The EPR spectra of the wheat (A, B, C, and D) seeds are displayed in Fig. 2. Under the measurement condition of 1 mW, the magnetic field (G) around 3350 G value indicates the presence of carbon radicals24,29,30which originate from proteins or other organic compounds. Wheat seeds were characterized by macroelement magnesium, whereas manganese, iron, copper, zinc, chromium, and nickel31 are among the microelements that are often present. The five unpaired electrons in manganese divide the spectral line into six (broad) because of its spin of 5/2 which is shown in Fig. 2. As shown in the literature, organic moieties, and manganese (II) metal may be detected interacting in the 3310–3410 G range8.As shown in Figure S2, the EPR spectrum was recorded at 10 mW. Since organic molecules have longer relaxation times concerning metals, the signal around 3350 G (corresponding to the carbon radicals) decreases and is dominated by the metals.The magnetic field (G) around 3350 G value in the EPR spectra of the seed E (Atriplex) shown in Fig. 3under 1 mW measurement conditions confirms the presence of carbon radical (as in seed A-D) with a sharpened peak. This is because there is a slightly higher presence of carbon radicals when compared to other species like A-D. The background of the spectrum is affected by the iron ion influence that is observed in the elevation of the peak around 3100 G. This effect either mask the presence of manganese ions or might indicate a lower quantity. Furthermore, an iron metal impact can be seen in the EPR because, when the measurement power is increased to 10 mW, the peak broadens and changes substantially below 3100 G (Figure S3)32.Fig. 2EPR spectra of seeds belonging to the wheat (T. durum L.) genotypes Mahmoudi (A), Maali (B), Karim (C), and Jnah Khotifa (D) were obtained by measuring with the microwave power of 1 mW at room temperature.Fig. 3EPR spectra of seeds belonging to A. halimus L. were measured with the microwave power of 1 mW at room temperature.For the seeds F and G (Fenugreek), the EPR spectra were observed under 1 mW overlay and show at G ~ 3350 a footprint for carbon radicals like the one observed with Atriplex (Fig. 4). However, these spectra show other significant differences, and the signal splitting at a G value of 3100, indicates the presence of copper and manganese metals in the seed33. Figure S4 illustrates the increased energy dispersion of metals that dominates the carbon radical peaks (organic polymers/compounds) under 10 mW measurement conditions. Moreover, the change in the background of the spectral line indicates the existence of iron metal33.Fig. 4EPR spectra of the seeds belonging to the trigonella (T. foenum graecum L.) genotypes F and G were measured with the microwave power of 1 mW at room temperature.Each of the following varieties (A, B, C, D, E, F, and G) were measured. We took into consideration the region of manganese and copper (3100 G, EPR M (metal region)) and carbon radicals (3350 G, EPR C (carbon region)) for PCA analysis, for both 1 and 10 mW.For the two regions, the value inserted is the maximum height of the EPR signal in the ranges, respectively, 3325.3-3385.3 G and 3081.3-3146.3 G. Values are normalized concerning the gain, number of scans, and seed weight, to have comparable results.From a brief comparison between the EPR spectra, seeds F and G contain higher amounts of copper, while seed E has a different spectrum with low manganese. Little differences are visible among spectrums A-D. We suppose that in the absence of an EPR spectrometer, metal content analysis could give similar results, except for the measurement of carbon radicals.High-pressure liquid chromatographyHPLC chromatograms were divided roughly into five zones which are shown in Fig. 5 as indicated in Table 3. The total area of the signal in those zones is considered for the PCA analysis. Since the system is working with a reverse phase column, the more the compounds are polar the less they are retained in the column. Zone 2 corresponds to mixtures of polar compounds like glycosides, phenols, and other polar compounds. Zone 3 is for intermediate polarity secondary metabolites while zones 4 and 5 are for lipids and derivatives.Table 3 Start and end retention times for the integrations of zones defined for PCA.From a brief comparison, seeds F and G represent a group having less polar compounds. Also, seeds E have a different profile at a retention time of 4 min. All the remaining seeds have similar chromatograms, but the D genotype (Jnah Khotifa) exhibits higher polar compounds and secondary metabolites, compared to the three other wheat genotypes. It is worth noting that Jnah Khotifa is an old durum wheat landrace that started to be cultivated in Tunisia in the first half of the 20th century34. This landrace was also reported to accumulate higher kernel weight and protein concentration, a relevant trait for the semolina industry34. Therefore, we cannot rule out a possible contribution of these polar compounds and secondary metabolites in this specific seed trait.Fig. 5HPLC chromatograms for the sample considered in the paper. Zones are divided into sections or bins, corresponding to the total area, in different regions, named: HPLC zone 1, HPLC zone 2, HPLC zone 3, HPLC zone 4, and HPLC zone 5.Principal component analysisPrincipal component analysis (PCA) is a technique used to understand the minimum number of components needed to describe the system under investigation. The technique is a very well-known method used for data analysis and reduction of complexity. We developed our procedure, written in Octave, and released it as an open source (S3). Data coming from observations are placed in a matrix m x n, where m is the number of columns corresponding to different measurements and n is the number of samples. In our case, 13 measurements were considered, named: HPLC zone 1, HPLC zone 2, HPLC zone 3, HPLC zone 4, HPLC zone 5, EPR C 1 mW, EPR M 1 mW, EPR C 10 mW, EPR M 10 mW, TGA 25–150, TGA 150–275, TGA 275–350, TGA 350–800 (described in the previous sections).Fig. 6A representation of the reduced space. The number on the axes title represents the number of the component.First, the matrix is normalized, in a way that all measurements will span the same maximum and minimum values. This is essential to obtain quantitative information from the subsequent analysis that permits performing comparisons between the different columns. PCA transformation can be used to reduce the dimensionality of the observation space. The results of PCA performed by using Octave are shown in Fig. 6. The key transformation is the Single Value Decomposition, from the observation matrix M, the decomposition M = USV* is created. The most important eigenvectors of U give a dimension reduction of the dataset. Data can be shown by using an arbitrary number of eigenvectors to understand similarity. As it is visible from Figure S5, it is sufficient to reduce the space to only one dimension to cluster wheat seeds, while the reduction to 3 dimensions is clearly showing the groups for the different seeds. As it is visible in Fig. 6, seeds A, C, D are in the left-low part of the diagram. Close to this group but in the same region of space are grouped seeds B, while E and F-G are in different volumes of the projection. It is rather unexpected that the seeds B are clustered a bit separately especially from seeds C, knowing that both genotypes Karim and Maali which were introduced relatively recently in Tunisia (in 1982 and 2003, respectively) for their improved yield are phylogenetically related as they share similar pedigrees35,36. However, based on several morphological and biochemical traits, they remain distinct. For instance, a significant variability in enzymatic activities such as lipoxygenase and polyphenol oxidase which were reported to play a relevant role in durum wheat flour darkening during pasta processing was observed between Karim and Maali35,36.Fig. 7Heatmap corresponds to the representation of the absolute value of U, the m x m matrix coming from PCA. Each row represents a PCA component, starting from the most important (top). Columns are the number of observations, that is the experimental measurements.For a more comprehensive representation, the projected vectors corresponding to observations are shown in supplementary Figure S7. The matrix U, coming from the PCA, is a representation of reduced components of the observation matrix (Fig. 7). In the figure, the columns represent the measurement while the rows are the components. As an example, for the first component, the most important observation is ‘HPLC zone 2’ followed by ‘EPR M 10 mW’. By analyzing the second component, ‘HPLC zone 4’ is the most important observation, in fact, the vector ‘HPLC zone 4’ points in the direction of the maximum variations of the data group (Figure S7). The absolute importance of each component (rows of U) is given by the value of diagonal elements of S, shown in Figure S6. Usually, the diagonal components of S decrease rapidly, and their magnitude is related to their importance in describing the initial dataset. The analysis of the curve in Figure S6 confirms that the addition of another component contributes negligibly to the description of the initial dataset.To understand the relative importance of the single analysis groups (HPLC, EPR, and TGA), groups were removed one by one, and PCA analysis was performed again. From that analysis, it was evident that HPLC measurements are playing a key role in the correlation. The removal of all HPLC components gave rise to an unclear classification.
