GC–MS data analysis

The GC–MS dataset was processed using the R statistics software as described previously⁴³. Briefly, the following processing steps were performed: (1) XCMS was used to defined a matrix of molecular features⁴⁴, (2) data in each sample were normalized to total ion current, (3) RAMClustR package for R clustered co-varying and co-eluting features into spectra⁴⁵, (4) RAMSearch software⁴⁶ (https://osf.io/x8bw5/) allowed annotation by searching spectra against internal and external databases. Databases used for annotations included in-house libraries, golm (http://gmd.mpimp-golm.mpg.de/), NISTv14 (http://www.nist.gov), and MassBank (http://www.massbank.jp). Principal component analysis (PCA) was performed using unit variance scaling in SIMCA v15.0.2 (Umetrics, Umea, Sweden). The MetaboAnalyst (http://www.metaboanalyst.ca/) was used to perform hierarchical clustering using the hclust function in the R package stat (euclidean distance measure, ward clustering algorithm) and to generate heatmap visualization⁴⁷. Univariate statistical analysis was performed in Prism (Version 8.2.1, GraphPad). Data was log transformed and then analyzed using a one way ANOVA with a Tukey’s multiple comparisons test.