Combining Electronic Tongue Array and Chemometrics for Discriminating the Specific Geographical Origins of Green Tea

The feasibility of electronic tongue and multivariate analysis was investigated for discriminating the specific geographical origins of a Chinese green tea with Protected Designation of Origin (PDO). 155 Longjing tea samples from three subareas were collected and analyzed by an electronic tongue array of 7 sensors. To remove the influence of abnormal measurements and samples, robust principal component analysis (ROBPCA) was used to detect outliers in each class. Partial least squares discriminant analysis (PLSDA) was then used to develop a classification model. The prediction sensitivity/specificity of PLSDA was 1.000/1.000, 1.000/0.967, and 0.950/1.000 for longjing from Xihu, Qiantang, and Yuezhou, respectively. Electronic tongue and chemometrics can provide a rapid and reliable tool for discriminating the specific producing areas of Longjing.


Introduction
Green tea, unfermented and made from the leaves of the Camellia sinensis plant, is one of the most popular beverages consumed across the world [1][2][3]. The property and chemical components of green teas are influenced by many factors, such as tea species, harvest season, climate, geographical locations, and processing. In China, among various factors, the geographical origin is recognized as an important aspect of tea. Because of the similar tea species, cultivation and processing conditions in a specific tea-producing area, many teas are named after their geographical origins.
Longjing tea is a green tea produced in Xihu and its surrounding areas (Hangzhou, China). As a famous green tea with Protected Designation of Origin (PDO), Longjing is recognized as one of the top green teas for its special appearance (flat and straight leaves), flavor, and taste. Various methods for distinguishing Longjing from other teas have been reported [4][5][6]. However, little information has been available on the feasibility of discriminating Longjing from its three specific subproducing areas, namely, Xihu, Qiantang, and Yuezhou. As the quality and prices of Longjing tea from the above three producing areas are different, it is necessary to develop effective analysis methods for discrimination of Longjing from different subproducing areas.
Because of the similarity (processing, appearance, and taste) among different subproducing areas, the specific geographical origins of Longjing are usually distinguished by sensory analysis. However, because it is very expensive and may take years to train a tea taster, it would be more efficient to use some nonhuman techniques. Recent years have witnessed increased applications of electronic tongue technology to analysis of wines, milk, tea, beer, juice, and so on [7][8][9][10]. In these applications, a very good time stability and sensitivity are obtained using electronic tongue sensors. Moreover, a good correlation between human and electronic tongue judgment has been observed, which makes it a promising alternative to human sensory analysis of teas.
This paper was focused on developing a rapid analysis method for discriminating specific subproducing areas of Longjing tea by electronic tongue and chemometrics. Robust principal component analysis (ROBPCA) [11,12] was used to detect outliers in each class. Partial least squares discriminant analysis (PLSDA) [13] was used to develop the classification model.

Tea Samples and Electronic Tongue Analysis. 155
Longjing samples were collected from the local tea plantations in Xihu (32 samples, class A), Qiantang (59 samples, class B), and Yuezhou (64 samples, class C). All the samples were preserved in a cool (about 4 ∘ C), dark, and dry place with integral packaging before preparation of tea extract. Six gram of each sample was added with 250 mL boiling deionized water and infused for 10 min. The infusion was then filtered into a beaker and cooled to the room temperature (25 ∘ C) by water bath for electronic tongue analysis.
The tea infusion was analyzed by an ASTREE Electronic Tongue system (Alpha M.O.S., Toulouse, France). The detection system consists of one reference electrode (Ag/AgCl) and 7 liquid cross selective sensors (ZZ, BA, BB, CA, GA, HA, and JB). The cross-sensitivity and selectivity of the sensor array are listed in Table 1. The sensors array analyzed the solutions of tea samples with sampling interval of 1 s. Each sample was measured for 150 s, which can ensure a stable response. SD and small OD), orthogonal outliers (with small SD and large OD), and bad PCA-leverage points (with large SD and large OD). With outliers deleted, the DUPLEX algorithm [14] was used to divide the measured data into a training set and test set. DUPLEX algorithm can obtain a test set of samples distributed uniformly in the range of training samples.

PLSDA.
For multiclass classification, PLSDA can be performed by regressing each column of a dummy response matrix Y on the measured data X by PLS. For the th ( = 1, 2, and 3) column in Y, an element is set a value of 1 if the corresponding object is from class ; otherwise, it is assigned a value of 0. For prediction, a new sample is classified into class when the th element of its predicted response vector is nearest to 1. Monte Carlo cross validation (MCCV) [15] was used to estimate the number of components in PLSDA by minimizing the mean percentage error of MCCV (MPEMCCV): where is the numbers of MCCV data splitting, is the number of prediction objects, and is the number of wrongly predicted for the th MCCV data splitting. Model sensitivity and specificity of prediction for each class were used to evaluate the performance of classification models: where TP, FN, TN, and FP represent the numbers of true positives, false negatives, true negatives, and false positives, respectively. For classification, objects in each class were denoted as positives and the other two classes were denoted as negatives.

Results and Discussion
The average electronic tongue features for each class of Longjing tea are shown in Figure 1. Seen from Figure 1 patterns. The features of class B and C are very similar and different from those of class A, especially in the responses of GA and HA. ROBPCA was used for outlier detection with a significance level of 0.05. Because each class of tea has a different probability distribution, ROBPCA models were performed separately on each class. Figure 2 Figure 3. The sensitivity/ specificity of PLSDA for classes A, B, and C was 1.000(10/10)/ 1.000(38/38), 1.000(18/18)/0.967(29/30), and 0.950(19/20)/ 1.000(28/28), respectively. The training accuracy was 1 and for prediction only one object from class C was wrongly assigned to class B, indicating the effectiveness of electronic tongue for classification of Longjing tea samples.

Conclusions
Rapid and effective discrimination of Longjing green tea from different subproducing areas was performed using electronic tongue and chemometrics. The sensitivity/specificity of PLSDA for classes A, B, and C was 1.000/1.000, 1.000/0.967, and 0.950/1.000, respectively. Electronic tongue and chemometrics can provide a rapid and reliable tool for discriminating the specific producing areas of Longjing. Compared with human sensory analysis, this method is easier to perform and the more attractive economically. In the future studies, the comparison of chemical methods, for example, LC/UV/MS, to the electronic tongue analysis will be performed to investigate the statistical correlation between the chemistry and the tastes of Longjing tea.