Machine Learning for the Identification of a Common Signature for Anti–SSA/Ro 60 Antibody Expression Across Autoimmune Diseases

• Published in: Arthritis & rheumatology (Hoboken, N.J.) N.J.), 2022-10, Vol.74 (10), p.1706-1719
• Main Authors: Foulquier, Nathan, Le Dantec, Christelle, Bettacchioli, Eleonore, Jamin, Christophe, Alarcón‐Riquelme, Marta E., Pers, Jacques‐Olivier
• Format: Article
• Language: English
• Subjects: Algorithms; Antibodies; Antibodies, Antinuclear; Antigens; Antigens, Nuclear; Arthritis; Autoantibodies; Autoantigens; Autoimmune diseases; Autoimmune Diseases - genetics; Autoimmunity; Chronic conditions; Connective tissue diseases; Connective tissues; Cross-Sectional Studies; Cytokines; Disease; DNA methylation; Flow cytometry; Full Length; Genes; Genetic diversity; Genome-Wide Association Study; Genomes; Humans; Inflammation; Interferon; Interferons; Learning algorithms; Lupus Erythematosus, Systemic - genetics; Machine Learning; Myxovirus resistance proteins; Patients; Phenotypes; Rheumatoid arthritis; Ribonucleoproteins - genetics; Scleroderma; Serology; Signatures; Sjogren's syndrome; Sjogren's Syndrome - genetics; Systemic lupus erythematosus; Systemic sclerosis; Transcriptomics; Undifferentiated Connective Tissue Diseases
• ISSN: 2326-5191; 2326-5205; 2326-5205
• DOI: 10.1002/art.42243

Objective Anti‐Ro autoantibodies are among the most frequently detected extractable nuclear antigen autoantibodies, mainly associated with primary Sjögren's syndrome (SS), systemic lupus erythematosus (SLE), and undifferentiated connective tissue disease (UCTD). This study was undertaken to determine if there is a common signature for all patients expressing anti–Ro 60 autoantibodies regardless of their disease phenotype. Methods Using high‐throughput multiomics data collected from the cross‐sectional cohort in the PRECISE Systemic Autoimmune Diseases (PRECISESADS) study Innovative Medicines Initiative (IMI) project (genetic, epigenomic, and transcriptomic data, combined with flow cytometry data, multiplexed cytokines, classic serology, and clinical data), we used machine learning to assess the integrated molecular profiling of 520 anti–Ro 60+ patients compared to 511 anti–Ro 60– patients with primary SS, patients with SLE, and patients with UCTD, and 279 healthy controls. Results The selected clinical features for RNA‐Seq, DNA methylation, and genome‐wide association study data allowed for a clear distinction between anti–Ro 60+ and anti–Ro 60– patients. The different features selected using machine learning from the anti–Ro 60+ patients constituted specific signatures when compared to anti–Ro 60– patients and healthy controls. Remarkably, the transcript Z score of 3 genes (ATP10A, MX1, and PARP14), presenting with overexpression associated with hypomethylation and genetic variation and independently identified using the Boruta algorithm, was clearly higher in anti–Ro 60+ patients compared to anti–Ro 60– patients regardless of disease type. Our findings demonstrated that these signatures, enriched in interferon‐stimulated genes, were also found in anti–Ro 60+ patients with rheumatoid arthritis and those with systemic sclerosis and remained stable over time and were not affected by treatment. Conclusion Anti–Ro 60+ patients present with a specific inflammatory signature regardless of their disease type, suggesting that a dual therapeutic approach targeting both Ro‐associated RNAs and anti–Ro 60 autoantibodies should be considered.