Key points

Introduction

Rheumatoid arthritis (RA) is a heterogenous disease and can range from a mild, self-limiting arthritis to rapidly progressive joint damage. Treatment is based on controlling inflammation, and early effective therapy reduces disability, joint damage, and mortality. A range of treatment options are available but none are universally effective, so treatment selection is based on a “trial-and-error” approach, trying different therapies until a drug that induces low disease activity or remission is identified. Time on multiple ineffective medications affects the patient’s quality of life, may lead to irreversible joint damage, exposes the patient to potential adverse events, and is a waste of health care resources. Therefore, considerable research effort has been applied to identifying predictors of drug response to allow more rational prescribing of the drug most likely to be effective in individual patients, an approach known as precision (or stratified) medicine.

Methotrexate (MTX) is the first-line therapy for RA, whereas biologic therapies target specific molecular pathways, including the tumor necrosis factor (TNF), interleukin-6, B-cell and T-cell costimulation pathways. The biologic drugs are typically reserved for those with an inadequate response to nonbiologic disease-modifying antirheumatic drugs, but there is currently no guidance on which biologic agent to use first. Each drug has a significant failure rate; for example, TNF inhibitors (TNFi) are ineffective in up to 30% patients, yet remain the most commonly prescribed first-line biologic. As most research has investigated biomarkers predictive of response to MTX and TNFi biologics, the current review limits the focus to these drug classes.

Treatment response is likely to be multifactorial and influenced by clinical, psychological, and biological factors. For example, robust clinical predictors of TNFi response include disease severity, smoking status, concomitant MTX, and patient disability, but account for a small proportion (r ² = 0.17) of the variance in response and so, alone, are not useful in informing therapy selection decisions. There is, therefore, a need for accurate predictors (biomarkers) of response to RA therapies to enable precision medicine, defined by National Academy of Sciences as the use of genomic, epigenomic, exposure, and other data to define individual patterns of disease, potentially leading to better individual treatment.

The use of genomic variants as predictors of response has several theoretic advantages. Genetic variants are stable and will not change because of the environment, unlike epigenetics or expression profiling. Genetic variants that are associated with response are likely to be involved in key molecular pathways and can therefore provide insight into the mechanisms of nonresponse. Whole-genome genotyping is now economically viable, and the assays are standardized, enabling their use in the clinical setting. Indeed, genetic biomarkers are already being used to personalize health care. In cystic fibrosis, for example, ivacaftor, a drug that targets the CFTR molecule, is recommended in the 4% of patients with the G551D mutation whereas in rheumatology, screening for the enzyme TPMT, responsible for the metabolism of 6-mercaptopurine and related compounds, is recommended to identify the 13% of the population with reduced activity and who are at increased risk of toxicity to azathioprine. There are currently more than 200 examples of US Food and Drug Administration–approved drugs that contain information on genomic biomarkers that may be used to inform treatment decisions. Although many of these are not commonly used in clinical practice, TPMT screening is frequently in the United Kingdom.

Studies investigating genomic predictors of methotrexate

Given that MTX remains the treatment of choice for patients with newly diagnosed RA, several studies have investigated genes involved in the key molecular pathways affecting MTX absorption, metabolism, or its target enzymes as predictive biomarkers of response ( Fig. 1 ).

The MTX metabolic pathway. 5-CH ₃ -THF, 5-methyltetrahydrofolate; 5,10-CH ₂ -THF, 5,10-methyltetrahydrofolate; AMP, adenosine monophosphate; DHFR, dihydrofolate reductase; FH ₂ , dihydrofolate; FH ₄ , tetrahydrofolate; FPGS, folylpolyglutamate synthetase; IMP, inosine monophosphate; MTHRF, methylene tetrahydrofolate reductase; MTXPG, methotrexate polyglutamate; SLC19A1, solute carrier family 19 member 1; THF, tetrahydrofolate.

The most consistent evidence for association is for the solute carrier family 19 member 1 (SLC19A1) gene, one of several transport carriers that allow MTX to enter cells. Studies have reported that the rs1051266 variant associates with intracellular MTX-polyglutamate levels and a recent meta-analysis of 12 studies (n = 2049) reported an association with MTX treatment response (odds ratio [OR] = 1.49 of AA genotype, P = .001). Methylene tetrahydrofolate reductase is another key enzyme in the MTX pathway and has also been extensively investigated with several studies reporting associations with efficacy and toxicity. However, a meta-analysis including 17 previous studies revealed no association with either outcome, and this finding has been replicated in 2 subsequent meta-analyses. MTX is thought to exert an anti-inflammatory effect through inhibiting aminoimidazole carboxamido tibo nucleotide (AICAR) transformylase (ATIC) leading to an increase in AICAR levels and the anti-inflammatory agent adenosine. Several studies have associated the (ATIC 347 C > G single nucleotide polymorphism [SNP] rs2372536) with toxicity, but this finding has not been consistently replicated.

As well as investigating MTX pathway genes, the major RA susceptibility gene, HLA-DRB1, has also been studied. As the gene is associated with more severe disease, it was hypothesized that carriers of the risk allele would be less likely to respond to MTX monotherapy. In a study of 309 patients from an early inflammatory polyarthritis inception cohort, the presence of the HLA-DRB1 allele was associated with MTX monotherapy inefficacy at 2 years (OR = 3.04, P = .02), but this finding requires replication in other data sets.

Studies investigating genomic predictors of response to tumor necrosis factor inhibitor

Early candidate gene studies investigating the pharmacogenomics of TNFi therapy revealed inconsistent findings, none of which have been robustly replicated. This review focuses on genome-wide association studies (GWAS): candidate gene studies where findings have been replicated by at least one group and candidate gene studies performed in sample sizes exceeding 1500 individuals ( Table 1 ).

Whole-genome studies

To date, 5 GWAS have been undertaken with the first including just 89 patients. Sixteen SNPs showed suggestive association ( P <5 × 10 ⁻⁵ ), but none exceeded genome-wide significance thresholds and none have been replicated in subsequent, larger studies.

A second GWAS undertaken by Plant and colleagues in 2011 included a 3-stage design with an initial GWAS investigating change in Disease Activity Score on 28 joints (DAS-28) over 6 months (n = 566); variants with P <10 ⁻³ were subsequently genotyped in an independent cohort with a subsequent meta-analysis. In stage 3, variants whereby the signal was strengthened were investigated in a third independent cohort, and finally, a second meta-analysis of the data was performed. The results demonstrated 7 loci associated with response, but 3 SNPs showed an opposite effect in the meta-analysis compared with the first stage and no SNP reached genome-wide significance ( P <5 × 10 ⁻⁸ ). Neither the Liu or Plant and colleagues results have been replicated subsequently.

In 2012, Krintel and colleagues performed a GWAS of 196 Danish RA patients treated with TNFi, most of whom were treated with infliximab, and performed a subsequent meta-analysis with the Liu and colleagues and Plant and colleagues datasets. Response was defined as the change in DAS-28 over 14 weeks. Suggestive association was detected at the PDE3A-SLC01C1 locus, where a C > T polymorphism at rs3794271 was associated with reduced efficacy according to the European League Against Rheumatism (EULAR) criteria (OR = 3.2, P = 3.5 × 10 ⁻⁶ ). A Spanish study by Acosta-Colman and colleagues tested the same variant in 315 RA patients and replicated the association (OR = 2.63, P = 1.74 × 10 ⁻⁵ ). The variant was associated with response to infliximab and etanercept but not adalimumab. A subsequent meta-analysis strengthened the association (OR = 2.91, P = 3.34 × 10 ⁻¹⁰ ). The PDE3A gene encodes a phosphodiesterase, inhibition of which suppresses TNF production in lipopolysaccharide-stimulated monocytes. The association was not reported in previous GWA studies, but the variant was not tested by the Plant and colleagues study. However, a subsequent study in a UK population found no evidence for association.

A multistage GWAS in 2013 recruited 882 Dutch patients and 2 further validation cohorts (n = 954 and 867, respectively) through international collaboration. Response was defined as 3-month change in DAS-28, a shorter time period than previous studies, but no variants were associated even at suggestive association thresholds ( P <5 × 10 ⁻⁵ ).

In 2013, Cui and colleagues performed the largest GWAS to date. Following international collaboration, 2706 RA patients from 13 different cohorts treated with etanercept, infliximab, or adalimumab were investigated. Response was defined as the change in DAS-28 at 3 to 12 months. No association reaching genome-wide significance was detected. A subset analysis revealed SNP rs6427528 nearing genome-wide significance ( P = 8 × 10 ⁻⁸ ) in the etanercept-treated group. rs6427528 is thought to disrupt transcription binding site motifs of CD84 and is associated with higher CD84 expression in peripheral blood mononuclear cells. CD84 is involved in T-cell activation and maturation and acts as a costimulatory molecule for interferon-γ secretion. Despite the strong initial association, the SNP failed to replicate in Portuguese and Japanese cohorts (n = 290).

Recently, a rigorous community-based assessment of the utility of SNP data for predicting anti-TNF treatment efficacy in RA patients was performed in the context of a DREAM Challenge ( http://www.synapse.org/RA_Challenge ). This approach enabled the comparative evaluation of treatment response predictions developed by 73 research groups using the most comprehensive available data on TNFi response and genome-wide data. Unfortunately, no significant genetic contribution to prediction accuracy was observed.