An expanding range of biological therapies is available for patients with rheumatoid arthritis. Clinical trials and real-life experience demonstrate significant interpatient heterogeneity in efficacy as well as important adverse effects of these treatments. In order to maximise their benefit:risk ratios and to minimise later joint damage, we need to define predictors of response and, ideally, of adverse effects for each of these drugs. There is huge interest in this field of ‘personalised medicine’, which should allow us to optimally match patient with treatment, providing the parallel benefit of reduced treatment costs. In this short article the current state of the art for licensed biological therapies is summarised. There have been some noteworthy discoveries but the challenge is now to design studies to confirm and validate these findings while also devising large, potentially international, collaborations to identify additional, robust biomarkers that predict outcome.
Statistics from Altmetric.com
Increased understanding of the molecular pathogenesis of rheumatoid arthritis (RA) has led to the development of a range of biological treatments targeted at specific components of the immune response. Although the quality of response to these drugs can be very high, the drugs fail to produce a response in a substantial proportion of patients. The high cost of these drugs suggests that a personalised approach to management is desirable, in which specific treatments are matched to individual patients. This would also minimise periods of disease activity and patient exposure to the potential side effects of an ineffective treatment.
Biomarkers of drug response may be present in a patient's serum, genetic material or, potentially, synovium. Clinical and demographic features may also provide clues. In this viewpoint, we briefly survey the literature and provide our opinion on future progress towards personalised medicine for RA. While we focus on drug-specific considerations, other features such as prognostic markers must also be taken into account when considering the therapeutic strategy for an individual patient.
Predicting response to tumour necrosis factor inhibitors
A large number of pharmacogenetic studies have so far failed to identify gene polymorphisms associated with response to tumour necrosis factor (TNF) inhibitors.1 For example, the –308 promoter polymorphism in the TNF gene has been studied by several groups with conflicting results. Small sample sizes and heterogeneity in study design may underlie some of the inconsistency, but registry studies and even meta-analyses have failed to provide a conclusive answer.2,–,5 Associations have also been sought with genes and pathways linked to disease pathogenesis or drug pharmacodynamics, such as the type II TNF receptor, Fc γ receptors and innate immune pathways.6,–,8 Associations that have been demonstrated await replication, and some are with one particular TNF inhibitor only, possibly reflecting pharmacodynamic differences between the drugs. A limitation is the need for large patient cohorts to demonstrate what are likely to be relatively small differences in treatment effects.9 Transcriptional profiling of peripheral blood cells may also yield useful information. A recent study demonstrated increased CD11c transcription in peripheral blood monocytes in responders to adalimumab.10 This association was lost in patients who were co-prescribed methotrexate, however, illustrating the importance of study design to allow for patient heterogeneity. Two further studies have identified transcriptional profiles associated with subsequent response to infliximab.11 12 As with pharmacogenetic studies, these findings require replication before being considered potentially useful in the clinic.
Demographic factors and baseline clinical characteristics have been shown to influence treatment response in some studies. Data from the British Society for Rheumatology Biologics Registry (BSRBR) suggest that a high level of disability at baseline is associated with a poor response to TNF inhibitors. Not administering non-steroidal anti-inflammatory drugs or concomitant methotrexate also reduces the likelihood of response, particularly to etanercept, and smokers are less likely to respond to infliximab. Women are less likely to achieve remission.13 Data from a registry in southern Sweden also link low baseline disability and concomitant disease-modifying antirheumatic drug treatment with a better outcome from TNF inhibition.14 In that study, baseline Disease Activity Score (DAS) 28 was positively associated with the odds ratio of an American College of Rheumatology criteria (ACR)70 response but negatively associated with the odds ratio of achieving DAS remission, emphasising the importance of the outcome measures used and their interpretation: if a patient with a high DAS starts treatment, it may be easier to obtain an ACR response but harder to achieve a low DAS score than for a patient with a low baseline DAS. In the Danish DANBIO registry, older age, prednisolone co-therapy and lower functional status each predicted a poorer response to the first anti-TNF treatment, whereas the Italian GISEA study identified rheumatoid factor (RF) negativity, a lower Health Assessment Questionnaire score, age <53 years and male sex as independent predictors of remission at 6 months.15 16
Autoantibodies provide potential serum biomarkers of treatment response. There is some inconsistency but several studies suggest that seropositive patients respond less well to TNF inhibition or that the likelihood of response correlates inversely with autoantibody titre. Data on 521 patients from the BSRBR demonstrated that, after 6 months of anti-TNF therapy (mainly infliximab and etanercept), RF-positive patients had a DAS28 improvement of 2.43 compared with 3.03 for RF-negative patients (p=0.02). Equivalent data for anticitrullinated peptide antibody (ACPA) status were 2.40 (ACPA positive) versus 2.90 (ACPA negative), p=0.02.17 In a study of 225 etanercept recipients, the odds ratio for lack of EULAR response was 4.61 (95% CI 1.55 to 13.7) for patients with high-titre ACPA (>1600 U/ml) compared with seronegative patients.18 High levels of RF isotype IgA have also been associated with a poorer outcome.19 A recent study defined a 24-biomarker autoantibody and cytokine ‘signature’ that associated with a positive clinical response to etanercept.20 All these findings require confirmation in other patient cohorts, ideally in well-designed, prospective studies.
The site of disease could yield important treatment-response biomarkers. A recent study of 143 patients demonstrated higher synovial TNF levels and more synovial T cells and macrophages at baseline in infliximab responders. However, there was significant overlap between responders and non-responders.21 In a further study, synovial lymphoid aggregates were a significant predictor of therapeutic response.22 Microarray analysis of baseline synovium has also highlighted features that may predict adalimumab responsiveness.23
In summary, certain demographic and clinical factors, together with autoantibody status, may identify patients more or less likely to respond to TNF inhibition, but current data do not provide a conclusive answer. More sophisticated approaches based on transcriptional profiles and synovial biopsies may yield useful information but are more technically challenging and therefore less generally available.
Predicting response to rituximab
Elevated baseline C-reactive protein may be associated with a greater response to rituximab and there is increasing evidence linking autoantibody status with response.24 Despite the limited power of substudies using patients with seronegative RA, cumulative data suggest a relative lack of response to rituximab in this subset. In the phase IIB DANCER study, the response of RF-negative patients did not differ from that of patients receiving placebo.25 In the phase III REFLEX study, a response was documented in RF-negative patients but x-ray progression was not inhibited.26 27 Subanalyses of the IMAGE study have also suggested lack of efficacy of rituximab in methotrexate-naïve seronegative patients.28 29 Most recently, pooled data from two phase III trials of rituximab in patients with an inadequate response to methotrexate suggested that seropositivity for RF and/or ACPA was associated with an increased probability of achieving ACR responses and low DAS compared with patients seronegative for both autoantibodies.30 Furthermore, any RF serotype and IgG-ACPA, especially in patients with high C-reactive protein, may be associated with enhanced clinical responses.24 Recently, the presence at baseline of a circulating IgG greater than the upper limit of normal was shown to be associated with a better response to rituximab and to be synergitic with seropositivity.31 Furthermore, a study of 148 strictly seropositive patients demonstrated, by multivariate analysis, that four baseline variables predicted EULAR response in this patient subset at 6 months: no steroids, a low lymphocyte count, high RF-IgG and low circulating levels of B-lymphocyte stimulator (BLyS).32 A further multivariate analysis also highlighted RF (and not ACPA), low baseline HAQ score and fewer previous anti-TNF drugs as independent predictors of response.33
Thus, seropositivity may prove to be a useful biomarker of rituximab responsiveness. Whether this reflects a specific pathogenic effect of autoantibodies, whose levels fall with treatment, or a non-specific indicator of disease that is more B-cell dependent, remains to be demonstrated.34 Although B cells are the precursors of autoantibody-producing plasma cells, they have other important roles, including antigen presentation and cytokine production. It has been suggested that the greater improvement seen in seropositive patients indicates more effective B-cell depletion in these patients,35 and the effectiveness of rituximab may relate to the depth of B-cell depletion.36 Although baseline characteristics of synovial membrane did not predict clinical response to rituximab, the subsequent decline in synovial plasma cells did correlate with efficacy.34 In another study, low synovial B-cell infiltration and low serum IgM ACPA levels at baseline predicted a better outcome.37 As with TNF blockade, transcriptional analysis has been applied to peripheral blood in an attempt to predict response to rituximab.38 These studies each used small numbers of patients, and should be viewed as hypothesis-generating for future work.
Levels of BLyS rise after rituximab treatment and drive B-cell reconstitution. A single nucleotide polymorphism in the BLyS gene (−871C/T) correlates with circulating BLyS levels, the CC genotype predicting low levels both at baseline and after treatment. In a recent study, patients with the CC genotype who were also seropositive for RF were more likely to respond to rituximab: 83% (19/23) achieved at least an ACR50 response compared with 29% (5/17) of RF-negative patients of CT/TT genotype.39 This is the first study to usefully combine serological and genetic biomarkers to predict treatment response to rituximab but again, these data require confirmation in prospective cohorts.
Predicting response to other biological agents
Abatacept is a T-cell costimulation modulator, so RA-related T-cell biomarkers or gene polymorphisms, such as those at the PTPN22 R620W allele, may provide useful response biomarkers. Tocilizumab is a monoclonal antibody against the interleukin-6 receptor. Interleukin-6 has characteristic systemic effects, such as stimulating the acute phase response and platelet production while impairing iron utilisation, leading to anaemia of chronic disease. Therefore, patients with active RA and these clinical characteristics, together with other systemic features such as weight loss and fever, may respond well to tocilizumab. However, there have been no reports to date that systematically analyse potential response predictors with either of these drugs, and these comments are necessarily speculative.
There is a huge need for robust biomarkers of response to biological treatments to improve response rate, preserve joint structure and function and reduce treatment cost. To date, a number of clinical features have been linked with response to TNF blockade, and the presence of serum autoantibodies predicts response to rituximab in several studies. It is important to recognise that most biomarker studies use change in DAS or European League Against Rheumatism response categories to define treatment response, but not the extremes of a good response versus no response. Consequently, current response indicators may predict the probability of responding to a drug, or the quality of the response, but may not accurately identify non-responders. In this sense they may become most useful in defining the order of biological treatment for a particular patient rather than defining patients in whom a particular treatment should be withheld. On the other hand, data are accumulating to indicate that rituximab has limited therapeutic utility in seronegative patients. While this is an important observation, the new ACR/EULAR diagnostic criteria for RA place considerable weight on autoantibody status, and it is conceivable that future RA cohorts will be enriched for seropositive patients.40 However, for the present, it seems sensible to use other drugs before rituximab in patients with seronegative RA.
Biomarkers may be of limited value in isolation and should ideally be studied in combination with one another. At present, all continue to require validation but, with the current interest in this field and the rapid pace of developments, regular incremental advances are likely to be made towards the development of clinically useful algorithms that predict biological drug responsiveness. The subtle influence of pharmacogenetic markers requires international collaboration in the field, and appropriate emphasis should also be placed on study design, to ensure comparability between studies. Wherever possible clinical trials should be powered to seek and confirm biomarkers of therapeutic response. Lastly, it is imperative that registries incorporate, as far as possible, the ability to seek predictors of response. Ultimately these will contain many more patients than clinical trial databases and, while a more heterogeneous mix of patients, they should provide the necessary power to seek and combine robust response biomarkers in ‘real-life’ scenarios.
Support for third-party writing assistance for this manuscript was provided by Hoffman La Roche.
Competing interests JDI has consulted for Hoffman La Roche and Bristol-Myers Squibb, and received educational or research grants from Centocor, Wyeth and Abbott pharmaceuticals. GFF received fees for lectures from Roche, Abbott, Wyeth, Shering Plough and grant for research support from Wyeth, BMS and Roche
Provenence and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.