Background: Rheumatoid arthritis (RA) and systemic lupus erythematosus (SLE) are heterogeneous autoimmune diseases with overlapping clinical and immunopathogenic features, yet distinct organ involvement, progression, and therapeutic response. This complexity challenges patient stratification and precision medicine. Multi-omics strategies integrating transcriptomic, proteomic, and clinical data can uncover molecular signatures to enhance prognosis and personalize therapy.

Objectives: 1) To define common and disease-specific molecular signatures in RA and SLE through integrative transcriptomic and proteomic analyses; 2) To identify cross-disease molecular patient subgroups associated with clinical profiles, organ damage, and cardiovascular (CV) risk; and 3) To develop machine learning-based predictive models for molecular stratification and personalized management.

Methods: The study included 96 RA patients, 66 SLE patients, and 42 healthy donors (HD). Detailed clinical characterization encompassed demographics, disease activity (DAS28, SDAI, CDAI in RA; SLEDAI in SLE), autoantibody profiles (rheumatoid factor, anti-citrullinated peptide antibodies, anti-dsDNA, antiphospholipid antibodies), organ damage indices, comorbidities, CV risk factors, and current treatments. Subclinical atherosclerosis was evaluated by carotid Doppler ultrasound, and renal involvement was assessed using clinical criteria and kidney biopsy. Peripheral blood mononuclear cells (PBMCs) were isolated for RNA sequencing, and serum levels of 92 inflammatory mediators were quantified using proximity extension assay (PEA). Analyses included differential gene expression, GSVA, clustering, MOFA, clinical correlations, and machine learning (XGBoost, Random Forest).

Results: Both RA and SLE exhibited markedly altered transcriptomic profiles compared with HD. In RA, 1,152 genes were upregulated and 1,990 downregulated, whereas SLE showed 1,942 upregulated and 2,045 downregulated genes. A substantial overlap of 1,356 differentially expressed genes was identified between diseases, reflecting shared pathogenic mechanisms. Unsupervised GSVA-based clustering identified two molecular clusters across RA and SLE patients, independent of clinical diagnosis. Cluster 2 exhibited upregulated modules for myeloid activation, inflammation, T/B-cell differentiation, and mitochondrial metabolism. PCA confirmed separation. In RA, Cluster 2 linked to higher DAS28/SDAI/CDAI, CRP, RF, and leukocyte counts; in SLE, to elevated SLEDAI, carotid intima-media thickness (atherosclerosis), and monocyte/lymphocyte counts.

Multi-omics integration using MOFA validated the molecular stratification into the previously identified clusters and revealed that the first latent factor was primarily driven by transcriptomic variation related to myeloid inflammatory responses, while the second factor captured proteomic variance dominated by key proinflammatory mediators, forming a highly interconnected and functionally coherent protein network. Relevant proteins included chemokines, interleukins, growth factors, and enzymes involved in immune activation, leukocyte migration, angiogenesis, and tissue remodelling. Correlation analyses demonstrated strong associations between transcriptomic modules, circulating inflammatory proteins, and clinical phenotypes in both RA and SLE. Myeloid/inflammation modules positively correlated with CRP/ESR, leukocyte/ monocyte/ neutrophil counts, and activity scores (DAS28/SDAI/CDAI in RA, SLEDAI in SLE) across diseases. Proteins like CXCL10, CXCL11, MCP3 (chemokines), IL-17A (myeloid activation), and ADA (metabolism) showed consistent positive links to inflammation and blood counts, indicating common innate immune drivers of clinical severity. In RA patients, gene modules involved in post-translational regulation/cell death strongly positively correlated with DAS28/SDAI/CDAI, CRP, RF, and neutrophils/monocytes count, underscoring synovial inflammation. Adaptive modules involved in B/T proliferation and adaptive immunity negatively correlated with activity/inflammation, suggesting regulatory roles. Proteins CD5/IL-2/SCF positively tied to SDAI/RF/leukocytes; negative links for regulatory elements like IL-10 indicated dampened responses in active disease. On the other hand, in SLE, the module involved in interferon response positively correlated with ANA/CRP/ESR but negatively with neutrophils; modules involved in mitochondrial activity/ immunoregulation tied to SLEDAI, proteinuria/creatinine (renal damage), and carotid intima-media thickness (CV risk). Proteins IL-12B/IL-18R1/IL-17C (JAK-STAT/cytokine signaling) positively associated with SLEDAI/proteinuria/atherosclerosis; chemokines CXCL10/11 reinforced systemic/multi-organ inflammation. These findings highlight shared inflammatory axes across diseases, as well as disease-specific regulatory patterns related to adaptive immunity and clinical expression. Finally, machine learning-based predictive models integrating clinical variables with selected proteomic signatures achieved robust performance. A combined RA–SLE model reached an AUC-ROC of 0.79, driven by proteins associated with leukocyte migration (CXCL10, CXCL11, MCP3), immune activation (CD5, IL17A), and metabolic regulation (ADA). Disease-specific models showed higher accuracy in RA (AUC-ROC 0.85), emphasizing markers related to T cell activation and cellular trafficking, while the SLE model (AUC-ROC 0.72) highlighted the relevance of JAK–STAT signalling and cytokine–receptor interactions.

Conclusions: This integrative multi-omics study reveals shared and disease-specific molecular architectures in RA and SLE and identifies clinically meaningful molecular subgroups across both diseases. The combination of transcriptomic, proteomic, and clinical data using machine learning provides a robust framework for molecular stratification, biomarker discovery, and personalized therapeutic targeting, supporting precision medicine approaches in systemic autoimmune diseases.

REFERENCES: NIL.

Acknowledgments: NIL.

Disclosure of Interests: None declared.