Abstract
Celiac disease is a lifelong immune-mediated disease triggered by dietary gluten and managed exclusively through a strict gluten-free diet. Differences in the gut microbiome are consistently observed in celiac disease, yet the associated metabolic consequences remain insufficiently defined. This study evaluated the stool microbiome-metabolome relationship in children from Canterbury, New Zealand, a region with a notably high prevalence of celiac disease (∼ 1 in 82). Eleven children with untreated celiac disease and 10 controls were assessed, with nine children resampled after 6 months of gluten-free diet adherence. Stool samples were subjected to 16S rRNA sequencing, targeted and untargeted metabolomic profiling, and multi-omics integration using cross-block and machine-learning approaches. Celiac disease was characterized by consistent reproducible changes in a small set of resampled-stable taxa and metabolites, rather than a uniform shift. The most prominent finding was rearrangement within Bacteroides rather than a uniform shift: B. ovatus increased reproducibly in both untreated and treated disease, while integration analyses associated B. fragilis with celiac disease and B. vulgatus with controls. Metabolite changes were observed in two linked axes: a butyrate-led fermentation signature and a redox and organic acid signature, both elevated in untreated disease and shifting toward control levels after 6 months of gluten-free diet. Recovery was partial, and the diet imposed its own signature, notably an increase in the fat-responsive pathobiont Bilophila wadsworthia. Machine learning discrimination was modest (ROC-AUC = 0.66–0.68, PR-AUC = 0.73–0.79), but the microbiome and metabolome aligned along one reproducible cross-omic axis. The results support prioritizing a concise list of reproducible candidate features for mechanistic validation and longitudinal treatment monitoring in more extensive cohorts, rather than for immediate diagnostic application.