Abstract
INTRODUCTION: Identifying the phytochemistry underpinning a plant's observed therapeutic benefits is essential for understanding mechanisms of action and developing novel therapeutics. More recent efforts fusing global metabolomics and multivariate predictive modeling have improved compound discovery; however, these models rely on chemical variations between samples, which often necessitates at least one round of fractionation and may result in compound loss or degradation. OBJECTIVES: This study uses multiple whole botanical extracts to explore whether a metabolome-wide association study approach can accurately identify bioactive phytochemicals without prior fractionation. METHODS: We employed 40 Ocimum extracts with a range of IC(50) levels against HT-29 cells in an in vitro MTT assay and combined this data with untargeted UPLC-MS/MS metabolomics for biochemometric modeling of the potential bioactives. Multiple chemometric tools and statistical filters were employed to improve feature selection. RESULTS: The metabolomic profiles resulted in ca. 1600 metabolite features; implementing source-based filters, followed by LASSO dimension reduction, improved the reliability of Partial Least Squares (PLS) bioactivity predictions. The resulting model highlighted four biomarkers positively correlated with activity, one of which was putatively identified as gallic acid. Gallic acid's cytotoxicity against HT-29 cells was confirmed with the purified compound. CONCLUSION: This study results demonstrated that predictive modeling of botanicals using a metabolome-wide association study of extracts with no fractionation was capable of identifying biologically active compounds.