Feasibility Study on the Use of Near-Infrared Hyperspectral Imaging for the Screening of Anthocyanins in Intact Grapes during Ripening

Depósito de investigación de la Universidad de Sevilla https://idus.us.es/ "This document is the Accepted Manuscript version of a Published Work that appeared in final form in JOURNAL OF AGRICULTURAL AND FOOD CHEMISTRY, copyright © American Chemical Society after peer review and technical editing by the publisher. To access the final edited and published work see https://doi.org/10.1021/jf4021637." 1 Feasibility study on the use of near infrared hyperspectral imaging for the screening of anthocyanins in intact grapes during ripening José Miguel Hernández-Hierro, Julio Nogales-Bueno, Francisco José Rodríguez-Pulido and Francisco José Heredia* Food Colour & Quality Laboratory, Department of Nutrition & Food Science, Universidad de Sevilla, Facultad de Farmacia, 41012 Sevilla, Spain. * Corresponding author: Francisco José Heredia Phone: +34 9545 56495 Fax: +34 9545 56110 E-mail: [email protected] Page 1 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 2 ABSTRACT 1 The potential of near infrared hyperspectral imaging to determine anthocyanins in intact 2 grape has been evaluated. The hyperspectral images of intact grapes during ripening 3 were recorded using a near infrared hyperspectral imaging covering the spectral range 4 between 900 and 1700 nm. Reference values of anthocyanins were obtained by HPLC-5 DAD. A number of spectral pre-treatments and different mask development strategies 6 were studied. Calibrations were performed by modified partial least squares regression 7 (MPLS) and present a good potential (RSQ of 0.86 and SEP values of 2.62 and 3.05 mg 8 g-1 of grape skin for non-acylated and total anthocyanins respectively) for a fast and 9 reasonably inexpensive screening of these compounds in intact grapes. 10 KEYWORDS: Phenolic compounds; anthocyanins; grapes; hyperspectral imaging; 11 near infrared; chemometrics. 12 13 Page 2 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 3 INTRODUCTION 14 Phenolics are characterized by presenting at least one aromatic group with one or more 15 hydroxyl groups attached. These plant bioactive compounds, commonly referred to as 16 phytochemicals, are present in the epidermis of leaves and the skin of fruits throughout 17 the plant kingdom. In plants, these compounds have important roles as secondary 18 metabolites such as UV protection, pigmentation, stimulation of nitrogen-fixing nodules 19 and disease resistance 1, 2. Among these phenolic compounds, anthocyanidins are vastly 20 distributed in fruit and flower tissue where they are responsible for red, blue and purple 21 colours 1. Grapes are a non-climateric fruit that follow three growing phases and contain 22 several phenolic compounds, which include anthocyanins in the case of red cultivars. 23 Anthocyanins are responsible for the colour of red wines and their interactions with 24 other phenolic compounds largely determine the colour changes observed during ageing 25 3-5. Furthermore, the influence of anthocyanins in the antioxidant activity of red wines 26 has also been previously reported 6. Biosynthesis and accumulation of anthocyanins in 27 red grape skins start at veraison (the interception between phases II and III) 7 and factors 28 such as cultivar, growing region, climate, and growth conditions may influence the 29 levels of anthocyanins 8-14. In warm climate regions, the aforementioned compounds are 30 negatively influenced by high solar radiation exposures and temperatures, which are 31 high not only during the day but also during the night 15. In that case, grape cultivar 32 selection is extremely important in order to face future problems with colour quality and 33 stability of red wines obtained from these berries 8. Moreover, deciding on the optimal 34 harvest time regarding anthocyanins is a great concern for wineries in order to obtain 35 optimal amounts of the aforementioned compounds. In this context, it may be important 36 to evaluate the changes that occur during ripening using rapid analytical methods in 37 order to decide the optimal harvest time. 38 Page 3 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 4 Hyperspectral imaging is an emerging and rapid technique for non-destructive food 39 analysis usually carried out in either the visible-short near infrared (vis-NIR; 400-1000 40 nm) or near infrared (NIR; 1000-1700 nm) spectral regions 16. The use of hyperspectral 41 analysis has risen considerably in the food sector in the recent past 16-19. Specifically, 42 hyperspectral image analysis in the visible-short wavelength near infrared (400-1000 43 nm) and adaptive boosting neural networks has been used to determine anthocyanins in 44 Cabernet Sauvignon grape obtaining promising results 20. Nonetheless, classic near 45 infrared spectroscopy has been used in the oenological sector to determine parameter in 46 grapes such as gluconic acid, glycerol, soluble solids and pH in grape juice 21, total 47 polyphenols, extractable anthocyanins, concentration of sugars, density 22, total 48 anthocyanins 23, 24 and the main families of phenolic compounds 25. Therefore, it would 49 be necessary to pay attention to the aforementioned electromagnetic range in the 50 development of hyperspectral imaging methods in grapes since near infrared 51 spectroscopy has been proved to be a powerful analytical tool to determine bioactive 52 compounds in a number of foodstuffs 26. 53 The aim of this study was to evaluate the potential of near infrared hyperspectral 54 imaging for the screening of anthocyanins in grapes during ripening. To our knowledge, 55 this is the first time that near infrared hyperspectral imaging has been applied to grapes 56 for this purpose. 57 MATERIAL AND METHODS 58 Samples 59 Vitis vinifera L. cv. Tempranillo and Syrah red grape samples were collected from two 60 vineyards located in the Condado de Huelva Designation of Origin (D.O.) (Andalusia, 61 Spain) which is under the typical climatic conditions of a warm area 8. Tempranillo is 62 the most often used variety to produce quality red wines in Spain and Syrah has been 63 Page 4 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 5 proven to be a well adapted cultivar to warm climatic conditions 8. Red grapes were 64 collected at different developmental stages during berry maturity in the 2012 vintage: 65 prior to veraison (July 16th) to over-ripening (September 6th). Sixteen dates were taken 66 into account for Tempranillo and seventeen for Syrah. Three groups of berries per 67 vineyard were collected at each date. Berries were collected from the top, middle and 68 bottom of the cluster and were immediately frozen and stored at -20 ºC until analyses 69 were performed. Two subsamples were randomly taken from each sample, one for the 70 HPLC analysis and the other one for the hyperspectral analysis. 71 Anthocyanins extraction and chromatographic analysis 72 Grape skins were then separated manually from the whole grapes. One gram of grape 73 skins was macerated at 4 ºC in methanol containing 0.1% of 12 M HCl. Methanolic 74 phases were centrifuged (3000 rpm, 10 min) and successively pooled, a few milliliters 75 of water were added and the extract was concentrated under vacuum at 30 ºC until 76 methanol was removed and finally made up to 10 mL with ultrapure water. The aqueous 77 extract was diluted 1:2 with 0.1 M HCl, filtered through 0.45 µm pore-size filters and 78 directly injected into the chromatographic system to determine the anthocyanins. 79 Anthocyanins chromatographic analysis was carried out following a modification of 80 García-Marino et al. 27. Chromatographic analyses were performed on a Hewlett-81 Packard 1200 Series HPLC equipped with an auto-injector, quaternary HPLC pump, 82 column heater, diode array detection (DAD) and data treatment station. A Zorbax SB 83 C18 column (4.6 mm x 250 mm, 4.6 µm particle size) thermostated at 35 °C was used. 84 Solvents were (A) 0.1% trifluoroacetic acid, and (B) 100% HPLC grade acetonitrile. 85 The elution profile was as follows: 10% B for 3.25 min, from 10 to 15% B for 12.37 86 min, 15% B for 5.21 min, from 15 to 18% B for 5.21 min, from 18 to 30% B for 20.84 87 min and from 30 to 35% B for 5.20 min. The flow-rate was 0.8 mL min-1 and the 88 Page 5 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 6 injection volume was 100 µL. The UV-vis spectra were recorded from 220 to 600 nm 89 with a bandwidth of 2 nm. The preferred detection wavelength was 520 nm. Up to 15 90 anthocyanins were identified according to their spectroscopic and chromatographic 91 features which had been previously acquired 27. The quantification was carried out from 92 the peak areas at the aforementioned preferred detection wavelength by an external 93 standard procedure. Results were expressed as malvidin-3-O-glucoiside equivalents. All 94 analyses were performed in duplicate. The standard error was generally around 10% so 95 the error and degree of accuracy of the reference method was considered appropriate to 96 use these data as reference values. 97 Total anthocyanins were expressed as the sum of the individual anthocyanins. 98 Moreover, anthocyanins were grouped taking into account their basic structures. Total 99 anthocyanins contents in the grape samples ranged from 0 to 23.8 mg g-1 of grape skin 100 with a standard deviation value of 5.6 mg g-1 of grape skin. Non-acylated anthocyanins 101 contents in the grapes samples ranged from 0 to 16.9 mg g-1 of grape skin with a 102 standard deviation value of 3.9 mg g-1 of grape skin. Acylated anthocyanins could be 103 calculated as the difference between total and non-acylated anthocyanins. 104 Hyperspectral imaging analysis 105 The main components of the hyperspectral imaging device (Infaimon S.L., Barcelona, 106 Spain) were the illumination source, optics (mirror scanner and lens), spectrograph, 107 camera and computer. The system comprised a Xenics® XEVA-USB InGaAs camera 108 (320 × 256 pixels; Xenics Infrared Solutuions, Inc., Leuven, Belgium), a spectrograph 109 (Specim ImSpector N17E Enhanced; Spectral Imaging Ltd., Oulu, Finland) covering the 110 spectral range between 900 and 1700 nm (spectral resolution of 3.25 nm), two 70W 111 tungsten iodine halogen lamps (Prilux ®, Barcelona, Spain) mounted as source light, a 112 mirror scanner (Spectral Imaging Ltd., Oulu, Finland) and a computer system. 113 Page 6 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 7 Hyperspectral images were recorded using a 50 Hz frame rate and an exposure time of 9 114 ms using the instrument acquisition software SpectralDAQ v. 3.62 (Spectral Imaging 115 Ltd., Oulu, Finland). 116 A two point reflectance calibration was used. A Spectralon ® ceramic tile (Labsphere 117 Inc., North Sutton, USA) was used as a white reference while dark current was recorded 118 by taking a measurement after covering the spectrograph lens with a cup and closing the 119 shutter. Corrected reflectance values (R) were calculated taking into account the 120 relationship between sample (S), white standard (W) and dark current (D) absolute 121 signal intensities using the following formula: 122 R=[(S-D)/(W-D)] (1) 123 Thereafter, the samples were thawed and tempered at room temperature and the 124 hyperspectral images of the intact grapes on a polyethylene plastic were recorded. The 125 characteristic spectral profile of this surface was useful in segmentation process for 126 recognising the region of interest. Only spectral data in the 950 - 1650 nm regions were 127 used in data analysis due to reduced efficiency outside this range. 128 Image processing and data analysis 129 Image processing. Image treatment was carried out using Matlab (R2010b; The Math 130 Works, Inc. USA). Figure 1 shows a calibrated image at 1360 nm, different parts of this 131 image and the spectrum of each part. Prior to the quantitative analysis, a thresholding 132 rule method was applied to the grape images to isolate the grapes from other parts of 133 image. Firstly, three regions of interest (ROIs) were selected (background, grape and 134 pedicel) to develop a stepwise lineal discriminant model. The aforementioned 135 discriminant model classified each pixel into two classes (grape or no grape pixel) using 136 the reflectance values from six wavelengths (979, 1034, 1073, 1314, 1386 and 1550 137 nm). After that, the average spectrum of the grape region was extracted and then 138 Page 7 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 8 transformed into Log (1/R) units. The procedure was repeated for each sample and the 139 obtained spectra were combined into the spectral matrix. A flowchart of the image 140 processing strategy employed in this study is shown in Figure 2. 141 Data analysis. Prior to quantitative analysis, an unsupervised pattern recognition 142 technique, principal component analysis (PCA), was used in order to provide 143 information about the latent structure of spectral matrix. This method provides not only 144 information related to spectral outliers and the distribution of samples in the newly-145 created space but is also an important source of knowledge with which to create cross-146 validation groups used in the calibration process 28, 29. 147 Using the raw spectral data and testing different spectral pre-treatments and allocating 148 the corresponding total anthocyanins or non-acylated anthocyanins values to each 149 sample, calibrations were performed by modified partial least squares regression 150 (MPLS). In this method, the group of calibration samples is divided into a series of 151 subsets in order to perform cross-validation to set the number of PLS factors, reduce the 152 possibility of overfitting 28 and remove chemical outliers. Using the T≥2.5 criterion, 153 samples that presented a high residual value when they were predicted were eliminated 154 from the set. Finally, validation errors are combined into a single figure, the standard 155 error of cross-validation (SECV). 156 Spectral pre-treatments are usually applied to NIR raw data; scattering effects were 157 removed using multiplicative scatter correction (MSC), standard normal variate (SNV), 158 and detrending 30, 31. Moreover, the effect of derivatisation and variations in spectral 159 ranges were tested in the development of the NIRS calibrations. 160 The software used was Win ISI® (v1.50) (Infrasoft International, LLC, Port. Matilda, 161 PA, USA). This software allowed the data pre-treatment and development of principal 162 components and quantitative models. From the three samples of each date, one (33%) 163 Page 8 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 15 Figure captions Figure 1. (A) Calibrated image at 1360 nm, (B) different parts of this image and (C) the spectrum of each part. Figure 2. Flow chart of the data processing strategy employed in this study. Figure 3. Average and standard deviation (10 times amplified) spectra of the whole group of grapes (99 samples) in the NIR zone between 950 and 1650 nm. Figure 4. Score plot of grape samples (calibration set) in the space defined by PC1 and PC2. Figure 5. (A) Average spectrum of the best of the different pre-treatment (SNV 2,5,5,1). (B) Loading plots of the MPLS model for total anthocyanins prediction. (C) Malvidin-3-O-glucoiside and its second derivative spectra. Page 15 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry 16 Table 1. Calibration Statistical Descriptors for the Models Developed in the NIR Zone Close to 950-1650 nm. Spectral pre-treatments Compounds T outliers PLS factors Na Est. Min SDb Est. Max SECc RSQd SECVe SEPf (mg g - 1 skin) (mg g - 1 skin) SNV 2 , 5 , 5 ,1 N on - ac ylated anthocyanins 4 5 62 0 3.42 15.6 1 1.27 0.86 1.70 2.62 SNV 2 , 5 , 5 ,1 Total anthocyanins 3 5 63 0 4.95 22.82 1.84 0.86 2.41 3.05 aN: number of samples (calibration set); bSD: standard deviation; cSEC: standard error of calibration; dRSQ: coefficient of determination (calibration set); eSECV: standard error of cross-validation; fSEP: standard error of prediction (external validation). Page 16 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry Figure 1. (A) Calibrated image at 1360 nm, (B) different parts of this image and (C) the spectrum of each part. 140x120mm (300 x 300 DPI) Page 17 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry Figure 2. Flow chart of the data processing strategy employed in this study. 119x212mm (96 x 96 DPI) Page 18 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry Figure 3. Average and standard deviation (10 times amplified) spectra of the whole group of grapes (99 samples) in the NIR zone between 950 and 1650 nm. 102x71mm (300 x 300 DPI) Page 19 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry Figure 4. Score plot of grape samples (calibration set) in the space defined by PC1 and PC2. 265x105mm (300 x 300 DPI) Page 20 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry Figure 5. (A) Average spectrum of the best of the different pretreatment (SNV 2,5,5,1). (B) Loading plots of the MPLS model for total anthocyanins prediction. (C) Malvidin-3-O-glucoiside and its second derivative spectra. 254x187mm (300 x 300 DPI) Page 21 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry TOC Graphic. 300x89mm (96 x 96 DPI) Page 22 of 22 ACS Paragon Plus Environment Journal of Agricultural and Food Chemistry