Journal: bioRxiv
Article Title: Patient-Derived Synovium-on-a-Chip Recapitulates Disease and Treatment Response and Permits Study of Resistance in Rheumatoid Arthritis
doi: 10.64898/2026.10.30.763140
Figure Lengend Snippet: (A) Workflow for treatment prediction using patient-derived RA synovium-on-a-chip. Patient-derived chips were subjected to parallel drug perturbation with tocilizumab (IL-6 receptor antagonist, IL-6Ra), infliximab (TNF inhibitor, TNFi) and tofacitinib (JAK inhibitor, JAKi). Four functional readouts were quantified: stromal (ECM remodeling), cytokine (IL-6 secretion), immune (CD45⁺ cell count) and recruitment (immune cell migration distance). Treatment-induced changes relative to matched untreated controls were calculated as fold changes (treated/untreated), standardized within each patient across the three drug treatments (IL-6Ra, TNFi and JAKi), and combined into a composite on-chip response score to predict therapeutic outcome; the four features are also displayed as radar plots for visualization. (B) Heatmap of treatment-induced changes in the four readouts across nine patients and three drug conditions per patient. Rows represent patients, and columns represent readout–treatment combinations. Colors indicate standardized treatment-induced changes, with red indicating increases and blue indicating decreases. Patients were hierarchically clustered using unsupervised Ward’s linkage (ward.D2). Top annotations indicate on-chip treatment; right annotations indicate clinical response history to TNFi and JAKi treatment. Each value represents the mean of 3 chips per condition. (C) Composite treatment response score (mean of the four z-score–normalized readouts) quantified for each patient across IL-6Ra, TNFi and JAKi treatments, demonstrating patient-specific and drug-specific variation in composite response magnitude. (D) Correlation the composite on-chip response score for the clinically matched treatment and clinical improvement measured as ΔDAS28–ESR. Six matched patient–treatment observations from 5 independent patients are shown. Circles indicate TNFi and diamonds indicate JAKi treatment. Pearson r = 0.844, P = 0.0348; Spearman ρ = 0.943, P = 0.0167; both tests were two-tailed. The line represents the fitted linear regression (y = 3.444x + 0.505), and the shaded band indicates 95% confidence interval of the fit. (E) Feature contribution to prediction assessed by LOOCV root mean square error (RMSE) for individual features (migration distance, IL-6, collagen IV and CD45⁺ count, grey), integrated composite score (red) and clinical predictors (baseline DAS28 and JIS-defined inflammatory endotype, blue). The dashed line shows the null model baseline; The integrated composite score achieved the lowest RMSE. The asterisk denotes a prediction error significantly lower than expected by chance for the composite score (permutation test, p = 0.021). (F) Chip-based predicted treatment-response categories across nine patients and three drug conditions. Composite on-chip scores were converted into predicted ΔDAS28–ESR values using the regression model in panel D and classified as good, moderate, or no response according to clinical standard (Good: ΔDAS28–ESR>1.2; Moderate: 0.6 <ΔDAS28– ESR ≤ 1.2; No response: ΔDAS28–ESR ≤ 0.6). Green, yellow, and gray indicate the respective predicted categories.
Article Snippet: Qualitative cytokine profiles were assessed using a Human Cytokine Array C5 membrane kit (catalogue AAH-CYT-5-8, RayBiotech) according to the manufacturer’s protocol.
Techniques: Derivative Assay, Functional Assay, Cell Characterization, Immune Cell, Migration, Two Tailed Test