Profiling of gene expression biomarkers as a classifier of methotrexate nonresponse in patients with rheumatoid arthritis

Background: Approximately 30-40% of rheumatoid arthritis (RA) patients started on low dose methotrexate (MTX) will not benefit from treatment. To date, no reliable biomarkers of inefficacy to MTX in RA have been identified. Here we analyse whole blood samples taken at two time-points (pre-treatment and following 4-weeks on drug), to identify gene expression biomarkers of MTX response.
Method: RA patients about to commence treatment with MTX were selected from the Rheumatoid Arthritis Medication Study (RAMS). Using EULAR response criteria, 42 and 43 patients were categorised as good or non-responders, respectively, following 6-months on drug. Whole blood transcript data were generated, and supervised machine learning methods were used to predict EULAR non-response. Models including transcripts were compared to models including clinical covariates alone (e.g. baseline disease activity, gender). Gene network and ontology analysis was also performed.
Results: By taking the ratio of transcript values (i.e. the difference in log2-transformed expression values between 4-weeks and pre-treatment), a highly predictive classifier of non-response was developed using L2-regularized logistic regression (ROC AUC 0.78±0.11). This classifier was superior to models including clinical covariates (ROC AUC 0.63±0.06). Pathway analysis of gene networks revealed significant over-representation of type 1 interferon signalling pathway genes in non-responders at pre-treatment (p=2.8e-25) and at 4-weeks (p=4.9e-28).
Conclusion: Testing for changes in gene expression between pre-treatment and 4-week post-treatment may provide an early classifier for patients who are unlikely to benefit from MTX by 6-months and who should, therefore, have their treatment escalated more rapidly, thus potentially impacting treatment pathways.