Background/Purpose: The variability in treatment response in people with rheumatoid arthritis (RA) warrants the prediction of patients at high risk of treatment failure. Identification of biomarkers linked to clinical remission in RA is currently a challenge. Metabolomics may help to identify such biomarkers as it allows for a comprehensive exploration of disease-related variations that extends beyond the genome and proteome. This study aimed to systematically profile serum metabolic alterations in early RA to identify potential biomarkers associated with disease activity and therapeutic response using metabolomics.

Methods: The study included 220 early RA participants from the NORD-STAR study, who have been randomized at baseline to four arms, ranging from conventional anti-rheumatic treatment to three biological drugs; methotrexate combined with prednisolone (1), certolizumab (2), abatacept (3), or tocilizumab (4). Untargeted metabolomics was performed in serum samples at baseline and 24-week follow-up. Participants achieving clinical disease activity index remission (CDAI ≤ 2.8) at 24 weeks were defined as responders. Machine learning models for treatment response were constructed using random forest, logistic regression, and extreme gradient boosting. The analyses were performed using MetaboAnalyst 6.0 and R v4.3.

Results: We identified 278 metabolites, of which 39 were associated with baseline disease activity, including several acylcarnitines and amino acids (Figure 1A). Results from the univariate analysis combining the outcomes from t-test (p < 0.05) and fold-change (> 1.2) analyses in a volcano plot, showed that cytidine was the only baseline metabolite being significantly upregulated in responders, while six other metabolites significantly downregulated in responders (Fig. 1B). In multivariable analysis, adjusted for several baseline confounders, we found 17 baseline metabolites associated with remission at 24 weeks in the overall cohort including malic acid (ß=-0.4), cytidine (ß=0.4), arginine (ß=0.3), and citrulline (ß=0.2) (Figure 1C). Pathway enrichment analysis indicated several metabolic pathways potentially associated with treatment response, including, among others, arginine and proline metabolism, carnitine synthesis, and urea cycle (Fig. 1D). Eleven baseline features were identified as biomarkers to discriminate responders from non-responders at 24 weeks in machine learning analysis. ROC analysis resulted in AUC_max of 0.71 in the test set across three machine learning algorithms.

Conclusion: Our study has identified several baseline metabolites and metabolic pathways associated with disease activity and response to different treatments in early RA. Moreover, by integrating metabolomics and clinical data, we have developed models to predict response to treatment in people with early RA.

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/biomarkers-for-disease-activity-and-response-to-treatment-in-early-rheumatoid-arthritis-metabolomics-and-machine-learning-analyses-in-nord-star-cohort/