Profiling Turkish honeys to determine authenticity using physical and chemical characteristics

Seventy authentic honey samples of 9 different floral types (rhododendron, chestnut, honeydew, Anzer (thymus spp.), eucalyptus, gossypium, citrus, sunflower, and multifloral) from 15 different geographical regions of Turkey were analyzed for their chemical composition and for indicators of botanical and geographical origin. The profiles of free amino acids, oligosaccharides, and volatile components together with water activity were determined to characterize chemical composition. The microscopic analysis of honey sediment (mellissopalynology) was carried out to identify and count the pollen to provide qualitative indicators to confirm botanical origin. Statistical analysis was undertaken using a bespoke toolbox for Matlab called Metabolab. Discriminant analysis was undertaken using partial least-squares (PLS) regression followed by linear discriminant analysis (LDA). Four data models were constructed and validated. Model 1 used 51 variables to predict the floral origin of the honey samples. This model was also used to identify the top 5 variable important of projection (VIP) scores, selecting those variables that most significantly affected the PLS-LDA calculation. These data related to the phthalic acid, 2-methylheptanoic acid, raffinose, maltose, and sucrose. Data from these compounds were remodeled using PLS-LDA. Model 2 used only the volatiles data, model 3 the sugars data, and model 4 the amino acids data. The combined data set allowed the floral origin of Turkish honey to be accurately predicted and thus provides a useful tool for authentication purposes. However, using variable selection techniques a smaller subset of analytes have been identified that have the capability of classifying Turkish honey according to floral type with a similar level of accuracy.