Abstract and Introduction
Abstract
Following a greater understanding of the pathogenesis of rheumatoid arthritis (RA), the treatment of this chronic disease has improved with the availability of biological agents targeting key molecules. Despite this, initial treatment produces an inadequate response in many patients and guidance on the optimal treatment for these patients is needed. Research in specific patient populations aims to define predictive biomarkers of response to identify those patients most likely to benefit from treatment with specific agents. Although there have been conflicting results from studies of various genetic markers, single nucleotide polymorphisms in the tumour necrosis factor (TNF) −308 promoter region may play a role in response to specific TNF inhibitors. Microarray analysis of mRNA expression levels has identified unique sets of genes with differentially regulated expression in responders compared with non-responders to the TNF inhibitor infliximab. Of the various protein biomarkers studied, rheumatoid factor and/or anticitrullinated protein autoantibodies may have a future role in predicting response or guiding the order in which to use biological agents. Further research is needed with larger, well-designed studies to clarify the current understanding on the role of biomarkers in predicting treatment response in RA to help guide clinical decision-making. Individualised treatment has the potential to improve the therapeutic outcomes for patients.
Introduction
The effective use of biological disease-modifying antirheumatic drugs (DMARDs), which block key molecules involved in the pathogenesis of rheumatoid arthritis (RA), has dramatically improved the treatment of this chronic disease over recent years. However, for at least one-third of patients with RA tumour necrosis factor (TNF) inhibitor treatment will produce an inadequate response. Furthermore, although it is difficult to compare data across heterogeneous studies, it is clear that, generally, responses to biological agents vary greatly across different patient populations, with particular differences according to duration and reversibility of disease.
Indeed, while the currently approved TNF inhibitors (adalimumab, etanercept, infliximab, golimumab and certolizumab pegol) have been shown to improve clinical signs and symptoms of RA in patients for whom DMARDs had produced an inadequate response (DMARD-IR population), the placebo-adjusted American College of Rheumatology criteria (ACR20) response rates for registered doses of the TNF inhibitors at 12 months range from 20% to 44%. Similarly, in patients with an inadequate response to TNF inhibitors (TNF-IR population), biological agents studied in this population have been shown to be effective, with placebo-adjusted ACR20 response rates at 24 weeks ranging from 17% with golimumab 50 mg to 40% with tocilizumab 8 mg/kg. Rituximab 2 × 1000 mg is the only biological agent to have shown significant inhibition of the progression of joint destruction in this population with highly refractory disease.
The differences in response rates among the same class of biological agents and the fact that not all patients respond to biological agents, indicate the heterogeneity of RA. Importantly, these differences increase with disease duration, and fewer patients with long-term refractory disease than those with early RA respond to these biological agents. Therefore, the use of robust predictive markers of response to identify individuals who are likely to respond to specific treatments will provide benefit. An individualised approach will allow patients to receive effective treatment without unnecessary exposure to potentially toxic side effects. Furthermore, doctors could abandon the trial and error approach to treatment in favour of evidence-based guidance (with the added advantage that the cost of treatment could be kept under control).
In order to achieve the goal of individualised treatment, identification of specific patient populations who show greater response to treatments may provide guidance in determining optimal treatment. Indeed, a greater understanding of the role of biomarkers in predicting treatment response has the potential to lead to the development of clinically useful algorithms.
This review examines the current evidence for various biomarkers with prognostic capabilities in RA and assesses their potential in predicting treatment response to biological agents for RA and for informing clinical practice. Where no clear response to treatment is shown in patients with a potential marker, those markers cannot be considered predictive of response. Where this occurs in the review, additional discussion of the marker's prognostic capabilities—that is, markers that are associated with better or worse disease activity or radiological damage over an extended period of time and independent of treatment—has been included. In this context, a clear distinction between prognostic and predictive response markers, not essentially restricted to biomarkers, can only be delineated by comparing patients in clinical trials where prognostic markers should be equally distributed among treated and placebo arms at baseline with a subsequent differentiation by response markers considered predictive. Most studies have not performed such rigorous analyses.
Patient Baseline Demographics and Clinical Characteristics
'Real-life' clinical data from registries have shown that high levels of disability at baseline and being a smoker were predictive of a poor response to TNF inhibitors. In addition, several studies, including an analysis of the British Society for Rheumatology Biologics Register, showed that baseline 28-joint Disease Activity Score (DAS28), lower Health Assessment Questionnaire score and concurrent use of DMARDs might be associated with improved TNF inhibitor responses (Table 1). Despite these findings, there has been conflicting evidence for the predictive value of other baseline characteristics for response to TNF inhibitors. Age, disease duration and the previous number of DMARDs were not predictive of TNF inhibitor response. Sex did not predict response to TNF inhibitor therapy in the Swedish Registry, whereas analysis of patients from the Gruppo Italiano per lo Studio delle Early Arthritis Italian registry indicated that male sex and rheumatoid factor (RF) negativity were positive predictors of remission at 6 months after TNF inhibitor therapy. Clearly, these baseline characteristics need confirmation as potential markers of response or simply classification as prognostic markers.
Defining response is important when considering predictive markers, as different outcome measures may influence the predictive value of certain baseline features. This is highlighted by the analysis of registry data. High disease activity at baseline was directly associated with favourable response to TNF inhibitor therapy when measured by ACR50 and ACR70, whereas baseline DAS28 was inversely associated with European League Against Rheumatism (EULAR) remission (Table 1). However, this result is not unexpected as ACR relies on a change from baseline and so a high level of disease activity at baseline provides greater opportunity to show a fall from this start point, irrespective of which treatment is used to treat these patients. In the case of EULAR remission (DAS28 <2.6), it is more difficult to achieve this outcome from high baseline disease activity levels, even though a substantial improvement in the percentage change from baseline may be apparent. Therefore, despite showing some ability to predict response to TNF inhibitors in one study, baseline DAS28 is not an appropriate predictive marker of response, as suggested by another analysis. However, baseline DAS28 (<5.1), with a good, early therapeutic response, has been shown to be associated with clinical remission after 2 years in a group of patients with early RA, which highlights the use of DAS28 as a prognostic marker.
A number of studies have looked at predictors of response to methotrexate, including C-reactive protein (CRP) levels, RF status, structural damage, female sex and baseline Health Assessment Questionnaire score. Also, some reports suggest that baseline characteristics, such as elevated DAS or CRP levels, are associated with greater responses to rituximab. However, for biological agents in RA, it would appear that there are few baseline and clinical characteristics sufficiently predictive of response to guide treatment at present. Further evidence on whether or not certain baseline characteristics could act as predictive markers for response to biological agents (rather than prognostic markers), such as data generated from case–control studies or detailed comparison of randomised, placebo-controlled trials where the placebo arm eliminates the impact of prognostic factors, is required to support the findings from the current cross-sectional analyses.
Genetic Markers
Despite genetic susceptibility markers such as the shared epitope (SE), recognition that variation in response to treatment may be linked to genetic traits led to the study of genetic markers as predictors of response to treatment. Such analyses provide a step towards using genetics in a fully translational approach, from both a screening and therapeutic response perspective, to informing clinical practice. The identification of clinically important hereditary traits associated with response would be a major advance towards individualised medicine. Genes encoding proteins involved in the immune response continue to be studied to determine whether they are robust markers to predict response to TNF inhibitors. Specifically, polymorphisms in genes encoding for TNFα, the major histocompatibility complex region, the p38 network, STAT4, PTPN22, PADI4, CTLA-4, Traf1/C5 and the Fcγ receptor IIIA (FCγRIIIA) have all been the focus of a number of studies. In non-Hodgkin's lymphoma and systemic lupus erythematosus the efficacy of rituximab treatment has been shown to be related to the FCγRIIIA gene polymorphism 158VF and recently in RA, a reduction in the major activating FCγR expressed on natural killer cells (FcγRIIIA/CD16), which may be important in the mechanism of action of rituximab, was associated with non-response to rituximab.
TNFα Gene Polymorphisms
The TNF gene loci are obvious candidates for influencing response to TNF inhibitors. Several polymorphic regions of the TNF locus have been identified and their potential as markers for TNF inhibitor response has been studied, including single nucleotide polymorphisms (SNPs) at positions −308, −238 and −857 of the TNF promoter genes and −676 and −196 of the TNF receptor genes.
The SNP at position −308 of the TNF promoter gene is the best studied potential genetic marker for response to TNF inhibitors. Several studies, including two meta-analyses, have suggested that the TNF −308GG genotype is associated (often significantly) with a better general response to TNF inhibitor therapy than the TNF −308AA polymorphism. However, when the association between TNF −308 genotype and response to specific TNF inhibitors was analysed, the results were inconsistent. Indeed, some studies report that TNF −308GG can predict a better response to treatment with etanercept, adalimumab and infliximab than TNF −308AA or −308AG. In contrast, the various polymorphisms have also been shown to have no significant influence on response to these agents (Table 2). The most compelling evidence comes from a meta-analysis, which includes the largest cohort of patients with RA that has been assessed for association between TNF −308 polymorphisms and response to TNF inhibitors. The outcome of this meta-analysis was that there is no association between these polymorphisms and response, therefore, these polymorphisms cannot be considered as predictive markers. While a large pharmacogenetics study also suggested no association between response to adalimumab and TNF −308 genotype, the presence of the −238G/–308G/–857C haplotype (the most frequent among Western populations) was associated with a lower response to adalimumab than for patients with other haplotypes.
However, −308 polymorphisms have been assessed for their potential as prognostic markers. One study in Mexican patients with RA showed an increased frequency of −308 T2 (A) allele in severe RA in comparison with non-severe disease and healthy controls. Another study showed an association between −308 polymorphism and radiographic damage. Therefore, these studies suggest that patients might be screened for −308 polymorphisms to assess their likelihood of worse outcome—that is, −308 polymorphisms may be useful as prognostic markers.
Studies of SNPs of TNF receptor genes have also failed to provide definitive proof of their predictive ability. The TNF receptor II 676TG genotype was associated with a significantly lower ACR response than the 676TT genotype in patients after 3 and 12 months of treatment with infliximab, etanercept or adalimumab. The TNF receptor superfamily 1B (TNFSF1B) gene 676TG polymorphism was shown not to be associated with TNF inhibitor (infliximab and adalimumab) response after 3 or 6 months of treatment. A significant correlation was found between TNFSF1B gene −196 polymorphisms and poor response to infliximab in a small cohort of patients with RA.
Thus, conflicting evidence has failed to provide a consistent genetic marker predictive for treatment response to TNF inhibitors. However, there may be some merit in further investigation of the SNP in the TNF −308 promoter region to confirm its potential. The data are not conclusive because studies to date have been limited by the small patient numbers, heterogeneity in the underlying genetic background, differences in study design such that different outcomes at different time points have been assessed and the low frequency of the TNF −308AA genotype, which is particularly rare. The meta-analysis by Pavy et al attempted to reduce the limitations associated with meta-analyses as they only included studies that shared a common outcome and assessed a larger sample size than previous meta-analyses in this area, with sensitivity analyses suggesting that the data and conclusions from their meta-analysis are robust. Larger, well-designed studies should help to determine if the TNF −308GG genotype is a positive marker for response to all TNF inhibitors or just individual TNF inhibitors.
Shared Epitope
Certain major histocompatibility complex alleles, such as the SE in the human leucocyte antigen (HLA) region, confer susceptibility to RA. Several studies have investigated carriage of the SE sequence, but most have found no association with response to TNF inhibitor therapy. Indeed, the proportion of SE carriers was similar in responders and non-responders to infliximab, while a similar lack of association was shown when patients with RA were treated with etanercept or adalimumab. In contrast, a study of 457 patients with early RA showed that the presence of two HLA-DRB1 alleles encoding the SE (compared with the presence of less than two alleles) was associated with ACR50 response at 12 months after treatment with etanercept. Another study showed that an increased number of HLA-DRB1 SE copies was directly correlated with an improved clinical response to adalimumab.
The most convincing results to date come from an analysis of the British Society for Rheumatology Biologics Register, which was sufficiently powered to detect an effect, and showed no association between the SE motif and response to TNF inhibitor therapy. This is particularly noteworthy as presence of the SE has been identified as being very highly correlated with the production of anticitrullinated protein antibodies (ACPAs). A recent study has indicated the relative role of SE and PTPN22, combined with additional loci, for predisposition to RA.
On the whole, these studies indicate that the SE motif is not a robust genetic marker for predicting response to TNF inhibitors, but may instead reflect susceptibility for RA and represent a classic example for use as a prognostic biomarker, which is closely related to ACPA production and, usually, equally distributed among patients in randomised clinical trials.
Other Gene Polymorphisms
p38 Mitogen-activated protein kinases have been implicated in many pathogenic processes in RA. In a large cohort of patients treated with TNF inhibitors, SNPs spanning several genes from the p38 network were associated with clinical response to the TNF inhibitors, with greater potential for response shown with the monoclonal antibody TNF inhibitors than with etanercept. However, a highly liberal significance level (p=0.1) was used in this study and so further investigation is needed to confirm these associations.
A variant in the PTPN22 gene is a recognised susceptibility factor for RA. In a cohort of UK patients treated with TNF inhibitors, linear regression analysis showed there to be no association between change in DAS28 and PTPN22*620W carriage. There has been very little research on the association between therapeutic response and the presence of this polymorphism.
TNF inhibitors may exert their effects in part via immunoglobulin (Ig)G1 Fc and therefore another potential gene marker is FcγRIIIA. Despite a small cohort study suggesting that FcγRIIIA genotype variants may be useful genetic markers for response to TNF inhibitors, larger studies support the contrasting viewpoint. However, the possibility cannot be excluded that this genetic marker is related to disease activity, which can be a confounding factor and has been identified by analyses of registry data as a potential marker of response.
Variants in several other genes have shown an association with susceptibility to RA, including STAT4, TRAF1-C5, PADI4 and CTLA-4, which are associated with increased autoantibody levels in patients with RA. However, there does not appear to be any research into the use of these genetic markers as predictors of response to RA treatment. Given that all the alleles associated with RA are common in healthy people of European ancestry, it is likely that additional, as yet undiscovered genetic variants, may be involved and may be better predictors of response.
Are Genetic Markers Predictive of Response to Biological RA Treatments?
Despite some promising genetic factors being identified as potential predictive markers for response, there are limitations to these analyses. For example, many of the studies are underpowered to detect what may be relatively small differences in treatment effects; they were performed in different background populations; differences in study design and response criteria have made it difficult to make comparisons across studies; and they have focused on only a small number of candidate genes, which is less effective than analysis of the whole genome. Moreover, the differing mechanisms of action of the three TNF inhibitors may explain why certain polymorphisms show variation in their ability to predict response to these agents. Alternatively, certain polymorphisms may simply be prognostic markers, as shown by their association with disease severity or radiological damage. Furthermore, a number of prognostic markers have been identified, mostly in patients with severe disease. Therefore, studies examining these different domains are needed, but so far the challenge to identify activity-independent predictive markers is still ongoing. Study of appropriately sized patient populations is an important factor in demonstrating an association between genetic markers and treatment response to provide the evidence for everyday clinical decision-making. Therefore, well-designed studies are required to define the potential of the most promising genetic markers. New methods to uncover biomarkers continue to be developed and studied.
mRNA Analysis
Analysis of mRNA expression is an emerging field in the identification of predictive biomarkers and, consequently, there have been few studies in this area to date. An analysis of TNFα mRNA expression in whole-blood samples before and after treatment with infliximab (at baseline and 22 weeks) found no significant difference between responders and non-responders. In another recent study, responders to adalimumab were identified as having enhanced CD11c mRNA levels in purified blood monocytes compared with non-responders.
A more promising approach involves simultaneous monitoring of the expression levels of thousands of genes to identify molecular patterns that predict therapeutic response. This high-throughput microarray analysis has identified unique sets of genes with differentially regulated expression in infliximab responders compared with non-responders. Analysis of peripheral blood cells found that ~15% of the total genes exhibited a >1.5-fold change in expression after infliximab treatment, including ribosomal, degradation/apoptosis and metabolism-related genes. Eighteen genes were identified that were differentially expressed between responders and non-responders, with most being interferon (IFN)-related inflammatory response genes. mRNA expression patterns in synovial tissue samples found 279 differentially expressed genes between responders and non-responders to infliximab, including matrix metalloproteinase-3 (MMP-3) upregulation in good responders.
Most of the limited number of mRNA expression studies to date have analysed samples from peripheral blood rather than tissue samples taken from the synovium, which is the active site of cytokines and effector molecules. This may partly explain the differences in associations between mRNA expression and clinical response. Despite this limitation, this new approach may have a future role in the identification of further candidate genes and protein biomarkers involved in therapeutic responses. However, to produce valid scientific conclusions, defined and purified peripheral mononuclear cell populations are necessary to study this type of analysis initially with the potential to expand to a global blood profile. Furthermore, the sophisticated techniques required to predict response based on transcriptional profiles are more technically challenging and therefore less widely available and less easily applied to clinical practice. Although such detailed studies are required to improve our understanding of the immunopathogenesis of RA and potentially may allow better prediction based on studies of individual cell types, it may also be sufficient for diagnostic purposes to use peripheral blood mononuclear cells. To date, such studies have not been thoroughly performed.