AJEV Papers in Press. Published online February 23, 2011.

Transcription

1 AJEV Papers in Press. Published online February 23, American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Research Note: Application of NIR-AOTF Spectroscopy to Monitor Aleatico Grape Dehydration for Passito Wine Production Andrea Bellincontro, 1 * Daniel Cozzolino, 2 and Fabio Mencarelli 1 1 Department of Food Science and Technology, University of Tuscia, Viterbo, Italy; and 2 The Australian Wine Research Institute, PO Box 197, Glen Osmond, Adelaide, 5064, Australia. *Corresponding author ( bellin@unitus.it; tel: ; fax: 7498) Acknowledgments: This research was supported by the Italian Ministry of Agriculture (MIPAAF), MUVON Project and Bilateral Integrated Action Italy-Spain 08. The authors thank Azienda Agricola Pacchiarotti Antonella (Grotte di Castro, Viterbo, Italy) for providing Aleatico grapes and Mrs. Maria Dolores García Salinas for technical support during experimental procedures. Copyright 11 by the American Society for Enology and Viticulture. All rights reserved. Abstract: Aleatico (Vitis vinifera L.) grapes were harvested at 21.3 Brix and dehydrated at C, 45% relative humidity, and 1.5 m/s air flow in a small scale thermo-conditioned tunnel. Postharvest grape drying was performed until the fruit lost an average of % of its initial weight. During dehydration, single destemmed grape berries were analyzed non-destructively using an Acousto Optically Tunable Filter (AOTF) Near-Infrared (NIR) spectrophotometer (10-20 nm) in reflectance. Total soluble solids (TSS, Brix) and moisture content (%) were measured on the same berries with the objective of using spectral and non-spectral information to develop regression models for predicting these parameters. Partial Least Square (PLS) applied after different statistical pre-treatments (Multiplicative Scatter Correction, 1

2 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Savitzky-Golay 1 st or 2 nd derivative filter) was tested on absorbance spectra in order to define the most effective approach. Two prediction models were obtained (n = 450 for TSS, and n = 600 for water loss) in which the coefficient of determination in cross-validation (r 2 ) and the root mean standard error of cross-validation (RMSECV) were 0.93 and 0.89 Brix for TSS and 0.92 and 2.16% for water loss, respectively. A model validation procedure was performed using separate sample sets (n = 170 for TSS, and n = 0 for water loss) with the following results: coefficient of determination (R 2 ) and standard error of prediction (SEP) of 0.92 and 0.72 Brix for TSS and 0.9 and 1.89% for water loss, respectively. Key words: AOTF-NIR spectroscopy, partial least square (PLS), grapes, dehydration, Total Soluble Solids (TSS), water loss (wl). Visible and near infrared (Vis-NIR) spectroscopy is a well-known technique for the nondestructive measurement of quality attributes of fruits and vegetables (Nicolai et al. 09, Lin and Yin 09). The NIR region contains information concerning the relative proportions of C-H, N-H, and O-H bonds, which are the primary structural components of organic molecules. The potential of NIR spectroscopy has been tested as an alternative method for enological parameters (Cozzolino et al. 05, 06, 08, Cynkar et al. 09), as well as discrimination of geographical provenance (Liu et al. 08, Le Moigne et al. 08). The use of an NIRacousto-optic tunable filter (AOTF) (Barbieri Gonzaga and Pasquini 05) has allowed discrimination among areas of production of Cabernet Sauvignon grapes based on total phenolics, anthocyanins, malvidin-3-glucoside, tartaric and malic acid estimations (Kay and Wample 05). Santos and Kaye (05) reported on NIR-AOTF application for potential grapevine leaf water detection. In intact olives, NIR-AOTF was recently applied for fruit 2

3 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 moisture, free acidity, and oil content prediction (Cayuela et al. 09). Previous studies (Cozzolino et al. 06) demonstrated how sugar (TSS) and moisture content can be assessed in grape berries using NIR spectroscopy together with the PLS (Partial Least Squares) regression method of multivariate statistical analysis (Wold et al. 01). The production of sweet Italian dessert wines (e.g. Passito wines, VinSanto, Recioto), or dry wines like Amarone, requires postharvest grape dehydration which can be obtained under uncontrolled or controlled conditions. Management of dehydration process controlling the is important because it affects grape metabolism, including the concentration of volatile aroma compounds after metabolism (Bellincontro et al. 04, 09, Costantini et al. 06, Chkaiban et al. 07). The influence of different thermo-hygrometric conditions during dehydration of grape varieties used for Amarone wine production has been reported (Barbanti et al. 08). At the same time, molecular profiling and gene expression, patterns explaining the metabolic mechanisms involved in postharvest water stress and the accumulation or loss of grape berry metabolites, were reported (Zamboni et al. 08, Rizzini et al. 09). In common practice, most wineries monitor grape dehydration by weighing to determine berry water loss and using destructive measures to determine the relative sugar concentration. In large dehydration facilities, drying up to 50,000 tons of grapes per year for Amarone and Recioto wines, analyzing sugars and weight loss is expensive in terms of both labor and time. For this reason, a rapid, non-destructive method for estimating these commercial parameters could be useful for commercial application. The use of NIR spectroscopy for sugar determination in grape (expressed as total soluble content or reducing sugars) (Jarén et al. 01, Fernández-Novales et al. 09) and moisture in raisins (Huxoll 00) has been tested, as well as NIR-AOTF for monitoring postharvest grape dehydration (Bellincontro et al. 09). In this note, we report the 3

4 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 results of using NIR-AOTF to predict the total soluble solids and moisture content of Aleatico grapes dehydrated under controlled environmental conditions for Passito wine production. Materials and Methods Experimental procedure and grape sampling. Aleatico (Vitis vinifera L.) grapes, a red variety typically grown for Passito wine production in the northern area of the Latium Region, Italy, were carefully harvested at 21.3 Brix. The fruit was in sound condition and without fungal contamination. The fruit was placed in a single layer in perforated plastic boxes (60xx cm) commonly used for grape dehydration. Berry manipulation and technological applications for dehydration under controlled conditions were performed following the method described in Bellincontro et al. (09). A temperature of C ± 2 C, 45% ± 5% of R.H. and air flow of 1.5 m/s were ensured throughout the dehydration process, which lasted until the bunches lost an average of % with respect to their initial weight. Grape berries were sampled every 2-3 days for spectra acquisitions, weight loss and TSS measurements. Spectra detections aimed at weight loss prediction were performed throughout the dehydration process on the same 50 berries which were initially randomly selected from the clusters and numbered. Spectra aimed at TSS prediction were detected on 25 different berries, which were randomly picked, for each sampling. The same 25 berries were subsequently crushed for Brix measurement, which was performed using a digital refractometer (Atago, Tokyo, Japan). Weight loss was carefully monitored by weighing the same 50 numbered grape berries which used for spectra detections. The weight was measured using a technical balance (Adam Equipment Co. Ltd., Milton Keynes, UK) and expressed as % of water loss from the initial weight {[(initial weight - daily weight) / initial 4

5 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 weight] x 0}. The loss in weight was attributed entirely to moisture loss, ignoring the impact of metabolic processes (Van Dijk et al. 06). Spectral acquisition and chemometric procedure. A Luminar 50 Miniature Hand-held NIR Analyzer (Brimrose Corporation, Baltimore, MD), based on the AOTF-NIR principle, was used for spectral detection. Two different measurements were performed on each intact grape berry through contact between the external gun of the NIR device and the epicarp of the fruit, using the diffuse reflectance method of detection. Detection was conducted in the nm range, with 2 nm wavelength increments and spectra per average, which represented a single measurement. The average of the two measurements was the spectral response of the berry. Raw spectra were statistically pre-treated for absorbance (log 1/R) transformation using SNAP! 2.03 software (Brimrose). Before the calibration and building up of the prediction models, the spectral variations of the data sets were analyzed through Principal Component Analysis (PCA). The absorbance data were mean normalized and treated by Multiplicative Scattering Correction (MSC), first order of Savitzky-Golay filter (6 points of smoothing) or second order of Savitzky-Golay filter (6 points of smoothing), respectively. Partial Least Squares (PLS) models were obtained on the full spectrum (10-20 nm) and considering the spectral significant variables at specific wavelength intervals. Internal full cross-validation was applied for model building and statistical analyses were carried out by applying the best models obtained for TSS and weight loss prediction, respectively. The models were optimized by outlier identification and elimination. The statistical indexes R 2 5 (coefficient of multiple determination of calibration) and r 2 (coefficient of multiple determination of cross-validation), Standard Error of Calibration (SEC), Standard Error of Prediction (SEP), Root Mean Square

6 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Error of Calibration (RMSEC), Root Mean Square Error of Prediction (RMSEP), Root Mean Square Error in cross validation (RMSECV) and Bias were used to determine the significance of the calculations. As regards the final models carried out for TSS and water loss estimation, the RPD ratio (SD/SECV) was also calculated. PCA, statistical pre-treatments and PLS models were performed using Unscrambler v9.2 software (CAMO ASA, Oslo, Norway). Results and Discussion The observation of near infrared absorbance mean spectra relative to a single intact berry (data not shown) revealed how the spectral trend is dominated by water contribution. This contribution is well expressed by the peaks of the OH stretch first overtone at 1450 nm, and the OH asymmetric stretching and bending combination at 19 nm (Osborne et al. 1993). In our study, the contribution of these wavelengths is directly related to the water loss measurement. At the same time, as previously observed by Cozzolino et al. (06), the most important contribution for the TSS calibration is the absorption around 10 nm, 1900 nm, and 2170 nm, which are the wavelengths related to O-H and C-H bonds. In Table 1 we report the mean, SD and the range (as minimum and maximum values) calculated by the reference methods (weighing and refractometry) for both calibration and prediction sets, relative to the TSS and water loss models. Grape TSS, employed in the calibration set, ranged from 18.7 to 32.4 Brix, while grape water loss ranged from 1.67 to 44.62%. For prediction models, sets of samples with values of TSS and % of water loss within the range used for the calibration sets were taken from grape samples which were not used for the calibration procedure. In all chemometric approaches it is well known that the variability in the concentration of measured 6

7 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 parameters is relevant for the modeling: the variability expressed by our grape sample sets can be considered suitable for the two NIR calibrations. Table 2 shows all the chemometric results relative to PLS applications which were not considered in the final models for TSS and % of water loss estimations. Specific wavelengths were not applied thus, the entire spectrum (10-20 nm) was considered to build the models and preliminary PCA was used for outlier selection. Result of the same PCA(data not shown) demonstrates that PC1 explained 93% of the variation in the sample set. In Figure 1 we report the PCA result obtained using selected spectral measurements referring to specific steps of berry water loss (5,,,, 25,, and %). Score plot with labeled data, in which PC1 explained 97% of the variation, clearly shows how the AOTF-NIR detection was able to monitor the water loss evolution. For regressive models, different pre-treatments were performed on the spectra sets, previously transformed in absorbance (log 1/R) and mean normalized (MN). Multiplicative scatter correction (MSC) filter, and Savitzky-Golay (SG) filter 1 st and 2 nd order derivate and six points smoothed were tested. The spectral variables thus obtained were utilized as X-block of each PLS matrix. In this matrix, TSS or % of water loss was used, respectively, for the Y- block (reference variable). The best results were obtained by using absorbance spectra mean normalized, without any other pre-treatment, for sugars, and absorbance spectra mean normalized + MSC for water loss, respectively. The calibration and cross validation results for both models were reported in Table 3 and represented graphically in Figure 2 (a and b) using scatter plots. As suggested by Shenk and Westerhaus (1996), an r 2 7 (coefficient of determination in cross-validation) value greater than 0.9 represents good quantitative information. We obtained results of 0.93 and 0.92 for TSS and water loss, respectively, indicating that correlation equations can provide significant quantitative results. The SECV

8 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 (0.9 Brix for TSS and 2.17% for water loss.), as reported by Fernández-Novales et al. (09), is a statistical efficiency parameter in association with the SD of the reference values, by considering the RPD ratio (SD/SECV). Specifically, Williams and Sobering (1996) recommended an RPD ratio as close to three as possible or, in any case, larger than 2.5. This result was achieved perfectly for the sugar model (3.31) and was confident for water loss (5.28) since, as reported by Smyth et al. (08), an RPD value greater than 5 (range 5-6.4) is considered good for quality control (Table 3). In calibration developments, numbers of latent variables (LVs) were calculated and selected in correspondence to the RMSECV minimization. As shown in Table 3, we obtained and 6 PLS terms for TSS and water loss model, respectively. PLS regression coefficients were calculated with respect to the specific wavelengths (data not shown). For TSS, a significant correlation was observed around 10 nm associated with the C-H stretching second overtone, around 13 nm associated with C-H stretching and C-H deformation, around 10 nm associated with the O-H second overtone, around 1900 nm associated with the O-H first overtone and between to 20 nm associated with combination tones, C-H, N-H. For water loss around 10 nm associated with the O-H second overtone, around 1900 nm associated with the O-H first overtone, and between to 20 associated with the combination tones, C-H, N-H. Following the model calibration procedure, external validation exercises were performed for both parameters using separate sample sets. Table 4 shows the results in terms of feasibility and accuracy in predicting the two parameters by applying the models previously carried out. Relatively high values for R 2 index (0.92 and 0.90 for TSS and % of water loss, respectively) and moderate errors in prediction, especially for TSS evaluation (RMSEP of 0.72 Brix and 1.89%), were obtained. 8

9 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Conclusions Near infrared spectroscopy is a non-destructive technology sensitive to the chemical changes in intact grapes observed during berry ripening. This AOTF-NIR application on wine grapes during postharvest dehydration, demonstrates the feasibility of this technology in monitoring sugar and water loss. Through PLS, two robust models based on absorbance spectra were established for TSS (Brix) and % of water loss determination. Other authors have reported that the application of NIR models can reduce the time spent for the analytical procedure in grapes, musts, and wines by 80% (Cozzolino et al. 08). The results obtained through this study suggest that the use of NIR technology could be utilized by wineries for monitoring grape berry dehydration during the postharvest period. Literature cited Barbanti, D., Mora, B., Ferrarini, R., Tornielli G. B., and Cipriani, M. 08. Effect of various thermo-hygrometric conditions on the withering kinetics of grapes used for the production of Amarone and Recioto wines. J. Food Eng. 85(3): 0-8. Barbieri Gonzaga, F., and Pasquini, C. 05. Near-Infrared emission spectrometry based on an acousto-optical tunable filter. Anal. Chem. 77(4): Bellincontro, A., Nicoletti, I., Valentini, M., Tomas, A., De Santis, D., and Mencarelli, F. 09. Integration of nondestructive techniques with destructive analyses to study postharvest water stress of winegrapes. Am. J. Enol. Vitic. 60(1): Bellincontro, A., De Santis, D., Botondi, R., Villa, I., and Mencarelli, F. 04. Different postharvest dehydration rates affect quality characteristics and volatile compounds of 9

10 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Malvasia, Trebbiano and Sangiovese grape for wine production. J. Sci. Food Agric. 84(13): Cayuela, J. A., García, J. M., and Caliani, N. 09. NIR prediction of fruit moisture, free acidity and oil content in intact olives. Grasas y Aceites 60(2): Cen, H., and He, Y. 07. Theory and application of near infrared reflectance spectroscopy in determination of food quality. Trends Food Sci. Technol. 18: Chkaiban, L., Botondi, R., Bellincontro, A., De Santis, D., Kefalas, P., and Mencarelli, F. 07. Influence of postharvest water stress on LOX, ADH and aroma biochemistry of Gewürztraminer grape dehydrated under controlled and uncontrolled thermohygrometric conditions. Aust. J. Grape Wine Res. 13: Costantini, V., Bellincontro, A., De Santis, D., Botondi, R., and Mencarelli, F. 06. Metabolic changes of Malvasia grapes for wine production during postharvest drying. J. Agric. Food Chem. 54: Cozzolino, D., Cynkar, W. U., Dambergs, R. G., Janik, L., and Gishen, M. 05. Effect of both homogenisation and storage on the spectra of red grapes and on the measurement of total anthocyanins, total soluble solids and ph by visual near infrared spectroscopy. J. Infrared Spectroscopy 13(4): Cozzolino, D., Cynkar, W. U., Dambergs, R. G., Mercurio, M. D., and Smith P. A. 08. Measurement of condensed tannins and dry matter in red grape homogenates. Anal. Chimica Acta 513: Cozzolino, D., Dambergs, R. G., Janik, L., Cynkar, W. U., and Gishen, M. 06. Analysis of grapes and wine by near infrered spectroscopy-a review. J. Near Infrared Spectroscopy 14:

11 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Cynkar, W. U., Cozzolino, D., and Dambergs, R. G. 09. The effect of sample storage and homogenisation techniques on the chemical composition and near infrared spectra of white grapes. Food Res. Intern. 42: Fernández-Novales, J., López, M. I., Sánchez, M. T., Morales, J., and González-Caballero, V. (09). Shortwave-near infrared spectroscopy for determination of reducing sugar content during grape ripening, winemaking, and aging of white and red wines. Food Res. Intern. 42: Huxoll, C. C. 00. Assessment of near infrared (NIR) diffuse reflectance analysis for measuring moisture and water activity in raisins. J. Food Process. Pres. 24: Jarén, C., Ortuño, J. C., Arazuru, S., Arana, J. I., and Salvadores, M. C. 01. Sugar determination in grapes using NIR technology. Intern. J. Infrared Mill. Waves 22(): 21-. Kaye, O., and Wample, R. L. 05. Using near infrared spectroscopy as an analytical tool in vineyards and wineries. Abstract. Am. J. Enol. Vitic. 56: 296A. Le Moigne, M., Maury, C., Bertrand, D., and Jourjonet, F. 08. Sensory and instrumental characterisation of Cabernet Franc grapes according to ripening stages and growing location. Food Qual. Pref. 19: Lin, H., and Ying, Y. 09. Theory and application of near infrared spectroscopy in assessment of fruit quality: a review. Sens. & Instrumen. Food Qual. 3: Liu, L., Cozzolino, D., Cynkar, W. U., Dambergs, R. G. Janik, L., O Neill, B. K., Colby, C. B., and Gishen, M. 08. Preliminary study on the application of visible near infrared spectroscopy and chemometrics to classify Riesling wines from different countries. Food Chem. 6:

12 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Nicolai, B., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W., Theron, K. I., and Lammertyn, J. 09. Nondestructive measurements of fruit and vegetable quality by means of NIR spectroscopy: A review. Postharvest Biol. Technol. 46: Osborne, B. G., Fearn, T., and Hindle, P. H Practical NIR Spectroscopy with applications in Food and Beverage Analysis. Longman Scientific and Technical: Essex, England, UK, p 227. Rizzini, F. M., Bonghi, C., and Tonutti, P. 09. Postharvest water loss induces marked changes in transcript profiling in skins of wine grape berries. Postharvest Biol. Technol. 52(3): Santos, O. A., and Kaye, O. 05. Grapevine water potential based upon near infrared spectroscopy. Sci. Agric. (Piricicaba, Braz.) 66(3): Shenk, J. S., and Westerhaus, M. O Calibration the ISI way. In A. M. C. Davies & P. C. Williams (Eds.), Near infrared spectroscopy: The future waves. Chichester: NIR Publications, pp Smyth, H. E., Cozzolino, D., Cynkar, W. U., Dambergs, R. G., Sefton, M., and Gishen, M. 08. Near infrared spectroscopy as a rapid tool to measure volatile aroma compounds in Riesling wine: possibilities and limits. Anal. Bioanal. Chem. 390: Van Dijk, C., Borieu, C., Peter, F., Stolle-Smits, T., and Tijskens, L. M. M. 06. The firmness of stored tomatoes (cv. Tradiro). 1. Kinetic and near infrared models to describe firmness and moisture loss. J. Food Eng. 77: Williams, P. C., and Sobering, D. C How do we do it: A brief summary of the methods we use in developing near infrared calibrations. In A. M. C. Davies & P. Williams 12

13 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 (Eds.), Near infrared spectroscopy: The future waves. Chichester: NIR Publications. pp Wold, S., Sjöström, M., and Eriksson, L. 01. PLS-regression: a basic tool of chemometrics. Chemom. Intell. Lab. Syst. 58, 9 1. Zamboni, A., Minoia, L., Ferrarini, A., Tornielli, G. B., Zago, E., Delledonne, M., and Pezzotti M. 08. Molecular analysis of post-harvest withering in grape by AFLP transcriptional profiling. J. Exp. Botany 59():

14 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev Figures and Tables PC2 (3%) PC1 (97%) Figure 1. Principal component score plot of the absorbance spectra selected at the specific steps of calculated water loss, and marked with the labels 5,,,, 25,, and %.

15 TSS predicted (Brix) % of water loss predicted American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev Calibration Validation Calibration Validation Calibration Validation Calibration Validation TSS measured (Brix) (a) Calibration n = 450 r 2 = 0.93 RMSECV = 0.89 SECV = 0.90 Validation n = 170 r 2 = 0.92 RMSEP = 0.73 SEP = % of water loss measured (b) Calibration n = 600 r 2 = 0.92 RMSECV = 2.16 SECV = 2.17 Validation n = 0 r 2 = 0.90 RMSEP = 1.89 SEP = 1.90 Figure 2 (a). Scatter plot relative to the PLS calibration and validation models for TSS (Brix) measurements on Aleatico grape berries during postharvest dehydration process under controlled conditions. Figure 2 (b). Scatter plot relative to the PLS calibration and validation models for water loss (%) measurements on Aleatico grape berries during postharvest dehydration process under controlled conditions.

16 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Table 1. Statistical analyses of calibrations and validation sets, i.e. mean, standard deviation (SD), range (min. and max.), for Total Soluble Solids (TSS, Brix) and water loss (%) evaluation. samples (n) mean SD min. max. TSS ( Brix) calibration validation water loss (%) calibration validation Table 2. Calibration procedures with different statistical pre-treatments applied on the two sample sets for TSS and water loss estimation. outliers R 2 RMSEC SEC Bias LVs TSS ( Brix) MN + MSC b E-06 9 MN + SG 1 st der. c E-07 MN + SG 2 nd der. d E water loss (%) MN a E-06 6 MN + SG 1 st der E-08 7 MN + SG 2 nd der E-07 9 a MN = absorbance spectra mean normalized b MN + MSC = absorbance spectra mean normalized + multiplicative scatter correction c MN + SG 1 st der. = absorbance spectra mean normalized + Savitzky-Golay filter (1st derivative, 6 points of smoothing) d MN + SG 2 nd der. = absorbance spectra mean normalized + Savitzky-Golay filter (2nd derivative, 6 points of smoothing) 16

17 American Journal of Enology and Viticulture (AJEV). doi:.5344/ajev..041 Table 3. Calibration and cross validation results for selected TSS and water loss models. Parameter Calibration Cross validation outliers R 2 RMSEC SEC Bias LVs r 2 RMSECV SECV RPD TSS ( Brix) n = E water loss (%) n = E Table 4. Results of validation procedure of PLS models for TSS and water loss. R 2 RMSEP SEP Bias TSS ( Brix) Water loss (%)