Introduction

Rheumatoid arthritis (RA) is a chronic inflammatory disease that affects ~1% of the Caucasian population.1 Disease onset typically manifests at age of 35–50 years, and females are affected 2.5 times more frequently than males. RA is characterised by synovial inflammation of joints most often affecting the joints of hands, wrist and feet, potentially leading to joint destruction, and functional disability. Furthermore, extra-articular manifestations may occur, for example, osteoporosis, vasculitis or interstitial lung disease. The manifestations are consequences of a chronically activated immune system. Both proinflammatory cytokines as tumour necrosis factor (TNF), interleukin (IL)-6, IL-8, GM-CSF, IL-1 and anti-inflammatory cytokines as IL-10 are involved. TNFα is a member of the TNF family of regulators of immune and inflammatory responses, which may also mediate cell death.2

In the 1980s, it was shown that TNFα has a prominent role in RA,3, 4, 5 and over the past decades, the availability of drugs targeting tumour necrosis factor α (anti-TNF) has improved the treatment of RA patients. Nevertheless, only 60–70% of patients have a good to moderate response to the anti-TNF treatment, whereas 30–40% have no or insufficient response.2, 6 Apart from anti-TNF drugs, biological compounds targeting CD20, T-lymphocyte antigen 4 immunoglobulin, interleukin 6 receptor and B-cells have been developed.7, 8 Until now, the treatment paradigm has been ‘one drug suits all’. Thereby, patients may remain in high disease activity, with irreversible joint damage as a possible consequence. Pharmacogenetics may identify the individual patient’s signature that may help guide the treatment selection (reviewed in refs 9, 10). Genetic variants may impact anti-TNF drug response.9, 10, 11, 12, 13, 14, 15, 16 They may therefore be utilised as biomarkers for treatment selection by stratifying patients according to the expected response following medical treatment. Furthermore, genetic biomarkers hold the advantage that they do not change over time.

Biomarkers able to predict treatment response will help optimising treatment, reduce adverse side-effects and avoid treatment with drugs without effect in the individual patients. In addition, such biomarkers will also help improving the use of health care resources. The expectations from patients, health care professionals and health authorities are high. ‘Personalised medicine represents one of the most innovative new concepts in health care. It holds real promise for more effective early diagnosis and more effective and less toxic treatments for patients, for improved medical service to citizens, and for improving the overall health of the population’ (http://permed2020.eu/.,2015). ‘Personalised medicine refers to a medical model using characterisation of individual’s phenotypes and genotypes (for example, molecular profiling, medical imaging, lifestyle data) for tailoring the right therapeutic strategy for the right person at the right time, and/or to determine the predisposition to disease and/or to deliver timely and targeted prevention’ (http://permed2020.eu/.,2015). Until now, most advances in applied pharmacogenetics have taken place in the field of anticancer therapy.17

Thus, we undertook to review case–control studies on genetic variants associated with anti-TNF treatment response in RA patients.

Materials and methods

A systematic review and meta-analysis were carried out according to the guidelines of ‘Preferred Reporting Items for Systematic Reviews and Meta-Analyses’ (PRISMA) statement.18 Three individual searches were performed in PubMed combining various alternative search terms for (1) ‘anti-TNF treatment’, (2) ‘genetic variation’ and (3) ‘autoimmune disease’, respectively, resulting in 669 abstracts (latest search date: 29th of August 2016). A full list of search terms is found in Supplementary Table 1. Figure 1 shows the flow diagram of included studies. All studies suggesting that they presented original data on associations between polymorphisms and anti-TNF treatment response in autoimmune diseases were retrieved (170 articles) and reviewed by three independent authors (SiB, JVN, VA). Exclusion criteria were: <100 cases available for treatment evaluation, missing data on treatment response, not reporting original data and not reporting data on anti-TNF response in RA (122 studies). In total, 47 studies reported association between genetic markers and anti-TNF response in RA. No further studies were identified by searching the literature list of the retrieved articles. Data on study design, number of patients, response criteria, odds ratios (OR) and 95% confidence intervals (95% CI) or numbers of good responders, moderate and non-responders, and genotypes were included.

Figure 1
figure 1

Flow diagram of included studies.

PowerPoint slide

Full size image

Statistics

Meta-analysis was performed on studies using EULAR response criteria.19 All polymorphisms studied in at least two studies (with a minimum of one significant association with response), and where data on genotypes and treatment response could be retrieved, were included in a meta-analysis (30 studies). The meta-analysis was based on the total number of patients in the cohorts. Statistical analyses were performed in Stata version 14 (StataCorp, Collage Station, TX, USA) using the meta-analysis plugin, metan. Random effects models were specified as the studies included were based on samples from heterogeneous populations. Heterogeneity is reported as I2.20

We also evaluated the potential for predicting treatment response based on genotyping using a data set (15–17) with information on a cohort of RA patients treated with anti-TNF. We first estimated associations between single nucleotide polymorphisms (SNPs) and non-response using logistic (EULAR) and linear regression (ΔDAS28) to identify significant associations and to determine dominance of alleles. After dichotomising SNPs based on allelic dominance, five genotypes were significantly associated with non-response. We finally tested the association between the number of risk genotypes and treatment non-response using logistic regression, and positive and negative predictive values of each level.

Results

In total, 47 studies were included in the analysis; 42 candidate gene studies14, 15, 16, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59 and 5 genome-wide association studies (GWAS)60, 61, 62, 63, 64 analysing responders versus non-responders from anti-TNF therapy in RA (Table 1). Two studies reported associations between polymorphisms and treatment response in juvenile idiopathic arthritis (JIA)37, 50 and the others on adult RA. The studies differed according to the studied population, response criteria and elapsed time before evaluation of response (Table 1).

Table 1 Description of 43 candidate gene studies (candidate) and 5 GWAS on associations between polymorphisms and response to anti-TNF treatment in RA patients
Full size table

Table 2 shows polymorphisms associated with response to anti-TNF treatment in RA identified by GWS. Response criteria as well as study design differed among the studies as described in Table 2. In total 19 polymorphisms, including polymorphisms in WDR27, GFRA1, MED15, LINC01387, LOC102723883, CNTN5, NUBPL, PDZD2, EYA4, TEC and C12orf79 were identified.

Table 2 Identified genetic markers associated with response after anti-TNF treatment of RA patients in GWS
Full size table

The polymorphisms investigated in candidate gene studies in relation to the outcome from anti-TNF treatment of patients with RA and JIA are shown in Supplementary Table 2. Hundreds of polymorphisms in various pathways have been selected for evaluation as candidate genes. Many of the assessed polymorphisms were found to be associated with response after anti-TNF treatment in one study. However, only few of these polymorphisms have been sought replicated in other candidate gene studies.

Supplementary Table 3 shows the ORs and 95% CI for the associations between polymorphism and treatment response for polymorphisms that were significantly associated with response in more than one cohort. In total, 23 polymorphisms in 21 genes were identified. These polymorphisms were selected for meta-analyses. Figure 2 shows the results for 6 polymorphisms in 6 genes (CHUK, PTPRC, TRAF1/C5, NFKBIB, FCGR2A and IRAK3) that were associated with treatment response in our meta-analyses. Supplementary Figure 1 shows the results for 17 polymorphisms in 16 genes (including FCGR3A, TNF, CD226, MAPKAPKA, RPS6KA5, MAP2K6, TLR5, TLR1, IFNG, IKBKB and TLR10) that were not associated with treatment response.

Figure 2
figure 2

Meta-analyses of 6 polymorphisms in 6 genes, which were associated with treatment response in rheumatoid arthritis (RA).

PowerPoint slide

Full size image

Next, to evaluate the current status of clinical use of the biomarkers we perform an explorative analysis of one cohort with available genotyping data.14, 15, 16 First, we used logistic regression to identify genotypes associated with non-response (risk genotypes) (CHUK rs11591741 (CC), IKBKB rs11986055 (CC), IFNGR2 rs17882748 (CT/TT), IL6 rs10499563 (CT/TT), NLRP3 rs4612666 (CT/TT)). Next, we calculated the OR and 95% CI based on the number of risk genotypes (Table 3; Supplementary Table 4). OR for non-response increased dose-dependently with the number of risk genotypes carried by the patients. For example, individuals with 4 out of 5 non-response-associated genotypes had an OR of 6.35 (95% CI: 1.32–30.48) and a negative predictive value of 0.5. The reference group of individuals with none of the five risk genotypes had the lowest odds (0.17) for non-response and a positive predictive value of 0.86 (indicating a somewhat higher chance of effective treatment than the first-best average (60–70%)).

Table 3 Positive and negative prediction values for selected genotypes in an exploratory analyses based on data from Sode et al.14, 15, 16
Full size table

Discussion

We identified polymorphisms associated with treatment outcome from anti-TNF treatment in RA patients from 47 studies with available data (Table 1). Among the 25 polymorphisms that were identified, 19 polymorphisms were found in GWS (Table 2). Our meta-analyses further identified 6 polymorphisms in 6 genes (Figure 2). Furthermore, we analysed the potential predictive power in an exploratory analysis of an available cohort.14, 15, 16 We found increasing OR for carrying increasing numbers of non-response associated polymorphisms (Table 3; Supplementary Table 4). However, the positive and negative predictive values were moderate.

Knowledge on the biological pathways involved in the treatment response in RA may allow for development of new treatment strategies. The results suggest that genetic variants in CTCN5, NUBPL, PD2D2, EYA4 and TEC (from the GWS), and CHUK, PTPRC, TRAF1/C5, NFKBIB, FCGR2A and IRAK3 (from our meta-analysis) may be implicated in treatment response to anti-TNF drugs in RA (Tables 2 and 4, Figure 2 and Supplementary Table 5). Some of the polymorphisms may indeed be functional or be linked to functional polymorphisms. Rs3761847 in TRAF1/C5 is associated with changes in mRNA levels. However, the direction of the effect differs between tissue types (GTEx, http://www.gtexportal.org). Likewise, rs9403 in NFKBIB has been associated with allele-specific mRNA levels with the variant alleles having the highest expression in liver (GTEx, http://www.gtexportal.org). FCGR2A rs1801274 is also a missense polymorphism resulting in a non-conservative amino acid substitution (His to Arg). The variant receptor has lowered affinity towards CRP.65 The lack of associations may suggest that the assessed genes are not of major importance for treatment response provided that the studies had sufficient power and the investigated polymorphisms are functional themselves or linked to functional polymorphisms. Our meta-analyses suggested that FCGR3A, TNF, CD226, MAPKAPKA, RPS6KA5, MAP2K6, TLR5, TLR1, IFNG, IKBKB and TLR10 were not associated with response after anti-TNF treatment in RA (Supplementary Figure 1).

Table 4 Proposed functions of selected polymorphisms that are identified in GWS or meta-analysis as associated with treatment response in RA
Full size table

Recently, we performed a review and meta-analysis of genes involved in response to anti-TNF treatment in patients with inflammatory bowel disease (IBD).12 SNPs involved in the TLR signalling pathway were found to be associated with anti-TNF treatment response in IBD, thus suggesting a significant role for the host–microbial interaction. Thus, different genes have been identified to be involved in RA and IBD treatment response to anti-TNF therapy. This may suggest that genes involved in the adaptive immune response may have a larger role in RA than in IBD treatment response to anti-TNF therapy. However, the role of host–microbial interactions in RA is not clear. Patients with active RA were found to have dysbiosis in the gut microbiota that partly resolved after medical treatment.70 The reason for this observation and how it may relate to treatment mechanism(s) is not known.

RA is a highly heterogeneous disease in terms of clinical presentation, prognosis and response to treatment.71 It is likely that this also applies to the pathogenesis of RA, in fact, studies have shown pronounced heterogeneity in RA synovial tissue of inflammatory cell types and gene expression.72 Through an improved discrimination of different RA subsets, SNP associations may prove to be more clinically useful, as they could at least in theory be very important for a certain subgroup while irrelevant for others.

An explorative approach was used when identifying potential candidate biomarkers in order not to overlook relevant candidates. Response criteria varied between the reviewed studies and more than one criterion were used in most studies. Our findings may furthermore be subject to bias from, for example, publication bias and selective reporting within studies. Replication of findings in other cohorts is of major importance in studies of genetic epidemiology. Therefore, replication of the findings in another cohort was chosen as criterion for association in the present review. Furthermore, environmental factors such as nutrition, smoking, lifestyle and other medication may impact genetic susceptibility and treatment outcome. These factors may not have been captured in the present studies.

Further evaluation of pharmacogenetics of anti-TNF treatment response in rheumatoid arthritis including gene–environmental interactions will require large cohorts of well-characterised patients and replication of positive findings in other cohorts. This work necessitates collaboration between researchers, for example, via International Consortia. Investigations of genomics combined with microbiome and mucosa expression profiles in each patient may thus allow us to understand which pathways and cytokines are deregulated in each case. Such knowledge may be utilised to select the best treatment for each patient.

However, at present, the pharmacogenomic basis for stratifying patients according to the expected response to anti-TNF treatment is not yet available.