The obtained results from all analyses simultaneously described the volatile profile of untreated pollen samples and pollen extracts of SMC, SCBD, and GEL cultivars of C. sativa L. To better compare the large amount of data obtained from chemical analyses performed by PTR-ToF-MS technique and to identify similarities and/or differences in chemical constituents, a multivariate statistical investigation was carried out. In particular, an unsupervised PCA analysis was carried out to determine how metabolites differ from samples, and which compounds contribute the most to this difference. Namely, the PCA unsupervised algorithm is an orthogonal linear transformation of possibly correlated variables into a smaller number of uncorrelated variables called principal components (PCs) by revealing group structure when within-group variation is sufficiently less than between-group variation. PCA allows us to determine the contribution of the original variables to the PC model. In the case of pollen samples, the performed PCA was based on the first two principal PCs scores: PC1 explained 78.3%, and PC2 explained 19.3%. The PCA score plot (Figure 3A) showed a separation of GEL in comparison with the other cultivars. GEL was the main contributor to the samples’ variability and the most dissimilar in term of percentages of its metabolites. SCBD and SMC were clustered within the same quadrant, suggesting a similarity in the levels of their metabolites. In addition, a biplot (Figure 3A) provides information on both metabolites (displayed as points) and samples (displayed as vectors) of a data matrix. Metabolites grouped at the origin of the graph did not contribute to samples’ variability. SCBD and SMC were mostly characterized by CH₃O⁺ (formaldehyde) and C₃H₇O⁺ (acetone), whereas GEL was mostly characterized by C₂H₅O₂⁺ (acetic acid), C₂H₃O⁺ (alkyl fragment (e.g., terpenes and other compounds)), CH₅O⁺ (methanol), and C₂H₅O⁺ (acetaldehyde).

Unlike the same PTR-ToF-MS analysis applied to pollen extract samples (Figure 3B), it was found that GEL and SMC were clustered within the same quadrant, suggesting a similarity in the levels of their metabolites, while SCBD, which was the main contributor to the samples’ variability, was the most dissimilar in term of the percentage of its metabolites. PC1 explained 87.0%, and PC2 explained 10.5%. GEL and SMC were mostly characterized by C₂H₃⁺ (acetylene), C₃H₇⁺ (alkyl fragment (propene)), and C₃H₅⁺ (alkyl fragments). CH₅O⁺ (methanol) was particularly represented in GEL. The main contributors to SCBD were C₂H₅O₂⁺ (acetic acid), C₄H₉⁺ (Alkyl fragment), C₂H₃O⁺ (alkyl fragment (e.g., terpenes and other compounds)), and C₃H₇O⁺ (acetone).

The complete list of compounds obtained by PTR-ToF-MS analysis of the pollen samples is reported in a heatmap (Figure 4A). Next to the heatmap, a hierarchical clustering (HCA) technique based on the Euclidean distance method is employed to analyze the similarities of metabolite trends in the different cultivars with a repetitive process that associates/dissociates (agglomerative/divisive methods) object by object until all are equally and completely processed [8,9]. In the first cluster, the most relevant compounds were shown to be C₃H₇O⁺ (acetone), CH₃O⁺ (formaldehyde), and CH₅O⁺ (methanol), identified in the three cultivars and sharing a common trend in which their levels were particularly elevated in SCBD. The second cluster defined by the dendrogram branch showed C₂H₃O⁺ (alkyl fragment (e.g., terpenes and other compounds)) and C₂H₅O₂⁺ (acetic acid) as metabolites with a high level in GEL. In the third cluster, we found only the compound C₂H₅O⁺ (acetaldehyde), which showed a singular trend and was higher in GEL and SMC but lower in SCBD. C₃H₅O⁺ (alkyl fragment (e.g., hexanal/1-butanol/1-octanol)) and C₃H₅⁺ (alkyl fragments) were gathered in the fourth cluster, with an elevated percentage in SMC. Compounds under the detection threshold are represented in grey.

Similarly, a heatmap and HCA (Figure 4B) were performed for pollen hexanoic extract samples deriving from the same three cultivars. The first cluster included a single metabolite C₃H₅⁺ (alkyl fragment) that was elevated in GEL and SMC. The second cluster was represented by the six metabolites showing a significant composition percentage (C₂H₅O₂⁺ (acetic acid), C₄H₉⁺ (alkyl fragment), C₃H₇+ (alkyl fragment (propene)), C₂H₃⁺ (acetylene), C₃H₇O⁺ (acetone) and CH₅O⁺ (methanol)). In the third cluster, the following metabolites were noticeable: C₂H₃O⁺ (alkyl fragment (e.g., terpenes and other compounds)), C₂H₅O⁺ (acetaldehyde), and C₄H₇⁺ (C4 aldehydes fragment).

Furthermore, Pearson’s correlation analysis was performed among various metabolites of C. sativa pollen. In Figure 5A, the complete dataset of pollen compounds is explored with a correlation matrix, a table representing correlation coefficients between variables. The dendrogram next to the correlation matrix clustered compounds sharing a similar correlation trend. The most important direct correlations were between the following sub-groups:

1° group: CH₃O⁺ (formaldehyde), C₃H₇O⁺ (acetone), C₃H₅⁺ (fragments), C₆H₁₃O₂⁺ (hexanoic acid/methyl valerate), C₃H₉O₂+ (propylene glycol), C₈H₇⁺ (fragments); and the 2° group: C₈H₁₃⁺ (sesquiterpenes/terpenes fragments), C₆H₁₁O₂⁺ (γ-hexalactone), C₁₁H₁₇⁺ (sesquiterpene fragments), C₂H₃⁺ (acetylene), C₃H₇O₂⁺ (methyl acetate/propanoates), C₅H₉O₂⁺ (hexanal), C₃H₅O⁺ (alkyl fragment (e.g., hexanal/1-butanol/1-octanol)), C₅H₅O₂⁺ (furfural), C₆H₉O₂⁺ (dimethyl-furanone/methyl-cyclopentanedione/cyclotene), C₉H₁₉O⁺ (nonanal), C₁₀H₁₇⁺ (monoterpenes), C₁₀H₁₇O⁺ (terpenoids), C₁₀H₁₅⁺ (p-cymene), and C₁₅H₂₅ (sesquiterpenes).

Another significant correlation emerged between the 1°subset C₃H₇⁺ (alkyl fragment (propene)), C₅H₁₁O₂⁺ (butyl formate), C₅H₁₁O⁺ (pentanal), C₄H₇O⁺ (2-butenal), and C₆H₉⁺ (C6 and terpenes fragments) and the 2° subgroup C₇H₁₁⁺ (terpenes fragments), C₇H₁₁O⁺ (dimethyl-2-cyclopenten-1-one), C₄H₉O⁺ (2-butanone/isobutyraldehyde), C₈H₁₇O⁺ (methyl hexyl ketone), C₂H₅O⁺ (acetaldehyde), C₅H₇⁺ (terpenes fragments), C₅H₁₁⁺ (3-methyl-1-butanol/alcohol fragment), C₆H₁₁O⁺ (hexenal/methyl-pentenone), C₄H₉⁺ (alkyl fragment), C₇H₉⁺ (terpenes fragments), C₄H₇⁺ (C4 aldehydes fragment), and C₄H₇O₂⁺ (2,3-butandione/diacetyl).

Similarly, a correlation matrix was also performed for pollen hexanoic extract samples (Figure 5B). The most important direct correlations resulted from the following sub-groups:

1° group: C₅H₁₁O⁺ (pentanal), C₅H₅O₂⁺ (pentanal), C₂H₃O⁺ (alkyl fragment (e.g., terpenes and other compounds)), C₆H₁₃O₂⁺ (hexanoic acid/methyl valerate), C₂H₅O⁺ (acetaldehyde), C₅H₉O₂⁺ (hexanal), C₆H₁₁O₂⁺ (γ-hexalactone), C₃H₅O₂⁺, C₄H₉⁺ (alkyl fragment), CH₃O₂⁺ (formic acid/formates), and C₄H₇O₂⁺ (2,3-butandione/diacetyl); and 2° group: C₅H₉O⁺ (methyl-butenal/pentenone), C₇H₁₃O⁺ (identification not assigned), C₇H₁₁O⁺ (dimethyl-2-cyclopenten-1-one), CH₃O⁺ (formaldehyde), C₄H₉O₂⁺ (ethyl acetate/methyl-propanoate), C₃H₇O₂⁺ (methyl acetate/propanoates), C₅H₇O⁺ (3-methyl furan), C₅H₁₁⁺ (3-methyl-1-butanol/alcohol fragment), and C₃H₉O₂⁺ (propylene glycol).

The other significant direct correlation was between 1° subgroup C₈H₁₁⁺ (terpenes fragments), C₁₀H₁₅O⁺ (monoterpenes oxygenate), C₅H₉⁺ (isoprene/terpenes fragments), C₈H₁₅O₂⁺ (lactone compound), C₅H₁₁O₂⁺ (butyl formate), C₈H₁₅O⁺ (1-Octen-3-one), C9H15⁺ (sesquiterpene fragments), and C₈H₁₃O⁺ ((E)-2-octenal) and 2° group C₆H₇O⁺ (phenol), C₄H₉O⁺ (2-butanone/isobutyraldehyde), C₉H₁₁⁺ (terpenes fragments), C₈H₁₇O⁺ (methyl hexyl ketone), C₄H₅⁺ (C4 fragments), C₂H₅⁺ (fragments), C₇H₁₁⁺ (terpenes fragments), C₉H₁₃⁺ (sesquiterpene fragments), and C₁₀H₁₇O⁺ (terpenoids).