Determination of Technological Maturity of Grapes and Total Phenolic Compounds of Grape Skins in Red and White Cultivars During Ripening by Near Infrared Hyperspectral Image: A Preliminary Approach

1 Determination of technological maturity of grapes and total phenolic 1 compounds of grape skins in red and white cultivars during ripening 2 by near infrared hyperspectral image: A preliminary approach. 3 4 5 6 Julio Nogales-Bueno, José Miguel Hernández-Hierro, Francisco José Rodríguez-Pulido 7 and Francisco José Heredia* 8 9 10 Food Colour & Quality Laboratory, Department of Nutrition & Food Science, 11 Universidad de Sevilla, Facultad de Farmacia, 41012 Sevilla, Spain. 12 13 * Corresponding author: Francisco José Heredia 14 15 Phone: +34 9545 56495 16 Fax: +34 9545 56110 17 18 E-mail: [email protected] 19 20 2 ABSTRACT 21 Hyperspectral images of intact grapes during ripening were recorded using a near 22 infrared hyperspectral imaging system (900 - 1700 nm). Spectral data have been 23 correlated with grape skin total phenolic concentration, sugar concentration, titratable 24 acidity and pH by modified partial least squares regression (MPLS) using a number of 25 spectral pre-treatments and different sets of calibration. The obtained results (RSQ and 26 SEP respectively) for the global model of red and white grape samples were: 0.89 and 27 1.23 mg g-1 of grape skin for total phenolic concentration, 0.99 and 1.37 ºBrix for sugar 28 concentration, 0.98 and 3.88 g L-1 for titratable acidity and for pH 0.94 and 0.12. 29 Moreover, separate calibration models for red and white grape samples were also 30 developed. The obtained results present a good potential for a fast and reasonably 31 inexpensive screening of these parameters in intact grapes and therefore, for a fast 32 control of technological and phenolic maturity. 33 KEYWORDS: Technological maturity, phenolic maturity, grapes, near infrared 34 hyperspectral imaging; chemometrics. 35 36 3 1. Introduction 37 Grape harvest time is one of the most fundamental aspects that have influence on the 38 future of wine quality. A number of factors have influence on this decision, among them 39 technological and phenolic maturity of grape, especially grape skins phenolic maturity. 40 Technological maturity is mainly connected with sugar concentration, titratable acidity 41 and pH. The sugar concentration determines the potential alcoholic strength. The 42 titratable acidity and pH help to control the wine quality and colour. Phenolic maturity 43 shows the ripeness degree for the skins, pulp and seeds taking into account its phenolic 44 composition (Meléndez, Ortiz, Sarabia, Íñiguez, & Puras, 2013; Ferrer-Gallego, 45 Hernández-Hierro, Rivas-Gonzalo, & Escribano-Bailón, 2012). 46 It is really important to the winemakers the determination of adequate technological and 47 phenolic maturities. Nowadays, at wineries, the maturity of grapes is usually controlled 48 using classic physical and chemical analyses. For determining the sugar concentration is 49 common to carry out density studies in the grape must since the specific gravity is 50 directly related to the contents of soluble solids (°Brix) or to the Baume scale. It is also 51 possible to measure the sugar concentration in the must by means of its refractive index. 52 The determination of the total acidity and pH is usually carried out by means of 53 volumetric titrations using NaOH and selective electrodes respectively. The detailed 54 phenolic profile is performed by High Pressure Liquid Chromatography (Xu et al., 55 2011) meanwhile spectrophotometric methods such as Folin-Ciocalteu are commonly 56 selected for total phenolic determinations (Singleton, 1985). 57 All methods above mentioned are time consuming or destructive, or both. It would be 58 interesting to replace these methods for new ones, not destructives and roughly reliable. 59 Near infrared spectroscopy (NIRS) has been also used as analysis tool to replace 60 traditional methods (Ferrer-Gallego, Hernández-Hierro, Rivas-Gonzalo, & Escribano- 61 4 Bailón, 2011). Moreover, hyperspectral techniques have been studied to replace 62 physicochemical analysis in several matrices (Baiano, Terracone, Peri, & Romaniello, 63 2012; Fernandes et al., 2011; Barbin, Elmasry, Sun, & Allen, 2013). 64 Characterizations of food quality, safety and composition have been accomplished using 65 the aforesaid analytical tool (Elmasry, Kamruzzaman, Sun, & Allen, 2012; Gowen, 66 O'Donnell, Cullen, Downey, & Frias, 2007; Lorente et al., 2012). 67 In the wine sector, it is really important to know critical parameters and attributes of 68 grapes, and is necessary to do it quickly and precisely. Near infrared hyperspectral 69 imaging could be an option to measure these parameters without sample destruction and 70 reagent consumption. In essence, hyperspectral imaging is a rapid, non-destructive, 71 rugged, multiparametric and flexible tool that potentially provides a suitable way to 72 analyse food (Gowen et al., 2007). 73 The aim of this study is to develop a useful and non-destructive hyperspectral method 74 for the determination of the principal parameters that compose phenolic and 75 technological maturity (i.e. pH, total acidity, sugar concentration and total phenols) in 76 white and red grapes. The samples used in this work have been collected in the 77 Condado de Huelva Designation of Origin D.O. (Andalusia, Spain) which is under the 78 typical climatic conditions of a warm area (Gordillo et al., 2012). To our knowledge, 79 this is the first time that near infrared hyperspectral imaging has been applied to grapes 80 to face the aforementioned goals. 81 2. Material and methods 82 2.1. Samples 83 Vitis vinifera L. cv. Zalema, Tempranillo and Syrah were collected from four vineyards 84 located in the Condado de Huelva Designation of Origin D.O. (Andalusia, Spain). 85 Zalema is a white cultivar autochthonous to the South of Spain where it represents over 86 5 90% of the overall production (Hernanz et al., 2009). Zalema grapes were collected 87 from two vineyards which present different types of soil, sand and clay. Tempranillo is 88 the most often grown red grape cultivar in Spain for producing quality red wines and 89 Syrah is a resistant cultivar to warm climatic conditions (Gordillo et al., 2012). 90 Both, white and red grapes were collected at different dates from mid-July to early 91 September in the 2012 vintage. In this way, grapes were collected at different stages of 92 maturity. There were different numbers of samples for each variety due to the earlier 93 ripening of the red cultivars. Sixteen dates were taken into account for Tempranillo, 94 seventeen for Syrah and eighteen for each Zalema type of soil. Three groups of berries 95 were collected for each date and vineyard. With the aim of achieving representative 96 samples, these were collected from the top, middle and bottom of the cluster and in the 97 sunlight and shade side of this. After that, the samples were immediately frozen and 98 stored at -20 ºC until analyses were performed. Two subsamples were taken from each 99 sample, one to determine the reference parameters and the other one for the 100 hyperspectral analysis. 101 2.2. Determination of reference parameters 102 Reference parameters were total phenolic concentration in grape skins, sugar 103 concentration, titratable acidity and pH. Total phenolic concentration was determined 104 using the Folin-Ciocalteu method (Singleton, 1985). In order to perform this 105 determination, grape skins were separated manually from the whole grapes. Afterwards, 106 one gram of grape skin was macerated in 10 mL of methanol containing 0.1% of 12M 107 HCl. Methanolic phases were centrifuged (3000 rpm, 10 min) and successively pooled, 108 approxymately 2 mL millilitres of water were added and the extract was concentrated 109 under vacuum at 30 ºC until methanol was removed and finally made up to 10 mL with 110 ultrapure water. The total phenolic concentration was determined using the Folin- 111 6 Cicalteu method in this aqueous extract and it was expressed as gallic acid equivalents 112 per gram of grape skin. 113 Technological maturity parameters were determined using the analytical methods 114 recommended by the O.I.V. (1990) using the must obtained after crushing the grapes. 115 2.3. Hyperspectral imaging analysis 116 Fig. 1 shows the main components of the hyperspectral imaging device (Infaimon S.L., 117 Barcelona, Spain) which were the illumination source, optics (mirror scanner and lens), 118 spectrograph, camera and computer. The system comprised a Xenics® XEVA-USB 119 InGaAs camera (320 × 256 pixels; Xenics Infrared Solutions, Inc., Leuven, Belgium), a 120 spectrograph (Specim ImSpector N17E Enhanced; Spectral Imaging Ltd., Oulu, 121 Finland) covering the spectral range between 900 and 1700 nm (spectral resolution of 122 3.25 nm), two 70 W tungsten iodine halogen lamps (Prilux ®, Barcelona, Spain) 123 mounted as source light, a mirror scanner (Spectral Imaging Ltd., Oulu, Finland) and a 124 computer system. Hyperspectral images were recorded using a 50 Hz frame rate and an 125 exposure time of 9 ms using the instrument acquisition software SpectralDAQ v. 3.62 126 (Spectral Imaging Ltd., Oulu, Finland). 127 A two point reflectance calibration was used. A Spectralon® ceramic tile (Labsphere 128 Inc., North Sutton, USA) was used as a white reference while dark current was recorded 129 by taking a measurement after covering the spectrograph lens with a cup and closing the 130 shutter. Corrected reflectance values (R) were calculated taking into account the 131 relationship between sample (S), white standard (W) and dark current (D) absolute 132 signal intensities using the following formula: 133 R=[(S-D)/(W-D)] (1) 134 Thereafter, the samples were thawed and tempered at room temperature and the 135 hyperspectral images of the intact grapes on a polyethylene plastic were recorded. The 136 7 characteristic spectral profile of this surface was useful in segmentation process for 137 recognising the region of interest. Noisy wavebands at both extremes of the spectra 138 range were removed and only spectral data in the resulting effective wavelength 950 - 139 1650 nm regions were used in data analysis due to reduced efficiency outside this range 140 in the used device. 141 2.4. Image processing and data analysis 142 Image processing. Image treatment was carried out using Matlab (R2010b; The Math 143 Works, Inc. USA). Prior to the quantitative analysis, a discriminant method was applied 144 to the grape images to isolate the grapes from other parts of image. Firstly, three regions 145 of interest (ROIs) were selected (background, grape and pedicel) to develop a stepwise 146 lineal discriminant model. The aforementioned discriminant model classified each pixel 147 into two classes (grape or no grape pixel) using the reflectance values from six 148 wavelengths (979, 1034, 1073, 1314, 1386 and 1550 nm). After that, the average 149 spectrum of the grape region was extracted and then transformed into Log (1/R) units. 150 The procedure was repeated for each sample and the obtained spectra were combined 151 into the spectral matrix. 152 Data analysis. Prior to quantitative analysis, an unsupervised pattern recognition 153 technique, principal component analysis (PCA), was used in order to provide 154 information about the latent structure of spectral matrix. The spectral matrix was 155 constructed from the red grape spectra, white grape spectra or both. This method 156 provides not only information related to spectral outliers, the distribution of samples in 157 the newly-created space and their possible separations in different spectral groups but is 158 also an important source of knowledge with which to create cross-validation groups 159 used in the calibration process (Shenk & Westerhaus, 1995; Brereton, 2003). 160 8 Calibrations were performed using modified partial least squares regression (MPLS). 161 For achieving this task, the corresponding total phenolic concentration, sugar 162 concentration, titratable acidity and pH values were allocated to the raw spectrum of 163 each sample, and then different spectral pre-treatments were tested. They were also used 164 different calibration sets (i.e red grapes, white grapes or both). In this method, the group 165 of calibration samples is divided into a series of subsets in order to perform cross- 166 validation to set the number of PLS factors, reduce the possibility of overfitting (Shenk 167 & Westerhaus, 1995) and remove chemical outliers. Using the T≥2.5 criterion, samples 168 that presented a high residual value when they were predicted were eliminated from the 169 set. Finally, validation errors are combined into a single figure, the standard error of 170 cross-validation (SECV). 171 Spectral pre-treatments are usually applied to NIR raw data; scattering effects were 172 removed using multiplicative scatter correction (MSC), standard normal variate (SNV), 173 and detrending (Geladi, MacDougall, & Martens, 1985; Dhanoa, Lister, & Barnes, 174 1995). Moreover, the effect of differentiation and variations in spectral ranges were 175 tested in the development of the NIRS calibrations. 176 The software used was Win ISI® (v1.50) (Infrasoft International, LLC, Port. Matilda, 177 PA, USA). This software allowed the data pre-treatment and development of 178 quantitative and qualitative models. From the three samples of each date, one (33%) 179 was randomly allocated to the validation set and the other two (66%) to the calibration 180 set. Consequently, from the 213 spectral samples (99 red and 114 white grape spectral 181 samples), 142 were allocated in the calibration set and the remaining 71 were allocated 182 in the validation set (two and one thirds of white and red grape spectral samples 183 respectively). 184 3. Results and discussion 185 9 3.1. Chemical analysis 186 Total phenolic concentration ranged from 2.2 to 15.8 mg g-1 of grape skin with a 187 standard deviation value of 2.9 mg g-1 of grape skin. 188 Sugar concentration, titratable acidity and pH ranged from 4.1 to 25.4 ºBrix, 2.7 to 52.9 189 g L-1 and 2.5 to 3.8 respectively. 190 3.2 Hyperspectral imaging analysis 191 It was expected a clear difference between red and white grape spectra, however the 192 average spectra of red and white grapes are really similar (Fig. 2). For this reason it was 193 initially decided to work with both groups together. This figure shows the average 194 spectra of red and white grapes over the 950-1650 nm range. Standard deviation spectra 195 for each group are also represented and for display reasons they have been multiplied by 196 a factor of 10. Spectral intensities were low and well within the linear response range of 197 the instrument detector range. A strong feature of the sample spectra was the absorbance 198 pattern around 1250 and 1450 nm wavelengths. 199 A SNV (2,5,5,1) spectral pre-treatment was applied to the spectra of both red and white 200 grapes in the 950-1650 nm regions, where the hyperspectral system has revealed greater 201 efficiency. Mathematical treatment is denoted as a,b,c,d, where the first digit is the 202 number of the derivative; the second is the gap over which the derivative is calculated; 203 the third is the number of data points in a running average or smoothing, and the fourth 204 is the second smoothing (Shenk & Westerhaus, 1995). This spectral pre-treatment was 205 applied only in the calibration set and after that, principal component analysis was 206 carried out in order to look for spectral outliers and create cross-validation groups. 207 Overall, the spectral variability explained was 99% using 15 principal components and 208 Mahalanobis distances for each sample were calculated. Samples were ranked in order 209 16 for Fruit and Vegetable Quality Assessment. Food and Bioprocess Technology, 352 5, 1121-1142. 353 Meléndez, E., Ortiz, M. C., Sarabia, L. A., Íñiguez, M., & Puras, P. (2013). Modelling 354 phenolic and technological maturities of grapes by means of the multivariate 355 relation between organoleptic and physicochemical properties. Analytica 356 Chimica Acta, 761, 53-61. 357 Organisation internationale de la vigne et du vin (O.I.V.) Le recueil des méthodes 358 internationales d'analyse des vibs et des moûts (1990). O.I.V. 359 Shenk, J. S. & Westerhaus, M. O. (1995). Routine Operation,Calibration, Development 360 and Network System Management Manual, NIRSystems. MD, USA: Silver 361 Spring. 362 Singleton, V. L. (1985). Citation Classic - Colorimetry of Total Phenolics with 363 Phosphomolybdic-Phosphotungstic Acid Reagents. Current 364 Contents/Agriculture Biology & Environmental Sciences, 18. 365 Xu, Y. P., Simon, J. E., Welch, C., Wightman, J. D., Ferruzzi, M. G., Ho, L. et al. 366 (2011). Survey of Polyphenol Constituents in Grapes and Grape-Derived 367 Products. Journal of Agricultural and Food Chemistry, 59, 10586-10593. 368