*2.3. Formatting of Mathematical Components Principal Component Analysis (PCA) and Orthogonal Projections to Latent Structures-Discrimination Analysis (OPLS-DA)*

Principal Component Analysis (PCA) is a multidimensional data statistical analysis method of unsupervised pattern recognition. Through principal component analysis of samples (including QC samples), we can preliminarily understand the overall metabolic differences among samples and the degree of variability between samples within groups. From the analysis results, it can be observed that there are significant differences among RL, YL and GL groups, but there is no significant difference within groups (Figure S2).

Although PCA can effectively extract the main information, it is not sensitive to variables with small correlation, and Partial Least Squares-Discriminant Analysis (PLS-DA) can solve this problem. Compared with PCA, PLS-DA can maximize the distinction between groups and facilitate the search for differential metabolites. Through the analysis of PLS-DA, the orthogonal variables, which are not related to the classification variables of metabolites are first eliminated, and then the differences of correlation between groups and within groups are analyzed. According to the OPLS-DA model, we analyzed the metabolic group data, draw the score chart of each group, and further showed the differences between each group. The prediction parameters of the evaluation model are R2X, R2Y and Q2, in which R2X and R2Y represent the interpretation rate of the model to X and Y matrix respectively, and Q2 indicates the prediction ability of the model. The closer these three indexes are to 1, the more stable and reliable the model is. Q2 > 0.5 can be regarded as an effective model, and Q2 > 0.9 is an excellent model. From the results, there are significant differences among the three groups of data, but there is no significant difference between groups (Figure S3a). The alignment verification of OPLS-DA was carried out (*n* = 200, that is, 200 permutation experiments were carried out). In the model verification, the horizontal lines correspond to R2 and Q2 of the original model, and the red dots and blue dots represent R2' and Q2' of the model after Y replacement, respectively. The results showed that R2' and Q2' of each group were smaller than R2 and Q2 of the original model, which indicated that the model was meaningful and the differential metabolites could be screened according to VIP value analysis (Figure S3b).

**Figure 3.** Hierarchical clustering analysis of all metabolites detected in this study. The abscissa indicates three biological replicates of red leaves (RL1, RL2, and RL3), yellow leaves (YL1, YL2, and YL3) and green leaves (GL1, GL2, and GL3), and the ordinate indicates the metabolites detected in this study. The red segments indicate a relatively high content of metabolites, while the blue segments indicate a relatively low content of metabolites. The relative metabolite contents represented by color segments at the corresponding locations are listed in Table S1.
