A Research On Techniques For Discriminating The Geographical Origin Of Wuyi Rock-essence Tea

Wuyi rock-essence tea, a famous Oolong tea which has been awarded with a protected geographical indication, is original cultivated in Mount Wuyi. Wuyi rock-essence tea is most favored by consumers for its pleasurable savor and functional effects. However, the actual yield of Wuyi rock-essence tea is limited and the authentic Wuyi rock-essence tea is expensive, a lot of Wuyi rock-labeled teas are adulterated with normal teas which cultivated outside the protected production area. Therefore, it is urgent to employ quality control of Wuyi rock-essence tea against various counterfeits. Compared with some western agricultural products such as olive oil, wine, cheese and honey, the provenance detecting research on tea is still in the primary stage, so many of the analytical techniques and classification models about provenance discrimination should be tested in tea products.In this study, authentic Wuyi rock-essence tea samples were collected from 33 different producing areas in Mont Wuyi, and some counterfeit rock-essence tea were collected from 11 different production sites such as Guizhou, Guangxi, Jiangxi and so forth. First of all, inter-simple sequence repeat(ISSR) molecular identification was investigated to distinguish the cultivars of Oolong teas. After that, analytical instruments like near infrared spectrum(NIR), isotope ratio mass spectrometer(IRMS), inductively coupled plasma mass spectrometry(ICP-MS), atomic absorption spectroscopy(AAS), high performance liquid chromatography(HPLC) and electronic tongue were performed in sample analysis, then principal component analysis(PCA), partial least square(PLS), artificial neural network(ANN) and support vector machine(SVM) were applied to model the complicated relationship between measured data and production site. Then, each feature of isotope, mineral element, catechin and amide data was ranked by its contribution to the SVM provenance prediction accuracy. In addition, the comparison of models based on different instruments and chemometrics methods was investigated. The results of this study are mainly as follows:6 primers with high polymorphism were picked out from 27 original ISSR primers to distinguish 8 different Oolong tea cultivars. As a result, a total of 52 bands were amplified using PCR. In each cultivar, the number of polymorphism bands is between 12 and 22(23%-42%), and all of the 8 different Oolong cultivars were successfully distinguished by ISSR fingerprinting.The result demonstrates NIR combined with PLS which provides a potential tool for provenance discrimination, for the best prediction accuracy of NIR-PLS model reached 0.9104. Then, 4 stable isotopes, 14 mineral elements, caffeine, 6 catechins and 27 amides of the samples were detected as the geographical origin indications, and each feature of isotope, mineral element, catechin and amide data was ranked by its contribution to the SVM provenance prediction accuracy. For isotopes, δ2H is ranked at the top, δ18O, δ15N and δ13C are then ranked in descending order. For mineral elements, Cs, Cu, Ca, Rb and Sr contents are ranked in the leading five. For catechins and caffeine, arranged in descending order, they are epigallocatechin, catechin, epigallocatechin gallate, gallic acid, epicatechin, epicatechin gallate and caffeine in turn. For amides, agedoite, proline, tryptophan, phosphorylethanolamine and carbamide are the top five indications in SVM model.Moreover, the usability of PCA was discussed in the work. PCA is most frequently used method in sample classification for geographical origins. However, for large number datas, the results of PCA classification become inaccurate and can hardly satisfy the demand of provenance detecting, so in this study, PCA was abandoned in further analysis.To compare the prediction accuracies of instruments and chemometrics methods, PLS, ANN and SVM classification models based on NIR, isotope, mineral element, catechin, amide and electronic tongue data were respectively established. For instruments, the models based on NIR data get the best performance. And for classification models, nonlinear models(SVM and ANN) achieve higher prediction accuracies than linear model(PLS). Therefore, we infer that nonlinear relationship may exist between these measured features and tea provenance.