Tool-based approach to explore metabolomic data and identify discriminant features: case study on the fecal metabolome of piglets exposed to a pathogen and/or a mycotoxin
Résumé
Swine are regularly exposed to exogenous biotic and abiotic agents susceptible to alter health or metabolic homeostasis. Metabolomics could provide a relevant means to capture the global impacts of such xenobiotics. To assess the potential of this approach, a study was conducted on feces and serum samples from piglets exposed to two foodborne hazards (individually or simultaneously): deoxynivalenol (DON, “D”), a cereal-derived mycotoxin, and/or the monophasic variant of Salmonella (S. Typhimurium, “S”), the second most prevalent zoonotic bacterial genus in Europe. Forty piglets were divided into four groups (S-D-, S-D+, S+D-, S+D+). Daily DON exposure began at D-7 (3 mg/kg) and was then maintained. Salmonella was administered orally once at D0 (10⁹ CFU/pig). Feces were collected at D15, and serum at D3 and D15. While serum metabolomic aspects were previously presented at an international congress (I3S, 2025), fecal metabolome responses and possible cross-compartment markers have not yet been described. Exploring metabolomic datasets requires robust acquisition methods and a reproducible sequence of tools to reveal biologically meaningful patterns. Following methanol extraction and HRLC-MS coupled with Q-Exactive Orbitrap MS² spectrometry under ESI, this study presents an approach to metabolomic data processing. This strategy combines tools from Workflow4Metabolomics (XCMS, CAMERA, filtering, normalization, log₂) with statistical analysis in RStudio, including dispersion (Betadisper on Euclidean distance matrix, ANOVA, Tukey-HSD, PCA) and multivariate difference in signal intensity (PERMANOVA 1,000 times with Euclidean distance, FDR correction). Discriminant metabolites (VIPs, score ≥ 1) were selected after denoising (Kruskal-Wallis, NearZeroVar) and pairwise sPLS-DA modeling (cross-validation, tuning, BER, AUC, Wilcoxon). Retained features were annotated using Sirius. Their roles were contextualized through literature. This approach revealed alterations of the fecal metabolome, mainly in negative ionization. One discriminant feature, 3-ethynyl-1H-indole, was specific to the unexposed group. Comparison of fecal and serum VIPs revealed no overlap, suggesting limited metabolic correspondence between compartments.