Authentication of the botanical origin of monofloral honey by dielectric barrier discharge ionization high resolution mass spectrometry (DBDI-HRMS). Breaching the 6 s barrier of analysis time

Abstract

Rapid volatile profiling of monofloral honey with dielectric barrier discharge ionization high resolution mass spectrometry (DBDI-HRMS) enabled the authentication of the studied honeys’ floral sources in as little as 6 s in a solventless manner. The overarching goal of this study was the set-up of a rapid and cost effective tool for the determination of the botanical origin of monofloral honey that can be easily accessed i) by beekeepers to authenticate and, thus, enhance the value of their own honey and ii) by the food industry for quality checks. To this aim, the volatile compounds of two independent batches of honeys with seven different botanical origins (acacia, dandelion, chestnut, rhododendron, citrus, sunflower, and linden) were captured by DBD-HRMS. A total of 112 monofloral honeys were analyzed by three different operators. Using the spectral data of the first batch of honeys, we built up and compared the performances of three different classification algorithms: least absolute shrinkage and selection operator (LASSO), partial least squares discriminant analysis (PLS-DA), and random forest (RF). The performances of the three classifiers were verified by repeated cross-validation, permutation test, resubstitution into the training set, and then finally validated with a second, independent set of honeys by a different, inexperienced operator. The outcomes of the tests were expressed by the area under the curve (AUC), Kappa statistic, overall accuracy, and sensitivity and specificity rates. The misclassification rates of the built classifiers were evaluated by computing different key indicators, and the repeatability of the analytical method was also verified by cosine similarity. These insights are relevant for future adoption of the method in routine work. We determined that the RF classifier was the most powerful in predicting the floral source of the honeys. The RF classifier produced high performance values when predicting an independent batch of samples analyzed by a different, inexperienced operator (AUC 82.91 %, overall accuracy 81.25 %, Kappa 77.78 %, sensitivity 81.05 %, and specificity 96.76 %). The misclassified honeys were those characterized by the presence of other nectars and or pollens, as previously pointed to by the sensory panelists This proof-of-principle work warrants a future large-scale study to validate and challenge the method with honeys from different harvests and of different geographical origins.