Article Text
PDF
Abstract
Background There is a need for biomarkers that can predict anti-tumour necrosis factor (anti-TNF) treatment outcome in patients with rheumatoid arthritis (RA). Several studies have suggested that the rare A allele of the tumour necrosis factor α (TNFA) −308G→A polymorphism could be associated with a poorer response to anti-TNF therapy. Nevertheless, these results remain controversial.
Objective To determine by a meta-analysis whether the TNFA −308G→A polymorphism is associated with response to anti-TNF treatment in patients with RA.
Methods A bibliographic search identified studies in which the TNFA −308G→A gene polymorphism was investigated in Caucasian patients with RA treated with anti-TNF agents. Complementary data were requested when the 28-joint count Disease Activity Score (DAS28) was not used as the primary outcome measure. Odds ratios (ORs) for response based on DAS28 and standardised mean difference (SMD) for mean improvement of DAS28 were calculated to assess the potential association between TNFA −308 genotypes and response to anti-TNF agents.
Results The bibliographic search yielded 12 studies that met the inclusion criteria, which were supplemented with the data from a large Dutch cohort (n=426). The OR based on the 12 studies including 1721 patients was 1.24 (95% CI 0.98 to 1.56) and the SMD based on 11 studies including 2579 patients was −0.18 (95% CI −0.36 to 0.1). Subgroup analysis based on the two classes of anti-TNF agents did not demonstrate any association between TNFA −308 genotypes and anti-TNF treatment outcome.
Conclusion According to this meta-analysis, the TNFA −308 polymorphism is not a predictor of the clinical response to anti-TNF treatment in RA.
Statistics from Altmetric.com
Introduction
Although the pathogenesis of rheumatoid arthritis (RA) is not yet fully understood, it is clear that tumour necrosis factor α (TNFα) has a key role in the inflammatory process of this common autoimmune-mediated disorder.1 2 In recent decades, TNFα has emerged as one of the most important targets for therapeutic interventions in RA and other autoimmune diseases. Three anti-TNF agents are currently used in daily clinical practice—etanercept, a human, soluble, fusion protein consisting of two p75 TNF receptors linked to an IgG1 Fc part and two monoclonal antibodies; infliximab, a chimeric IgG1 antibody; and the fully humanised IgG1 antibody, adalimumab. All three anti-TNF drugs bind to TNF with high affinity leading to inactivation of TNF by preventing TNF binding to its receptor.3 4 Though, TNF neutralisation is highly effective and widely used in RA, the response is heterogeneous and approximately 30% of patients treated with TNF-blocking agents fail to show clinical improvement.5 This fact, together with the potential toxicities and the high costs of TNF-blocking agents, has driven the search for genetic markers to predict treatment outcome.6 7 To date, several variants in genes potentially involved in the action of anti-TNF therapy and/or in the pathogenesis of RA have been analysed3 but no highly specific and/or sensitive genetic predictors of anti-TNF therapy outcome have been identified, yet.
The most thoroughly investigated polymorphism is the −308G→A (rs1800629) polymorphism in the promoter of the tumour necrosis factor α (TNFA) gene which codes for TNFα.6 8,–,14 The relation between the −308G→A promoter polymorphism and anti-TNF response has been investigated by several groups with inconsistent results.6 8,–,10 12,–,24 Two meta-analyses, performed by Lee et al in 20066 and O’Rielly et al in 2009,25 including data of 311 and 692 patients, respectively, suggested that patients with RA carrying the rare A allele had a worse response to anti-TNF therapy than those carrying the common G allele. However, these meta-analyses pooled data with varying response criteria and curiously, did not include studies that did not support a significant effect of this polymorphism on anti-TNF response.16 26 Furthermore, the meta-analysis from O’Rielly did not include three recently published studies accounting for 1551 additional patients with RA treated with anti-TNF agents.20,–,22
Therefore we considered that a new meta-analysis of all published data could help in reaching more definitive conclusions on the role of the −308G→A TNFA polymorphism in predicting anti-TNF treatment outcome in patients with RA.
The present meta-analysis included a total of 2579 patients, which is four times more than the previous meta-analysis 2009.25
Materials and methods
The meta-analysis was performed according to the recommendations of the Cochrane Collaboration.27
Search strategy
A bibliographic search was performed on the Medline (January 1966 to May 2009), Embase and Cochrane databases by two investigators (SP and ET) using the medical subject heading key words: polymorphism genetic OR single nucleotide polymorphism AND rheumatoid arthritis AND (TNFR-Fc fusion protein OR adalimumab OR infliximab). Bibliographic references contained in the articles and abstract publications from the ACR Annual Scientific Meeting (2006–8) and the EULAR Annual European Congress of Rheumatology (2006–8) were also included. Results from all searches were combined and duplicate references excluded. Inclusion and exclusion criteria were checked for every article independently by two reviewers (SP and ET). In cases of disagreement, articles were re-examined and discussed until consensus was achieved.
Study selection
In order to be included in this meta-analysis, studies had to (a) be clinical trials or cohort studies using anti-TNF treatment for at least 3 months; (b) include adult patients with RA according to the American College of Rheumatology (ACR) criteria and (c) investigate the role of the TNFA −308G→A gene polymorphism. Studies on non-Caucasian patients were excluded from this analysis in order to evaluate a more genetically homogenous population.
Data collection
The primary outcome measure of the meta-analysis was the number of patients achieving clinical response to anti-TNF treatment defined as the number of patients who achieved an improvement of the DAS28 (28-joint count Disease Activity Score)28 ≥1.2 after at least 12 weeks of treatment. This is a dichotomous variable—that is, response/non-response. The secondary outcome measure was a continuous variable consisting of the mean improvement of the DAS28 in the same period of time. We chose to focus on DAS28 because it is the main outcome measure used in daily clinical practice. Moreover, it is the outcome variable most frequently reported in the selected studies. Upon identification of papers fulfilling the inclusion and exclusion criteria, complementary information was requested from the corresponding author if needed. This was the case if the number of patients who had an improvement of DAS28 ≥1.2 was either not used as outcome measure or not reported in the paper. Data collection and data extraction were performed by one reviewer (SP) using a predefined form.
Quality assessment
The influence of individual study quality on the results of meta-analysis has been well described.29 Although the use of assessment scales is recommended in systemic reviews and meta-analysis to undergo a quality review, it has been demonstrated that use of such scoring methods may not accurately assess the quality measures of interest.30 Furthermore, those scales assess the quality of the data reporting, but not that of the study, especially in the case of brief reports or letters. According to the recommendations of the Cochrane collaboration,31 we used study design as quality variable (prospective clinical trials vs cohort designs) and performed subgroup analysis to evaluate its impact on effect measures.
Statistical analysis
The odds ratio (OR) for the dichotomous outcome measure of efficacy (number of patients achieving an improvement of DAS28≥1.2) and the standardised mean difference (SMD) for the continuous outcome measure of efficacy (mean improvement of DAS28) were the effect measures of interest for assessing the association between the TNFA −308G→A gene promoter polymorphism and the response to anti-TNF treatment. For the meta-analysis, summary ORs and summary SMDs were computed using either fixed effects models (for data that did not demonstrate significant heterogeneity, I2<50%) or random effects models (for data demonstrating significant heterogeneity, I2≥50%).32 To assess potential heterogeneity across studies, we used the I2 statistic based on Cochran's heterogeneity statistic (Q).33 Meta-analyses were computed using RevMan analyses software (Review Manager (RevMan, version 5.0; Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008).
Sensitivity analysis
Assessment of the effect of TNF agents, of the potential excessive influence of varying studies and of potential publication bias
This sensitivity analysis was performed on the data from the current meta-analysis including 12 studies. The OR and the SMD of the different classes of anti-TNF agents (TNF receptor fusion protein and antibodies against TNF) were calculated separately in order to determine whether the results were affected by the agent used. Furthermore, to assess whether a specific study might have exerted excessive influence on the final results, we repeated analyses after excluding each study and after excluding studies of varying designs.
Potential publication bias (ie, the association of publication probability with the statistical significance of study results) was investigated using visual assessment of the funnel plot calculated by using RevMan analysis software. Funnel plots are plots of effect estimates on the horizontal axis against sample size on the vertical axis and publication bias may lead to asymmetrical funnel plots.
Assessment of the effect of covariates related to the characteristics of the disease and co-medication
As part of the sensitivity analysis, we performed a separate statistical analysis on individual patient's data from two large studies. This analysis aimed to investigate whether potentially important covariates might influence the association between the −308G→A promoter polymorphism and the response to anti-TNF treatment. These covariates consisted of gender, age, disease duration, classes of anti-TNF drug, co-medication with disease-modifying antirheumatic drugs (DMARDs), DAS28 at baseline and Health Assessment Questionnaire at baseline. This analysis was performed in a subset of 814 well-characterised patients (426 patients from The Netherlands and 388 French patients from the large international ReAct clinical trial).21 34 In this subset of patients, the TNFA −308G→A polymorphism was genotyped by allelic discriminating TaqMan PCR by use of the PreDeveloped TaqMan assay kit C_7514879 (PE Applied Biosystems, Foster City, California, USA), according to the manufacturer's protocol. The primary clinical outcome for this analysis was the absolute change in DAS28 at 12 weeks (±2 weeks for infliximab treatment). Bivariate and linear regression analyses were performed to assess association between DAS28 and the TNFA −308 genotype using R software (R-Development-Core-Team. R: A Language and Environment for Statistical Computing. In. Vienna, Austria: R Foundation for Statistical Computing, 2008).
Results
Selected studies
The literature search yielded 32 citations (figure 1). Twelve published studies met our inclusion criteria and could be included in the meta-analysis after requesting and receiving complementary results from five groups who did not use or report the DAS28 in detail in the original publications.
Study selection. DAS28, 28-joint count Disease Activity Score; RA, rheumatoid arthritis; TNF, tumour necrosis factor.
The 20 remaining, non-selected papers encompassed studies in which the TNFA −308 gene polymorphism was not investigated, duplicate publications, review and comment articles, and one study on non-Caucasian patients from Korea. A small study with only nine patients with RA was also excluded.15 In addition to the 12 selected studies, unpublished data from a total of 426 Dutch patients were included in this meta-analysis. Two studies20 26 were included in only one of the two analyses (for the dichotomous or the continuous variable used as outcome measure) owing to the lack of accessible data. In addition to the 12 selected studies, unpublished data from a total of 426 Dutch patients were included in this meta-analysis. Detailed information of the studies included is presented in table 1.
Characteristics of studies included in the meta-analysis of the association of tumour necrosis factor (TNF) −308 gene polymorphism with response to anti-TNF treatment in rheumatoid arthritis (RA)
Meta-analysis
As the TNFA −308 AA genotype is rare, we compared the TNFA −308 GG genotype group with the combined AG and AA genotype groups. Considering all anti-TNF agents together, we did not observe any significant association between the TNFA −308 G→A polymorphism and number of patients responding to TNF inhibitors (decrease of DAS28 ≥1.2): OR=1.24 (95% CI 0.98 to 1.56) (figure 2A). The test for heterogeneity was I2=31% with a p value of 0.07. Similar results were found when the mean improvement of the DAS28 was compared between genotypes with a non-significant effect size of −0.18 point in favour of TNFA −308 GG genotype (95% CI −0.36 to 0.01) (figure 2B).
Meta-analysis of the TNFA –308 genotype with response to anti-tumour necrosis factor (TNF) treatment in rheumatoid arthritis for the outcome: (A) number of responders according to the 28-joint count Disease Activity Score (DAS28; ≥1.2) and (B) mean improvement of the DAS28. M–H, Mantel–Haenszel statistical method.
Sensitivity analysis
Analysis of a potential effect of the class of anti-TNF agent anti-TNFα antibodies (infliximab and adalimumab) versus the soluble TNF receptor (etanercept) did not show any association between the TNFA −308G→A promoter polymorphism and response to treatment (figure 3A,B).
(A) Meta-analysis of the TNFA –308 genotype with response to tumour necrosis factor (TNF) antibody drugs in rheumatoid arthritis for the outcome: number of responders according to the 28-joint count Disease Activity Score (DAS28; ≥1.2) and (B) number of responders according to the DAS28 (≥1.2). M–H, Mantel–Haenszel statistical method.
To assess the potential influence of study quality, we repeated the analysis excluding one study at a time. Only exclusion of the largest studies11 20 21 34 would have modified the statistical significance of the main results. To assess the influence of study quality, we repeated the analysis excluding those studies that were part of randomised clinical trials. This approach was considered because inclusion criteria in a randomised clinical trial might have selected for a more severe RA phenotype with a higher chance of responding to anti-TNF treatment. Nonetheless, the estimate generated was similar to the estimate found in the primary analysis (OR=1.30, 95% CI 0.97 to 1.73).
The analysis of the funnel plot to check for potential publication bias suggested a random distribution around an OR of 1, consistent with the lack of association between TNFA −308G→A polymorphism and response to treatment (figure 4).
Funnel plot: assessment of potential bias of publication according to the odds ratios (ORs) of association between TNFA –308 gene polymorphism and response to anti-tumour necrosis factor treatment in rheumatoid arthritis.
The effect of potentially important covariates, on the association between the TNFA −308G→A polymorphism and response to treatment, was analysed using individual patient data from subset database including 814 patients from The Netherlands and France. The patients' characteristics are presented in the online supplementary table. The covariates studied encompassed the class of anti-TNF drug, concomitant use of DMARDs, disease duration, Health Assessment Questionnaire and DAS28 at baseline, rheumatoid factor positivity, age and gender.
Bivariate and linear regression analysis did not demonstrate an association between the AA genotype versus the AG or GG genotypes and response to treatment. The mean absolute variations of the DAS28 after 12 weeks of treatment were −1.38, −1.76 and −1.76 for the TNFA −308 genotypes AA (n=28 (3.5%)), AG (n=207 (26.1%)) and GG (n=559 (70.4%)), respectively (p value=0.33). In the subgroup of patients with RA treated by adalimumab (n=526), the association between the TNFA −308G→A polymorphism and number of responders to TNF inhibitors was not significant: OR=1.20 (95% CI 0.80 to 1.81). None of the covariates studied was statistically associated with the TNFA −308 genotypes or the response to treatment.
Discussion
The meta-analysis did not show any association between this polymorphism and response to anti-TNF in the largest cohort of patients with RA analysed up till now. This was still the case after analysis for important covariates such as anti-TNF agents, disease duration, DMARD use.
As inherent in all meta-analyses, especially those of observational or experimental studies, the validity of the results is limited by the differences in study designs, the conduct and the reporting of the data which were extracted and pooled. In order to reduce the heterogeneity of the analysis, we used the same response measurements based on the DAS28 for all studies that met the inclusion criteria defined by our study protocol. To explore the heterogeneity due to the assessment of the treatment response at variable time points, we performed a time-course analysis on the data from the DREAM study (3, 6, 9 and 12 months). The ORs generated were similar at the different end points (data not shown).
Our results are not in line with the results presented in two previous meta-analyses including fewer patients. The study Lee et al in 20066 and O’Rielly et al in 2009,25 included 311 and 692 patients with RA from six to nine different studies, respectively. They suggested that the TNFA −308 A allele predisposes to a poorer response to TNF-blocking therapy compared with the −308 G allele.6 25 Several factors may explain the observed difference between these previous reports and our study. First, one of the obvious differences in favour of our study is that sample sizes were very different. Our meta-analysis involved between 2.5 and 4 times more patients (depending on the outcome) than the previous ones. Moreover, our study of the funnel plot showed that the largest cohorts did not show any association between TNFA −308 GG versus GA/AA genotypes, whereas the smallest studies did. Those large studies, published in 2008, were not included in the previous meta-analysis.20,–,22 We identified these studies and we asked the authors for details of their negative results. Second, the previous meta-analyses disregarded other studies that showed no significant effect of this polymorphism on anti-TNF response, which may have introduced publication bias. Finally, we used response measurements based only on the DAS28 whereas Lee et al6 and O’Rielly et al25 combined two distinct response measurements, the ACR response criteria35 and the DAS2827 in their meta-analysis.
Since the last study included in the meta-analysis of O’Rielly and colleagues, three studies investigating the TNFA −308G→A polymorphism and anti-TNF treatment outcome in large cohorts of patients with RA have appeared. Altogether, six9 10 12,–,14 17 of the 13 studies reported an association between the polymorphism and treatment outcome, whereas seven did not.11 16 20,–,22 26 34 These inconsistent results are largely explained a statistical power which was too low and small sample sizes. The problem of low power is illustrated even by the largest study published up to date (n=1041 patients with RA treated with anti-TNF agents).20 The authors reported an association only between the very rare homozygous −308AA genotype (n=7) and response to etanercept (n=444) but not to other TNF neutralising agents. Comparison of the GG genotype versus the combined AA and AG groups in this and all other studies including our meta-analysis (figure 3B) showed no difference in the response to any anti-TNF therapy, including etanercept. Combining the AA and AG group is in line with the overall hypothesis that the A allele may be associated with anti-TNF non-response.21
Besides the overall meta-analysis, we also investigated whether potentially important covariates might influence the association of the TNFA polymorphism with treatment outcome, but this was not the case. Such type of analysis is crucial to assess possible interactions between genetic and environmental factors.
Our meta-analysis does not support an association between the TNF −308G→A promoter polymorphism and response to anti-TNF therapy. This does not fully rule out the possibility that the gene encoding for TNF might influence anti-TNF treatment outcome as we focused on only one genetic variant in the gene. In fact, linkage disequilibrium is strong in this area and it is difficult to study the role of a single nucleotide polymorphism in isolation. Therefore, other single nucleotide polymorphisms of the TNFA gene, not in complete linkage disequilibrium with the −308G→A polymorphism, deserve further study.36 37 Haplotype analysis of the gene might be an alternative approach to investigating anti-TNF treatment outcome. An example of such an analysis can be found in the recently published study of our group.21 This showed that neither the number of HLA-DRB1 shared epitope copies nor the presence of three TNF polymorphisms (−238A→G, −308G→A and −857C→T) tested separately was significantly associated with response to adalimumab therapy. However, haplotype reconstruction of the TNF locus showed that the GGC haplotype (−238G, −308G, −857C) in a homozygous form was significantly associated with lower ACR50 response to adalimumab at week 12.21 Nevertheless, owing to the multiple interactions of the TNF ‘system’, the effect of the TNFA haplotypes, as a potential predictor of clinical response to treatment, needs to be confirmed by functional studies that correlate this haplotype with the level of TNF production.
In conclusion, we observed no effect of the TNFA −308G→A promoter polymorphism on response to anti-TNF treatment in the largest cohort of patients with RA analysed so far. We conclude that this polymorphism is not a suitable predictor for response to anti-TNF in clinical practice. Nonetheless, the gene encoding for TNFα may still be an interesting candidate gene, and other polymorphisms, haplotypes and interactions with environmental factors need to be investigated for their role in anti-TNF response.
Acknowledgments
We thank Dr A A den Broeder, C M A De Gendt, Professor Dr M A F J van de Laar, Dr T L Th A Jansen and Dr H L M Brus for their contribution to the DREAM registry, and we are grateful to their patients for participating. We are indebted to Joao E Fonseca, Alessandros Drosos, Anthi Chatzikyriakidou, Pierre Miossec and Hubert Marotte who generously contributed data from their independent research for inclusion in this meta-analysis. We thank Christophe Combescure for his insight into the methodological issues of the study.
References
Supplementary materials
Web Only Data ard.2009.117622
Files in this Data Supplement:
Footnotes
SP, EJMT, XM and MJHC contributed equally to the study.
For the Dutch Rheumatoid Arthritis Monitoring (DREAM) Register: Piet L C M van Riel.
-
Provenance and peer review Not commissioned; externally peer reviewed.