Blood protein signatures may hint at which rheumatoid arthritis drug will work first

Choosing the first targeted therapy for rheumatoid arthritis has long been a matter of educated guesswork. Doctors weigh comorbidities, safety warnings, patient preferences and cost, but no validated molecular test tells them whether a newly diagnosed patient will respond better to a Janus kinase inhibitor (JAKi) or a tumour necrosis factor inhibitor (TNFi). A new study published in Biomarker Research reports an exploratory framework that could change that, by deriving drug-specific protein signatures from immune cells in the laboratory and testing whether those signatures, measured in a patient’s blood before treatment, track with later clinical response.

The research team, led by investigators at Reina Sofía University Hospital in Córdoba, Spain, worked with 271 rheumatoid arthritis patients across five independent cohorts. The core discovery cohort comprised 48 patients who had never received a biologic or targeted synthetic DMARD, meaning their immune systems had not yet been reshaped by prior advanced therapy. From these patients, the researchers isolated peripheral blood mononuclear cells (PBMCs) and neutrophils, the two cell types that drive much of the inflammation and joint damage in rheumatoid arthritis, and cultured them with the patients’ own serum.

Into these cultures the team added either baricitinib, a JAK inhibitor, or etanercept, a TNF inhibitor, both at 10 micromolar concentrations. Over 24 hours for PBMCs and 12 hours for neutrophils, they measured how the drugs affected cell proliferation, adhesion and neutrophil extracellular trap formation, a inflammatory cell-death process known as NETosis. They also profiled 92 inflammation-related proteins in the culture supernatants using proximity extension assay technology. Both drugs broadly suppressed the tested cell functions, but the protein programmes they engaged were strikingly different: only six proteins were modulated by both treatments, while the vast majority of changes were specific to one drug class.

The JAK inhibitor signature included downregulation of cytokine signalling proteins such as CSF-1, IL-10RA and IL5, proliferation markers like CD40 and FGF-21, and adhesion and migration proteins including CCL28, CD5, CD6 and uPA, with CXCL5 the only protein significantly increased in PBMCs. The TNF inhibitor signature instead featured reductions in inflammatory mediators such as IFN-γ, IL-1α, IL-17A, IL6, OSM and TNF itself, along with matrix-remodelling and tissue-damage proteins including MMP-1, MMP-10 and CASP-8. Enrichment analysis linked the JAKi profile to JAK-STAT signalling and cell proliferation, while the TNFi profile mapped onto MAPK cascade regulation, leukocyte migration and responses to interferon-γ.

The critical question was whether these laboratory-derived signatures meant anything in real patients. The researchers measured the same proteins in baseline serum collected before treatment from four independent cohorts: 32 non-naïve and 36 treatment-naïve patients starting JAK inhibitors, and 30 non-naïve and 125 treatment-naïve patients starting TNF inhibitors. In previously treated patients, unsupervised clustering found no clear molecular subgroups. But in b/tsDMARD-naïve patients, both signatures cleanly separated each cohort into two clusters, one with low and one with high baseline expression of the drug-specific proteins.

Those clusters tracked with clinical outcomes. In the naïve JAKi cohort, patients in the high-expression cluster showed larger improvements in disease activity, with mean DAS28-CRP reductions of 2.89 versus 1.65 points at three months, 3.38 versus 2.03 at six months, and 2.84 versus 1.89 at twelve months, all statistically significant. A parallel pattern appeared in the naïve TNFi cohort, where high-signature patients achieved greater disease-activity reductions at six and twelve months. The clustering persisted after the researchers statistically adjusted each protein for baseline disease activity, with cluster assignments remaining concordant in 83.3 percent of JAKi-treated and 93.6 percent of TNFi-treated naïve patients, indicating that the stratification was not simply a proxy for how sick patients were at the start.

Cross-therapy analyses sharpened the treatment specificity of the signal. When the TNFi signature was applied to naïve patients who actually received JAK inhibitors, those in the high-expression cluster were significantly less likely to reach remission or low disease activity at twelve months. The mirror image held for the JAKi signature in TNFi-treated patients. In other words, a baseline profile resembling the response to one drug class predicted not only benefit from that class but also a reduced likelihood of remission with the alternative, an opposing pattern that underpins the idea of biomarker-guided selection between the two therapies.

To formalise this, the team built machine-learning models using the 50 proteins from the combined ex vivo signatures to distinguish two clinically opposite groups among naïve patients: those who achieved six-month remission with TNFi but not JAKi, and those who achieved remission with JAKi but not TNFi. Under repeated nested cross-validation, with feature selection and tuning confined to training data, logistic regression performed best, achieving a mean area under the ROC curve of 0.70, with a spread of 0.65 to 0.75 across the 2.5th to 97.5th percentiles. Neural network and random forest models scored 0.66 and 0.64, respectively. The most consistently selected proteins included CDCP1 and PD-L1, retained in every training partition, along with 4E-BP1, IL-24, NT-3, IL-17C, EN-RAGE, LAP TGF-β1 and TSLP, drawing from both the JAKi- and TNFi-specific signatures. In previously treated patients, the same framework collapsed, with an AUC of just 0.43, essentially no discrimination.

The study also probed what remission looks like biologically. In a subset of naïve patients who achieved remission at six months, paired baseline and follow-up sera revealed largely treatment-specific circulating protein changes, with only a small shared set including CXCL10, IL-17A, IL6, LIF, MCP-3, NT-3, SCF and TWEAK modulated by both drugs. Thirteen proteins overlapped with the ex vivo signatures, linking the cellular drug response to systemic change. When these paired sera were applied to a three-dimensional synovial micromass model built from fibroblast-like synoviocytes and endothelial cells, baseline serum thickened the synovial lining, while six-month serum from remission achievers of either therapy significantly reduced lining thickness, showing that distinct molecular routes converged on similar functional improvement in synovial tissue.

The authors are careful about what these results do and do not establish. The cohorts were small and imbalanced, no formal sample-size calculation was performed, the design was observational and non-randomised, and the model has not been validated externally in an independent naïve population. Each patient received only one treatment, so individual counterfactual responses cannot be assessed, although treatment choices were made independently of the proteomic analyses. The findings support further investigation of a baseline biomarker panel to inform first-line selection between JAK and TNF inhibitors, but prospective validation of a locked model in larger independent cohorts is required before any clinical application.

Subject of Research: Ex vivo-derived JAK and TNF inhibitor protein signatures as candidate biomarkers for first-line therapy selection in rheumatoid arthritis

Article Title: Ex vivo JAK and TNF inhibitor–modulated immune signatures guide first-line therapy selection in rheumatoid arthritis

Article References: Corrales, S., Pérez-Sánchez, C., Cerdó, T., Merlo, C., Malvar-Fernández, B., García, S., Muñoz-Barrera, L., Sanchez-Pareja, I., Formanti, L., Moreno-Caño, E., Ortega-Castro, R., Calvo, J., Aranda-Varela, C., Ladehesa, L., Font, P., Abalos-Aguilera, M. C., Ruiz-Vilchez, D., Aguirre, M. Á., Barbarroja, N., … López-Pedrera, C. (2026). Ex vivo JAK and TNF inhibitor–modulated immune signatures guide first-line therapy selection in rheumatoid arthritis. Biomarker Research, 14(1), Article 146. https://doi.org/10.1186/s40364-026-01029-4

Image Credits: AI Generated

DOI: 10.1186/s40364-026-01029-4

Keywords: rheumatoid arthritis, JAK inhibitors, TNF inhibitors, proteomics, biomarkers, machine learning, immune cells, PBMCs, neutrophils, 3D synovial model, remission, precision medicine