Evaluation of saffron adulteration by means of thin layer chromatography-image analysis and chemometrics methods

Saffron, the dried stigmas of Crocus sativus L., is one of the most precious spices in the world, due to its delicate aroma and special color. Because of its limited production owing to the laborious process required, saffron become an expensive spice and therefore is considered as one of the major candidates for economically motivated fraud [1-2]. In the present work, thin layer chromatography-image analysis (TLC-IA) combined with chemometrics has been employed for detection of saffron adulteration and identification of the adulterant. Seven saffron natural and artificial adulteration including sumac, turmeric, safflower, common madder, quinoline yellow, sunset yellow and tartrazine were used to simulate artificial counterfeit mixtures of saffron. Overall, 6 mixtures containing saffron and 5-30% (w/w) of plant adulterants were prepared and analyzed for each adulterant and thus eight classes, including the authentic saffron samples, were defined. The saffron metabolites were extracted with a mixture of ethanol/water and analyzed by TLC. The images of the TLC plates were recorded under visible light and subsequently imported into the MATLAB software. The images were compressed, inverted, baseline corrected using asymmetric least squares (AsLS), aligned by correlation optimization warping (COW) and finally converted to RGB chromatograms. To eliminate the artificial sources of variations, the chromatograms were normalized toward total peak area. The data were mean centered and Pareto-scaled prior to data analysis [3]. Eventually, the preprocessed data were randomly divided into training and test sets for further classification analysis. Different pattern recognition techniques including principal component analysis (PCA) and K-means were used to explore the general structure of the obtained data. Moreover, supervised pattern recognition was carried out through partial least squares discriminant analysis (PLS-DA). The model parameters for fitting, cross-validation and external validation including sensitivity, specificity and accuracy for all the groups were equal to 1.00. In order to obtain a better understanding of the model, two different variable selection methods including variable importance projection (VIP) and PLS loading weights were used. The results indicated that even after reducing a large number of the variables, the model performance was acceptable. Finally, the model was successfully utilized to predict the potential adulterants in real samples.