*3.1. Overall Metabolic Profiles of the Bee Pollen*

To obtain an overview of grouping patterns of the bee pollen samples of different botanical origins, 670 valid entities were submitted with which to perform unsupervised PCA. The first two principal components of PCA explained 74.46% of the total variance (PC1 = 65.70% and PC2 = 8.76%). In the PCA score plots (Figure 1A), a tight clustering of the pollen samples from the same botanical origins was observed, and their distribution patterns were found to be affected by their botanical origins. Remarkably, a clear separation between the CBP and non-CBP samples was observed. Specifically, the CBP samples were all distributed in the negative part of the PC1 axis, while the non-CBP samples were located in the positive part of the PC1 axis and separated along the PC2 axis.

**Figure 1.** Multivariate and univariate analysis of CBP and non-CBP (rose, apricot, lotus, rape, and wuweizi bee pollen) metabolomics data. PCA score plots were generated from all valid entities (**A**); score plots (**B**) and loading plots (**C**) of OPLS-DA for the 54 identified compounds. The ellipses indicate 95% confidence limits according to Hotelling's T2 statistics. VIP score and log fold change (CBP/non-CBP) of the identified compounds are shown (**D**).
