3.4. Metabolite Analysis by NMR

The pH of each sample prepared for NMR analysis was adjusted to 6.00 ± 0.02, and the samples were dried under vacuum and then dissolved in 800 µL of deuterated solvent prepared in a mixture of (1:1) Methanol-d4: KH₂PO₄ buffer (0.1 M) in D₂O at pH 6.0 with TMSP (0.0125%), NaN₃ (0.6 mg/mL), and 15 mM EDTA-d12. Then, samples were briefly vortexed, sonicated, and centrifuged. The supernatant was placed in 5-mm NMR tubes and then used for NMR analysis.

All NMR spectra were acquired at 300 K with a Bruker Avance III 600 MHz spectrometer operating at 600.13 MHz for ¹H, and 150.91 MHz for ¹³C, using a 5-mm multinuclear broadband, equipped with z-gradient (TXI 5 mm tube). CD₃OD was used as the internal lock. All 1D ¹H-NMR spectra were collected using 32 scans of 131 K data points and a spectral width of 8417 Hz with a relaxation delay of 13 s, and a water suppression pulse sequence. The resulting spectra were automatically phased and baseline corrected, using Bruker Topspin software (version 3.5) [51]. 2D J-resolved NMR spectra were processed with a 2 s relaxation delay using 16 scans per 64 increments collected into 64K data points, with spectral widths of 8417.5 Hz in F2 and 50 Hz in F1. 2D HSQC spectra were recorded with a 2 s relaxation delay, using 64 scans per 256 increments that were collected into 4 K data points using spectral widths of 8417.5 Hz in F2 and 50 Hz in F1. All spectra were calibrated to TMSP at 0.0 ppm by the Topspin v3.5 (Bruker) software. Each of the 2 D acquisitions were performed by analyzing one control and one stressed samples of each part of plant corresponding to each of the two lines (Pi1AM and LuT).

¹H-NMR spectra were automatically converted into ASCII format and the data from each part of plant (roots, stems and leaves) were collected and imported into Matlab software (version 2018a, the Mathworks Inc, Natick, MA, USA) where baseline correction was performed with the algorithm “airPLS 2.0” [53,54,55]. All spectra were aligned using the icoshift algorithm (v 1.2.1) with manually defined alignment bins. Then specific integration intervals of the spectra ‘‘buckets’’ were defined manually, and each bucket was integrated. Regions of the NMR spectra corresponding to the methanol-d4 (3.34–3.30 ppm), to residual water (4.85–4.70 ppm), to TMSP (0.01–0 ppm) and to PEG-6000 (3.68–3.64 ppm), were removed. All signals with intensities lower than 3.3 times the mean variance from such a noise region were considered to be noise and were also removed. The obtained datasets were then used for statistical analysis. Metabolites in the 1D and 2D NMR spectra of flax extracts were identified based on comparison with spectra and chemical shifts of reference compounds of database previously developed in the laboratory.

Extracts of stems and leaves were diluted 5 and 10 times, respectively, with methanol/water (50/50), while no dilution was performed for root extracts. All samples were filtered through 0.22 µm PTFE membrane filters and placed in glass vials for further LC-MS analysis. For each plant part, a QC sample were prepared; 10 µL from 20 different samples were taken and thoroughly mixed, reaching a total volume of 200 µL. For roots, stems, and leaves, analysis samples were run in a randomized order.

The metabolomics analysis was performed using ACQUITY UPLC H-Class system (Waters Micromass, UK), coupled to a SYNAPT G2-Si Q-TOF mass spectrometer (Waters-Micromass, Manchester, UK), which was equipped with an electrospray ion source (ESI). UPLC separation was performed using a Kinetex C18 (1.7 µm, 100 mm × 2.1 mm, Phenomenex, Torrance, CA, USA) column. The column temperature was maintained at 50 °C. The injected volume was 1 µL. Water (A) and methanol (B), both supplemented with 0.1% formic acid, were used as mobile phases. A stepwise gradient method was used for elution at a flow rate of 0.4 mL/min, with the following conditions: 5–95% B (0–7 min), followed by 3 min of isocratic 95% solvent B, and 1 min gradient to 95% (A), followed by 5 min of re-equilibrium at 100% A. MS data was collected in the negative ion mode, over a m/z range of 50 to 1150. The parameters of electrospray ionization (ESI) source were set as follows: capillary voltage at 3 keV, the cone voltage at 3 V, source temperature at 120 °C, desolvation at 450 °C, the cone gas flow 6.5 bar and the desolvation gas flow 800 L/h. Analyses of the samples were carried out in a mode of a full MS survey, at a resolution of 2000 (FWHM). MSMS scans for most intense peaks were performed to produce high resolution MSMS spectra, with a collision energy of 30 eV. Data acquisition was performed by MasslynxTM v4.1 software (Waters, Milford, MA, USA). The QC samples were injected at the beginning of the run to set up the system and then every eight samples, so they were used to ensure system conditioning within the analytical sequence.

The acquired spectral data were converted to mzXML format using the Proteowizard MSConvert tool from 0 to 10 min RT in order to avoid features coming from cleaning step of the gradient.

Then, the mzXML files data of each part of the plant were loaded and pre-processed with the XCMS package (v3.0.2) in the open-source R software (v3.2.2).

The centWave algorithm was used for peak detection with the following optimized parameters: minimum peak width = 3 s, maximum peak width = 15 s, ppm = 5, threshold = 2, mzdiff = 0.005 and prefilter: (4,100,000), and noise filter = 1000. Peaks were well aligned by XCMS, using the following parameters: bw = 5.0 and mzwid = 0.025. Retention time correction was performed by the Obiwarp algorithm, which aligns multiple samples by using a center sample.

A filling step was included to reduce the number of missing peaks, using the fillPeaks tool. For each sample, Peak area, retention time and peak widths, calculated as the difference between the end and start of the integration points, were extracted from XCMS data.

After the whole data processing, a table (matrix) including retention time and m/z, sample names, and ion intensities, was obtained for each sample set. Then, before the statistical analysis, these matrices were prepared in order to remove the features before the injection peak (less than 1 min).

The percentage of relative standard deviation (%RSD) was calculated for all metabolic features in QC samples and the features with %RSD greater than 30% were removed due to its variability.

Metabolites were principally identified by matching masses, retention times and fragment patterns of pure standards.

The obtained datasets for NMR and LC-MS were imported into SIMCA-P software (version 15.0; Umetrics, Umea, Sweden), and applied to principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) for multivariate statistical analysis. The UV scaling method was used.

The Wilcoxon Rank Sum test was used in R’s statistics base-package in order to test the significant difference in metabolite content, between analyzed groups, with different p-values (p < 0.05; p < 0.01; p < 0.001).