Serum Proteomic Signatures Before the Diagnosis of Rheumatoid Arthritis: Evolving Biologic Pathways and Specific Periods of Disease Development

Objective This longitudinal case‐control study evaluated serum proteomics before a clinical diagnosis of rheumatoid arthritis (RA) (ie, pre‐RA) to evaluate biologic pathways of disease development and inform prediction of timing of onset of future disease. Methods Patients (n = 213) meeting the 1987 American College of Rheumatology classification criteria for RA and matched controls without RA (n = 215) were identified in the Department of Defense Serum Repository. Serum samples from patients before and after RA diagnosis and controls were tested for RA‐related autoantibodies (anti–cyclic citrullinated peptide‐3 [anti‐CCP3] and rheumatoid factor [RF] isotypes IgM and IgA) and 197 proteins using a commercial platform (Olink). We applied linear mixed effect models to identify biomarkers distinguishing patients from controls before RA diagnosis and analyzed longitudinal patterns of enriched pathways; in addition, models were developed to classify the time of a sample in relationship to the time of RA diagnosis. Results Levels of anti‐CCP3, RFIgA, and RFIgM demonstrated the greatest differences between patients and controls ≤5 years before RA diagnosis. Longitudinal analyses identified 104 proteins that were differentially expressed between patients and controls; 60 proteins were differentially expressed ≤5 years before diagnosis, 42 proteins were differentially expressed within and before five years of diagnosis, and 2 proteins were differentially expressed >5 years before diagnosis. Kyoto Encyclopedia of Genes and Genomes analyses identified that these proteins were associated with 32 pathways, including 21 pathways that were enriched ≤5 years before diagnosis. Within the anti–citrullinated protein antibody–positive samples from before RA diagnosis and controls, a set of features classified if that sample was from a period <3 years before RA diagnosis, with an area under the receiver operating characteristic (ROC) curve of 0.78 (95% confidence interval 0.67–0.89) in a training set and 0.80 (0.68–0.92) in a validation set. Conclusion Autoantibodies and protein signatures evolve in distinct stages before a diagnosis of RA. Furthermore, protein biomarkers may identify biologic pathways relevant to specific stages. These can be further explored to potentially improve prediction of disease onset and identify stage‐specific biologic pathways to target with preventive interventions.