*3.6. UPLC-ESI-QTOF-MS Conditions*

*Cytinus* extracts were redissolved at 5 mg/mL in the same extraction solvent and filtered by 0.22 μm. The compounds were separated using an ACQUITY UPLC H-Class System (Waters Corp., Milford, MA, USA) with a reversed-phase column (ACQUITY UPLC BEH Shield RP18, 130Å, 1.7 μm, 2.1 mm × 150 mm) at a flow rate of 0.7 mL/min and using a injection volume of 10 μL. The mobile phases were acidified water (0.5% acetic acid, *v*/*v*) and acetonitrile as solvents A and B, respectively. The following multi-step linear gradient was used in order to achieve an efficient separation: 0.00 min [A:B 99/1], 2.33 min [A:B 99/1], 4.37 min [A:B 93/7], 8.11 min [A:B 86/14], 12.19 min [A:B 76/24], 15.99 min [A:B 60/40], 18.31 min [A:B 2/98], 21.03 min [A:B 2/98], 22.39 min [A:B 99/1] and 25.00 [A:B 99/1].

The UPLC was coupled to an electrospray-quadrupole-time of flight mass spectrometer (ESI-QTOF-MS) Synapt G2 (Waters Corp., Milford, MA, USA) working in negativeionisation mode in a *m*/*z* range from 50 to 1200 *m*/*z*. The MS acquisition was based on two parallel scan functions switching between them continuously. The first function was operated at low collision energy in the gas cell (4 eV) and the other at an elevated collision energy (MSE energy linear ramp: from 20 to 60 eV). Leu-enkephalin was injected for mass calibration continuously. Other MS parameters were, as follows, source temperature 100 ◦C; scan duration 0.1 s; resolution 20,000 FWHM; desolvation temperature 500 ◦C; desolvation gas flow 700 L/h; capillary voltage 2.2 kV; cone voltage 30 V; cone gas flow 50 L/h. Finally, the acquired data were processed using MZmine 2.53 open-source software and Sirius 4.4.29.

#### *3.7. UPLC-ESI-QTOF-MS Data Processing*

Firstly, the raw data files were transformed to. mzML format using MSConvert software. The converted data were processed using the open-source software MZmine 2.53 (Pluskal et al., 2020). A noise level of 1.0 × <sup>10</sup><sup>3</sup> was selected. ADAP chromatogram builder method was used under the following parameters: MS level: 1; min number of scans: 9; group intensity threshold: 1.0 × <sup>10</sup>3; min highest intensity: 1.0 × <sup>10</sup>4; *<sup>m</sup>*/*<sup>z</sup>* tolerance: 10 ppm. After that, the chromatogram was deconvoluted using the Wavelets (ADAP) algorithm and the following parameters: S/N threshold: 50; min feature height: 5E4; coefficient/area threshold: 110; peak duration range: 0.05–0.3 min; RT wavelet range: 0–0.30. An isotopic peak grouper algorithm was also applied (*m*/*z* tolerance: 10 ppm, RT tolerance: 0.02 min, maximum charge: 2). The obtained features were aligned between samples using the "Join Aligner" algorithm using a *m*/*z* tolerance of 10 ppm and a RT tolerance of 0.1 min. The molecular features, which were also detected in blank samples, were filtered from the final dataset. Finally, the molecular formulas of the final features were predicted using Sirius 4.4.29 (Dührkop et al., 2019) and the biological identities were annotated by comparing the MS/MS spectra of different databases (e.g., MoNA, Massbank, HMDB, FoodDB, etc.), with the fragments detected in the MSE scans.
