A genetic risk score composed of rheumatoid arthritis risk alleles, HLA-DRB1 haplotypes, and response to TNFi therapy – results from a Swedish cohort study

By linking clinical data from the Swedish Rheumatology Register to genetic data from the EIRA study, we evaluated whether known RA susceptibility genes, HLA-DRB1 amino acids, and haplotypes, either individually or integrated into a risk score, predicted treatment response to TNFi therapy. Although all were strong predictors of RA, a high genetic risk score was not associated with good EULAR response in either overall RA or stratified by ACPA status.

Our results are in line with, but also extend, the previous GWAS and candidate gene approach findings, and indicate that although treatment response in RA has been reported to be somewhat heritable [35], the genetic variants that influence disease onset do not necessarily influence TNFi treatment response to the same extent. This is also supported by a recent study showing that family history of RA did not predict RA TNFi treatment response [27], although this on its own could have been due to the information on genetic risk contained in a family history being too slight among patients who are all at high genetic risk, as evident from them having already developed the disease.

The SNP that was found to be significantly associated with both RA risk and TNFi treatment response (PTPRC/CD45 rs10919563 at chromosome 1) in a previous larger sample [25] could unfortunately not be replicated in our study because of lack of proxies (the closest possible SNP to the PTPRC region available in our material is 100 kbp distant). Even though weak yet nominally significant associations were indeed identified on a handful of individual genetic markers in the current study, they were neither strong enough to withstand correction for multiple testing nor close enough to be clinically informative. Unfortunately, this has also been the case in previous studies. For example, the first TNFi treatment response GWAS performed in 89 RA patients by Liu et al. [5] provided a reference list of 16 candidate SNPs with suggestive significance; none was replicated in a subsequent separate study with slightly larger samples (n?=?151) [36]. Plant et al. [6] performed a multistage GWAS in 1285 RA patients and found seven genetic loci that might influence treatment response; none survived the two additional replication attempts [4, 7]. Similarly, the majority of the markers identified by Krintel et al. [4] in 196 Danish RA patients did not withstand replication in another 315 Spanish subjects [37]. However, one SNP (rs3794271) at PDE3A-SLCO1C1 reached genome-wide significance in the meta-analysis combining the Danish and Spanish cohorts. Further investigations are needed for this pharmacogenetics biomarker of interest. GWASs performed after Liu et al. attempted to both identify new predictors and replicate previous findings, neither of which succeeded. This may be due to the small sample size for GWASs, where often 2–5 million SNPs were analyzed with an average sample size less than 1000. We attempted to increase power by combining many SNPs into a single score, yet still failed to reveal any significant associations. We considered that there may still be a genetic overlap among the as yet unidentified RA alleles, but found no evidence for this because we did not observe any apparent increment of variance explained in any of the disease activity measures when a genome-wide polygenic risk score model was performed. This might indicate that RA risk alleles do not necessarily have an effect on TNFi treatment, despite most of the current treatment target genes/pathways being involved in general inflammatory response; it may be reasonable to expect more and different biological pathways via which TNFi exerts its effect. In light of the limited statistical power in the current study, however, RA risk SNPs with modest effects in TNFi treatment response cannot be ruled out. One nominally significant finding may deserve mention: rs629326 (located 23.61 kb 5? of TAGAP, involved in T-cell activation) was fairly strongly associated with EULAR response in our material, with OR?=?0.31 (0.15–0.62), and FDR-adjusted (for 76 tests) p?=?0.08. Further, the HLA-DRB1 SE alleles were associated with reduced EULAR response when restricting the time window to 2–5 months (OR?=?0.69 (0.49–0.97), unadjusted p?=?0.0347), although this was a post-hoc analysis and the p value would not remain significant after adjusting for multiple testing. Interestingly, specific HLA-DRB1 amino acids have been associated previously with TNFi treatment response, where valine at amino acid position 11 was reported to be associated with a smaller change in Larsen score, and improved EULAR response [24]. We unfortunately lacked data on joint erosions. Our results for treatment response, however, did not immediately support the finding that the valine-containing VKA haplotype was associated with a good EULAR response (OR in [24]?=?1.23 (1.06–1.43)), while the OR was 1.06 (0.58–1.94) for our data among ACPA-positive RA patients. The confidence intervals overlap greatly, and we are thus unable to either refute or confirm this previous finding. We did find a nominally significant association of residue serine (OR?=?1.96 (1.14–3.37)) at position 13 and tightly linked with valine at position 11. This should be interpreted with caution, however, because it was not statistically significant after correcting for the multiplicity of tests. When the haplotype analyses were restricted to patients with high disease activity RA (baseline DAS??5.1), to maximize comparability with the UK study, the association with valine further diminished (Additional file 11: Table S10).

When addressing continuous measures of treatment response, interestingly, all of the risk scores seemed to consistently explain a small yet significant proportion (5%) of the variance in HAQ changes in ACPA-positive RA. The clinical value of such results warrants further confirmation. It has been suggested that the modest influence of genetic effects on treatment response is due to the “composite” traits of the DAS28 score that most outcomes are based upon, and which relies on information from both subjective and objective measures [26]; and that using a well-defined phenotype could aid in disclosing the true genetic effects [26]. We examined each component of the DAS28 without seeing any remarkable associations, except for a small proportion of variance in ESR change explained by amino acid risk score in ACPA-negative RA, which suffered markedly from limited power.

One strength of the current study is the accuracy of both exposures and outcomes. The genetic data were genotyped and quality controlled, and the amino acid data, although imputed, presented a high concordance rate as compared with true genotyping data (a 97.3% concordance rate for two-digit SE and 95.0% for four-digit SE) [30]. We used clinically relevant outcome measures recommended by the European guidelines collected as part of clinical practice in an unselected manner. Because no environmental confounders could in practice influence the genetic markers, we performed the analyses without including adjustment for many covariates to preserve power. One limitation is insufficient power, which could cause false positive findings, and we consider the overall lack of association to be the main conclusion of this study. Another weakness of this study is the time point at which clinical response to TNFi therapy is determined. In a highly heterogeneous disease like RA, time of clinical assessment is crucial and can substantially modify response classification. However, our register-based data could not provide evaluation at time points as exact as those usually obtained in controlled trials, but rather in a time window, reflecting the Swedish clinical practice. We performed a post-hoc sensitive analysis restricting the time window for the evaluation visit to 2–5 months, and the results remained the same.