Dissection of quality variation and hierarchical quality evaluation of Evodiae Fructus under multidimensional factors

Introduction Considerable quality fluctuation and heterogeneity are commonly observed in Evodiae Fructus due to multiple confounding factors. Accordingly, this study performed multi-factor analysis to explore potential key factors associated with its chemical quality, aiming to provide scientific evidence and theoretical support for the quality evaluation of this medicinal material. Methods First, for the HPLC-derived multi-component quantitative dataset, the Kruskal-Wallis H test followed by post-hoc Dunn’s tests with Benjamini-Hochberg FDR correction was performed, and generalized linear models (GLMs) specified with a Gamma distribution and log-link function together with random forest regression were further constructed to quantify the associations of botanical origins, geographical origins, commercial specifications, and maturity stages with the quality of Evodiae Fructus. Second, for the multi-dimensional UPLC-MS/MS metabolomic profiles, principal component analysis (PCA), orthogonal partial least-squares discriminant analysis (OPLS-DA), Kruskal-Wallis H tests, and fold-change analysis were applied to screen differential metabolites across the predefined comparison groups, with metabolites satisfying VIP > 1, |log2FC| > log2(1.5) , and unadjusted p < 0.05 retained as differential metabolites. GNPS-based feature-based molecular networking (FBMN) was utilized to visualize the corresponding chemical space. Subsequently, Pearson correlation analysis (|r| > 0.6, P < 0.05) and random forest modeling were adopted to filter candidate core markers, which were further confirmed by receiver operating characteristic (ROC) curve analysis (AUC > 0.8). Finally, comparative back-propagation neural network (BPNN) models, with hyperparameters optimized by grid search using 10-fold cross-validation, were established to discriminate different commercial specifications of Evodiae Fructus using both pharmacopoeia-referenced indicators and the selected core biomarkers as input, and their predictive performance was evaluated on an independent test set. Results A significant correlation was found between commercial specifications and the differential accumulation of active constituents in Evodiae Fructus. ROC curve validation confirmed that 2-hydroxy-4-methoxy-3-(3’-methylbut-2’-enyl) quinolin-4-one and 1-methyl-[(3E,6Z,9Z,12E)-pentadeca-3,6,9,12-tetraen-1-yl]-4(1H)-quinolone, two alkyl quinolone alkaloids, could be regarded as specific potential biomarkers for quality assessment of Evodiae Fructus. The two candidate markers enable the discrimination of Evodiae Fructus with distinct specifications. Conclusions Collectively, this study disentangled the independent and interactive associations between commercial specification and chemical quality of Evodiae Fructus. The candidate markers we obtained enabled BPNN models to effectively discriminate different commercial specifications. These findings provide valuable references for the hierarchical quality assessment of this medicinal herb.