Abstract
Objective The objective of this study was to discover differential metabolites and pathways underlying infrequent gout flares (InGF) and frequent gout flares (FrGF) using metabolomics and to establish a predictive model by machine learning (ML) algorithms.Methods Serum samples from a discovery cohort of 163 patients with InGF and 239 patients with FrGF were analyzed by mass spectrometry-based untargeted metabolomics to profile differential metabolites and explore dysregulated metabolic pathways using pathway enrichment analysis and network propagation-based algorithms. ML algorithms were performed to establish a predictive model based on selected metabolites, which was further optimized by a quantitative targeted metabolomics method and validated in an independent validation cohort with 97 participants with InGF and 139 participants with FrGF.Results A total of 439 differential metabolites between InGF and FrGF groups were identified. Top dysregulated pathways included carbohydrates, amino acids, bile acids, and nucleotide metabolism. Subnetworks with maximum disturbances in the global metabolic networks featured cross-talk between purine metabolism and caffeine metabolism, as well as interactions among pathways involving primary bile acid biosynthesis, taurine and hypotaurine metabolism, alanine, aspartate, and glutamate metabolism, suggesting epigenetic modifications and gut microbiome in metabolic alterations underlying InGF and FrGF. Potential metabolite biomarkers were identified using ML-based multivariable selection and further validated by targeted metabolomics. Area under receiver operating characteristics curve for differentiating InGF and FrGF achieved 0.88 and 0.67 for the discovery and validation cohorts, respectively.Conclusion Systematic metabolic alterations underlie InGF and FrGF, and distinct profiles are associated with differences in gout flare frequencies. Predictive modeling based on selected metabolites from metabolomics can differentiate InGF and FrGF.