Metabolomics-driven classification and ratio-based quality control of polygonum multiflorum processing degree using UHPLC-MS/MS

Polygonum multiflorum Thunb. (PM) is a widely used herbal medicine whose efficacy and safety depend critically on the degree of processing; however, current pharmacopeial standards rely on subjective visual inspection, leading to inconsistent product quality. To address this, we combined UHPLC-Q-Exactive MS/MS-based untargeted metabolomics with time-series clustering and machine learning to objectively classify PM processing stages. Samples from three geo-authentic origins were collected across a 32 h standardized steaming process. Among 22,639 detected molecular features, 155 were significantly altered during processing, 87 were putatively annotated, and 14 were unequivocally identified using reference standards. Spearman correlation analysis identified 58 metabolites with strong monotonic relationships with processing time (|ρ| > 0.7, p < 0.01), from which 20 key compounds were prioritized based on monotonic trends, low inter-batch variation (CV<20%), and chemical class representativeness. A decision tree model built on these 20 compounds achieved 100% classification accuracy on an independent set of 50 commercial samples, discriminating raw, intermediate, and fully processed PM. The peak area ratio of hydroxymusizin O-glucoside (most negatively correlated) to a putatively identified deoxy and dimethylated emodin O-glucoside (most positively correlated) decreased monotonically with processing time. A validated threshold of < 0.1 was established for fully processed PM to prioritize safety by ensuring absolute discrimination from raw and intermediate samples. When applied to 20 commercial products pre-classified as “processed” by traditional criteria, 16 (80%) met this standard; the remaining four exhibited ratios of 0.10-0.11 and were re-classified as intermediate, reflecting a prudent safety margin. This work provides a practical, chemistry-driven framework for standardizing herbal processing, combining a robust multivariate model with a simple, transferable ratio to replace subjective visual endpoints. The strategy bridges traditional knowledge and modern analytical science, offering a template for evidence-based quality control of other processed botanicals.