Data analysis
Complex chromatograms were deconvoluted and peaks were aligned, and a data matrix was constructed using MS‐DIAL (v.2.56, http://prime.psc.riken.jp/compms/msdial/main.html). In MS‐DIAL setting, the range for mass detection was 50 - 650 Da and the minimum peak height for trustable feature detection was chosen to be1000 amplitude. The preferred tolerance limit for retention time was 0.05 min. The identification cut-off score was 70%. Identificaiton of the metabolites was performed by using the Fiehn Retention Index database.
The data matrix normalization was based on the sum of the total peak area for individaul samples. In the data matrix, metabolites that had traits more than 50% of the values missing were excluded. Missing values in the data matrix were filled with the half value of the lowest concentration in the metabolite group. PLS‐DA analysis in Metaboanalyst 4.0 platform was used for class discrimination, simplifying interpretation and finding candidate biomarkers. The variable importance in projection (VIP) value is estimated to discriminate the most significant metabolites for stratified groups. Student’s unpaired t‐test was used to compare changes in mean expression per metabolites between groups. p < 0·05 was accepted to be statistically significant. Altered metabolites were evaluated in pathway analysis within the Metaboanalyst 4.0 platform.