Annotation and pathway analysis

We annotated features based on three levels. First, significant features were matched to a database of authenticated chemical standards previously characterized in the Emory laboratory, i.e., metabolites confirmed using MS/MS and authentic standards, providing the strongest level of annotation [18, 19]. The error tolerance was set to 5 ppm and 30s for m/z and retention time. Additional m/z feature mapping was done based on mummichog annotations and xMSannotator. mummichog is a computational algorithm which uses metabolic pathways and networks to predict functional activity from untargeted metabolite feature tables, including providing annotations of features based upon predicted ions and pathway associations [20]. With xMSannotator, accurate mass m/z for adducts formed under positive/negative ESI mode were matched to HMDB, KEGG, and LipidMaps with a mass error threshold of 10 ppm [21]. xMSannotator uses correlations of intensities and retention time and assigns confidence scores based on a multilevel scoring algorithm (0–3, a higher score representing higher-confidence result), ensuring annotation accuracy. Only results with scores ≥2 were considered for annotations.

For pathway enrichment analysis we used metapone, which uses a permutation-based weighted hypergeometric test with joint pathway analysis using positive and negative ion mode data to avoid double counting and account for multiple-matching uncertainty with a weighting factor [22]. Metabolic pathways were compiled from KEGG, mummichog, and the small molecule pathway database (SMPDB).