Predicting clinical response in psoriatic arthritis through integrative analysis of transcriptomics and proteomics
Arthritis Research & Therapy, 2026
Bentvelzen M., el Bouhaddani S., Spierings J., Concepcion A., Vonkeman H., Mooij S., Schipper L., Herman A., Vreugdenhil S., Bhansing K., van Bijnen S., Bisoendial R., van Bon L., Jansen T., van Kuijk A., Kok M., Leijten E., Jahangier Z., van Tubergen A., Wijngaarden S., Kadir S., Comarniceanu A., Mastbergen S., Welsing P.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
Immunological & Inflammatory Diseases Dermatological Diseases | Patient Stratification | Serum |  Olink Target 96 |
Abstract
Background
The therapeutic response to disease-modifying antirheumatic drugs (DMARDs) remains relatively low in psoriatic arthritis (PsA), leading to delayed disease control and frequent treatment switches. Predictive biomarkers may enable personalized treatment and earlier disease control. We aimed to identify transcriptomic and proteomic markers for tofacitinib or comparator treatment outcomes and develop a prediction model to support treatment decisions in PsA patients.
Methods
Baseline CD4+ T-cell transcriptomics and proteomics data from 80 PsA patients in the development cohort of the TOFA-PREDICT trial were analyzed. The TOFA-PREDICT trial is a four-arm randomized trial that was designed to discover profiles of PsA patients that predict response to tofacitinib as compared with methotrexate or etanercept. Forty DMARD-naïve patients were randomized to tofacitinib or methotrexate, and 40 patients who failed DMARD-treatment were randomized to add-on tofacitinib or etanercept. Treatment response was defined as reaching minimal disease activity at 16 weeks. Feature selection was performed in the full cohort and in each treatment subgroup using XGBoost and sPLS-DA. Using different modeling strategies, prediction models were developed that combine clinical variables with transcriptomic, proteomic, or integrated multi-omics predictors. The models were cross-validated and compared using AUC-ROC and their ability to identify the most promising treatment (tofacitinib versus control) per patient.
Results
Fifty percent of patients responded to treatment. Eighteen transcriptomic, ten proteomic, and two clinical predictors were selected. The integrated multi-omics model incorporating treatment–predictor interactions achieved the highest performance (AUC = 0.70 ± 0.19; variation (SD) in treatment-effects in patients 15.2% ± 14.8%). The selected proteins were significantly interconnected (p-value = 3.41E-5) and related to immune system processes.
Conclusions
Integrated baseline gene and protein expression profiles combined with clinical variables can predict treatment response and identify differential treatment effects between individual patients. These findings demonstrate the potential of omics-guided personalized treatment for patients with PsA.