Development of a colorimetric taste sensor based on dye-bead conjugated array and imaging system for white wines Soo Chung1 Soo Hyun Park1 Tu San Park2 Seongmin Park1 Daesik Son1 Seong In Cho1,3* 1 2 3 Department of Biosystems & Biomaterials Science and Engineering, Seoul National University, Seoul, 151-921, Republic of Korea. Agricultural and Biosystems Engineering, The University of Arizona 1177 E 4th Street, Room 403 Tucson, AZ 85721, United States. Research Institute for Agriculture and Life Sciences, Seoul National University, Seoul, 151921, Republic of Korea. * Corresponding Author Tel: +82-2-880-4603, Fax: +82-2-873-2049, Email : sicho@snu.ac.kr Abstract Introduction: The development of a colorimetric sensor is based on a concept different from that of conventional electronic tongue systems, and the method is easier and fast. Objective: A taste sensor based on a colorimetric array with beads conjugated chemical dyes was developed to estimate flavors such as sweetness, sourness, and astringency, and applied to predict flavors of white wines. Method: Chemical dyes which were decided to use for conjugations were alizarin, calconcarboxylic acid, cresol red, crystal violet, fluorescein, methylthymol blue, phenol red, and xylenol orange. They were analyzed the spectral characteristic by wavelengh in visible ray and colorimetric characteristic. R, G, B color values extracted from images of a sensor array were used to develop estimation models using principle component regression (PCR). Results: The best prediction ability of PCR model showed R2 = 0.85, RMSEP = 11936 for sweetness flavor, R2 = 0.83, RMSEP = 59 for sourness flavor, R2 = 0.73, RMSEP = 0.88 for astringency flavor. Conclusion: The difference map of a sensor array was used to develop PCR model to estimate flavors in white wines. The model performs enough to estimate tastes within 10% error. It is determined that basic flavor information is possible to be provided to the consumers before their choice of a white wine. Keywords: White wine, Taste sensor, Dye-bead conjugation, PCR, Wine sensor. Introduction The sense of taste is responsible for detecting and distinguishing among sweet, bitter, sour, salty, and amino acid (umami) stimuli (1). However, the variability between taste panelists can sometimes give rise to more than 50% variation in terms of flavor units and there are more extra tastes such as pungent, astringent, acrid, cool, metal, colloidal, and so on (2). The determination of the quality of foodstuffs and beverages and the facilitation of their conformation standards are important and different matters. Usually, such studies are carried out with a scarcity of sample preparations and rather complex and expensive laboratory analytical tools like various chromatographs, spectrometers, etc. These types of analysis may be in-depth and reliable but the necessary procedures can hardly be done. Also for quality control, it is necessary to monitor the evolution of a group of certain components that reflect the process of ageing and spoilage of a food product. These components can be numerous or unknown and the problem appears to be quite difficult. Further, it is very hard to compare the results of instrumental analysis to biological sensing (3). The development of taste sensor is based on a concept very different from that of conventional chemical sensors, which selectively detect specific chemical substances such as glucose or ethanol (4-6). Taste cannot be measured even if all the chemical substances contained in foodstuffs are measured. Humans do not distinguish each chemical substance, but express the taste itself; thus, the relationship between chemical substances and taste is obscure. One of the goals of taste sensing should be to reproduce the five basic kinds of tastes, which humans perceive, based on the quantity of each taste substance (6). In this case, the required abilities of sensing are the following: high sensitivity for quantitative analysis on each taste ingredient, high selectivity among combined taste substances, and sensing repeatability. Applying sensors which possess high sensitivity and selectivity to each taste could be difficult. Taste sensors should have hundreds of combined sensor arrays. It is impossible to integrate hundreds of highly selective sensors. Sensors which have specificity and selectivity on each taste material should be developed (4, 5). Therefore, the development of a taste sensor by using an individually selective or a specific taste material is not appropriate. Generally, a non-selective sensor array system is applied on taste sensor development. The signal from the sensor system is analyzed in a determination analysis to a group taste pattern in a certain way. By comparing those patterned signal tastes, samples are recognized and classified (7, 8). In addition, it is determined that the recent colorimetric sensor arrays are considered as an inexpensive disposables method to be completed easily with imaging technology (9, 10). The objectives of this study are to fabricate a colorimetric sensor array with dye-bead conjugates and to apply for sensing the taste of white wines by optical taste sensor. In case of using the dye-bead conjugate, several advantages follow such as being able to control the amount of dye entering a sensor chip and rapidly measuring because usage of quantitative dye could be used by adjusting the amounts of uniform beads, and the reaction between dye and liquid happens in a moment. For the analysis of performance, the models using partial component regression (PCR) were developed to estimate the flavors such as sweetness, sourness, and astringency. Materials and Methods Sample and sensor array Twenty different commercial white wines were used in this experiment. For additional samples, those white wines were mixed, diluted, and added with other components to make 25 more samples. Forty-five white wines were prepared totally. The dyes used for research were alizarin, calconcarboxylic acid, xylenol orange, phenol red, methyltymol blue, crystal violet, cresol red, and fluorescein (Sigma-Aldrich, MO, USA). These dyes were conjugated with micro-bead (BeadTech, Seoul, Korea) using 1-Ethyl-3-(3dimethylaminopropyl) carbodiimide (EDC) cross-linker purchased from Sigma-Aldrich. Sensor array is fabricated with a silicon isolator (GRACE BIO-LABS, OR, USA) and dyebead conjugates were placed in each hole respectively shown as Fig. 1. Figure 1. Sensor array with dye-bead conjugates. Quantitative analysis The amount of sucrose, glucose, fructose, citric acid, malic acid, tartaric acid, succinic acid, lactic acid, acetic acid of wine was quantitatively analyzed using HPLC (high performance liquid chromatography, Agilent, HP1100, USA). As a preprocessing step, samples were filtered through 0.45μm syringe filter (WhatmanTM, Buckinghamshire, UK), so that 2ml of pure liquid was used in quantitative analysis. The summation of analyzed sucrose, glucose, and fructose amount were converted to value of sweetness flavor. The sourness flavor was calculated using the sum of molar mass of organic acid. Tannin wasn’t quantitatively analyzed using HPLC, because its molecular weight was too large; instead tartaric acid iron colorimetric method was used to analyze tannin by using the transmittance intensity at 415nm. Using Unscrambler(v. 7.5 CAMO A/S, Norway) PLSR (partial least square regression) was performed with the several different concentration tannin solution to obtain the equation that can predict the amount of tannin. The accuracy of tannin prediction model is high enough shown as Fig. 2 it showed determination of coefficient (R2) was 0.9854. Figure 2. Developed equation for tannin quantitative analysis. A camera (A601fc Basler, Ahrensburg, Germany) was set up to take photo of the sensor array. The color value of the dye-bead conjugate before the reaction was measured in the sensor array made up of silicon isolator using image acquisition system shown in Fig. 3 below. Then, 100μl of white wine was reacted with dye-bead conjugates for 20 minutes and was put into 60℃ oven for 30 minutes to evaporate the water. The reason why the dryinf process is necessary is because it is difficult to develop a model with high accuracy due to influence of the color of sample and light glares in wet condition. The each R, G, B color value of 8 holes in the sensor array was extracted using MATLAB (MathWorks, Natick, MA, USA). Figure 3. Features of the measurement system by color change of dye. Data analysis The difference of color value before and after the reaction was used to develop a flavor prediction model using PCR, which can estimate the degree of each taste, using Unscrambler (CAMO PROCESS AS, Oslo, Norway). Evaluation of each model was performed using cross validation, and the performance result was represented as the determination of coefficient (R2) and root mean square error of prediction (RMSEP). Results and Discussion PCR result for sweetness The PCR performance analysis with the color values in the difference map and sweetness flavor were represented R2 = 0.7769, RMSEC = 14680 for calibration and R2 = 0.7769, RMSEP = 16159 for validation. Since error bound went over 10%, to develop a model with better estimation, three data considered as extreme image distortion and outliers were removed for reanalysis. When three data were eliminated, R2 = 0.8848, RMSEC =11428 for calibration, and R2 = 0.8273, RMSEP = 13388 for validation as performance results. Also eliminating two more data resulted even better, which showed R2 of 0.8959, RMSEC of 10017 for calibration and R2 of 0.8528, RMSEP of 11936 for validation. The prediction error was less than 8%, which means accurate enough to estimate sweetness flavor of wines. Figure 4 is the graphs of calibration and validation plot with 40 white wines excluding the five outlier data for sweetness flavor. Figure 4. Predicted vs. measured sweetness flavor with 40 white samples; (A) calibration plot and (B) validation plot. PCR result for sourness As for sourness flavor analysis, the estimation model performed similar as the model for sweetness flavor. The PCR performance with the color values in the difference map and sourness flavor were represented R2 = 0.7950, RMSEC = 66.20 for calibration and R2 = 0.7575, RMSEP = 72.25 for validation. R2 of 0.8303, RMSEP of 59.97 for calibration and R2 of 0.8131, RMSEP of 64.57 for validation were showed when three outliers were excluded. Additional removal of two more data showed a better performance result, which were R2 = 0.8777, RMSEC = 49.48 for calibration and R2 = 0.8263, RMSEP = 59.09 for validation. It represented less than 7.5% error, accurate enough to estimate sourness flavor of wines. Figure 5 is the graphs of the best performed calibration and validation plot with 40 white wines excluding the five outlier data for sourness flavor. Figure 5. Predicted vs. measured sourness flavor with 40 white samples; (A) calibration plot and (B) validation plot. PCR result for astringency As for astringency flavor analysis, the PCR performance analysis with the color values in the difference map and astringency flavor were represented R2 = 0.6706, RMSEC = 1.10 for calibration and R2 = 0.4754, RMSEP = 1.47for validation. When three data was removed, R2 = 0.8758, RMSEC = 0.59 for calibration and R2 = 0.7113, RMSEP = 0.94 for validation. Astringency gave better results when five data was removed, as well. R2 = 0.9077, RMSEC = 0.51 for calibration and R2 = 0.7435, RMSEP = 0.88 for validation. The error showed about 11%, which are slightly less accurate results comparing to sweetness or sourness estimating model. Figure 6 is the graphs of calibration and validation plot with 40 white wines excluding the five outlier data for astringency flavor. Figure 6. Predicted vs. measured astringency flavor with 40 white samples; (A) calibration plot and (B) validation plot. Conclusions In this study we calculated flavors representing as the white wine’s sweetness, sourness and astringency through examining the color change of dye-bead conjugate when reacted with a sample and we developed regression model to estimate flavors. The estimated amount of quantitatively analyzed sweetness, sourness, and astringency and the change in color value were used to develop the model, using PCR. The best result with 40 samples showed R2 = 0.8528, RMSEP = 11936 for sweetness flavor, R2 = 0.8263, RMSEP = 59.09 for sourness flavor, R2 = 0.7343, RMSEP = 0.88 for astringency flavor. There was a slight error in estimating the exact amount, however, it seemed no problem representing roughly what flavors classify to 3 ~ 4 degrees. It is determined that essential flavor information is able to be provided to the consumers before their choice of a white wine. In further experiments, such as the artificial neural network (ANN) method is needed to estimate the degree of taste more precise. Acknowledgement This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF2012R1A1B3004) References 1. Lorenz, J. K., Reo, J. P., Hendl, O., Worthington, J. H., and Petrossian, V. D. (2009). Evaluation of a taste sensor instrument (electronic tongue) for use in formulation development. International Journal of Pharmaceutics 367(1): 65-72. 2. Sun, H. Z. H. M., Mo, Z. H., Choy, J. T., Zhu, D. R., and Fung, Y. S. (2008). Piezoelectric quartz crystal sensor for sensing taste-causing compounds in food. Sensors and Actuators B: Chemical 131(1): 148-158. 3. Leonte, I. I., Sehra, G., Cole, M., Hesketh, P., and Gardner, J. W. (2006). Taste sensors utilizing high-frequency SH-SAW devices. Sensors and Actuators B: Chemical 118(1): 349-355. 4. Toko, K. (2000). Taste sensor. Sensors and Actuators B: Chemical 64(1): 205-215. 5. Toko, K. (1996). Taste sensor with global selectivity. Materials Science and Engineering: C 4(2): 69-82. 6. Hsueh, T. J., Chang, S. J., Hsu, C. L., Lin, Y. R., and Chen, I. (2007). Highly sensitive ZnO nanowire ethanol sensor with Pd adsorption. Applied Physics Letters 91(5): 053111. 7. Ouyang, Q., Zhao, J., Chen, Q., & Lin, H. (2012). Classification of rice wine according to different marked ages using a novel artificial olfactory technique based on colorimetric sensor array. Food chemistry 84: 77-83 8. Kim, N. S. (2005). Discriminant Analysis of Marketed Liquor by a Multi-channel Taste Evaluation System. Food Science and Biotechnology 14(4): 554-557 9. Zhang, C., and Suslick, K. S. (2007). Colorimetric sensor array for soft drink analysis. Journal of agricultural and food chemistry 55(2): 237-242. 10. Zhang, C., Bailey, D. P., and Suslick, K. S. (2006). Colorimetric sensor arrays for the analysis of beers: a feasibility study. Journal of agricultural and food chemistry 54(14): 4925-4931.