NOF) fracture. The cartilage of the OA patients had visible lesions and these patients were screened to exclude OA due to trauma or other pathologies. The NOF patients showed no signs or symptoms of hip OA, with the cartilage being macroscopically intact and with no lesions. The cartilage was collected from the tibial plateau and the lateral and medial femoral condyles. For the OA patients, the cartilage was collected at sites distal to the OA lesion. Informed written consent for tissue use and data publication was provided by all donors.

Cartilage specimens were snap-frozen at −80 °C on the day of surgery. The cartilage was ground to a powder under liquid nitrogen and genomic DNA and total RNA were extracted using an E.Z.N.A.® DNA/RNA Isolation Kit (Omega Bio-Tek, Georgia, USA), following the manufacturer’s guidelines. Nucleic acid was quantified and cDNA was synthesised as previously described [12].

Custom-designed, quality-controlled PrimeTime® TaqMan primers and probes (Integrated DNA Technologies, USA; Additional file 1) were used in real-time qPCR to measure the expression of COL12A1, COX7A2, TMEM30A, FILIP1, SENP6 and MYO6 in OA and NOF donor cartilage. Patient details can be found in Additional file 2 and n numbers can be found in Additional file 3. Reactions were performed in triplicate on an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems, UK), and the data normalised to that of the housekeeping genes 18S, GAPDH and HPRT1 using the 2^-ΔCt method. A Mann–Whitney U test was performed to assess if gene expression significantly differed between OA and NOF cohorts, and between rs9350591 genotypes. Given the rarity of T allele homozygotes, gene expression was compared between CC homozygotes and T carriers only.

Following a search of publicly available databases for known transcript SNPs, we selected two transcript SNPs for allelic expression imbalance analysis per gene for COL12A1, SENP6, and MYO6, and one transcript SNP for TMEM30A (Additional file 4). We excluded COX7A2 as there were no known transcript SNPs, and FILIP1 because the transcript SNPs had low minor allele frequencies (MAF; < 5 %).

rs9350591 was genotyped by restriction fragment length polymorphism (RFLP) analysis. All transcript SNP genotypes were determined by pyrosequencing. The primer sequences and rs9350591 restriction enzyme used are listed in Additional file 5.

We used pyrosequencing to assess allelic expression imbalance (AEI) as previously described [8]. Patient details can be found in Additional file 2 and n numbers can be found in Additional file 3. Each pyrosequencing assay was validated using known artificial allelic ratios prior to use in genotyping and AEI analysis. We performed three technical replicates, normalising the cDNA allelic ratios to the mean of the corresponding genomic DNA ratios. A Mann–Whitney U test was performed to assess the association between AEI ratios and genotype at rs9350591.

MSCs from three young human donors were purchased from Lonza, UK. MSCs from three OA patients were harvested from femoral neck aspirates following total hip replacements. Briefly, trabecular bone was extracted and passed through a 100 μm cell strainer in a solution of Dulbecco’s PBS (Life Technologies, UK), 100 U/ml penicillin and 100 μg/ml streptomycin. The cell mixture was layered over 10 ml Ficoll-Paque (GE Healthcare, UK) and centrifuged at 800 g for 40 min at room temperature. The buffy coat, containing mononuclear cells, was washed in a solution of Dulbecco’s PBS containing 5 mM EDTA, 0.2 % BSA, 100 U/ml penicillin and 100 μg/ml streptomycin. Cells were pelleted by centrifugation at 200 g for 10 min at room temperature. The wash and centrifugation were repeated once again. The cell pellet was resuspended in Dulbecco’s PBS, 100 U/ml penicillin and 100 μg/ml streptomycin. Cells were cultured in a 25 cm² cell culture flask (Sigma-Aldrich, UK) for 24 h before the media, supplemented with 8 ng/ml bFGF (Millipore, UK), was replenished. Additional file 6 provides details regarding the Lonza donors and OA patients from whom MSCs were derived.

Chondrogenesis was induced by culturing the MSCs in a cocktail of media and additives including 10 ng/ml TGF-β3 and 100 nM dexamethasone in a well-established in vitro differentiation model [13]. In this model, by day 14 the MSCs have created a cartilaginous disc with an extensive extracellular matrix. RNA was extracted at days 3, 7 and 14 using TRIzol® Reagent (Life Technologies, UK) following the manufacturer’s guidelines and cDNA synthesised as described above.

Although there are no publicly available bioinformatics data pertaining specifically to cartilage, we explored online databases to interrogate the region in related cell lines. We found evidence of functionality, including transcription factor binding and regions with regulatory activity, for SNPs in perfect or high LD with rs9350591 (Additional file 7). All of the 39 SNPs with an r ² of 0.8 or above with rs9350591 are intronic or intergenic, supporting our hypothesis that rs9350591 marks a cis-eQTL that accounts for the OA signal at the 6q14.1 locus.

We quantified COL12A1, COX7A2, TMEM30A, FILIP1, SENP6, and MYO6 and compared the expression levels between OA cartilage (knee and hip OA patients combined; n ≥ 68) and NOF cartilage (n ≥ 14; Additional file 8). This revealed that the expressions of COL12A1 (p = 0.011) and COX7A2 (p = 0.03) were significantly increased and the expression of SENP6 (p = 0.001) significantly decreased in OA. As rs9350591 was identified as being associated with OA in the hip stratum of the arcOGEN study, we next compared gene expression between OA hip and OA knee (data not shown). We observed a reduced expression of MYO6 (p = 0.047) in OA hip relative to OA knee. Finally, we removed the OA knee samples from the analyses and directly compared OA hip (n ≥ 19) and NOF (n ≥ 14; Fig. 1). The expressions of both SENP6 (p = 0.005) and MYO6 (p = 0.026) were reduced in OA hip relative to NOF.

As highlighted above, all of the SNPs that can account for the association signal—that is, those that are in an LD of ≥ 0.8 with rs9350591—are intronic or intergenic and we can therefore discount the action of a non-synonymous SNP altering protein structure or function. This implies that rs9350591 marks a functional role for one of the 39 polymorphisms in the form of a cis-eQTL. The fact that the SNPs are in perfect or high LD with the association signal (all have a D’ of 1 with rs9350591; Additional file 7) negates the need to genotype all of the polymorphisms.

To assess if the OA association was acting as an eQTL on one or more of the six genes, we therefore quantified overall levels of gene expression in OA cartilage (knee and hip OA patients combined) and stratified the data by rs9350591 genotype. Of the 74 OA donors whom we assayed, only two were homozygous for the minor T allele of rs9350951. This is in close agreement with the calculated HapMap CEU rs9350591 TT genotype frequency of 1.7 %. For a robust analysis, we therefore grouped all T allele carriers (n ≥ 23), that is, minor allele homozygotes (TT) and heterozygotes (CT), and compared these to major allele homozygotes (CC; n ≥ 45). We observed no significant differences in cartilage gene expression that correlated with rs9350591 genotype (Additional file 9).

Again, we removed the OA knee data to allow for the comparison of OA hip donors only (CC n ≥ 12; T carriers n = 7; Fig. 2). We observed no significant differences that correlated with rs9350591 genotype. To corroborate these findings, we replicated the analysis in an independent cohort of 25 additional OA hip donors (CC n ≥ 7; T carriers n ≥ 4; Fig. 3), and again observed no significant correlations of expression with rs9350591 genotype for any gene.

Correlating genotype with overall gene expression is vulnerable to the natural fluctuation in gene expression, such that subtle genotypic effects on gene expression may be hidden by inter-individual variation in gene expression, generating false-negative results. Directly testing for allelic expression differences within an individual can overcome this. We therefore identified patients who were heterozygous for transcript SNPs, measured allelic expression by pyrosequencing in cartilage from these individuals and then stratified this data by genotype at rs9350591. If this association SNP correlates with gene expression, then AEI would be observed in compound heterozygotes, i.e. individuals heterozygous for both the transcript SNP and rs9350591, and the AEI ratios would be significantly different to any AEI ratios observed in individuals homozygous for rs9350591 (Additional file 10). We were able to design and validate AEI assays for transcript SNPs within COL12A1 (rs594012 and rs240736), TMEM30A (rs41269315), SENP6 (rs71561434 and rs17414687) and MYO6 (rs1045758 and rs699186). We were unable to perform AEI analysis for COX7A2, which has no known transcript SNPs, or for FILIP1, whose transcript SNPs have MAFs < 5 %, which precluded the identification of a sufficiently large number of heterozygotes to do a meaningful analysis upon. As we aimed to investigate the effect of rs9350591 genotype specifically on allelic output, it was not necessary to stratify the data by disease state or joint site; however to ensure a thorough investigation, we have included this analysis in Additional file 11. None of the patients used in this AEI analysis were TT homozygotes at rs9350591 and our analysis of each of the four genes therefore involved a comparison of patients who were all heterozygous at a transcript SNP for that gene and either homozygous CC (n ≥ 7) or heterozygous CT (n ≥ 7) at rs9350591. We identified a spread of allelic expression ratios for all four genes (Fig. 4), which therefore supported the activity of eQTLs operating at this locus. The AEI was greatest for MYO6, with allelic ratios in excess of 1.5 for some individuals. However, these eQTL activities were observed in cartilage from individuals homozygous for rs9350591 as well as heterozygotes, and as such are not dependent on genotype at the association SNP. There were no significant differences (p < 0.05) in the AEI ratios between CC and CT patients for COL12A1, TMEM30A or SENP6. There was a significant difference (p = 0.019; Additional file 11) for MYO6 AEI in OA hip cartilage, which remained significant upon combining the data with OA knee and NOF (p = 0.012; Fig. 4), although this was accounted for by a large allelic ratio spread in those patients homozygous CC for rs9350591. Therefore, the AEI of MYO6 is not correlating with the eQTL marked by rs9350591. Combined with the large number of statistical tests that we have performed, a p of 0.012 would not endure if corrected for multiple testing; this implies that rs9350591 does not correlate with AEI at any of the tested genes.

Our data thus far suggests that the OA association marked by rs9350591 does not correlate with eQTLs operating in the tested genes in the patient cartilage tissues analysed. Although OA is a disease in which age is a predominant risk factor [1], it has been suggested that several OA genetic susceptibility loci may act either during skeletal development or throughout life [14]. The OA association marked by rs9350591 could therefore be exerting its effect on one or more of the nearby genes at an earlier stage of cartilage development. Thus, to assess their potential activity during development, we tracked the expression of the six genes throughout chondrogenesis in a well-established in vitro chondrogenesis model. We studied MSCs from three OA and three non-OA (Lonza) donors (Additional file 6). All six genes were expressed throughout chondrogenesis for both types of donor (Fig. 5). Although there were inter-individual variations in the levels of expression, there were no consistent differences between OA and non-OA. The six genes are therefore active during cartilage formation.

The raw data supporting the findings presented here are available from the corresponding author upon request. They are not stored in a publicly available repository.