Food Chemistry 298 (2019) 125046

Contents lists available at ScienceDirect

Food Chemistry

journal homepage: www.elsevier.com/locate/foodchem

Highly discriminant rate of black grades based on fluorescent T probes combined with chemometric methods ⁎ Jing Zhu1, Fengyuan Zhu1, Luqing Li, Linlin Cheng, Liang Zhang, Yue Sun, Xiaochun Wan , ⁎ Zhengzhu Zhang

State Key Laboratory of Tea Plant Biology and Utilization, and Department of Applied Chemistry, Anhui Agricultural University, Hefei 230036, Anhui,

ARTICLE INFO ABSTRACT

Chemical compounds studied in this article: We established a novel Dianhong grades discriminant analytic technique based on a fluorescence image Phenol (PubChem CID: 996) along with carbon quantum dots (CDs) as fluorescent probes. Different grades of Dianhong black tea contain (PubChem CID: 1203) different various amounts of tea polyphenols. Tea polyphenols can quench the fluorescent intensity ofCDs, (PubChem CID: 439378) resulting in different fluorescent peaks; Dianhong black tea grades can then be discriminated through theuseof (PubChem CID: principal component analysis and Bayesian analysis. Compared with the additional data processing required in 65064) other methods, the advantage of our method is that the fluorescence curve can be used directly, and it achieves (PubChem CID: 11980943) Ethanol (PubChem CID: 702) satisfactory results. We firstly used CDs combined with chemometrics to identify eight grades of Dianhong black Cobalt Nitrate Hexahydrate (PubChem CID: tea, and we also provide a new method that improves the identification rate using nanotechnology to avoid 24821) performing complex data processing. Sodium Citrate (PubChem CID: 6224) Carbamide (PubChem CID: 1176) Thiourea (PubChem CID: 2723790)

Keywords: Fluorescence spectroscopic analysis Grades discriminant Carbon quantum dots Principal component analysis Bayesian discriminant analysis

1. Introduction discriminate the grades on the basis of their own experience. It is ne- cessary to establish an objective analytic procedure for evaluating and Dianhong black tea is a representative and influential black tea discriminating the grades of Dianhong black tea (Pang et al., 2012). brand distributed throughout the world and it is prepared from Camellia Electronic tongue is a new detection method for analysing and re- sinensis grown in , where the plant’s bud leaves are soft and cognising the taste of liquid (Huo et al., 2014; Qin et al., 2013). Spectral polyphenol content is high. Because of the abundance of polyphenols in technology combined with chemometrics methods to assess and analyse Dianhong black tea, it possesses antibacterial property (Beresniak, tea grades is a promising area for investigation (He, Li, & Deng, 2007; Duru, Berger, & Bremond-Gignac, 2012) and can lower blood sugar and Hu et al., 2018; Li, Xu, Zhang, Sun, & He, 2017; Luypaert, Zhang, & blood pressure (Ramadan, El-Beih, & El-Ghffar, 2009). Because of Massart, 2003; Ojha & Roy, 2018; Wang, Zheng, Liu, & Fang, 2016). various processing techniques are employed for different grades of tea, Carbon quantum dots (CDs) as a preeminent fluorescent nanoma- obvious differences are perceptible between grades of tea on the basis terial with low cytotoxicity and excellent biocompatibility, are well of their qualities (Wang et al., 2017; Xiao et al., 2017). The price of suited for use in chemical analyses of appointed targets (Xu, Liu, Gao, & Dianhong black tea varies on the basis of its quality grade, and can Wang, 2014; Zong et al., 2014). Their unique characteristics improve range from a few hundred to several thousand renminbi (RMB) per the safety of fluorescence detection. CDs can be synthesized in alab, kilogram. Customers typically cannot evaluate the quality and which makes the method more convenient and cheaper (Zheng, Than,

⁎ Corresponding authors. E-mail addresses: [email protected] (X. Wan), [email protected] (Z. Zhang). 1 These authors contributed equally. https://doi.org/10.1016/j.foodchem.2019.125046 Received 7 November 2018; Received in revised form 18 February 2019; Accepted 18 February 2019 Available online 20 June 2019 0308-8146/ Published by Elsevier Ltd. J. Zhu, et al. Food Chemistry 298 (2019) 125046

Fig. 1. Co-CDs system for sensing Dianhong black tea grades.

Ananthanaraya, Kim, & Chen, 2013). Many studies have expended 2.2. Extraction of tea polyphenols considerable effort to use CDs in applications such as fluorescence sensing, bioimaging, and drug delivery. CDs have abundant and low- We used solvent extraction, which exhibits the characteristics of cost carbon precursors (e.g. ground (Hsu & Chang, 2012), used stability, reliability, and simple operation, to extract tea. We used tea (Hsu, Chen, Ou, Chang, & Chang, 2013), candle soot (Liu, Ye, & ethanol 70% or water as solvents. We followed the typical rank order of Mao, 2007), grass (Liu et al., 2012), thus diversifying carbon prepara- Dianhong black tea (Super, First, Second, Third, Fourth, Fifth, Sixth, tion. and Outside), and we collected 15 tea samples (0.2 g per sample) of In this work, our objective was to design a discriminant model to each tea grade, yielding a total of 120 samples. Each sample was placed identify Dianhong black tea grades by using Co-modified CDs combined in a 10-mL centrifuge tube and labelled with a corresponding number in with chemometrics methods (Fig. 1). No study has performed an ex- the range of D1–D120 (Super: D1–D15, First: D16–D30, Second: periment to identify eight grades of a type of tea by using Co-modified D31–D45, Third: D46–D60, Fourth: D61–D75, Fifth: D76–D90, Sixth: CDs as fluorescent sensors because of higher number of grades signifies, D91–D105, and Outside: D106–D120). Subsequently, 8 mL of 70% smaller differences, and thus greater difficulty in tea identification. We ethanol solution was added to each sample, and the sample was then also used PCA and Bayes discriminant analysis to analyse the data ac- placed in an ultrasonic cleaning instrument at 70 °C. Ultrasonic ex- quired from the application of the CDs fluorescence sensor to Dianhong traction was performed for 45 min, which entailed shaking the cen- black tea. PCA can identify the grade information of the sample, and trifuge tube at intervals of 15 min to prevent the accumulation of tea Bayes discriminant analysis can predict each grade of Dianhong black powder at the bottom of the tube, which would affect the extraction tea accurately. effectiveness. The extract was then centrifuged at a speed of2191g- force for 10 min, and the upper liquid was filtered using a 0.22-μm filter membrane. The filtrate was collected into a corresponding E1–E120 5- 2. Methods mL centrifuge tube and stored at 4 °C without light. The procedure for extracting tea polyphenols with water followed 2.1. Materials the same method: 0.2 g of each sample was placed in a 10-mL centrifuge tube labelled W1–W120. Next, 8 mL of ultrapure water was added to All chemicals and reagents used in this work were of analytical each sample, and the sample was then placed in a constant-temperature grade. Sodium citrate (C6H5NaO7·2H2O), cobalt nitrate hexahydrate water bath of 70 °C to extract for 45 min. The extraction process also (Co(NO3)2·6H2O), iron chloride hexahydrate (FeCl3·6H2O), thiourea entailed shaking the centrifuge tube at intervals of 15 min. After ex- (CS(NH2)2), carbamide (CN2H4O), and D-glucose anhydrous (C6H12O6) traction, centrifugation and filtration were performed as in the previous were purchased from Sinopharm Chemical Reagent Co. Ltd (Shanghai, operation. The filtrate was collected into the corresponding W1–W120 China). The resistivity of the test water was greater than 18 MΩ·cm−1. 5-mL centrifuge tubes and stored at 4 °C without light.

Chitosan (C6H11NO4)n was procured from Beijing Solarbio Science & Technology Co. Ltd (Beijing, China). Acetic acid glacial (C2H4O2) was 2.3. Synthesis of fluorescent CDs purchased from Tedia Company (Ohio, America). Fluorescence measurements were performed using a Carry Eclipse We used a one-pot hydrothermal synthetic approach to obtain CDs. Fluorescence Spectrophotometer (Aglient Technologies). Dianhong Carbon sources for synthesizing are extensive, and we selected sodium black tea samples were collected from Yunnan Dianhong Black Tea citrate and chitosan as carbon sources in this experiment. Briefly, Group Co. Ltd. To reduce heterogeneity, the black tea samples were 0.28 mmol sodium citrate was added into 30 mL of ultrapure water and grinded ground into powder by a soymilk machine, sieved through a mixed with 1.26 mmol carbamide and 0.42 mmol thiourea. After com- 40–80-mesh sieve, and stored in a refrigerator at 4 °C. plete dissolution, the mixture was transferred into a 50-mL PTFE

2 J. Zhu, et al. Food Chemistry 298 (2019) 125046

Fig. 2. (A). Fluorescence spectra of CDs system reacted with ethanol-extracted samples; (B). Content of tea polyphenols in eight grades of Dianhong black tea samples.

reactor and placed in an oven at 185 °C. After 6 h of reaction, the re- variables from raw data (Bilge, Velioglu, Sezer, Eseller, & Boyaci, 2016; actor was cooled to room temperature. The CDs solution, a brownish Bro & Smilde, 2014; Gurdeniz & Ozen, 2009). PCA focuses on the vector yellow liquid, was purified using filtration through a 0.22-μm filter space for a superior description of the raw data, whereas LDA for the membrane and stored at 4 °C for further use. In another synthetic op- vector space that distinguishes the data most effectively, taking full eration, 50 mmol chitosan was fully dissolved in 20 mL of 2.5% acetic advantage of the class information of the training samples (Li, Xie, acid solution. Subsequently, the solution was transferred to a 50-mL Ning, Chen, & Zhang, 2018). We used PCA and LDA to analyse the reaction kettle and heated in an oven at 200 °C. The reactor was then fluorescence spectrum, and determine whether the fluorescence spec- cooled to room temperature after 4 h of reaction. The nitrogen-carbon- trum data can fully represent the tea information. In recent years, SVM doped CDs (NCDs) were obtained and purified using a 0.22-μm filter has been applied in pattern classification and nonlinear regression membrane, and stored in a refrigerator at 4 °C. problems because of its nonlinear and high-dimensional pattern-re- cognition abilities (Granato et al., 2018). We used SVM to verify the 2.4. Reaction of CDs with tea extraction solution analysis results. A discriminant model for Dianhong black tea was es- tablished using Bayesian discriminant analysis. Bayesian discriminant The tea solution extracted by the ethanol solvent reacted with CDs. analysis can obtain posterior probability from prior probability on the For tea samples E1-E120, we prepared a new set of 5-mL centrifuge basis of on new information (AlHichri et al., 2013; Chen & Wang, 2010; tubes: E-CDs-1 to E-CDs-120. Then, 2.85 mL of ultrapure water was Dong, Zhang, Zhang, & Wang, 2012). In this study, we used SPSS sta- added into the 5- mL centrifuge tubes, after which 0.1 mL of CD solu- tistics 22.0 software for statistical analysis. tion and 0.05-mL tea samples extracted by ethanol were added. The tubes were centrifuged at a speed of 2380g-force for 5 min. The fluor- 3. Results and discussions escence curve of the upper solution was measured using the fluores- cence spectrophotometer, and the average value of three measurements Tea polyphenols constitute a key chemical substance in tea, ac- of each sample was reserved for analysis. All optical measurements counting for a large proportion of the extracts. Hence, we selected the were performed at room temperature under natural conditions. method of biased tea polyphenol extraction. During extraction, ethanol For tea samples E1–E120, we prepared another new set of 5-mL can inhibit the oxidation of tea polyphenols, especially catechins. We centrifuge tubes: E-CoCDs-1 to E-CoCDs-120. As in the previous op- selected 70% ethanol as extraction solvent. Many experiments on ul- eration, 2.75 mL of ultrapure water, 0.1 mL of CDs solution, 0.1 mL of trasonic extraction of tea polyphenols have demonstrated that when the 10−3 mol/L Cobalt (Co) solution, and 0.05-mL tea samples extracted by solid-liquid ratio was in the range of 1:30–1:50, the extraction rate of ethanol were added into the 5-mL centrifuge tubes. After centrifuga- tea polyphenols was higher; thus, the soli-liquid ratio of this experiment tion, the fluorescence curves of the upper layer liquid were measured. was 0.2 g: 8 mL. We selected 45 min for the extraction time. For tea samples E1–E120, we prepared another new set of 5-mL cen- Differences exist in quality among the grades of Dianhong black tea, trifuge tubes: E-FeNCDs-1 to E-FeNCDs-120. Per the aforementioned and the content of tea polyphenols must thus vary as well. The CDs can steps, 2.75 mL of ultrapure water, 0.1 mL of NCD solution, 0.1 mL of be quenched by tea polyphenols, which causes subtle differences in the 10−3 mol/L Iron (Fe) solution, and 0.05-mL tea samples extracted by positions and intensities of peaks in the CDs system (Fig. 2). We only us ethanol were added into the 5-mL centrifuge tubes. The fluorescence CDs as probe (Fig. 2A), the Fifth grade, Sixth grade, and Outside grade spectra were recorded after centrifugation. For tea samples W1–W120, tea samples exhibit high fluorescence intensity peaks and their spectral we prepared sets of 5-mL centrifuge tubes: W-CoCDs-1 to W-CoCDs-120 positions are at the top; Third grade and Fourth grade samples are and W-FeNCDs-1 to W-FeNCDs-120. The routine procedure was fol- clustered together in the middle position; First grade and Second grade lowed, and fluorescence spectra information was recorded. samples are located at the bottom. However, the interlacing phenom- enon is so severe that the human eye can hardly determine the rules to 2.5. Statistical analysis judge the grade. The content of tea polyphenols exhibits a downward trend as the The human eye cannot ascertain the characteristic information of Dianhong black tea grade decreases (Fig. 2B). From Fig. 2B, the higher fluorescence spectrum data collected by a fluorescence photometer, fluorescence peaks are the results of the reaction of CDs withlower Multi resolution signal analysis and correction in chemometrics is a polyphenol content . By contrast, the positions of fluorescence very active field. Chemometrics offers many methods for multivariate peaks are lower after the reaction of CDs and teas with higher poly- resolution and calibration; one commonly used method is PCA. PCA, a phenol content. The reason for this phenomenon is that tea polyphenols linear projection statistical methods for handling multivariate data, can inhibit the fluorescence phenomenon of CDs. In order to improve the sufficiently compress raw data and extract high-quality feature accuracy of discrimination, the image data will be optimized, but it has

3 J. Zhu, et al. Food Chemistry 298 (2019) 125046

Fig. 3. Fluorescence spectra of (A). Co-CDs system, (B). Fe-NCDs system reacted with ethanol-extracted samples, and (C). Co-CDs system, (D). Fe-NCDs system reacted with water-extracted samples.

certain requirements for the ability of using software (Hu et al., 2018; a k-dimensional space (k < n); the resulting k-dimension is a new or- Liu et al., 2017; Orzel et al., 2014; Tzouros & Arvanitoyannis, 2001). thogonal feature, and the k-dimensional feature is called the principal Our aim is to improve the response to make the original reaction data component (PC) (Iliopoulos, Pitsianis, Sun, Yin, & Ren, 2016; Zhang & better, avoiding the operation of image optimization processing. Wang, 2011). PCi represents the ith principal component. The propor- We introduced Co2+ to form Co-CDs. Co2+ can combine with tion of PCi variance to the total variance is called the contribution rate. polyphenols to form a complex; the precipitation was very weak when A higher value of the contribution rate indicates a stronger ability to only tea polyphenols and Co ions were in the system. When tea poly- synthesise the information of the PCi. Bayesian discriminant analysis is phenols, and Co-CDs all coexist, a large amount of reddish brown a statistical method used in multivariate statistical analysis to identify a precipitate is produced. Therefore, CDs play a key role in the formation type of sample. It establishes the function on the basis of properties of of reddish brown matter. Fig. 3A reveals that the Fifth grade tea sam- the known category, and then makes a decision concerning new items ples have high fluorescence intensity peaks, the spectral map ofthe in an unknown category in order to subsume the new items into the Third grade samples is located below Fifth, the spectral maps of the known class. The principle of ‘leave-one-out’ is used in the cross-vali- Second grade and Fourth grade samples are located at the bottom. dation; if the original data have N samples, then during the validation However, some spectra of tea samples cannot be recognised by the of one sample, the remaining the N-1 samples are used as the training human eye: Sixth grade and Outside grade tea spectra are clustered set (Lamnisos, Griffin, & Steel, 2012; Vehtari, Gelman, & Gabry, 2017). together. This approach has two clear advantages: First, almost all the samples in We also used Fe3+ to conduct the experiments. As shown in Fig. 3B each round are used for the training model; thus, so the training model the range of peaks is very broad, and the degree of severity of spectral can closely match the distribution of the original sample, and the interlacing is greater than Fig. 3A that human eye is unable to discern evaluation results are more reliable. Second, random factors that the difference between the both samples. emerge in the course of the experiment do not affect the experimental For the extraction methods, we used constant-temperature extrac- data, which ensures that the experimental process can be duplicated. tion as a comparison. The extraction yield of tea polyphenols increases We used SPSS software perform the PCA and Bayesian discriminant at first but then decreases with the increase in extraction temperature. analysis of spectral information, and the results are presented in However, when the temperature is higher than 70 °C, the high tem- Table 1. The cumulative contribution rate of PCA is the sum of the first perature promotes the oxidation and decomposition of tea polyphenols, three PC contribution rates, and all samples reached 99.9%. The pre- which reduces the extraction yield; thus, the maximum extraction sented results indicate that the first three PCs can fully reflect the temperature is 70 °C. Tea polyphenol is a type of polyhydroxyl sub- sample information. However, major differences appear in the dis- stance that easily oxidises. We selected 45-min as extraction time. We criminant analysis results. When only ethanol-extracted tea samples recorded the fluorescence spectra after reaction with CDs(Fig. 3C and and CDs are extracted in the detection system, the classification accu- D). Fig. 3C illustrates that the spectral distributions of different levels of racy and cross-validation accuracy of the Bayesian discriminant ana- tea occupy different wavelength regions, but considerable overlap is lysis are 82.5% and 73.3%, respectively. This classification accuracy is nevertheless evident. barely acceptable, and the cross-validation accuracy is unacceptable. PCA is use linear transformation to map an n-dimensional feature to However, the accuracy of the Bayesian discriminant analysis is

4 J. Zhu, et al. Food Chemistry 298 (2019) 125046

Table 1 Discrimination of fluorescence spectral images using Bayesian discriminant analysis.

Extraction Detection Bayesian discriminant solvent system Classification Cross Validation accuracy % accuracy %

Ethanol CDs 82.50 73.30 Ethanol Co-CDs 98.30 95.00 Ethanol Fe-CDs 61.70 55.80 Water Co-CDs 70.00 62.50 Water Fe-CDs 65.00 60.80 improved by introducing Co ion into the reaction system. The classifi- cation accuracy and cross-validation accuracy, 98.30% and 95.00%, are shown to be very satisfactory and to meet the requirements for estab- lishing a discriminant model. The effect of introducing Fe ion into the system is very poor. When the extraction solvent is water, the amount of the substance dissolved in water may be excessively high and con- siderably interfere with the system, thus resulting in less satisfactory discriminant results compared with results obtained using ethanol ex- traction. In addition, tea polyphenols may be oxidised at a constant temperature, but ethanol inhibits the oxidation of polyphenols. Many factors cause the extraction effect of water to be less effective than ethanol. The contents of theaflavin and amino acids should also be taken into account in the evaluation of black tea grade. Theanine and epigallo- gallate (EGCG) are the highest contents of amino acids and tea polyphenols in tea. It can be seen from the Fig. 4 that theaflavin has very little response to the Co-CDs system, and theanine has little effect on Co-CDs system. The response of EGCG to Co-CDs system is very Fig. 5. (A). Fluorescence spectra of Co-CDs system reacted with ethanol-ex- obvious, which shows that the constructed system does have a certain tracted samples; (B). Spatial distribution of Co-CDs system reacted with selectivity to tea polyphenols. ethanol-extracted samples based on PCA. We used the fluorescence spectral information from the Co-CDs

Fig. 4. Fluorescence spectra of Co-CDs system reacted with (A). theaflavin, (B). theanine, and (C). EGCG.

5 J. Zhu, et al. Food Chemistry 298 (2019) 125046

Table 2 Acknowledgements Classification results of Co-CDs system reacted with ethanol-extracted samples in LDA, and SVM. This study was financially supported by the National Key Research Models Parameters Calibration Accuracy Prediction Accuracy and Development Program of China (2017YFD0400800) and the (%) (%) Program for Changjiang Scholars and Innovative Research Team in University (No. IRT_15R01). Natural Science Foundation of China LDA PCs = 5 100 80 (51702004, 31701600), Anhui Provincial Natural Science Foundation SVM c = 197.012, 90 87.5 g = 0.435275 (1808085QE158, 1808085QC77), Introduction and Stabilization of Talent Projects of Anhui Agricultural (yj2017-06, yj2017-04).

Declaration of Competing Interest system reacted with ethanol-extracted samples (Fig. 5A) as the basis of the discriminant model. SPSS software was used to analyse the fluor- The authors have no conflicts of interest to declare. escence spectra of the samples. The cumulative contribution rate of the first three PCi (PC1, PC2, and PC3) is 99.97%; this rate can represent Mendeley data the aforementioned spectral information and be used for subsequent data analysis. Fig. 5B displays the spatial distribution of the Dianhong Original fluorescence spectral data: https://data.mendeley.com/ black tea grades based on PCA, where the first three PC explain 99.97% datasets/wf7frjr3zk/draft?a= beac9670-c90f-4ba4-b993- of all components. Clearly, each grade of tea sample has its own cluster 01df15bb28c0. area. We verified the data of this sample by LDA and SVM, the results are presented in Table 2 and also relatively acceptable. Appendix A. Supplementary data We used the aforementioned spectral information to perform a Bayesian discriminant analysis and construct discriminant functions. Supplementary data to this article can be found online at https:// The discriminant functions are expressed as follows: doi.org/10.1016/j.foodchem.2019.125046. Super Y1 = −1824.903 + 456.846X1 − 328.685X2 + 6.040X3 − 74.539X4 − 58.226X5 + 282.021X6 + 665.671X7 References First Y2 = −1626.083 + 464.204X1 − 319.689X2 − 17.431X3 − 68.944X4 − 40.354X5 + 254.927X6 + 603.998X7 AlHichri, H., Bazi, Y., Alajlan, N., Melgani, F., Malek, S., & Yager, R. R. (2013). A novel Second fusion approach based on induced ordered weighted averaging operators for che- mometric data analysis. Journal of Chemometrics, 27(12), 447–456. Y3 = −1508.008 + 435.134X1 − 278.207X2 − 30.396X3 − 66.146X4 Beresniak, A., Duru, G., Berger, G., & Bremond-Gignac, D. (2012). Relationships between −42.075X5 + 259.010X6 + 603.119X7 black tea consumption and key health indicators in the world: An ecological study. Third BMJ Open, 2(6), 10. Bilge, G., Velioglu, H. M., Sezer, B., Eseller, K. E., & Boyaci, I. H. (2016). Identification of Y4 = −1818.324 + 496.710X1 − 337.825X2 − 18.356X3 − 79.489X4 meat species by using laser-induced breakdown spectroscopy. Meat Science, 119, − 33.808X5 + 263.090X6 + 639.320X7 118–122. Fourth Bro, R., & Smilde, A. K. (2014). Principal component analysis. Analytical Methods, 6(9), Y = −1729.896 + 475.929X − 314.492X − 20.121X − 76.056X 2812–2831. 5 1 2 3 4 Chen, T., & Wang, B. (2010). Bayesian variable selection for Gaussian process regression: −40.880X5 + 261.312X6 + 665.300X7 Application to chemometric calibration of spectrometers. Neurocomputing, 73(13–15), Fifth 2718–2726. Y = −2129.954 + 541.882X − 388.712X − 35.500X − 73.033X Dong, W., Zhang, Y. Q., Zhang, B., & Wang, X. P. (2012). Quantitative analysis of adul- 6 1 2 3 4 teration of extra virgin olive oil using Raman spectroscopy improved by Bayesian −26.187X5 + 254.666X6 + 656.509X7 framework least squares support vector machines. Analytical Methods, 4(9), Sixth 2772–2777. Granato, D., Putnik, P., Kovačević, D. B., Santos, J. S., Calado, V., Rocha, R. S., ... Y7 = −1958.496 + 489.223X1 − 334.444X2 − 24.145X3 − 78.133X4 Pomerantsev, A. (2018). Trends in chemometrics: Food authentication, microbiology, − 33.087X5 + 268.274X6 + 643.912X7 and effects of processing. Comprehensive Reviews in Food Science and Food Safety, Outside 17(3), 663–677. Gurdeniz, G., & Ozen, B. (2009). Detection of adulteration of extra-virgin olive oil by Y8 = −1743.528 + 490.493X1 − 334.217X2 − 55.411X3 − 57.301X4 chemometric analysis of mid-infrared spectral data. Food Chemistry, 116(2), 519–525. −21.377X5 + 225.965X6 + 580.644X7 He, Y., Li, X. L., & Deng, X. F. (2007). Discrimination of varieties of tea using near infrared X1,X2,X3,X4,X5,X6, and X7 represent the fluorescence intensity spectroscopy by principal component analysis and BP model. Journal of Food values at 396, 398, 410, 426, 478, 518, and 554 cm−1, respectively. Engineering, 79(4), 1238–1242. Hsu, P. C., & Chang, H. T. (2012). Synthesis of high-quality carbon nanodots from hy- When we obtain a new sample and, insert the corresponding value into drophilic compounds: Role of functional groups. Chemical Communications, 48(33), the aforementioned functions, whichever function yields the largest 3984–3986. value can determine which type of sample the new sample it belongs to. Hsu, P. C., Chen, P. C., Ou, C. M., Chang, H. Y., & Chang, H. T. (2013). Extremely high inhibition activity of photoluminescent carbon nanodots toward cancer cells. Journal of Materials Chemistry B, 1(13), 1774–1781. Hu, Y., Wu, H. L., Yin, X. L., Gu, H. W., Liu, Z., Xiao, R., ... Yu, R. Q. (2018a). A flexible 4. Conclusions and novel strategy of alternating trilinear decomposition method coupled with two- dimensional linear discriminant analysis for three-way chemical data analysis: Characterization and classification. Analytica Chimica Acta, 1021, 28–40. We successfully developed a discriminant model of Dianhong black Hu, O., Xu, L., Fu, H. Y., Yang, T. M., Fan, Y., Lan, W., ... She, Y. B. (2018b). “Turn-off” tea grades based on Co-CDs coupled with chemometric methods. The fluorescent sensor based on double quantum dots coupled with chemometrics for detection system is shown to demonstrate excellent capacity for sensing highly sensitive and specific recognition of 53 famous green teas. Analytica Chimica Acta, 1008, 103–110. differences in the fluorescence responses of different Dianhong black Huo, D. Q., Wu, Y., Yang, M., Fa, H. B., Luo, X. G., & Hou, C. J. (2014). Discrimination of tea grades. In particular, the established fluorescent sensor achieves a Chinese according to varieties and grade levels using artificial nose and high degree of classification accuracy for different Dianhong black tea tongue based on colorimetric sensor arrays. Food Chemistry, 145, 639–645. Iliopoulos, A. S., Pitsianis, N., Sun, X., Yin, F. F., & Ren, L. (2016). An extension of PCA- grades. The operating procedure for this method is convenient, and no based respiratory deformation modeling via multi-linear decomposition. Medical additional processing of spectral data is required. The results reveal that Physics, 43(6), 3439. classification accuracy and cross- validation accuracy in the modelare Lamnisos, D., Griffin, J. E., & Steel, M. F. J. (2012). Cross-validation prior choicein Bayesian probit regression with many covariates. Statistics and Computing, 22(2), 98.3% and 95%, respectively. 359–373.

6 J. Zhu, et al. Food Chemistry 298 (2019) 125046

Li, L., Xie, S., Ning, J., Chen, Q., & Zhang, Z. (2018). Evaluating green tea quality based on Research International, 53(2), 864–874. multisensor data fusion combining hyperspectral imaging and olfactory visualization Ramadan, G., El-Beih, N. M., & El-Ghffar, E. A. A. (2009). Modulatory effects of blackv. systems. Journal of the Science of Food and Agriculture. green tea aqueous extract on hyperglycaemia, hyperlipidaemia and liver dysfunction Li, X. L., Xu, K. W., Zhang, Y. Y., Sun, C. J., & He, Y. (2017). Optical determination of lead in diabetic and obese rat models. British Journal of Nutrition, 102(11), 1611–1619. chrome green in green tea by fourier transform infrared (FT-IR) transmission spec- Tzouros, N. E., & Arvanitoyannis, I. S. (2001). Agricultural produces: Synopsis of em- troscopy. PLoS One, 12(1), 14. ployed quality control methods for the authentication of foods and application of Liu, L., Fan, Y., Fu, H. Y., Chen, F., Ni, C., Wang, J. X., ... She, Y. B. (2017). Turn-off“ chemometrics for the classification of foods according to their variety or geographical fluorescent sensor for highly sensitive and specific simultaneous recognition of29 origin. Critical Reviews in Food Science and Nutrition, 41(4), 287–319. famous green teas based on quantum dots combined with chemometrics. Analytica Vehtari, A., Gelman, A., & Gabry, J. (2017). Practical Bayesian model evaluation using Chimica Acta, 963, 119–128. leave-one-out cross-validation and WAIC. Statistics and Computing, 27(5), 1433. Liu, S., Tian, J. Q., Wang, L., Zhang, Y. W., Qin, X. Y., Luo, Y. L., ... Sun, X. P. (2012). Wang, C., Zhang, C. X., Kong, Y. W., Peng, X. P., Li, C. W., Liu, S. H., ... Xu, Y. Q. (2017). A Hydrothermal treatment of grass: a low-cost, green route to nitrogen-doped, carbon- comparative study of volatile components in Dianhong teas from fresh leaves of four rich, photoluminescent polymer nanodots as an effective fluorescent sensing platform tea cultivars by using chromatography-mass spectrometry, multivariate data analysis, for label-free detection of Cu(II) ions. Advanced Materials, 24(15), 2037–2041. and descriptive sensory analysis. Food Research International, 100, 267–275. Liu, H. P., Ye, T., & Mao, C. D. (2007). Fluorescent carbon nanoparticles derived from Wang, J. M., Zheng, P. C., Liu, H. D., & Fang, L. (2016). Classification of candle soot. Angewandte Chemie-International Edition, 46(34), 6473–6475. leaves using laser-induced breakdown spectroscopy combined with the discriminant Luypaert, J., Zhang, M. H., & Massart, D. L. (2003). Feasibility study for the use of near analysis method. Analytical Methods, 8(15), 3204–3209. infrared spectroscopy in the qualitative and quantitative analysis of green tea, Xiao, Z. B., Wang, H. L., Niu, Y. W., Liu, Q., Zhu, J. C., Chen, H. X., & Ma, N. (2017). (L.). Analytica Chimica Acta, 478(2), 303–312. Characterization of aroma compositions in different Chinese black teas using Ojha, P. K., & Roy, K. (2018). PLS regression-based chemometric modeling of odorant GC-MS and GC-O combined with partial least squares regression. Flavour and properties of diverse chemical constituents of black tea and coffee. RSC Advances, Fragrance Journal, 32(4), 265–276. 8(5), 2293–2304. Xu, Y. H., Liu, J. Q., Gao, C. L., & Wang, E. K. (2014). Applications of carbon quantum Orzel, J., Daszykowski, M., Kazura, M., de Beer, D., Joubert, E., Schulze, A. E., ... Walczak, dots in electrochemiluminescence: A mini review. Electrochemistry Communications, B. (2014). Modeling of the total antioxidant capacity of rooibos (Aspalathus linearis) 48, 151–154. tea from chromatographic fingerprints and identification of potential an- Zhang, R., & Wang, W. J. (2011). Learning linear and nonlinear PCA with linear pro- tioxidant markers. Journal of Chromatography A, 1366, 101–109. gramming. Neural Processing Letters, 33(2), 151–170. Pang, X. L., Qin, Z. H., Zhao, L., Cheng, H., Hu, X. S., Song, H. L., & Wu, J. H. (2012). Zheng, X. T., Than, A., Ananthanaraya, A., Kim, D. H., & Chen, P. (2013). Graphene Development of regression model to differentiate quality of black tea (Dianhong): quantum dots as universal fluorophores and their use in revealing regulated traf- Correlate aroma properties with instrumental data using multiple linear regression ficking of insulin receptors in adipocytes. ACS Nano, 7(7), 6278–6286. analysis. International Journal of Food Science and Technology, 47(11), 2372–2379. Zong, J., Yang, X. L., Trinchi, A., Hardin, S., Cole, I., Zhu, Y. H., ... Wei, G. (2014). Carbon Qin, Z. H., Pang, X. L., Chen, D., Cheng, H., Hu, X. S., & Wu, J. H. (2013). Evaluation of dots as fluorescent probes for “off-on” detection of Cu2+ and L-cysteine inaqueous Chinese tea by the electronic nose and gas chromatography-mass spectrometry: solution. Biosensors & Bioelectronics, 51, 330–335. Correlation with sensory properties and classification according to grade level. Food

7