Ex vivo JAK and TNF inhibitor-modulated immune signatures guide first-line therapy selection in rheumatoid arthritis.

BACKGROUND: Rheumatoid arthritis (RA) is a heterogeneous autoimmune disease for which selecting the most appropriate first targeted therapy remains challenging. We integrated ex vivo immune-cell functional assays, targeted proteomics, clinical outcomes, and a three-dimensional (3D) synovial model to identify treatment-associated molecular signatures and evaluate their relationship with subsequent response to Janus kinase inhibitors (JAKi) or tumour necrosis factor inhibitors (TNFi). METHODS: Peripheral blood mononuclear cells and neutrophils from 48 b/tsDMARDs-naïve RA patients were cultured with autologous serum and treated ex vivo with Baricitinib or Etanercept to assess proliferation, adhesion, NETosis, and 92 inflammation-related proteins. The resulting JAKi- and TNFi-modulated protein signatures were subsequently evaluated in pretreatment serum from four independent RA treatment cohorts (n = 223; naïve and non-naïve). Unsupervised clustering was used to identify molecular subgroups, which were related to longitudinal DAS28-CRP outcomes at 3, 6, and 12 months. Logistic regression, neural network, and random forest models were evaluated using repeated nested cross-validation in naïve patients to discriminate treatment-specific 6-month remission outcomes. In a subset of naïve remission achievers, paired baseline and 6-month sera were applied to a 3D synovial micromass model. RESULTS: Ex vivo JAKi and TNFi comparably modulated immune cell functions but induced distinct protein signatures with limited overlap. When evaluated in pretreatment baseline serum, these signatures stratified b/tsDMARD-naïve patients into clusters with different baseline inflammatory profiles and treatment-specific ΔDAS28-CRP. High baseline expression of a drug-specific signature was associated with remission with the corresponding therapy and failure to achieve remission with the alternative treatment. Under repeated nested cross-validation, logistic regression showed the highest mean discrimination among the models, with an AUC of 0.70 (SD, 0.03; empirical 2.5th-97.5th percentile, 0.65-0.75) for distinguishing treatment-specific remission outcomes in b/tsDMARD-naïve RA. In remission achievers, 6-month sera from either therapy significantly reduced synovial lining thickness in 3D micromasses versus baseline. CONCLUSION: Ex vivo-derived JAKi- and TNFi-associated immune-related protein signatures, integrated within a machine-learning framework, showed moderate discriminatory ability for treatment-specific remission patterns in b/tsDMARD-naïve RA and, together with complementary evidence of reduced synovial lining thickness in remission achievers, support further investigation of biomarker-informed selection between JAKi and TNFi as first targeted therapies.