4.3.7. Raw Data Processing by Compound Discoverer

The raw files obtained in the positive and negative ion modes were separately processed using Compound Discoverer™ 2.0 (Thermo Scientific™). Four output tables (RP+, RP-, HILIC+, and HILIC-) were generated, including *m/z*, retention time, and peak intensity, for all the analyzed samples. An untargeted metabolomics workflow for retention time alignment, component detection, elemental composition prediction, and gap-filling was used. The workflow tree included the following nodes: input files, select spectra, align retention times, detect unknown compounds, group unknown compounds, fill gaps, normalization areas, and mark background compounds. The raw files were aligned with an adaptive curve setting with a 5 ppm mass tolerance and a 0.4 min retention time shift. Unknown compounds were detected with a 5 ppm mass tolerance, signal to noise ratio of 3, 30% of relative intensity tolerance for isotope search, and 10,000 minimum peak intensity, and then they were grouped with 5 ppm mass and 0.3 min retention time tolerances. A procedural blank sample was used for background subtraction and noise removal during the pre-processing step. Peaks were removed from the list if they showed less than a 3-fold increase compared to blank samples or if they were detected in less than 50% of QCs and/or with relative standard deviation (%RSD) of the QCs greater than 50%. To balance differences in intensities that may have arisen from instrument instability, a normalized area across all samples was provided for each detected metabolic feature by normalization to the periodically analyzed QC samples (pooled QC).

Finally, the hit intensities of each sample were multiplied by the dilution factor used for pre-normalization. Thus, un-normalized data were used to ensure a better degree of correlation between NMR and MS.

#### *4.4. H-NMR Spectroscopy*
