B-194 Bile Acid Metabolomics Reveals Distinct Immunometabolic Niches and Enables Accurate Diagnosis of AQP4-IgG–Seronegative NMOSD

Abstract

Distinguishing neuromyelitis optica spectrum disorder (NMOSD) from multiple sclerosis (MS) remains clinically consequential, particularly in patients seronegative for anti–aquaporin-4 IgG (AQP4-IgG), for whom objective biomarkers are limited. Because bile acids (BAs) mediate gut microbiota–host signaling and immunoregulation, we investigated whether serum BA profiling could provide diagnostically actionable signatures for NMOSD and MS, including serostatus-specific patterns. In a cross-sectional study, we consecutively enrolled 112 NMOSD patients, 50 MS patients, and 66 healthy controls (HC) at a single tertiary center (August 31, 2024–August 31, 2025). NMOSD and MS were diagnosed using the 2015 International Consensus criteria and 2018 McDonald criteria, respectively. Blood samples were obtained during relapse prior to acute-phase therapy, following =8 h fasting, and serum BAs were quantified by stable isotope–dilution LC–MS/MS. BA pools and compositional indices were derived a priori. Unsupervised PCA and nonparametric group comparisons were performed. For classification, we implemented a leakage-controlled pipeline with an 8:2 stratified split, nested cross-validation, stability selection (LASSO, ?.1se; stability score =50%), and model comparison across logistic regression, SVM, random forest, and XGBoost; performance was assessed on an independent hold-out test set with bootstrap confidence intervals. Associations with disability were examined using Spearman correlations with Expanded Disability Status Scale (EDSS) scores. Serum BA profiles differed substantially across NMOSD, MS, and HC, with disease-level separation observed in PCA. Compared with MS and HC, NMOSD showed an expanded total BA pool dominated by primary conjugated BAs, whereas MS exhibited enrichment of secondary BAs (notably deoxycholic acid [DCA], lithocholic acid [LCA], and related conjugates). A stability-selection procedure identified six robust BA features, and the logistic regression model built on these stable features achieved strong discrimination between NMOSD and MS on the independent test set (AUC 0.874; sensitivity 0.900; specificity 0.826). In serostatus-stratified analyses, AQP4-IgG–seronegative NMOSD demonstrated a pronounced suppression of secondary BAs in both absolute concentrations and proportional representation, enabling highly accurate separation from MS; DCA was the strongest single discriminator (AUC 0.965). A Youden-optimized cutoff of DCA =0.068 µmol/L yielded 100% sensitivity and 90% specificity for identifying AQP4-IgG–seronegative NMOSD in this cohort, and resampling/permutation procedures supported internal robustness. BA–disability relationships were disease-specific: UDCA-related metrics tended to inversely associate with EDSS in NMOSD, whereas secondary BA measures showed positive associations with disability in MS. Serum BA metabolomic signatures differentiate NMOSD from MS and further stratify NMOSD by AQP4-IgG serostatus. A marked reduction in secondary BAs—captured most strongly by DCA—characterizes AQP4-IgG–seronegative NMOSD and may aid differential diagnosis in this diagnostically challenging subgroup. These findings support BA profiling as a noninvasive, biologically grounded adjunct for diagnostic refinement and disability-linked metabolic phenotyping, while external multicenter validation and longitudinal sampling are required before clinical translation.