Background:Rheumatoid arthritis (RA) is an aggressive immune-mediated joint disease characterized by synovial proliferation and inflammation, cartilage destruction, and joint destruction1. Despite efforts to characterize the disease subsets and to predict the differential prognosis in RA patients, disease heterogeneity is not adequately translated into the current clinical subclassification2.Objectives:To develop and validate an integrative system approach for stratifying patients with RA according to disease status and whole-genome gene expression data.Methods:An RNA sequencing dataset of synovial tissues from 124 RA patients (including 57 patients with early RA, 95 with established RA) and 15 healthy controls (HC) was imported from the Gene Expression Omnibus (GEO) database (GSE89408) by software package R (version 4.0.3). After filtrating of differentially expressed genes (DEGs) between RA and HC, non-negative matrix factorization, functional enrichment, and immune cell infiltration were applied to illustrate the landscapes of these patients for classification. Clinical features (age, gender, and auto-antibodies) were also compared to discover the signatures of these classifications.Results:A matrix of 576 DEGs from RA samples was classified into 5 subtypes (early/C1–C3, established/C4-C5) with distinct molecular and cellular signatures and two sub-groups (S1 and S2) (Figure 1A-1D). New-onset patients (early C2) and established C4 patients were named as S1, they shared similar gene signatures mainly characterized by prominent immune cells and proinflammatory signatures, and enriched in the chemokine-mediated signaling pathway, lymphocyte activation, response to bacterium and Primary immunodeficiency. S2(C1, C3 and C5) were more occupied by synovial fibroblasts of destructive phenotype. They were mainly enriched in the response to external factors and PPAR signaling pathway (Figure 1E-1H). Interestingly, combined with clinical information, S1 and S2 had no significance in age and gender (P > 0.05). But patients in S1 had a stronger association with the presence of anti-citrullinated protein antibodies (ACPA) (P < 0.05) (Figure 1I-1J).Conclusion:We successfully deconvoluted RA synovial tissues into pathobiological discrete subsets using an unsupervised machine learning method and described their distinct molecular and cellular characteristics. These results provide important insights into divergent and shared mechanistic features of RA and serve as a template for future studies to guide drug tar-get discovery by synovial molecular signatures and de-sign stratified approaches for patients with RA.References:[1]Smolen JS, Aletaha D, McInnes IB. Rheumatoid arthritis. Lancet 2016;388(10055):2023-38. doi: 10.1016/S0140-6736(16)30173-8 [published Online First: 2016/10/30][2]Jung SM, Park KS, Kim KJ. Deep phenotyping of synovial molecular signatures by integrative systems analysis in rheumatoid arthritis. Rheumatology (Oxford) 2020 doi: 10.1093/rheumatology/keaa751 [published Online First: 2020/11/25]Acknowledgements:This project was supported by National Science Foundation of China (82001740), Open Fund from the Key Laboratory of Cellular Physiology (Shanxi Medical University) (KLCP2019) and Innovation Plan for Postgraduate Education in Shanxi Province (2020BY078).Disclosure of Interests:None declared