Genome wide association studies (GWAS) have heralded a major breakthrough in the search for genes underlying rheumatoid arthritis (RA). Indeed, GWAS in UK, US and Swedish populations have identified a number of confirmed RA loci (1). Recently, a modestly-sized GWAS was undertaken in a Spanish population comprising 400 RA patients and 400 controls (2). Association with the KLF12 gene was detected (p = 6.5 x 10-6) and subsequently replicated in an independent cohort of 410 Spanish RA patients and 394 controls (p = 0.01). The most strongly associated single nucleotide polymorphism (SNP) was rs1324913 mapping to intron 1 of the KLF12 gene. The KLF12 protein is a repressor for a transcriptional regulatory factor, AP-2á, which has been implicated in the control of inflammation. Other SNPs at this locus also showed weak evidence for association with RA in UK (rs1884596, p = 0.005) and US (rs9318225, p = 0.002) GWAS (3;4) lending support to the candidacy of this locus; therefore, we tested association to SNPs at this locus in a large independent cohort of UK RA cases and controls.
Two SNPs, rs1887346 and rs9565072 mapping to the KLF12 gene, were tested. The former was selected because it showed weak evidence for association in the UK GWAS, whilst the latter is perfectly correlated (r2=1, HapMap CEU population) with rs1324913, the SNP reported to be most associated with RA in independent Spanish cohorts but which failed assay design. The SNPs were genotyped using the Sequenom platform after imposing quality control thresholds of >90% SNP and sample success rates and no significant deviation from Hardy-Weinberg expectations (p>0.05). However, in an available cohort of 3962 cases and 3531 controls recruited from throughout the UK as reported previously (5;6), no evidence for association with either SNP was detected at a significance threshold of p < 0.025, corrected for multiple testing (Table 1). Similarly, stratification analysis by gender and autoantibody status revealed no evidence for association (data not shown).
It is well-recognised that individual GWA studies have limited power to replicate association at all truly associated loci meaning that failure to detect association in independent cohorts does not necessarily mean that the association is not real. However, the sample size used in the current study had >95% power to detect an effect size of 0.77 reported for rs1324913 intron 1 KLF12 SNP in the Spanish cohort at the 1% significance level. Effect sizes can be over-estimated in the first study to report an association (winner’s curse); hence power calculations were made based on the Spanish validation cohort effect size.
The KLF12 gene is large (~448Kb) and contains several linkage disequilibrium (LD) blocks. Although meta-analysis incorporating data from UK, US and Swedish populations did not highlight this locus as showing evidence for association (7), different SNPs showed weak association in the UK and US GWA series raising the possibility that the true effect may be due to a SNP not directly genotyped by any of these studies but in LD with the genotyped variants. if the extent of the LD varied between populations, this may explain the different associations reported. However, imputed data for all HapMap phase 2 SNPs across this locus from the UK GWAS (433 SNPs, 81% coverage at r2 > 0.8) does not reveal increased evidence for association with any other SNP in the region (data not shown).
In summary, in a large UK RA cohort, we have failed to add to the evidence for association at the KLF12 locus, reported to be associated with RA in independent Spanish cohorts. Analysis in other large cohorts and subsequent meta-analysis will be required before association at this locus can be confidently excluded.