Rheumatoid Arthritis and Predicting Treatment Response

Molecular signature models accurately predicted patient responses to adalimumab and etanercept treatment, paving the path toward personalized rheumatoid arthritis therapy.

As rheumatoid arthritis (RA) treatment continues to improve, more patients are experiencing disease remission, but even with new biologic agents, about 30% of patients do not respond well to them. With advances in machine learning, however, personalized RA treatment is becoming an increasingly reachable goal.

Researchers at Utrecht University in The Netherlands are studying how responses to treatment with tumor necrosis factor (TNF) inhibitors – drugs that help suppress the inflammatory response – can be predicted in patients diagnosed with rheumatoid arthritis.¹ Using gene expression and/or DNA methylation profiling on immune cells and peripheral blood mononuclear cells, along with profiling of clinical factors, a patient’s response to treatment may be predicted before treatment is initiated.

A chronic autoimmune disease that leads to joint inflammation and destruction, RA is typically treated initially with methotrexate or another conventional synthetic disease-modifying antirheumatic drug (csDMARD) to lower disease activity. If csDMARD therapy fails or loses efficacy, patients may be switched to a biologic DMARD (bDMARD) such as a TNF inhibitor.

Predicting response to treatment is, therefore, a strong unmet need in rheumatoid arthritis; machine models may soon pave the way. (Image: iStock)

Different TNF inhibitors do not have the same clinical effect in all patients, leading some to try a second or even a third medication before they experience relief. Being able to predict which TNF inhibitor would be effective and, therefore, should be the first choice of treatment would be highly beneficial in reducing the time to achieve effective treatment and disease remission. Furthermore, biologic agents are costly, and treatment failure elevates the risk of adverse events, such as infections.

Predicting response to treatment is, therefore, a strong unmet need in rheumatoid arthritis.

Rheumatoid Arthritis Treatment and Gaps in Clinical Predictors

Researchers have identified several potential predictive factors of RA remission and response to biologic therapy – including age, sex, disease duration, disease activity, smoking status, and concomitant methotrexate therapy.^2,3 Others developed a matrix tool to predict remission and low disease activity in patients treated with golimumab that was based on sex, the Health Assessment Questionnaire, presence of comorbidities, age, tender joint count, and erythrocyte sedimentation rate.^4,5

These studies did not illuminate the biological mechanisms that underlie the differential response to different TNF inhibitors, however, nor did they examine potential treatment responses to individual TNF inhibitors.¹

Genetic Profiling and TNFi Response: One Inhibitor at a Time

To understand biologic processes associated with the anti-TNF response, more recent studies have been conducted using the synovium and blood from patients diagnosed with RA. This focused research has been able to show that transcriptomic and epigenetic profiling may be able to predict a patient’s response to anti-TNF therapy before treatment. However, the gene signature for predicting response is unique for each TNF inhibitor and it has been impossible to predict if patients who fail to respond to one TNF inhibitor will respond to another one.

Investigating the role of different cell types, especially immune cells, in people with RA who are receiving different anti-TNF therapies could reveal the biologic process involved in a specific TNF inhibitor.

Machine Learning Predicts Response in Two TNFis

In their research, Weiyang Tao and colleagues at Utrecht University sought to generate cell-specific profiles that could predict the response to two TNF inhibitors – adalimumab (ADA) and etanercept (ETN) – prior to treatment initiation for rheumatoid arthritis.¹ ADA is the first fully human therapeutic anti-TNF monoclonal antibody. ETN is a recombinant human TNF receptor (p75)-Fc fusion protein that competitively inhibits TNF.

Researchers identified 80 patients diagnosed with RA who were eligible for treatment with both TNF inhibitors and non-TNF inhibitors and were enrolled in Utrecht University’s Medical Center BiOCURA cohort. Blood samples were obtained at baseline, and patients were treated with subcutaneous ADA or ETN for 6 months. Peripheral blood mononuclear cells (PBMCs) were isolated from the blood samples, and genome-wide gene expression and DNA methylation profiling using RNA sequencing and methylation identification technology were performed on the cells.

CD4+ T cells and CD14+ monocytes were then isolated from the PBMCs for RNA sequencing to gain additional insight into different responses to anti-TNF. Researchers then analyzed these responses to identify transcriptional and epigenetic signatures that were clearly associated with a response to ADA or ETN treatment and used them to build machine learning models to predict patient responses. Nine patients were used to validate the predictions made by the machine learning models.

Patients were classified as responders or non-responders according to the European League Against Rheumatism (EULAR) criteria after 6 months of treatment. Non-responders to ADA were switched to ETN, and non-responders to ETN were switched to ADA for another 6 months of treatment.

At the 6-months follow-up, 53% of patients receiving ADA and 45% of patients receiving ETN were considered to be responders. The researchers found that patients classified as responders had significantly different molecular signatures than did the non-responders, yet at baseline, no significant differences in clinical parameters were observed between the two groups. There was very little overlap in differently expressed genes, or DEGs, between the ADA and ETN cohorts, suggesting that patients’ responses to ADA and ETN are defined by distinct gene signatures. Neither group, though, showed an association between TNF expression and response to the TNF inhibitor.

TNF Inhibitors: Individual Inhibitor Studies Suggest Differences within the TNF Class

Responders to ADA and ETN also showed distinct differences in hypermethylation patterns. DNA methylation has been shown to play an important role in the progression of RA. A previous study did not detect such differences,⁶  which Tao attributed to differences in study design and use of older analysis technology. In this study, ETN responders showed strong hypermethylation, which suggests that epigenetics also has a role in defining responses to ADA and ETN in the blood cells.

These differences suggested that the two drugs may have different mechanisms of action. The majority of the patients did not respond to both ADA and ETN but had the potential to respond to one of the drugs. The genetic and epigenetic differences between individual patients therefore determine the drug response.

“Given such different molecular signatures, ADA and ETN should be studied and considered differently in the future although both are TNF inhibitors,” wrote Tao.¹

TNF Inhibitors for Rheumatoid Arthritis Treatment: Practical Takeaways

Tao’s study found, for the first time, that transcription signatures in ADA and ETN responders are divergent in PBMCs, as well as in monocytes and CD4+ T cells. In addition, differentially methylated positions (DMPs) of responders to ETN but not ADA were strongly hypermethylated. The machine learning models based on these molecular signatures could accurately predict patients’ responses to ADA and ETN even before treatment was initiated, which, according to the study authors, paves the path toward personalized treatment strategies with TNF inhibitors.    

See also, a commentary on the US rheumatology shortage.