Plant material used in this study involves 22 genotypes of fig trees from an ex-situ collection located in the Sais plain in Morocco and managed by the National Institute for Agricultural Research (INRA Morocco). The collection is 14 years old, trained to a cup shape, and planted in a randomized complete block design with three trees per genotype and a spacing of 5x3 meters. The collection comprised a diverse range of both exotic and locally sourced genotypes, as detailed in Table 1.

Fig fruits were harvested at their full ripening stage in the summer. The genotypes were carefully selected based on the quantity and size of seeds in their fruits. To determine full ripeness, the figs were evaluated for a reddish-purple coloration that covered three-quarters of the receptacle and ensured that the fruit could be easily separated from the twig. The fruits were randomly picked from various positions around the canopy at a height of 160 cm, ensuring that the sample was representative of the entire tree.

Manually peeling figs, the seeds were separated from the pulp through a technical ethanol solution (10%). The mixture was stirred for about 10 minutes, then allowed to settle until the seeds separated and floated to the surface. The yellow, round-shaped seeds were collected, thoroughly washed with distilled water, and left to dry at room temperature for 24 hours. For each sample, the extracted seeds were ground into a fine powder using an IKA A11 Basic type grinder. The fig seed oil was then extracted chemically using a Soxhlet apparatus. To extract the oil, 20g of the powder was mixed with 150ml of 99% pure n-hexane using cellulose cartridges. The extraction process took up to 4 hours, after which the solvent was evaporated using a Buchi rotavapor R-200 at 40°C. The oil weight was determined through the relationship provided by Chougui et al. (2013):

Where M₀ refers to empty flask weight (g), M₁ is the flask weight after solvent evaporation (g), and M₂ is the seed powder weight (g). At the same time, the average oilcake weight was determined, and extracted oils were stored in dark glass bottles at 4°C until analysis.

The Total Nitrogen (NT) content was quantified using the Kjeldahl method (Buchi, Switzerland). The Nitrate content was measured by complexing with chromotropic acid and analyzing the absorbance in a spectrophotometer (Spectronics, USA) set at 410 nm (Hadjidemetriou, 1982). The protein content was calculated based on the Nitrogen content, using the formula P = 6.25 * NT. The Phosphorus (P) content was determined using the yellow color method with NaOH to neutralize excessive acidity, measured using a spectrophotometer (Shimadzu, China) (Gupta et al., 1993). Other elements (K, Na, Ca, Mg, Fe, Cu, Zn, and Mn) were analyzed through a dry incineration method carried out at 50°C for 48 hours in an oven. The macroelements (K, Na, Ca, and Mg) were evaluated using a flame photometer (Model CL 378, Elico, India) after extraction with nitric acid from 250mg of dry samples (Knudsen et al., 1982). The microelements (Fe, Cu, Zn, and Mn) were evaluated by atomic absorption spectrometry (AAS) (Perkin Elmer Analyst 300, USA) (Estefan et al., 2013). Three measurements were taken for each sample to ensure the reliability of the analysis.

To validate the results of our chemometric analysis, we first checked the normality and homogeneity of variance in our data. Next, we used IBM SPSS Statistics v22 software to conduct a one-way ANOVA to determine significant differences (P<0.05) between fig seed samples. The ionomic and FTIR data were then subjected to a principal component analysis (PCA) to uncover variables and spectral signatures that best explain the phenotypic diversity between samples. The PCA was performed using OriginLab Pro v2019 and 2D scatterplots were created to illustrate the classification patterns based on genetic seed variability. The PCA was significant because it reduced the dimensionality of the data set and allowed us to identify the signatures that reflect shared variance within the data set. Pearson’s correlation coefficient (r) was used to evaluate the associations between traits. Lastly, we performed a hierarchical cluster analysis (HCA) using Ward’s method based on Euclidean distance to cluster fig seed genotypes based on mineral composition and protein content. The results were displayed using a tree diagram (dendrogram) after standardizing the data.

Fig seeds are typically round and yellow in appearance. According to Hssaini et al. (2020), the average weight of a thousand seeds is around 1.14 ± 0.01g. However, as shown in Table 2, the weight of seeds per fig fruit can vary significantly based on the genotype. There were differences ranging from a minimum of 1.14 g/fruit for the “Grosse Dame Blanche” genotype to a maximum of 2.45 g/fruit for the “El Qoti Lezreq” genotype. On the other hand, few studies have evaluated the oil yield from fig seeds. Yarosh and Umarov (1971) found that the extraction produced 29.4% oil, while seeds from various regions in Turkey had oil yields ranging from 23.06 to 23.67% for the “Sarilop” cultivar (Nakilcioğlu-Tas, 2019). Joseph and Raj (2011) reported that the dried fig seeds had a fixed oil content of 30%. Hssaini et al. (2020) found that the oil content of four different fig seed cultivars from Morocco ranged from 21.54 to 29.65% and was characterized by a high content of linolenic acid. This high yield compared to the oil from other fruits makes fig seed oil a valuable component of the human diet and its commercial production should be encouraged due to its health benefits (Hssaini et al., 2020; Hssaini et al., 2021a). Despite these benefits, there are limited studies on fig seeds and the impact of cultivar on oil yield has not received much attention. However, our results are in line with previous studies. As seen in Table 2, the oil yield from fig seeds can vary greatly based on the cultivar, with the highest yields recorded by the “Bourqui” (39.68%) and “Bourjassate Noire” (38.44%) genotypes, while the genotypes “INRA 2105” and “INRA 2802” had lower yields of 15.06 and 16.66% respectively. These results suggest that the variability in yield can be attributed to the genetic diversity of the cultivars. Further research is necessary to fully understand the impact of cultivar diversity on the chemical composition and other qualities of this oil.

Richness of fig seeds in proteins and mineral elements has already been reported (Duman et al., 2018; Nakilcioğlu-Tas, 2019; Bölek, 2020); also, the effect of pollination on their content (Hssaini et al., 2022). Ustun-Argon et al. (2021) report that Potassium, Magnesium and Calcium were the main mineral compounds in the dried fig seeds. Our investigation showed similar results. Nevertheless, identification of genetic material that combines both agronomic performance and good nutritional quality is a major objective of genetic improvement programs for agricultural organisms and nurseries. During the past ten years, biofortification has gained prominence as a way of enhancing the nutritional profile of food crops (Jha and Warkentin, 2020). This is how this investigation was initiated, which remains to the best of our knowledge the first to highlight the genotypically effect of fig seeds diversity on the protein composition and mineral elements. A potential effect of biotic and abiotic factors, as well as essential chemical inputs, on fruit and seed cultivation and quality exists. To this end, genotypes were methodologically cultured under similar experimental conditions to circumvent environmental variability, and the discrimination collected is potentially linked to the genetic factor of the matrix. Thus, analysis of results presented here, first, joined the previous investigations on the presence in figs seeds ten mineral elements including Nitrogen (N), Phosphorus (P), Potassium (K), Calcium (Ca), Sodium (Na), Iron (Iron), Zinc (Zn), Manganese (Mn), Copper (Cu) and Magnesium (Mg), which showed as expected statistically significant (P<0.05) and non-significant differences between genotypes. Among macro-elements (P, K, Na, Ca and Mg) analyzed, calcium and potassium were the most abundant elements in the figs seeds genotypes studied (Figure 1). As the visual representation alone did not provide a comprehensive understanding of the phenotypic diversity, we recognized the need to supplement it with numerical data. In order to achieve a more thorough analysis, we included Tables 3A, B which presents the experimental data for the nutritional elements found in fig seeds, combined with the ANOVA results for fig samples mineral composition. This additional data not only supports our findings, but also offers a more detailed perspective on the observed phenotypic diversity.

According to the data presented in Table 3A, it’s evident that the Ca content in the different fig seed genotypes varied significantly. The highest concentration of Ca was observed in the “Aicha Moussa” genotype, with a value of 32.18 ± 0.50 (g/kg), whereas the lowest concentration was found in the “Nabout” genotype, with a value of 7.28 ± 0.08 (g/kg). Similarly, Potassium was present in significant amounts in all the sampled genotypes, with values ranging from 1.84 ± 0.00 (g/kg) in the “INRA 2105” genotype to 5.8 ± 0.28 (g/kg) in the “Melissosyki” genotype. In addition to the macro-elements, the results from the study also revealed that the sampled genotypes had a substantial content of Iron compared to other micro-elements (as depicted in Figure 2). The Iron content ranged from a minimum of 38.98 ± 6.33 mg/kg observed in the “INRA 2105” genotype to a maximum of 332.58 ± 16.88 mg/kg recorded in the “Aicha Moussa” genotype.

Furthermore, Nakilcioğlu-Tas (2019) reported the protein content of fig seeds between 14.74-15.07%, and Ustun-Argon et al. (2021) found 14.65%. Whereas, in our investigation, protein content varied between a minimum of 6.08 ± 0.04% recorded in the “INRA 2105” genotype and a maximum of 24.11 ± 2.32% recorded in the “Rhoul” genotype. As a comparison with the content of other fruit seeds, protein content of mangosteen seeds is 6.57%, 7.6% for guava, 12.4% for rambutan, 25.0% for honeydew, 25.63% for papaya, 25.2 to 37% for watermelon and 39.25% for pumpkin (Raihana et al., 2015). Moreover, between nitrogen and proteins, proteins are logically the most abundant in the fig seed genotypes studied (Figure 3). In agreement with the finding of Hssaini et al. (2022), it seems that minerals and proteins content of fig seeds depends on pollination and then on the source of pollen, which requires further studies able to investigate the qualitative performance of female genotypes sampled by compared to crossing with any caprifig tree. Since fig seeds have not received particular attention until now compared to the other parts of the species, results reported here constitute a solid base which will allow the orientation of new investigations.

To explore potential connections between protein content and mineral levels in fig seeds, we conducted a pairwise correlation analysis using the Pearson coefficient and presented the results in Figure 4. Weak correlations between the variables are represented by a low color intensity, while strong correlations are indicated by a high color intensity. Only correlations that were statistically significant at the 0.001 level with correlation coefficients greater than or equal to |0.5| were considered noteworthy. The matrix heatmap generated from the correlation analysis revealed a highly significant positive correlation between protein content and phosphorus content. We also observed a significant positive correlation between copper and iron levels. Additionally, bivariate correlations of the variables showed positive weakly correlated associations between manganese and protein as well as potassium. However, a highly significant negative correlation was detected between the calcium content and both phosphorus and sodium levels. Interestingly, our results differ from those of previous studies that found a significant positive correlation between iron and zinc concentrations in lentil seeds (Choukri et al., 2020). No such correlation was observed in the analyzed fig seed genotypes. Our findings align with those of other researchers, such as Kumar et al. (2014) and Darai et al. (2020), who reported similar observations of no correlation between iron and zinc levels. These results hold significant implications for future fig seed breeding programs, as they can serve as biomarkers to predict the level of significance, strength, and direction of associations between different variables. Examining potential correlations in the data set provides valuable insights into the most critical factors for evaluating and classifying fig seed genotypes.

The hierarchically clustered heatmap is a useful tool for analyzing complex biological data without reducing the dimensionality. This technique displays network connections in a symmetric adjacency matrix, providing a two-dimensional color-coded heatmap (Clark and Ma’ayan, 2011). The heatmap was formed with two groupings using the Euclidean distance and the Ward’s method. The first grouping was oriented for fig seed genotypes, while the second was for mineral elements and proteins, with a color indicating numerical differences in the correlation coefficient (Figure 5). The heatmap showed that Calcium and Iron had the highest scores in the data set, meaning they had the greatest impact on genotype clustering. However, other variables also showed an impact, although they were less important than these two elements. The combined dendrogram in Figure 5 differentiated the fig seed genotypes into four main clusters. The first cluster included two genotypes; “Hayoul” and “Aicha Moussa”, which were particularly very rich in Ca, Fe and Cu, and moderately rich in Zn, K, proteins and poor in P and Na. The second cluster included nine genotypes composed of “INRA 2105”, “INRA 2802”, “Breba Blanca”, “Kamalata”, “Nabout”, “Hafer Jmel”, “INRA 2603”, “White Adriatic” and “Embar El Khal”, which were rich in Mg, Cu and poor in proteins, P and Na. The third cluster consisted of genotypes including “Ahra”, “Rhoul”, “Amtalaa Arch”, “Breval Blanca”, “Lerida”, “El Qoti Lezreq”, “Bourjassate Noire” and “Melissosyki”, which were grouped based on their poverty in Ca compared to other clusters but were rich in proteins, P, Na, and Zn compared to other clusters. The last cluster included the genotypes “Grosse Dame Blanche”, “Adroulaniki” and “Bourqui”, which were mainly grouped by their richness in proteins, P, K and moderate values of Zn, Mn, Ca, and Fe but poor in Na and Mg. Finally, the heatmap combined to PCA (Principal Component Analysis) below allowed a better understanding of the relationship between the variables and the resulting clusters. The heatmap was a supervised variable construction method, while the PCA was an unsupervised method.

Based on the correlation coefficient, a principal component analysis (PCA) was performed to determine the discriminating variables in the data set. This analysis aimed to define the main factors that contribute to the discrimination of fig seeds. In our study, only a principal component loading of more than |0.35| was considered as being significant for each factor. The analysis of nutritional data using PCA revealed valuable insights into the minerals and protein content of fig seed cultivars. The results showed that three principal components were enough to explain 73.44% of the total variance, with the first component accounting for the majority at 38.69% (Figure 6). With reference to Table 4, the first component (PC1) consisted of the variables, N, P and Zn, for which the scores were respectively 0.45075, 0.39814 and 0.35065. It explained about 38.69% of the total variance observed, which means that these attributes had the highest impact on the discrimination. The second component (PC2) accounted for about 22.60% of the total variance and was defined by Fe, Cu and Ca, for which the respective scores were 0.53085, 0.49685 and 0.43542. The third component (PC3) represented 12.15% of the total inertia and was mainly correlated with the amount of Mg (0.7528) and Na (0.39577). This information reported here could be used to inform on the distribution of these compositions between all genotypes studied, and could serve as a starting point for further research into the relationships between fig seed phenotypic attributes and nutritional properties.

For evaluation of fig seeds chemotypic properties, FTIR-ATR fingerprinting has proven to be a powerful and interesting approach for screening and discriminating the characteristic signatures present in the analyzed samples (Hssaini et al., 2021a). Also, this technique has provided in previous studies satisfactory results with a high-throughput screening framework providing numerous biomolecules in several food samples (De la Mata et al., 2012; Anjos et al., 2015; Cassani et al., 2017). In this sense, FTIR spectroscopy coupled with an ATR was used in this work to assess relevance of chemotypic discrimination, and to identify the characteristic molecular signatures of biochemical substances present in fig seeds in order to study their sensitivity to phenotypic diversity. Therefore, seeds samples were directly scanned in the wavenumber range of 4000 to 450 cm^-1 with a spectral resolution of 4 cm^-1. Figure 7 illustrates spectra of the 22 fig seeds cultivars. Obviously, as expected for all seeds, all the spectra are similar. However, using chemometric tools, analysis of these spectra reveals important differences, particularly by evaluating their integrated intensities for each major vibrational region. These differences, even if sometimes considered minor, indicate molecular dissimilarities between samples. Overall, the peaks of each spectrum were identified at different wavenumbers, assigned to specific functional groups (Koch et al., 2014) and summarized in Table 5. The latter reports the detailed attributions of the bands, which reveal fifteen identifiable footprints in Figure 8. Indeed, Figure 8 represents a unitary case for the genotype “El Qoti Lezreq”, which perfectly shows values of the wavelengths from 4000 to 450 cm^-1 of the large vibratory regions with their corresponding peaks.

The first FTIR fingerprint appeared around 3417.51 cm^-1 which would be assigned to O–H stretching vibration, most likely attributed to fibers, which have been reported to be present in large quantities in fig seeds (Bölek, 2020). This flattened peak appeared in the vibrational region from 3750 to 3100 cm^-1 associated with intramolecular hydrogen bonding in cellulose between C(3)OH•••O(5) and C(6)O•••O(2)H according to Schwanninger et al. (2004); Oh et al. (2005) and Cassani et al. (2017). A second vibration can be detected between 3250 and 3000 cm^-1. This occurred around 3071.26 cm^-1, which is most likely attributed to C–H stretching of olefinic double bonds bonded to long carbon chains containing a relatively high number of unsaturated fatty acid CH₂ groups (Oh et al., 2005; Bouafif et al., 2008; Gopalakrishnan and Raghu, 2014; Niu et al., 2017). Absorbance in the vibrational region from 3100 to 2950 cm^-1 has been attributed to symmetric and asymmetric stretching of C–H, O–H and NH₃. The peak appears around 3012.70 cm^-1, which is probably typical for carboxylic acids, carbohydrates and phenolic compounds (Oh et al., 2005; Bouafif et al., 2008). Peaks in the range of 2950 to 2800 cm^-1 are identified as absorbances related to asymmetric and symmetric stretching of aliphatic C–H in the −CH₂ and terminal −CH₃ groups, respectively (Niu et al., 2017; Tulukcu et al., 2019). A first sharp and distinct peak is at 2929.11 cm^-1 and a second occurred at 2860.06 cm^-1. Bands in the range of 1775 to 1725 cm^-1 are attributed to stretching of ester carbonyl groups of triglycerides (Tulukcu et al., 2019). This absorbance region is related to C=O elongation of carboxylic ester type (Ahmed et al., 2005; Vlachos et al., 2006). At 1746.56 cm^-1, a sharp peak was identified, and is probably due to C=O stretching vibration of carbonyl groups belonging to triacylglycerols. It’s probably associated with lipids and fatty acids contained in seeds (Oh et al., 2005; De la Mata et al., 2012). Nevertheless, several authors report that this peak could be attributed to proteins (Cai and Singh, 2004; Cocchi et al., 2004; Cocchi et al., 2006; Terpugov et al., 2016).

Amide I and Amide II bonds usually resulting from protein-bound structures. Vibrations in the wavenumber range of 1700 to 1550 cm^-1 shows two peaks appeared at 1648.75 cm^-1 and 1544.32 cm^-1 which have been assigned to the Amide I and Amide II bands respectively (Ellepola et al., 2005; Tulukcu et al., 2019). According to several studies, this band between 1500 and 1100 cm^-1 are the result of several weak peaks wouldn’t be differentiated in the analyzed samples, which would correspond to mixed vibrations resulting from bending modes of groups >CH₂ and –CH₃ in proteins, fatty acids and phosphate bearing compounds (Tulukcu et al., 2019). The spectral band at 1458.15 cm^-1 could probably be associated with CH₂ bending and methylene deformation of lipids, proteins or cholesterol esters (Ahmed et al., 2005; Vlachos et al., 2006; Liu et al., 2021). Vibrations occurred around 1386.02 and 1239.31 cm^-1 could probably be assigned to CH₂ shearing and C–O stretching vibrations attributed to esters (Ahmed et al., 2005). Indeed, absorptions of certain bending vibrations of the methylene group occur between 1350 and 1150 cm^-1 (Guillén and Cabo, 1997). On the other hand, C–O bond stretching vibrations of esters are composed of two coupled asymmetric vibrations C–C(=O)–O and O–C–C, whose the former being larger (Sylverstein et al., 1991); these bands occur in the region between 1300 and 1000 cm^-1. The C–C(=O)–O band of saturated esters appears between 1240 and 1163 cm^-1, and in unsaturated esters, the vibration is produced at lower frequencies (Guillén and Cabo, 1997). Thus, the band appearing around 1165.91 cm^-1 can be assigned to bending vibration of C–O ester group (Ahmed et al., 2005; Niu et al., 2017). However, O–C–C band of esters derived from primary alcohols appears in the range between 1064 and 1031 cm^-1, whereas for those derived from secondary alcohols, the band appears at approximately 1100 cm^-1; what can probably correspond to vibration appeared around 1102.25 cm^-1 (Guillén and Cabo, 1997). Vibrational region ranging from 850 to 700 cm^-1 is marked by a sharp peak appearing at 721.19 cm^-1 which would probably be assigned to C–H deformation in aromatic hydrocarbons (Guillén and Cabo, 1997; Gopalakrishnan and Raghu, 2014). The last vibrational region ranging from 700 to 400 cm^-1 shows a peak around 611.93 cm^-1 attributed to overlap of bending vibrations of methylene −CH₂ and out-of-plane vibrations of C–H bonds of disubstituted olefinic cis-alkenes (Van de Voort et al., 1995; Vlachos et al., 2006). As expected, although the fig seeds spectra appear to be similar and generally overlap in Figure 7, they show differences in integrated intensity of their bands as well as in the exact frequency at which maximum absorbance is produced in each case; probably because of genotypically differences in nature and composition.

The marginal boxplot chart in Figure 9 provides a clear visualization of the phenotypic differences between the seed cultivars based on their vibrational intensities in the 3500-1000 cm^-1 range. The chart displays the distribution of integrated intensities for each genotype, as well as the median, interquartile range, and outliers. The one-way analysis of variance conducted on the data revealed statistically significant differences in the integrated intensities of major peaks among the sampled fig seeds. For instance, the “Hafer Jmel” genotype had an integrated area of approximately 2250, while the “White Adriatic” genotype had an integrated area of around 2500. Interestingly, the “Breval Blanca” genotype had the lowest integrated area of about 1500, making it unique among the cultivars studied. These findings highlight the significant impact of genotype on the molecular signatures of fig seed cultivars. Although the differences may seem minor, they are crucial in determining the distribution of nutritional composition among the various genotypes. This is consistent with previous studies that have demonstrated a similar pattern of phenotypic effect on fig seed quality, including nutritional composition. Overall, the results suggest that FTIR-ATR spectroscopy could be an effective and non-destructive method for rapid analysis of the structural characteristics and functional properties of various genotypes of fig seeds. This could have important implications for improving breeding programs and enhancing the nutritional value of fig seeds for human consumption.

Infrared spectroscopy or FTIR is a powerful analytical tool that has been widely used with chemometric methods such as PCA. By using PCA, the complexity of multi-wavelength FTIR data can be reduced to a limited number of parameters, with minimal loss of total variance. This process results in the creation of a scatterplot, which reveals the resolution of the classification of fig seed samples based on their phenotypic diversity. This unsupervised variable construction method was applied to spectra to unveil the resolution of fig seed classification based on phenotypic diversity. This technique was also employed to evaluate if spectra data could produce a discrimination pattern similar to that of nutritional composition. Figure 10 presents FTIR data in the 4000-450 cm^-1 wavenumber range, exhibiting a total variance of 91.62%. The first component (PC1) accounted for 64.09% of the variance, while the second component (PC2) contributed 27.53% to the total variance. The PCA plot displayed a high level of discrimination resolution, clustering cultivars with similar FTIR profiles, as spectra are a qualitative method that precisely measures functional groups.

The scatter plot of the PCA model reveals four main clusters, with clear classification of the “Breval Blanca” genotype, which showed the lowest vibration intensities among all cultivars studied. These four groups were differentiated based on the total intensity area in the entire vibration range. This result is in line with previous studies that used FTIR spectra and chemometric approaches to distinguish several fruit seeds, including apples and grapes. Previous studies using FTIR spectra coupled with chemometric approaches have demonstrated the excellent discrimination capacity of this method in differentiating various fruit seeds, such as apples (Anang et al., 2019) and grape (Lucarini et al., 2019). The results suggest that FTIR spectroscopy is a useful tool for the screening and discrimination of a large sample of fig seeds based on chemotypic attributes. It’s an accurate, rapid, inexpensive, and environmentally friendly technology that requires minimal sample preparation, making it a valuable tool for the management of fig seed diversity and for use in breeding and industrial development programs. This study is the first of its kind to assess the chemotypic attributes of a wide range of fig seed genotypes and has important implications for maintaining the longevity and genetic diversity of the species.