Samples collection: six genotype of camelina were grown at the experimental farm of Bologna University (Italy) located in Cadriano (Bologna, Italy, 44°30′N, 11°23′E, 32 m a.s.l.) during five consecutive years (2015 - 2019). The six tested genotypes were: OMEGA (University of Poznan, Poland), WUR (Wageningen University and Research, The Netherlands), and 787-08, 789-02 and 887 cultivars (Smart Earth Camelina, Saskatoon, Canada). Camelina was sown each year in late winter/early spring accordingly to the specific meteorological conditions. The same agronomic management for all the trials was adopted consisting in typical tillage system (ploughing + harrowing), a seeding rate of 500 seeds m-2, no irrigation, a top-dressing fertilization with N (50 kg N ha-1, as urea at bolting stage), with no pest, disease nor weed chemical control. The experimental design was a randomized complete block with 3 or 4 replicates depending on the year. Representative seed samples from individual plots were cleaned and used for omic analyses. Only OMEGA and 789-02 seeds were used for LC-MS/MS untargeted lipidomic analyses.
Extraction: Metabolites were extracted from 60 mg of camelina dry mature seeds. Briefly, 1 mL of MeOH: Methyl-tert-butyl: H2O (1:3:1), conserved at 4°C, and 200 ng of Apigenin (used as internal standard) were added to each sample, which were then homogenized in 2ml tubes using a FastPrep instrument (1 min, 14,000 rpm). The mixtures were then shaken for 30 min at 4ºC using a ThermoMixer™ C (Eppendorf), placed in an ice cooled ultrasonication bath for 15 min and centrifuged for 1 min at the maximum speed to remove debris. The extracted samples were then transferred in new tubes containing 650 mL of MeOH: water (1:3), previously placed at -20ºC. The mixtures were centrifuged for 1 min at 14,000 rpm. The addition of MeOH: water (1:3) and the centrifugation led to a led to a phase separation, providing the upper organic phase, containing the lipids, a lower aqueous phase, containing the polar and semi- polar metabolites, and a pellet of starch and proteins. The phase containing the lipids was dried down in a SpeedVac vacuum concentrator (o/n) and resuspended in 200 μL of ULC/MS grade water (Biosolve).
Chromatography: an EC 100/2 Nucleoshell Phenyl-Hexyl column (2 x 100 mm, 2.7 μm; Macherey-Nagel) was used for chromatographic separation. The mobile phases used for the chromatographic separation were (A) H20 + 0.1% ammonium formate in H2O + 0.1% formic acid and (B) Acetonitrile : isopropanol (7 : 3) + 0.1% of 10mM ammonium formate + 0.1% formic acid. The flow rate was of 400 μL/min and the following gradient was used: 45% of A for 1-min, followed by a linear gradient from 45% A to 30% A from 1 to 2-min, then a linear gradient from 30% A to 15% A from 2 to 7-min, a linear gradient from 15% A to 10% A from 7 to 15-min, a linear gradient from 10% A to 6% A from 15 to 19-min, a linear gradient from 6% A to 2% A from 19 to 26-min. 0% of A was hold until 27-min, followed by a linear gradient from 0% A to 45% A from 27 to 35-min (35-min total run time).
Mass spectrometry: Data-dependent acquisition (DDA) methods were used for mass spectrometer data in positive and negative ESI modes using the following parameters: capillary voltage, 4.5kV; nebuliser gas flow, 2.1 bar; dry gas flow, 6 L/min; drying gas in the heated electrospray source temperature, 140ºC. Samples were analysed at 8Hz with a mass range of 100 to 1500 m/z. Stepping acquisition parameters were created to improve the fragmentation profile with a collision RF from 200 to 700 Vpp, a transfer time from 20 to 70 µs and collision energy from 20 to 40 eV. Each cycle included a MS fullscan and 5 MS/MS CID on the 5 main ions of the previous MS spectrum.
Data transformation: The .d data files (Bruker Daltonics, Bremen, Germany) were converted to .mzXML format using the MSConvert software (ProteoWizard package 3.0). mzXML data processing, mass detection, chromatogram building, deconvolution, samples alignment, and data export, were performed using MZmine 2.52 software (http://mzmine.github.io/) for both positive and negative data files. The ADAP chromatogram builder method was used with a minimum group size of scan 3, a group intensity threshold of 1000, a minimum highest intensity of 1500 and m/z tolerance of 10 ppm. Deconvolution was performed with the ADAP wavelets algorithm using the following setting: S/N threshold 8, peak duration range = 0.01-1 min RT wavelet range 0.02-0.2 min, MS2 scan were paired using a m/z tolerance range of 0.01 Da and RT tolerance of 0.4 min. Then, isotopic peak grouper algorithm was used with a m/z tolerance of 10 ppm and RT tolerance of 0.4. All the peaks were filtered using feature list row filter keeping only peaks with MS2 scan. The alignment of samples was performed using the join aligner with an m/z tolerance of 10 ppm, a weight for m/z and RT at 1, a retention time tolerance of 0.3 min.
Metabolites identification:
Metabolites annotation was performed in four consecutive steps. First, the obtained RT and m/z data of each feature were compared with our home-made library or experimental common features (RT, m/z). Second, the ESI- and ESI+ metabolomic data used for molecular network analyses were searched against the available MS2 spectral libraries (Massbank NA, GNPS Public Spectral Library, NIST14 Tandem, NIH Natural Product and MS-Dial), with absolute m/z tolerance of 0.02, 4 minimum matched peaks and minimal cosine score of 0.8. Third, not-annotated metabolites that belong to molecular network clusters containing annotated metabolites from step 1 and 2 were assigned to the same chemical family. Finally, for metabolites that had no or unclear annotation, Sirius software (https://bio.informatik.uni-jena.de/software/sirius/) was used. Sirius is based on machine learning techniques that used available chemical structures and MS/MS data from chemical databanks to propose structures of unknown compounds.
