Detection of High Fructose Corn Syrup in Honey by Fourier Transform Infrared Spectroscopy and Chemometrics

DE ZAN, MERCEDES; MARCELO BELLO; GOICOECHEA H C; VERONICA FUSCA

Oslo

Congreso; XVI Scandinavian Symposium on Chemometrics; 2019

The National Service of Agri-Food Health and Quality (SENASA), controls honey to detectcontamination by synthetic or natural chemical substances and establishes and controls thetraceability of the product. The utility of near infrared spectroscopy for the detection ofadulteration of honeys with high fructose corn syrup (HFCS) was investigated. First of all,a pool of different argentinian honeys was prepared and then, several mixtures wereprepared by adding different concentrations of high fructose corn syrup (HFCS) to samplesof the honey pool. 237 samples were used, 108 of them were authentic honeys and 129samples corresponded to honeys adulterated with HFCS between 1 and 10%. They werestored ambient temperature from time of production until scanning. Immediately prior tospectral collection, honeys were incubated at 40°C overnight to dissolve any crystallinematerial, manually stirred to achieve homogeneity and adjusted to a standard solids content(80° Brix) with distilled water. Adulterant solutions used as standard were also adjusted to80° Brix. Samples were measured by Microscopy coupled Fourier Transform InfraredSpectroscopy in the range of 650 to 7000 cm-1. The technique of specular reflectance wasused, with a lens aperture range of 150 mm. A pretreatment of the spectra was performedby Standard Normal Variate (SNV). The ant colony optimization genetic algorithm sampleselection (ACOGASS) graphical interface was used, using MatLab version 5.3, to selectthe variables with the greatest discriminating power. The data set was divided into avalidation set and a calibration set, using the Kernnard-Stone (KS) algorithm. A combinedmethod of Potential Functions (PF) together with Partial Least Squares-DiscriminantAnalysis (PLS-DA) was chosen. Different estimators of the predictive capacity of themodel were compared, which were obtained using a decreasing number of groups, whichimplies more demanding validation conditions. The optimal number of latent variables wasselected as the number associated with the minimum error and the smallest number ofunassigned samples. Once the optimal number of latent variables was defined, the modelwas applied to the training samples. With the model calibrated with the training samples,the validation samples were studied. The calibrated model that combines the potentialfunction methods and PLSDA can be considered reliable and stable, since its performancein future samples is expected to be comparable to the one achieved for the training samples.By use of Potential Functions (PF) and Partial Least Square Linear Discriminant Analysis(PLS-DA) classification, authentic honey and honey adulterated with HFCS could bediscriminated with a correct classification rate of 97.9%. The results showed that NIR incombination with the PT and PLS-DS methods can be a simple, fast and low cost techniquefor the detection of HFCS in honey with high sensitivity (LoD=1%) and specificity.