Article Text
Abstract
Objective To determine whether genetic variation within genes integral to the Toll-like receptor (TLR) and NFκB signalling systems, two cardinal regulators of inflammatory and immune responses, contributes towards the observed variation in response to tumour necrosis factor (TNF) blocking agents in patients with rheumatoid arthritis (RA).
Methods Pairwise-tagging single nucleotide polymorphisms (SNPs) spanning 24 candidate genes were selected and genotyped in a large UK cohort of patients receiving anti-TNF therapy for RA. Multivariate regression analyses were performed to test association between individual genotypes, under an additive model, and treatment response at 6 months' follow-up assessed using both the absolute change in 28-joint count Disease Activity Score (DAS28) and the European League Against Rheumatism (EULAR) response criteria. Analyses were performed across subgroups comprising etanercept-, infliximab- and infliximab/adalimumab-treated patients as well as the combined anti-TNF-treated cohort. p Values <0.05 were considered statistically significant.
Results A total of 187 SNPs were successfully genotyped and analysed in 909 patients. Eight SNPs spanning six genes demonstrated nominal evidence of association with response (DAS28) across the anti-TNF-treated subgroups, six of which were restricted to etanercept-treated patients. Twelve SNPs spanning nine genes demonstrated nominal evidence of association with treatment response (DAS28 and/or EULAR) across the combined anti-TNF cohort. These included SNPs mapping to MyD88 (rs7744) and CHUK (rs11591741), which were associated under each model applied (etanercept-treated and combined anti-TNF cohort analysis (DAS28 and EULAR)).
Conclusions Several SNPs mapping to the TLR and NFκB signalling systems demonstrated association with anti-TNF response as a whole and, in particular, with response to etanercept. Validation of these findings in an independent cohort is now warranted.
Statistics from Altmetric.com
Introduction
The introduction of tumour necrosis factor antagonists (anti-TNF) has improved the treatment and management of rheumatoid arthritis (RA) for many patients. However, these agents are expensive and a substantial proportion of patients do not respond.1 Even in responders there exists a spectrum of efficacy ranging from complete remission of symptoms to modest improvement. Such variation in treatment response is unlikely to be random, but influenced by genetic, environmental and psychological factors. Identified clinical predictors of response include baseline Health Assessment Questionnaire (HAQ) score, administration of concurrent disease-modifying antirheumatic drugs, gender and seronegativity.1,–,4 Whereas some genetic factors have been proposed to influence treatment response (eg, TNF, type 2 TNF receptor, Fcγ receptors), few have been consistently replicated across study cohorts, indicating a complex relationship with treatment outcome (or simply, a small effect).5,–,10 Even in combination these factors have not yet been proved to be clinically useful and the identification of additional genetic determinants of treatment response could provide enormous clinical and economic benefit.
The Toll-like receptors (TLRs) and NFκB signalling systems are cardinal regulators of inflammatory and immune responses. For instance, TLRs sense exogenous and endogenous antigens and downstream pathways (eg, NFκB, JNK and p38, in particular) signal the production of proinflammatory cytokines (eg, TNF and interleukin 1 (IL1)), chemokines and matrix metalloproteinases (MMPs). In recent years evidence has emerged implicating these signalling systems in the pathogenic, inflammatory and destructive processes characteristic of RA.11 12 For instance, increased TLR expression (types 2 and 4, in particular) has been repeatedly described in synovial tissue from patients with RA and associated with the upregulation of cytokines and MMPs.13 14 In addition, ex vivo cultures have linked other components of the TLR signalling pathway (eg, MyD88 and TIRAP/MAL) to the spontaneous and increased production of TNF, other cytokines and MMPs.15 Activated NFκB has also been found in RA synovial tissue and may provide a conduit for proinflammatory signals initiated by various receptors also implicated in RA (eg, TLR, IL1β receptor and TNF receptors).16 17
Critically, both systems could contribute towards chronic inflammation in RA via positive feedback loops. First, endogenous markers of tissue injury that are present in RA joints can induce TLR signalling, resulting in further inflammation. Second, NFκB is an important activator of TNF-mediated proinflammatory signalling but is also stimulated by TNF. Since persistent inflammation defines anti-TNF non-response, these systems provide a logical focus for pharmacogenetic studies of this drug class. Specifically, polymorphisms within the TLR and NFκB pathways that promote more active/proinflammatory signals may reduce the efficacy of anti-TNF therapy. We tested this hypothesis in a large UK cohort of patients with RA.
Methods
Participants
DNA samples from 923 patients receiving anti-TNF therapy for the treatment of RA were available from the Biologics in Rheumatoid Arthritis Genetics and Genomics Study Syndicate (BRAGGSS) (http://www.medicine.manchester.ac.uk/epidemiology/research/arc/genetics/pharmacogenetics/braggss/, accessed 19 March 2010). This syndicate incorporates collaborations between multiple UK-wide rheumatology clinics from which eligible patients have been recruited, DNA and serum samples obtained and extensive clinical data, taken at baseline and 6 months' follow-up, compiled. The criteria for selection and recruitment of patients have been described in full elsewhere.4
Candidate gene and SNP selection
Twenty-four candidate genes were selected for investigation (table 1). In addition to TLR-2 and TLR-4, critical downstream adapter proteins and signalling intermediaries were selected (TLR-1 and TLR-6, MyD88, Mal, IRAK-1, IRAK-4 and IRAK-M, TAB1 and TAB2, TAK1). NFκB-1 and NFκB-2, and their key regulators were chosen (IKK1 and IKK2, IkBα, IkBβ, IkBε, SUMO4, Rel, RelA, RelB), plus one key NFκB target not previously investigated (COX2). Marker coverage for each gene included the 10 kb upstream and downstream flanking regions. Pairwise tagging single nucleotide polymorphisms (SNPs) were selected from the CEPH/CEU HapMap dataset (phase II, release 23a/March 08) using Haploview software (minor allele frequency (MAF) >0.05, Hardy–Weinberg p value >0.05, Min genotype percentage >90, Max Mendel errors >1, pairwise r2>0.8).18 19 Additional SNPs were included if they were non-synonymous or known to have functional effects (based on information available on NCBI: PubMed, Entrez Gene and dbSNP).
Genotyping
A total of 199 SNPs were selected and genotyped using Sequenom's MassARRAY iPLEX system (Sequenom, Cambridge, UK). Multiplex assays were designed and performed according to the manufacturer's specifications. SNP genotype cluster plots for each assay were manually checked. Quality control (QC) procedures before analysis removed any samples with a call rate <80% and any assay with a call rate <95%. Hardy–Weinberg equilibrium was assessed to identify potential genotyping errors. For SNPs that failed assay design and/or genotyping, alternative tagging SNPs were selected and genotyped, where available.
Statistical analyses
The primary outcome measure was absolute change in the 28-joint count Disease Activity Score (DAS28) between baseline and 6 months' follow-up.20 Multivariate linear regression analyses were performed to investigate association between change in DAS28 and individual SNP genotypes under an additive model. Adjustments were made for independent clinical predictors of anti-TNF response identified in the study cohort—namely, baseline DAS28, baseline HAQ score, administration of concurrent disease-modifying antirheumatic drugs and gender (online supplementary table 1). Anticyclic citrullinated peptide antibody and rheumatoid factor were also significant predictors of outcome but were not included in the final analysis as data were only available for 75% of patients. Analyses were initially performed according to treatment received (etanercept, infliximab and combined monoclonal antibody group). Adalimumab recipients were not analysed alone owing to small numbers. Subsequently, the combined cohort was studied to identify potential class effects, applying the same multivariate model.
In addition, the European League Against Rheumatism (EULAR) response criteria were assessed as a secondary outcome measure (none vs moderate/good responders) using multivariate logistic regression analyses, applying the same model as described above.21 This secondary analysis was only performed across the combined anti-TNF drug cohort owing to the limited power of the smaller subgroups. All analyses were performed in Stata (StataCorp, Texas, USA). Power calculations were performed using the Quanto software package.22 23
Results
Genotyping results
A total of 199 SNPs spanning the 24 candidate gene regions were initially genotyped in the 923 patients. These included 93 pairwise tagging SNPs (tagging ~600 polymorphisms) and 95 singleton SNPs selected from the phase II HapMap dataset plus 11 SNPs selected from NCBI (PubMed, Entrez Gene and dbSNP) owing to known or potential functional effects (online supplementary table 2). Despite the redesigning and genotyping of alternative assays, 12 SNPs (6%) failed to meet QC thresholds (>95% genotyping success) and were excluded from subsequent analyses (online supplementary table 2). In addition, 14 patients (2%) were removed from analysis owing to failure to meet QC thresholds (>80% genotyping success). Consequently, multivariate regression analyses were performed on data generated for 187 SNPs in 909 patients.
For the linear regression analyses across the combined cohort (n=909), there was >90% power to detect a difference ≥0.6 in the absolute change in DAS28 under an additive model, given a minimum MAF of 5%, at the 5% significance level. In the etanercept, infliximab and combined monoclonal antibody subgroups, there was 68%, 64% and 76% power, respectively. Similarly, by comparing EULAR non-responders with moderate/good responders (~20 vs ~80% of patients, respectively) using logistic regression, the combined anti-TNF group had >80% power to detect an odds ratio ≥2 under the same model.
Characterisation of study cohort and response to anti-TNF therapy
Baseline characteristics for the individual subgroups are presented in table 2. Forty-two per cent of patients had received etanercept, 44% infliximab, and 14% adalimumab. Overall, these patients had longstanding, active disease (mean duration 14 years, mean DAS28 6.7) with a high degree of disability (mean HAQ score 2.1). At 6 months' follow-up, 20% of patients were non-responders, 53% moderate responders and 27% good responders according to the EULAR criteria. The mean change in DAS28 was an improvement of 2.5 points.
Pharmacogenetic predictors of response to individual drugs
Linear regression analyses were first performed according to drug received (table 3). A total of eight SNPs spanning six genes demonstrated evidence of association with the absolute change in DAS28. Interestingly, six of these were limited to the etanercept-treated subgroup. Significant differences in response to etanercept were demonstrated with the rs11591741 (CHUK) (p=0.008), rs7744 (MyD88) (p=0.006) and rs5743704 (TLR-2) (p=0.006) polymorphisms. For rs11591741, the linear regression coefficients suggested that, relative to the major homozygotes, heterozygotes and minor homozygotes demonstrated a smaller reduction in DAS28 of 0.267 and 0.534 units on average, respectively. The same effect was demonstrated for rs5743704, with heterozygotes and minor homozygotes demonstrating a smaller reduction in DAS28. However, this could be an artefact of the low MAF (4.8%) calculated across this study cohort. The rs7744 SNP showed the opposite effect, with heterozygotes and minor homozygotes demonstrating a mean greater improvement in DAS28 of 0.390 and 0.780 units compared with major homozygotes, respectively. The remaining five SNPs demonstrated nominal evidence of association with treatment response (p<0.05).
Pharmacogenetic predictors of an anti-TNF class effect
Linear regression analyses were subsequently performed across the combined cohort, under the assumption that all three anti-TNF agents act through shared pathways (class effect). Seven SNPs spanning five genes demonstrated nominal evidence of association (p<0.05) with the absolute change in DAS28 (table 3). These included SNPs mapping to CHUK (rs11591741, rs2230804) and MyD88 (rs7744) identified in the etanercept-treated subgroup analysis, which demonstrated the same pattern of association but with smaller effect sizes in the combined cohort. Statistically significant evidence for association was lost across the combined cohort for the other five polymorphisms, suggesting drug-specific associations. In contrast, four additional SNPs demonstrated association in the combined cohort, which may reflect its greater power. Patients carrying the minor alleles at rs11986055 (IKBKB) and rs11541076 (IRAK-3) demonstrated a greater improvement in DAS28 compared with major homo-zygotes. For rs11595324 (CHUK) and rs11574851 (NFκB-2), patients carrying the minor allele demonstrated a smaller reduction in DAS28.
Finally, logistic regression analyses were performed in the larger combined dataset using the EULAR response criteria as a secondary outcome measure. Of the seven SNPs associated with DAS28 in the combined anti-TNF dataset, four were also associated with EULAR response (table 4). In particular, rs11591741 (CHUK) and rs7744 (MyD88) were again associated with this secondary outcome measure. Consistent with the linear regression analysis, for SNPs rs11986055 (IKBKB), rs11541076 (IRAK-3) and rs7744 (MyD88), the odds ratios indicated that patients carrying the minor allele at these loci had a greater chance of achieving a moderate/good EULAR response compared with major homo-zygotes. Carriage of the minor allele of rs11591741 (CHUK) was again associated with a lower probability of achieving a moderate/good EULAR response. Additional nominally significant associations were demonstrated between EULAR response and five additional SNPs (rs3136645 and rs9403 (NFkBIB), rs2206593 (PTGS2), rs2289318 (TLR-2), rs11096957 (TLR-10/1/6)) (table 4).
Discussion
Analyses were performed to identify common variants spanning 24 candidate genes involved in the NFκB, TLR-2 and TLR-4 signalling pathways that associate with response to anti-TNF treatment in patients with RA. Eight SNPs spanning six genes (CHUK, IKBKB, MyD88, NFkBIA, TLR-2, TLR-4) demonstrated nominally significant associations with absolute change in DAS28 after treatment with individual anti-TNF agents. In addition, four of the same genes (CHUK, IKBKB, MyD88, TLR-2) plus an additional five genes (IRAK-3, NFκB-2, NFkBIB, PTGS2, TLR-10/1/6) demonstrated nominal evidence of association with the absolute change in DAS28 and/or EULAR response rates across the combined anti-TNF cohort.
It is important to emphasise that these results arise from exploratory analyses and are reported uncorrected. Indeed, none of the present associations would remain significant after Bonferroni correction. However, such corrections are often considered overly stringent. For instance, although in this study tagging SNPs were selected as genetic markers, these SNPs demonstrate modest linkage disequilibrium (r2 = 0.4–0.8) and therefore cannot be considered completely independent. Nonetheless, and notwithstanding the large size of our cohort, these findings should be considered preliminary, awaiting confirmation in other datasets.
In order to thoroughly explore the data, we chose to analyse and present the results in several ways. Given similarities and differences between agents, pharmacogenetic factors could influence response to individual agents and/or anti-TNF drugs as a group. Hence, we analysed according to both paradigms, acknowledging the reduced power of the smaller drug-specific subgroup analyses. In fact, eight SNPs demonstrated nominal evidence of association with individual drugs, six of which were restricted to the etanercept-treated subgroup. This apparent dichotomy of association according to drug has previously been reported for both genetic and non-genetic factors.1 7 Three of these associations (CHUK (×2), MyD88) remained significant (with diluted effect sizes) in the combined anti-TNF cohort but the others (IKBKB, NFkBIA, TLR-2, TLR-4) did not. Although this would be consistent with stronger effects linked to specific drugs, the reduced power and increased risk of both type I and type II errors in smaller subgroup analyses, emphasise the importance of replication in larger datasets. However, to our knowledge BRAGGSS is currently the largest dataset available for such analyses world wide. Four of the seven associations demonstrated across the combined anti-TNF cohort were not detected in the individual subgroup analyses, which again may reflect limited power. Smilarly, there is continuing debate about the benefits of switching from one anti-TNF drug to another. Although only a small proportion (6% overall) of our patients had previous exposure to a different anti-TNF agent, exclusion of these patients from the analysis did not change the overall conclusions of the study (data not shown).
There is currently no consensus about the best measure of treatment response in such pharmacogenetic studies. The absolute change in DAS28, which measures response at the level of the individual patient, is, statistically speaking, more powerful owing to its continuous scale. In contrast, the EULAR improvement criteria, which measure response at the group level, are more clinically meaningful. We therefore analysed according to both outcomes, although EULAR improvement criteria analyses were limited to the combined cohort because of power limitations. Differences between these two measures may explain why some polymorphisms did not demonstrate associations in both analyses.
Our most robust findings involve associations of MyD88 and CHUK with response in both the etanercept-treated subgroup and combined anti-TNF cohort. Both associated markers were selected as tagging SNPs and, if replicated, these results could present a number of functional possibilities. In MyD88 the associated SNP rs7744 maps to the 3′UTR and could influence mRNA stability. Alternatively, rs7744 is in strong linkage disequilibrium with a SNP, rs156265 (r2=0.9), which maps to the upstream promoter region and could therefore influence gene expression. In CHUK, the associated marker rs11591741 maps to intron 9, is perfectly correlated with an intron 18 SNP (rs11597086) and demonstrates modest linkage disequilibrium (r2>0.5) with six other polymorphisms spanning the gene region. Hence, more detailed analyses, including the assessment of functional effects, may be required to identify the true pharmacogenetic association.
With the increasing feasibility of genome-wide association studies, there is hope that such designs will aid the identification of markers predictive of anti-TNF treatment efficacy. For instance, although we selected crucial adapter proteins, signalling intermediaries and regulators involved in the TLR and NFκB pathways, there are a large number of genes implicated in these pathways that we did not investigate. However, this does not detract from the power and success of well-designed candidate gene association studies, particularly when drug pathways and mechanisms are often well characterised.24 25 MyD88 is the foremost adapter protein essential for TLR intracellular signalling, whereas the IκB kinase CHUK (alias IKK1) is responsible for the phosphorylation and degradation of NFκB inhibitors (IκBs), resulting in NFκB activation and subsequent TNF induction.26 Many components involved in these pathways, in particular MyD88, have been implicated in the chronic inflammatory and destructive processes characteristic of RA.13,–,15 Furthermore, early downregulation of genes within the NFκB signalling system has recently been associated with anti-TNF efficacy.27 Hence, genetic variation that alters the expression and mediation of these pathways, and consequently TNF activation, may contribute towards anti-TNF treatment response.
In summary, several SNPs mapping to genes involved in the TLR and NFκB signalling pathways demonstrated association with response, particularly to etanercept, but also to anti-TNF drugs as a group. Our data relate to the largest reported cohort of patients with RA treated with TNF inhibitors. Nonetheless, as with all pharmacogenetic studies, it is now necessary to validate them in independent cohorts of equal or greater size.
Acknowledgments
The authors would like to thank staff at the Arthritis Research UK Epidemiology Unit, Manchester University—in particular, Dr Steven Eyre and Edward Flynn for all their support while allowing the authors access to their genotyping facilities.
References
Supplementary materials
Web Only Data ard.2009.117309
Files in this Data Supplement:
Footnotes
Members of BRAGGSS are given in the online supplementary data.
-
Funding The authors thank JGW Patterson Foundation for funding this work. Work in Newcastle's Musculoskeletal Research Group is supported by the UK NIHR Biomedical Research Centre for Ageing and Age-related disease award to the Newcastle upon Tyne Hospitals NHS Foundation Trust
-
Competing interests None.
-
Ethics approval This study was conducted with the approval of the UK Central Office of Research Ethics Committees (COREC) approval (04/Q1403/37).
-
Provenance and peer review Not commissioned; not externally peer reviewed.