Semi-polar compounds were extracted, including primary and secondary metabolites, using automated high-throughput ethanol extraction procedures at the MetaboHUB-Bordeaux Metabolome (https://metabolome.u-bordeaux.fr/) from 35 mg of fresh powder, following previously established protocols (Luna et al., 2020). All samples were randomised and injected alternately with extraction blanks (prepared without plant material and used to rule out potential contaminants detected by untargeted metabolomics), and 13 Quality control samples that were prepared by mixing 10 µL from each sample. Quality control samples were injected every 8 runs and used for the correction of signal drift during the analytical batch, and the calculation of coefficients of variation for each metabolomic feature so only the most robust ones are retained for chemometrics (Broadhurst et al., 2018). Untargeted analysis was performed on a UHPLC Vanquish (Thermo Fisher Scientific) coupled to a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). One µL of sample was injected on a Phenomenex Luna® Omega Polar C18 column (50 x 2.1 mm, 1.6 µm) at 40°C and a gradient of solvent A (milliQ water – 0.1 % formic acid) and solvent B (acetonitrile – 0.1% formic acid) with a flow of 0.5 mL min-1 was used. The gradient elution was set as follows: 0-11.5 min: 1-40% solvent B; 11.5-12.5 min: 40-95% solvent B; 12.5-14 min: 95% solvent B; 14.5-16 min: 1% solvent B. The mass spectrometry data was acquired in negative polarity at 140.000 FWHM resolution with an automatic gain target at 3e6 and maximum IT of 100 ms. The source conditions were as follow: Spray voltage: 3000 V; Sheath gas: 45 a.u; Auxiliary gas: 15 a.u; Capillary temperature: 320°C; Probe heater temperature: 250°C; S-lens RF level: 100. The experiments were in full scan (mass range: 70-1050 m/z) – data depending MS2 with top three precursors and normalized collision energies of 15, 30, 45 using a dynamic exclusion of 5 s.