Revealing Metabolic Reprogramming Genes as Biomarkers of Rheumatoid Arthritis: A Bioinformatics Study Based on Ensemble Learning Approach and Single Cell RNA Sequencing

Introduction: Rheumatoid Arthritis (RA) is a chronic autoimmune disease characterized by persistent synovial inflammation and progressive joint destruction. Emerging evidence suggests that metabolic reprogramming plays a pivotal role in rheumatoid Arthritis (RA) pathogenesis. However, existing studies have focused predominantly on isolated pathways. This study aimed to systematically investigate the molecular interplay between metabolic reprogramming and immune dysregulation in RA. Methods: Twelve metabolism-related gene sets from MSigDB were analyzed across two bulk RNA-seq datasets (GSE93272: 232 RA/43 controls; GSE110169: 84 RA/77 controls). Based on bulk RNA-seq datasets related to RA, feature genes were selected using machine-learning algorithms, and an ensemble-learning model for RA diagnosis was developed. Subsequent analyses included Gene Set Enrichment Analysis (GSEA), immune cell infiltration profiling, drug-target prediction, and gene-disease network construction to further explore the potential functions of the feature genes. Additionally, a single-cell RNA sequencing (scRNA-seq) dataset containing a single RA patient (GSE159117) was used to identify key cell populations and investigate the interactions between cell populations. Finally, six candidate genes were validated by qPCR (RA n=5; control n=6). Results: Three high-performance diagnostic models were developed by combining the LASSO regression and XGBoost algorithms. The ensemble model (AUCs of 0.960–0.970 in training and 0.819 in external validation) integrating the 3 individual models outperformed the individual model, and 13 feature genes were identified via the ensemble model. Functional enrichment revealed significant associations with metabolic and immune pathways. Immune infiltration profiling revealed notably elevated T-cell subsets in RA patients, with multiple feature genes strongly correlated with T-cell activation. Single-cell analysis confirmed T cells as key mediators, with IRF7 regulating TXN and related genes. qPCR validated significant dysregulation of AKR1C3, ARG1, TXN, and C1QB in RA patients compared with controls (p< 0.05). Discussion: These findings suggest that metabolic reprogramming is not merely a consequence but may be a driver of immune dysfunction in RA. The strong association between metabolic gene signatures and T cell activity raises the possibility that targeting metabolic pathways could modulate immune responses in RA. Further studies are needed to explore the causal relationships and therapeutic potential of these interactions in clinical settings. Conclusion: Our study provides a comprehensive characterization of metabolic-immune crosstalk in RA, identifying novel diagnostic biomarkers and therapeutic targets. Our findings advance the understanding of RA pathogenesis and may facilitate the development of precision medicine approaches.