The datasets were normalized by centering and pareto scaling, then analyzed by PCA and PLS using SPSS 19.0 (IBM, USA) and Matlab (Version R2008b, MathWorks, USA) for subsequent statistical
analysis. In PCA, function of princomp was used to judge the differences among samples and process data into a two-dimensional map, which reserved most of the characteristics of raw data. PLS was carried out to investigate the correlations between metabolism and fumaric acid accumulation, and identify the key metabolites that mainly resulted in metabolic discrimination. Mean values and relative standard deviations of metabolites were calculated from four replicates of each sample. Significance level of metabolites abundance in xylose medium relative to the data in glucose medium was identified by a two-tailed Student’s t-test performed with SPSS 19.0 (IBM, USA).