Background: Rheumatoid arthritis (RA) is a remarkably heterogeneous autoimmune disease whose clinical outcomes with disease-modifying antirheumatic drugs (DMARDs) remain unpredictable in patients. Biomarker identification and personalised medicine is an imperative need.

Objectives: To characterize the molecular landscape of RA patients, by using a multi-omic approach involving transcriptomics and proteomics and assess its association with disease status and clinical response.

Methods: PBMCs from 149 subjects, including 27 healthy donors and 123 RA patients underwent RNAseq on Illumina platforms. The RA cohort included 39 biologics-naïve patients before receiving TNFi and 26 patients before receiving JAKinibs. Clinical outcomes were assessed after 3 months following EULAR criteria. Gene expression data were projected into gene pathway modules (gene signatures) using a validated functional annotation approach (‘scoring personalized molecular portraits’). Hierarchical clustering was conducted to identify patients’ subgroups based on transcriptomic profiles. Concurrently, 92 inflammatory mediators in RA serum were analyzed using Olink platform. Machine learning models assessed molecular signatures for predicting therapeutic response.

Results: Unsupervised clustering identified three clusters (CL). CL1 exhibited a pronounced myeloid and inflamed profile, characterized by diminished T-cell levels, in stark contrast to CL2, which mirrored a healthy profile. CL3 showed moderate changes, with increased B-cell signature in RA patients. Clinically, CL1 showed the highest status of disease activity, longest evolution time and greatest innate immune cells count. Correlation analysis of deregulated gene modules and disease features underscored a positive relationship between the disease severity and elevated levels of myeloid and inflammation modules, while inversely correlating with reduced T-cell modules. This observation suggests a potential link to the phenomenon of T-cell exhaustion. Alterations in 22 inflammation-related proteins were found among clusters, which exhibited positive correlation with expression levels of gene modules associated with inflammation, IFN pathway and myeloid cell activity. Association analysis found baseline gene modules differences between responders (R) and non-responders (NR) to JAKi or TNFi at 3 months. Heightened gene modules associated with T cell activation were linked to NR to JAKi, while increased gene modules tied to the IFN pathway were associated with NR to TNFi. Two gene signatures (involving the top 10 differentially expressed genes in these modules) distinguished R from NR to TNFi and JAKi. Machine learning approaches demonstrated the great potential of these signatures as predictive models of response to these therapies. Lastly, by analyzing the top 15 inflammation-related proteins at baseline between R and NR for both therapies, consistent opposing patterns emerged. Elevated chemokines and cytokines were found in R to JAKi but NR to TNFi, while reduced levels appeared in NR to JAKi but R to TNFi.