OP0089 RNAseq PROFILING OF RHEUMATOID ARTHRITIS PATIENTS TO PREDICT RESPONSE TO TOFACITINIB

Background Rheumatoid arthritis (RA) is characterized by chronic inflammatory synovitis and progressive disability. Nowadays the natural history and the prognosis of the disease has deeply changed thanks to the availability of conventional and targeted disease modifying anti-rheumatic drugs (DMARDs). Nevertheless, clinical response to treatment is highly variable and a considerable proportion of patients do not - or only partially - respond to the treatment. The stratification of patients through the identification of biological predictors of good clinical response to treatment could allow the selection of the best therapeutic strategy for each patient. Recently, precision medicine in rheumatology could rely on the development of new transcriptomics techniques, such as RNA sequencing (RNAseq), that permit to explore complex cellular and molecular networks associated with pathophysiological aspects. Objectives To characterize the transcriptomic profiles (RNAseq) of whole-blood samples of moderate to severe RA patients to predict the clinical response to tofacitinib. Methods We selected patients with active RA, candidate to receive tofacitinib after failure of a previous therapy with conventional or biologic DMARDs. RNAseq profiling on whole-blood samples (PAX gene tubes) was performed at baseline and after 24 weeks of treatment with tofacitinib. Treatment response was evaluated at week 24 by CDAI EULAR response criteria as drug responder (major or moderate response) or non-responder (minor or no response). Differential expression of gene ontology (GO) biological processes (BP) pathways was analyzed to identify the response-associated pathways. Machine learning models were built by different data mining algorithm to predict response to tofacitinib. Results The study population included 33 patients, of whom 7 discontinued prematurely the treatment (5 because of adverse events and 2 because of inefficacy) and 2 were lost to follow-up. At week 24, 10 patients (38.5%) had major, 4 (15.4%) moderate, 6 (23.1%) minor, and 6 (23.1%) no clinical response to tofacitinib. No significant difference was observed between responders and non-responders in terms of baseline characteristics (age, sex, rheumatoid factor and anti–cyclic citrullinated peptide [ACPA] positivity, DMARDs and prednisone exposure, Table 1). Overall, 307 out of 2137 coding transcripts and 85 GO BP pathways were differentially expressed after treatment in responders vs non-responders (Figure 1). In detail, an up-regulation of JAK-dependent pathways, including cation transport, metabolism of membrane lipids, calcium-mediated signaling, and osteoblast proliferation was observed in responders. Conversely, in non-responders, processes related to B cell activation, proliferation, and signaling along with mRNA epigenetic modifications were increased. After extensive internal validation (50 runs of 80:20 hold-out sampling) with univariate selection of GO terms, the accuracy of Support vector Machine (SVM) learning models to predict the correct clinical response was equal to 88.3% (AUROC = 0.940). Table 1. Demographic and baseline clinical characteristics of the study population* Non responders(n = 12) Responders(n = 14) Female, n (%) 11 (91.7%) 11 (78.6%) Age, years (±SD) 52.6±11.2 54.3±10.5 Rheumatoid factor, n (%) 7 (58.3%) 7 (50%) ACPA, n (%) 9 (75%) 8 (57.1%) Concomitant csDMARDs, n (%) 5 (41.7%) 8 (57.1%) bsDMARDs naïve, n (%) 12 (100%) 11 (78.6%) Prednisone ≥ 5mg/day, n (%) 4 (33.3%) 6 (42.8%) * There were no significant differences, determined by Fisher’s exact test for categorical variables Figure 1. Heatmap representation of selected GO BP pathways in responders/non-responders. Conclusion Machine learning models based on transcriptomic functional pathways can accurately predict response to tofacitinib. Our study could contribute to improve the treatment customization and the optimization of RA treatment strategy toward a personalized approach. Furthermore, these findings may help to understand the mechanisms underlying the clinical response to JAK inhibitors. Disclosure of Interests None declared.