Investigation of rheumatoid arthritis susceptibility genes identifies association of AFF3 and CD226 variants with response to anti-tumour necrosis factor treatment

Introduction

Rheumatoid arthritis (RA) is a chronic potentially disabling disease caused by autoimmune destruction of the synovial joints which affects approximately 1% of the Caucasian population.1 The introduction of anti-tumour necrosis factor (anti-TNF) biological therapies has dramatically altered the treatment of RA as they show good efficacy in patients resistant to disease-modifying antirheumatic drugs (DMARDs) and superior efficacy in the suppression of erosive damage compared with standard DMARDs.2 However, there remains a significant non-response rate (in the region of 30–40%). The reasons for this remain largely unknown.3 Furthermore, anti-TNF therapy is associated with expensive annual treatment costs, leading to restrictions in the numbers of patients who may be prescribed these drugs. The identification of predictors of treatment response could potentially reduce the number of non-responding patients, improving the cost-effectiveness of anti-TNF therapies.

Several clinical predictors of response have been determined, including the level of disability at the onset of treatment as measured by the Health Assessment Questionnaire (HAQ) (patients with higher levels of disability at the outset of therapy respond less well); concurrent therapy with DMARDs (co-administration of DMARDs improves response); and the presence of autoantibodies (presence of rheumatoid factor or anticyclic citrullinated peptide antibodies is associated with a poorer response).45 However, even when these factors were combined, they accounted for less than 20% of the variance in response to anti-TNF agents in one study.5

In other complex diseases, polymorphisms in susceptibility genes have been shown to be associated with treatment response. For example, two variants in the established type 2 diabetes (T2D) susceptibility gene TCF7L2 have been shown to influence the response to treatment with sulfonylurea drugs.6 In the current study we hypothesised that polymorphisms known to have a role in susceptibility to RA may also influence the response to anti-TNF treatment.

We have previously investigated—and found no evidence for—an association of the two major RA susceptibility loci: HLA-DRB1 shared epitope alleles and the PTPN22 R620W polymorphism.5 However, with the advent of genome-wide association (GWA) studies, there has recently been enormous progress in the identification of RA susceptibility genes. There are now at least 11 additional loci for which association with RA susceptibility has been confirmed in independent data sets, and the aim of the current study was to test the association of these markers with anti-TNF treatment response.

Methods

Results

A total of 18 SNPs mapping to the 11 loci investigated were selected for genotyping (table 2). These were polymorphisms with previous evidence for association with RA susceptibility including some proxy SNPs in case of assay failure, selected using SNAP.19

The initial test cohort comprised 1092 samples; 80 samples were excluded by the <80% QC measure, leaving 1012 samples available for analysis. One SNP (rs13207033) in TNFAIP3 was excluded from analysis due to <80% genotyping success rate, although a perfect proxy for this variant was successfully genotyped (rs13192841).

Two variants mapping to AFF3 and one to the CD226 locus demonstrated statistically significant evidence for association under an additive model (table 3) (AFF3: rs10865035, allele G coefficient −0.16 (95% CI −0.29 to −0.03), p=0.018; rs1160542, allele G coefficient 0.15 (95% CI 0.02 to 0.29), p=0.022; CD226: rs763361, allele C coefficient 0.16 (95% CI 0.03 to 0.29), p=0.016). rs1160542 served as a proxy (r²=0.97) for rs10865035, so these two associations represent a single effect at AFF3.

A SNP at the STAT4 locus rs7574865 (along with the proxy SNP rs10181656) reached statistical significance under a dominant model but not in the genotypic or additive model (table 3). The association appears to be driven by the reduced response conferred by the heterozygous genotype, suggesting that this association may be a spurious finding.

In order to increase confidence in the association at these three loci, they were genotyped in an additional 338 anti-TNF-treated RA samples which were reduced to 322 samples after the 80% QC measure. The clinical characteristics of this cohort are shown in table 1 and are similar to the initial cohort, allowing the data from both cohorts to be combined for analysis. The results of analysis of the additional samples alone are given in table S1 in the online supplement.

Power calculations performed in QUANTO computer program (2006) showed that, under an additive or a dominant model, the sample size in the combined cohort (n=1334) provided >99% power to detect a difference in DAS28 score of ≥0.6 units (a clinically important change) at minor allele frequencies of ≥0.05.

The SNPs mapping to AFF3 and CD226 remained statistically significantly associated with response under an additive model (AFF3: rs10865035, allele G coefficient −0.14 (95% CI −0.25 to −0.03), p=0.015; CD226: rs763361, allele C coefficient 0.11 (95% CI 0.00 to 0.22), p=0.048) (table 4). However, the association at the STAT4 locus continued to be driven by the heterozygous genotype in the combined data. Since this seems a biologically implausible model for association, we believe that the association at STAT4 probably represents a false positive. In a separate analysis using the European League Against Rheumatism (EULAR) response criteria as the outcome measure, only rs10865035 in AFF3 was associated, with good versus poor response (OR 1.46 (95% CI 1.13 to 1.88), p=0.0036) (see table S2 in online supplement).

It is possible that polymorphisms may have different effects on the treatment response depending on which of the three anti-TNF drugs was used. Despite apparent drug-specific effects upon stratification by drug type, interaction analysis revealed no statistically significant difference in treatment response between the three anti-TNF biological agents for either of the associated SNPs (additive model, AFF3: rs10865035, p=0.26; CD226: rs763361, p=0.51) (see table S3 in online supplement).

Discussion

This investigation of RA susceptibility loci in response to anti-TNF treatment is the largest study of genetic predictors of anti-TNF response performed to date. We have detected nominally significant effects at RA susceptibility variants mapping to the AFF3 and CD226 genes.

The identification of an effect on treatment response conferred by polymorphism within a susceptibility gene is not surprising as there are several examples in the literature where complex disease susceptibility genes encode therapeutic targets. For example, in T2D, the established susceptibility gene PPARG encodes a protein which is a target for the thiazolidinedione drugs.20 In RA the drug abatacept, a CTLA4 analogue, was shown to be an effective therapy before the CTLA4 gene was unequivocally demonstrated as associated with susceptibility to RA.71521 There are now examples of polymorphisms within susceptibility genes that influence response to treatment, such as variation in the T2D susceptibility gene TCF7L2 which predicts response to sulfonylureas.6

However, the association of CD226 and AFF3 variants with the anti-TNF response is weak, and the addition of these markers into predictive models including clinical variables has only a modest effect, increasing the r² value from 15.7% to 17.0%. This is in contrast to the large genetic effects seen in studies such as those of response to warfarin therapy. For example, genetic variants in the two genes VKORC1 and CYP2C9 account for about 40% of the variance in warfarin dose.22 In that case the genes were originally targeted as candidates because they were known to lie on the warfarin metabolic pathway, and it may be argued that major genetic effects on the anti-TNF response might be expected to arise from variation within genes implicated in the TNF pathway. However, previous investigations by our group have failed to detect an association between a number of such genes and treatment response.23 One notable exception is the association between the TNF −308 SNP and the response to etanercept, but not infliximab.24

Even with the warfarin story, although major genetic effects have been identified, inclusion of these into models with clinical variables remains only moderately predictive of warfarin dose required or time to stabilise international normalised ratio.25 A subsequent GWA study confirmed an association with VKORC1 and CYP2C9 loci and identified numerous signals which may represent other loci with smaller effects on warfarin requirements. Incorporation of these smaller effects may be required to develop accurate models of prediction.26 Hence, the weak effects detected in the current study may yet prove to be clinically important when combined with other predictors of response to anti-TNF therapy.

In order to detect subtle effects, studies must be adequately powered. One of the most important strengths of this study is the large sample size employed; almost all similar investigations (with notable exceptions52324) have focused on <500 patients with RA. The current sample size provided very high power (>99%) to detect a change in DAS28 of 0.6 units at minor allele frequency ≥0.05. We are therefore confident in excluding modest effects at the variants that did not demonstrate evidence of association in our study.

The associated rs10865035 SNP maps to the 5′ upstream region of AFF3 located on chromosome 2q11. Interestingly, the SNP was associated not only with change in the DAS28 but also with EULAR response criteria; indeed, for the latter analysis, the association remained statistically significant even after applying a stringent Bonferroni correction (p_c=0.049). The gene, also known as AF4/FMR2, is preferentially expressed on lymphoid cells and encodes a family of transcription factors that are thought to be implicated in the function of the lymphoid system.27 We speculate that variation in AFF3 may lead to an upregulated inflammatory response by lymphocytes, resulting in more circulating proinflammatory molecules and leading to a reduced response to TNF antagonists; further studies will be required to explore this.

The CD226 gene maps to chromosome 18q22 and encodes a type I membrane protein molecule expressed on the surface of haematopoietic cells which is involved in the triggering of both T and NK cell cytotoxicity. The associated variant (rs763361) is a non-synonymous SNP encoding a glycine to serine substitution and carriage of the minor allele has previously been reported to be associated with RA susceptibility.17 Alteration of T and NK cell cytotoxicity could once again lead to greater proinflammatory molecule production, thereby explaining why there is a reduced response to anti-TNF drugs.

A limitation of the current investigation is that no correction for multiple testing was applied and, if it was applied, the associations with change in DAS28 would not remain statistically significant (AFF3, p_c=0.195; CD226, p_c=0.624). It is therefore important that the findings of this study are validated in an independent cohort, but such validation was beyond the scope of the current study. Our strategy was to maximise sample size by genotyping all available DNA samples rather than splitting the cohort into test and confirmatory data sets. It is therefore possible that the associations may have arisen due to a type I error. However, our findings are in keeping with those in other complex diseases, in that susceptibility genes may also influence treatment response.6

In summary, we provide evidence for a weak association between SNPs in the AFF3 and CD226 RA susceptibility loci and response to anti-TNF treatment in patients with RA. The percentage of the variance explained by these genetic markers is only 1.3%. It is too early to say whether the response to anti-TNF treatment will be conferred through a number of genes, each with a small effect size, or whether genes exist that predict a large percentage of variance to treatment. Candidate gene studies have had limited success, however, in identifying predictors. We hypothesise that the response to treatment is polygenic and that well-powered GWA studies should be able to identify a genetic signature to identify those patients most—or, indeed, least—likely to benefit from these expensive but effective therapies.