Use of Visible and Short-Wave Near-Infrared ... - ACS Publications

Sep 22, 2016 - Use of Visible and Short-Wave Near-Infrared Hyperspectral Imaging. To Fingerprint Anthocyanins in Intact Grape Berries. Maria P. Diago,...
0 downloads 0 Views 1MB Size
Subscriber access provided by CORNELL UNIVERSITY LIBRARY

Article

Use of visible and short wave near-infrared hyperspectral imaging to fingerprinting of anthocyanins in intact grape berries MARIA PAZ DIAGO SANTAMARIA, Juan Fernandez-Novales, Armando M. Fernandes, Pedro Melo-Pinto, and Javier Tardaguila J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jafc.6b01999 • Publication Date (Web): 22 Sep 2016 Downloaded from http://pubs.acs.org on September 22, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Agricultural and Food Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 35

Journal of Agricultural and Food Chemistry

1

Use of visible and short wave near-infrared hyperspectral imaging to

2

fingerprinting of anthocyanins in intact grape berries

3

Maria P. Diago1*, Juan Fernández-Novales1, Armando M. Fernandes2, Pedro Melo-

4

Pinto3,4, Javier Tardaguila1.

5

1

6

La Rioja). Finca La Grajera, Ctra. Burgos Km. 6, 26007, Logroño, Spain.

7

2

INOV – INESC Inovação, Rua Alves Redol, 9, 1000-029 Lisboa, Portugal.

8

3

CITAB-Centre for the Research and Technology of Agro-Environmental and

9

Biological Sciences. Universidade de Trás-os-Montes e Alto Douro, Quinta de Prados,

Instituto de Ciencias de la Vid y del Vino (University of La Rioja, CSIC, Gobierno de

10

5000-911 Vila Real, Portugal.

11

4

12

de Prados, 5000-911 Vila Real, Portugal.

13

*Corresponding author: Maria Paz Diago; [email protected]; Tel: +34

14

941894980 (ext.410065); Fax: +34 941899728.

Departamento de Engenharias, Universidade de Trás-os-Montes e Alto Douro, Quinta

15

16

17

18

19

20

21

1 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

22

ABSTRACT

23

In red grape berries, anthocyanins account for about the 50% of the skin phenols and are

24

responsible for the final wine colour. Individual anthocyanin levels and compositional

25

profiles vary with cultivar, maturity, season, region and yield and have been proposed

26

as chemical markers to differentiate wines and to provide valuable information

27

regarding the adulteration of musts and wines. A fast, easy, solvent-free, non-

28

destructive method based on visible, short wave and near infrared hyperspectral imaging

29

(HSI) in intact grape berries to fingerprinting the color pigments in eight different grape

30

varieties was developed and tested against HPLC. Predictive models based on modified

31

partial least squares (MPLS) were built for 14 individual anthocyanins with coefficients

32

of determination of cross validation (R2cv) ranging from 0.70 to 0.93. For the grouping

33

of total and non-acylated anthocyanins, external validation was conducted with

34

coefficient of determination of prediction (R2P) of 0.86. HSI could potentially become

35

an alternative to HPLC with reduced analysis time and labour costs, while providing

36

reliable and robust information of the anthocyanin composition of grape berries.

37

38

KEYWORDS: grape color, anthocyanin profile, non-destructive method, contactless

39

spectroscopy, chemometrics

40

41

42

43

44

2 ACS Paragon Plus Environment

Page 2 of 35

Page 3 of 35

Journal of Agricultural and Food Chemistry

45

INTRODUCTION

46

Anthocyanins are key secondary metabolites in grape berries which determine grape

47

and wine color1.They are included in the group of flavonoids, a type of phenolic

48

compounds whose base structure (2-phenyl-benzopyrilium) consists of two aromatic

49

rings joined via a pyran ring. Although fifteen anthocyanidins are known to occur in

50

nature2, only five are generally present in red grape berries, primarily in the first cellular

51

layers of the hypodermis3. These anthocyanidins include delphinidin, peonidin,

52

cyanidin, petunidin and malvidin, which is the most abundant in Vitis vinifera L.

53

varieties. These pigments exist primarily as the corresponding monoglucosides in red

54

grape berries, with only small quantities of acylated forms involving phenolic acids. In

55

these cases, the sugar molecule is esterified with caffeic, acetic or coumaric acid4, the

56

latter two being the most common.

57

In red varieties, anthocyanins account for about the 50% of the skin phenols.

58

However, individual anthocyanin levels and compositional profiles vary greatly with

59

cultivar, maturity, season, region and yield5. Qualitative and quantitative anthocyanin

60

patterns in grape skins have been widely studied by chromatographic techniques,

61

initially by paper chromatography6 and later by HPLC7,8. Anthocyanins have been

62

postulated as chemical markers to differentiate wines made from different varieties,

63

providing valuable information regarding the adulteration of juices and wines9. HPLC

64

methods require berry processing and solvent extraction of the anthocyanin compounds

65

prior to analysis, which are time consuming tasks that require skilled operators. In this

66

context, it may be valuable for the wine industry to have fast, easy, solvent-free, non-

67

destructive methods to assess the anthocyanin changes and accumulation that occur in

68

the berries during ripening as well as to fingerprinting the color pigments in different

69

grapevine varieties. 3 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

70

Hyperspectral imaging is a non-destructive spectroscopic technique that

71

measures hundreds of narrow wavelength bands and spatial positions. At each spatial

72

location, the interaction of the electromagnetic radiation with the matter at many

73

different wavelengths is recorded. In recent years, hyperspectral imaging in the visible

74

and near infrared (400-1700 nm) has been adopted for fruit and vegetable quality

75

assessment10-12, food safety control13-15 and classification tasks16,17. In the field of

76

viticulture, varietal18,19 and clone discrimination within a given grapevine variety20 was

77

achieved by visible (VIS) hyperspectral imaging. Likewise, total anthocyanin

78

concentration21-24, technological maturity and total phenolics25, 26 in grape berries and

79

several phenolic substances in grape seeds and pomace27,28 were determined using VIS-

80

NIR hyperspectral imaging.

81

Using a small number of grape berries might be interesting for some wineries

82

aiming to produce high quality wines by selecting, within each cluster, the berries with

83

highest quality29-30. Automating the berry selection procedure might already be possible

84

with existing destemmers that extract berries one-by-one from clusters31 and afterwards

85

place them on a conveyor belt that passes under a hyperspectral camera. However,

86

quantification of individual anthocyanin compounds and profiling of the anthocyanin

87

pattern in intact berries has rarely been attempted.

88

Therefore, the aim of the present work was to evaluate the capability of VIS

89

short wave NIR hyperspectral imaging in intact berries to anthocyanin fingerprinting in

90

several grapevine varieties. This could become an alternative method to HPLC methods,

91

which could be very useful for the wine industry.

92 93

MATERIALS AND METHODS

4 ACS Paragon Plus Environment

Page 4 of 35

Page 5 of 35

Journal of Agricultural and Food Chemistry

94

Berry Samples

95

Grape clusters of eight red Vitis vinifera L. varieties (three clusters per variety),

96

Cabernet Franc, Cabernet Sauvignon, Grenache, Merlot, Petit Verdot, Pignolo, Rebo

97

and Sangiovese were handpicked at commercial harvest (October 2011) in two different

98

sites located in the DOCa Rioja (Spain). Site 1 was a commercial vineyard in Tudelilla

99

(Spain) (42º 18' 49.1'' N, -2º 8' 9.2'' W, 579 m above sea level) and vineyard 2 was an

100

experimental vineyard located in Mendavia (Spain) (42º 27' 53.7'' N, -2º 17' 28.2'' W,

101

343 m above sea level). After harvesting, clusters were immediately transported in

102

portable refrigerators to the University of La Rioja and kept in a cool room at 5 °C until

103

imaging. Ten berries per cultivar (80 berries in total) were randomly detached from the

104

clusters and allowed to warm for 15 minutes at room temperature before hyperspectral

105

imaging.

106 107

Hyperspectral Image Acquisition

108

Hyperspectral images of the individual grape berries were acquired using a

109

hyperspectral camera under controlled illumination conditions in the laboratory. The

110

hyperspectral camera was comprised of the Specim Imspector V10E (Specim, Oulu,

111

Finland) spectrograph that decomposes light into its different wavelengths, and a JAI

112

Pulnix (JAI, Yokohama, Japan) black and white camera. This imaging system covered

113

the spectral range of 380 – 1028 nm with the spectral channels having a 0.6 nm spacing.

114

The hyperspectral camera acquired 1392 pixels in the spatial dimension and 1040 pixels

115

(channels) in the wavelength dimension. The length of the imaged line over the sample

116

was 110 mm, and the distance between the hyperspectral camera and the sample to be

117

imaged was 400 mm. Image acquisition was done using Coyote software (Version

118

2.2.0, JAI, Japan) at a frequency of eight images per second. Spectral calibration was

5 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

119

conducted following the manufacturer procedure (EHE 2006), and spatial calibration

120

using a black and white target. The lighting set up was comprised of four 20 Watts, 12

121

Volts halogen lamps and two 40 Watts, 220 Volts reflector lamps (Spotline, Philips,

122

Eindhoven, Nederlands) powered by continuous current supply to avoid light flickering.

123

The blue spot lamps were necessary for improving the illumination quality at

124

wavelengths between 800 and 1000 nm, a region where halogen lamp illumination was

125

insufficient. Each berry was placed on a fruit holder and positioned at the center of the

126

field of view of the camera, with the pedicel perpendicular to the camera lens to avoid

127

any discrepancy between the berry surface and the pedicel. Each berry was

128

photographed on four sides (Figure 1A), by rotating the berry 90º between each image

129

acquisition. For each side of the berry, 32 images were taken. In total, for each berry

130

128 images were acquired. All imaged berries were individually weighed, stored in a

131

plastic bag and frozen at -20 ºC prior to chemical analysis of their anthocyanin

132

composition.

133 134

Spectral data collected from the CCD device provided digital numbers

135

corresponding to the signal intensity and not actual reflectance values. Therefore, the

136

raw data were transformed into reflectance (R) units in a first step and subsequently into

137

absorbance (A) units (A=Log 1/R). Image correction was carried out by acquiring white

138

and dark reference images. The dark current reference image was obtained by

139

completely closing the lens of the camera with its opaque cap, while the white reference

140

image was acquired from a uniform, stable and a high white reflectance reference

141

(99.9% diffuse reflectance), called Spectralon® (Specim, Oulu, Finland). Reflectance

142

for a certain position ṝ and wavelength  was calculated according to equation 1:

143

6 ACS Paragon Plus Environment

Page 6 of 35

Page 7 of 35

Journal of Agricultural and Food Chemistry

144

(ṝ, ) =

(ṝ, ) (ṝ, ) (ṝ, ) (ṝ, )



(1)

145 146

Where G is the intensity of the light reflected by the berry, W is the intensity of the light

147

coming from the white reference, and D is the dark current. Ambient light was

148

eliminated by performing the experiments in a dark room. The reference point for ṝ was

149

located outside the berry that was being imaged and corresponded to the position where

150

the hyperspectral image started for image segmentation. To facilitate segmentation the

151

berries were imaged using the Spectralon target as a background. The segmentation was

152

done automatically by analysing the variation of reflectance over the spatial dimension

153

at the wavelength of approximately 630 nm where there is a sharp difference between

154

the Spectralon and the berry reflectance.

155 156

Anthocyanin Extraction and Chromatographic Analysis

157

Extraction and analysis of the anthocyanins for the identification of the anthocyanin

158

profile and their quantification was conducted separately for each individual berry. The

159

extraction protocol was adapted for individual berries from the procedure described

160

elsewhere32. The first step involved the dissection of each berry. While still frozen, the

161

skin of each berry was manually detached using a scalpel, and subsequently blotted dry

162

with a kimwipe. The skin was then placed in a 15 mL Falcon centrifuge tube, added

163

some milliliters of liquid nitrogen and milled for 10 seconds with a glass stirring bar

164

until a powder was obtained. For the anthocyanin extraction, 10 mL of methanol

165

(Methanol LC-MS grade, HiperSolv, VWR, Radnor, PE, USA) acidified at 0.1% (v/v)

166

with HCl 12M were added to the centrifuge tube. This was then vortexed and placed in

167

the refrigerator (at 5ºC) covered with aluminium foil for 24 hours. After this time, the

168

supernatant was moved to a 50 mL Falcon tube and 4 mL of the acidified methanol 7 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

169

were added to the 15 mL Falcon tube containing the berry skin. This procedure was

170

repeated during eight consecutive days (all methanolic phases were pooled into the 50

171

mL Falcon tube), until the absorbance value of the supernatant determined at 520 nm

172

using a UV-VIS spectrophotometer (DR 5000, Hach-Lange, Loveland, CO, USA) was

173

less or equal to 0.002 AU. The 50 mL Falcon tube containing the anthocyanin extract

174

was kept inside the refrigerator at 5ºC covered with aluminium foil at all times until

175

their content was transferred to an evaporation flask and concentrated under vacuum at

176

30ºC using a rotary evaporator (Buchi Heating Bath B-490, Buchi Rotvapor R-200,

177

Flawil, Switzerland) until methanol was removed. The extract was then transferred to a

178

2 mL volumetric flask with at Pasteur pipet and rinsed with several fractions of 100µL

179

of Milli-Q™ water (Millipore, Bedford, MA, USA), that were also transferred to the

180

volumetric flask until a final extract volume of 2 mL. The 2 mL extracts of the 80

181

berries were filtered through 0.45 µm pore-size PTFE filters to 2 mL Eppendorf vials

182

and stored at -20ºC until chromatographic analysis.

183 184

Chromatographic analyses were carried out on a Shimadzu Nexera (LC-30AD),

185

equipped with auto-injector (SIL-30AC), quaternary HPLC pump, column heater (CTO-

186

20-AC), diode array (DAD; SPD-M20A) and mass-spectrometer (MS) (AB-Sciex 3200

187

Q-Trap) detectors. A LiChrosphere 100 RP-18 reverse phase column (5 µm packing,

188

250 x 4 mm i.d.) protected with a guard column of the same material (Scharlab,

189

Barcelona, Spain) thermostated at 35 ºC was used. The solvents used were: (A) Milli-Q

190

water at 2% formic acid (w/w), (B) acetonitrile/solvent A (80:20, v/v) establishing the

191

following gradient: isocratic 100% A in 2 min, from 100 to 92% A in 3 min, from 92 to

192

86% A in 12 min, from 86 to 82% A in 5 min, from 82 to 79% A 7.5 min, from 79 to 67

193

% A 25.5 min, from 67 to 50 % A in 15.5 min, 50 to 20 % A in 2.5 min, isocratic 20%

8 ACS Paragon Plus Environment

Page 8 of 35

Page 9 of 35

Journal of Agricultural and Food Chemistry

194

A during 5 min, from 20 to 100% A in 3 min, and isocratic 100 % A in 9 min at a flow

195

rate of 1 mL min–1. Quantification was accomplished by HPLC-DAD at 520 nm using

196

an external calibration curve of Malvidin-3-O-glucoside chloride (> 95% HPLC,

197

Extrasynthèse, Genay, France). Fifteen individual anthocyanins were identified

198

according to their order of elution and MS transitions. These were: Malvidin 3-O-

199

glucoside (MvGl); Malvidin 3-O-(6-acetyl)glucoside (MvGlAc); Malvidin 3-O-(6-p-

200

coumaroyl)glucoside (MvGlCm); Petunidin 3-O-glucoside (PtGl); Petunidin 3-3-O-(6-

201

acetyl)glucoside (PtGlAc); Petunidin 3- O-(6-p-coumaroyl)glucoside (PtGlCm);

202

Delphinidin 3-O- glucoside (DpGl); Delphinidin 3-O-(6-acetyl)glucoside (DpGlAc);

203

Delphinidin 3-O-(6-p-coumaroyl)glucoside (DpGlCm); Peonidin 3-O-glucoside (PnGl);

204

Peonidin 3-O-(6-acetyl)glucoside (PnGlAc); Peonidin 3-O-(6-p-coumaroyl)glucoside

205

(PnGlCm);

206

(CyGlAc); Cyanidin 3-O-(6-p-coumaroyl)glucoside (CyGlCm). Total anthocyanins

207

(Total Anth) were calculated as the sum of the fifteen individual compounds, while

208

total-3-glucoside (Total Gl Anth), total 3-O-(6-acetyl)glucoside anthocyanins (Total

209

GlAc Anth) and total 3- O-(6-p-coumaroyl)glucoside anthocyanins (Total GlCm Anth)

210

contents involved the addition of the 3-glucosidated, 3-O-(6-acetyl)glucoside

211

anthocyanins, and 3-O-(6-p-coumaroyl)glucoside anthocyanins, respectively.

212

Data Analysis

213

The averaged spectrum for each grape was used as input for a principal component

214

analysis (PCA), an unsupervised pattern recognition technique, in order to provide

215

information about the latent structure of spectral matrix, to find spectral differences

216

among all spectral samples and also to visualize the presence of outliers

217

outlier detection was performed based on Global Mahalanobis (GH) distance analysis,

218

i.e. Mahalanobis distance between the center of the population and each sample in the

Cyanidin

3-O-glucoside

(CyGl);

Cyanidin

9 ACS Paragon Plus Environment

3-O-(6-acetyl)glucoside

33,34

. Spectral

Journal of Agricultural and Food Chemistry

219

space defined by a PCA based on the spectra information. Samples with GH values

220

larger than 3 were considered outliers

221

Multiplicative Scatter Correction (MSC), Standard Normal Variate (SNV) and

222

Detrending to remove the effects of scattering

223

treatments were tested in the development of the calibrations and denoted as a four-

224

figure-code (a, b, c, d), where a was the number of the derivative; b was the gap over

225

which the derivative is calculated; c was the number of data points in a running average

226

or smoothing, and d corresponded to the second smoothing 38.

35

. The spectral data were pre-treated with

36,37

. Moreover, several mathematical

227 228

Modified Partial Least Squares (MPLS)39 regression was tested for the

229

prediction of the 15 individual pigments and the four groups of anthocyanins using the

230

average spectra from four different positions of the grape berries in the 380 - 1028 nm

231

range. In order to prevent overfitting, 6-fold cross-validation was conducted. For this,

232

the calibration set was divided into six groups and each group was then validated using

233

the calibration developed with the other samples. In this process another type of outliers

234

were analyzed using Student’s T statistic, which indicates the difference between the

235

reference and the predicted value. A critical limit of T > 2.5 was used to identify

236

samples as chemical outliers38. Different pre-processing combinations based on

237

derivatives, window-wise filtering and scatter correction methods were evaluated for

238

building the calibration models.

239 240

The following statistics were used to select the most adequate models: Standard

241

Error of Calibration (SEC), Standard Error of Cross-Validation (SECV), and

242

Determination Coefficient of Cross-Validation (R2cv). Additionally, the Residual

243

Predictive Deviation (RPDCV), calculated as the ratio between the standard deviation of

10 ACS Paragon Plus Environment

Page 10 of 35

Page 11 of 35

Journal of Agricultural and Food Chemistry

244

the reference data for the training set and the Standard Error of Cross-Validation

245

(SECV) was also computed. According to Williams and Sobering40 a model is suitable

246

for screening purposes – i.e. greater precision yields – if the RPD value is greater than

247

3.

248 249

WinISI II software package version 1.50 (Infrasoft International, Port Matilda,

250

PA, USA) and the Unscrambler X software package version 10.3 (CAMO ASA, Oslo,

251

Norway) were used for chemometric analysis. Samples for the calibration and external

252

validation sets were selected using the CENTRE algorithm included in the WinISI II

253

software package. This algorithm performs a principal component analysis, reducing the

254

original spectral information (log 1/R values) to a small number of linearly independent

255

variables, thus facilitating the calculation of spectral distances. These new variables

256

were used to calculate the center of the spectral population and the distance (expressed

257

as the Mahalanobis distance) of each sample in the calibration set from that centre.

258

Having ordered the samples from the two vineyard sites (N = 80) by spectral distances

259

(from smallest to greatest distance to the centre), the 20 samples forming the external

260

validation set were selected by taking one sample out of every four in the global set,

261

leaving a calibration set comprising 60 samples. No anomalous spectra or outliers were

262

identified38. The best-fitting equations, obtained from the four anthocyanin groups for

263

the calibration set, were subsequently evaluated by external validation using samples

264

not involved in the calibration procedure.

265 266

RESULTS AND DISCUSSION

267 268

Anthocyanin Composition of Individual Berries 11 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

269

Fifteen individual anthocyanins were identified and the sums of the groups with

270

different glycosylated moieties were computed (Table 1). Total Anth ranged from

271

505.82 mg L-1 in Petit Verdot berries, to 2463.18 mg L-1 in Rebo berries. On average,

272

Total Gl Anth represented 91.2% of Total Anth, while Total GlAc Anth 1.0% and Total

273

GlCm Anth a 7.8%. Of the 3-glucosidated anthocyanins, MvGl was the majority

274

pigment and CyGl the minority one, across all studied varieties, while differential

275

profiles for PtGl, DpGl and PnGl were found among varieties (Table 1). In absolute

276

terms, DpGlAc was the anthocyanin with the smallest concentration of the 15 identified

277

pigments, with values below the quantification limit in Cabernet Franc, Cabernet

278

Sauvignon, Grenache, Petit Verdot and Pignolo.

279

The correlation coefficients between the concentrations of all individual and

280

groups of anthocyanins were also computed and are shown as supporting information in

281

the Supplementary Table 1 (Table S1).

282 283

Spectral Features

284

The spectra recorded from the imaged four positions of the berries are displayed in

285

Figure 1B. The absorbance values of the spectra corresponding to two of the four grape

286

berry rotating positions were substantially different than those of the other two positions

287

(Figure 1B). Within a cluster, given the spherical shape of the berries, their whole

288

surface is not equally exposed to the solar radiation. Therefore, spectra with higher

289

absorbance values may correspond to the part of the berry exposed, while those spectra

290

with lower absorbance would potentially correspond to the part of the berry facing the

291

inner section of the cluster. For this reason the spectra from all pixels in the 128 images

292

per berry were averaged to a unique mean spectrum per berry, as this is meant to more

12 ACS Paragon Plus Environment

Page 12 of 35

Page 13 of 35

Journal of Agricultural and Food Chemistry

293

accurately represent the average anthocyanin content and distribution of the whole

294

berry.

295 296

Typical absorbance spectra collected of the whole set of berries in the zone

297

between 380 and 1034 nm are shown in Figure 2A. The effect of derivative was mostly

298

apparent for the first derivative of the spectrum, which enabled separating the

299

overlapping absorption bands, and revealed certain characteristic absorbance peaks

300

around 550 nm, 696 nm, 720nm and 990 nm wavelengths and an inflection point around

301

600 nm (Figure 2A, B). The spectral regions between 400 and 550 nm substantially

302

contributed to the loadings of the model and are characteristic of chemical compounds

303

whose structure involves the presence of aromatic ring compounds, such as the

304

anthocyanins, which exhibit their maximum peak of absorbance around 520 nm,

305

although small shifts may occur for each specific anthocyanin compound41. Also in this

306

region the main absorbance bands of chlorophyll a (maximum at 430 nm) and b

307

(maximum at 453 nm), as well as those of carotenoids can be found 42, 43. Additionally,

308

the high absorbance values between 600 and 700 nm were also indicative of an

309

abundance of chlorophylls, as both chlorophyll a and b are described to exhibit a local

310

maximum at 662 and 642 nm, respectively41. The absorbance band observed between

311

960 and 990 nm could also be related to the water content in the berries45, which

312

constitutes 70 to 80 % of the grape berry mass.

313 314

Development of Calibration, Cross Validation and Prediction Models

315

Table 2 summarizes the range, mean value and standard deviation of each individual

316

anthocyanin concentration and those of the groups, for the global set of samples of all

317

varieties. All individual anthocyanins were well represented with a high variability,

13 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

318

having a similar number of samples (between 74 and 80), with the exception of

319

MvGlAc, DpGlAc and CyGlAc, with cardinality values equal or less than 62. Special

320

attention deserves the DpGlAc (Table 2), whose range was inferior to 1mg L-1. For this

321

anthocyanin, only 15 values could be recorded, as in the remaining 65 berry samples

322

DpGlAc concentration was below the quantification threshold.

323 324

Table 3 shows the calibration, cross-validation and external prediction statistics

325

of the best models for the prediction of 15 individual anthocyanins and four anthocyanin

326

groups using the combination of signal pretreatments that yielded the best results in

327

each case. In this table, N, minimum and maximum refer to the number of samples,

328

minimum and maximum values of each dataset used in the model after removal of

329

chemical outliers, which accounted for 3-16% of initial data. The model of DpGlAc

330

could not be built due to its reduced number of samples. Additionally, the loadings for

331

all calibration models for individual and groups of anthocyanins are plotted in Figure S1

332

of the Supporting Information.

333 334

While the calibration models for all anthocyanins and groups showed

335

determination coefficients (R2c) above 0.90, with the exception of PnGlAc and CyGlAc,

336

the best prediction models exhibited a determination coefficient of cross validation

337

(R2cv) between the 0.77 and 0.93 marks (Table 3). The SECV values must be analyzed

338

in association with the SD and range of the parameters studied. The ratio of

339

performance to deviation (RPD) was used to indicate the prediction capacity of the

340

models. The RPD values ranged from 2.33 to 6.51, indicating that the performance of

341

the calibration models for most of the individual pigments and groups of anthocyanins

342

was very satisfactory, and fitted to the recommended values (close or larger than three).

14 ACS Paragon Plus Environment

Page 14 of 35

Page 15 of 35

Journal of Agricultural and Food Chemistry

343

Moreover, the relative low number of PLS factors (between four and seven) for most of

344

the compounds and the fact that the models were built using grape berries of eight

345

different varieties contributed to the robustness of the models. These results are very

346

remarkable, as they indicate that from a unique spectrum between 380 and 1028 nm,

347

obtained from hyperspectral imaging of a berry, the whole anthocyanin profile and

348

composition of that berry can be reliably obtained without the need of wet chemistry

349

extraction using solvents and further HPLC analysis. Other works have also reported

350

very good linear models to predict the total anthocyanin content in intact grape berries,

351

with R2CV above 0.95, using spectral measurements in the VIS (400 to 800nm) and NIR

352

range (1100 to 2000nm)32 or the total phenolic content, with R2CV = 0.89, using spectral

353

measurements in the NIR range of 900 to 1700 nm26, 46. Non-linear models were also

354

successful in predicting the total anthocyanin content in grape berries of Cabernet

355

Sauvignon using hyperspectral imaging in the range of 400 to 1000 nm21. Additionally,

356

in the current study both the spectroscopic and HPLC measurements were carried out on

357

a single berry, instead of multiple-berry samples21,23,32. This is relevant because the

358

smaller the berry sample size, the harder is to obtain accurate enological results. The

359

approach of analyzing berries individually would mimic a potential scenario of a sorting

360

table where berry selection (after destemming) could be driven from the total

361

anthocyanin content provided by a contactless, hyperspectral camera above the

362

conveyor belt. This approach can be considered to have a potentially large impact on

363

grape composition monitoring and control of continuously operating sorting belts which

364

deliver the berries to one fermentation tank or another depending on the type of wine

365

aimed.

366

15 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

367

Nevertheless, modelling and profiling of individual anthocyanin compounds was

368

seldom reported. The closest approach, but focused on the phenolic constituents other

369

than anthocyanins was attempted using NIR hyperspectral imaging (950 to 1650nm) in

370

samples of Vitis vinifera L. Zalema grape skins, stems and seeds, separated from

371

winemaking pomace28. In this study, R2cv values above 0.90 were obtained for

372

epicatechin, individual hydroxicinnamic acids, such as caffeic, caftaric and trans-

373

coutaric acids, and kaempferol derivatives, but previous freeze-drying of the samples

374

was necessary to obtain reliable results.

375

In order to evaluate the best models an external validation was developed to

376

predict the two main global anthocyanin groups. The robustness of the selected models

377

was tested using external validation set samples, which did not belong to the internal

378

validation set (Table 3). For the internal validation (data not shown) N = 60 samples

379

while for the external validation sets N = 20 samples. A relatively wide range was

380

present in both data sets. Structured selection with the CENTRE algorithm - only using

381

spectral information treatment algorithms - proved adequate, since the internal and

382

external validation sets displayed similar values for mean, range and standard deviation

383

for all the parameters studied. Regarding the external validation, the determination

384

coefficients of prediction (R2P) were above 0.80 for eight individual anthocyanins as

385

well as for Total Anth and Total Gl Anth (Table 3). These results are in agreement with

386

previous works22-24 who examined the potential of NIR (900-1700 nm) hyperspectral

387

imaging for the screening of total and non-acylated anthocyanins in intact berries of

388

several cultivars. The best predictive models for the two major anthocyanin groups,

389

Total Anth and Total Gl Anth for both internal and external validation sets along with

390

the prediction lines at 95% of confidence are displayed in Figure 3. A high data

391

dispersion together with a good fit can be seen along the correlation lines. 16 ACS Paragon Plus Environment

Page 16 of 35

Page 17 of 35

Journal of Agricultural and Food Chemistry

392 393

The potential of VIS-NIR hyperspectral imaging for the screening and profiling

394

of acylated and non-acylated individual anthocyanins and their sums, as well as total

395

anthocyanins in intact grape berries was investigated and confirmed. The developed

396

models were robust and applicable to berries of eight different varieties, but further

397

examination of the influence of more grapevine cultivars, seasons and origins should be

398

conducted with the aim of developing even more robust, global predictive models. This

399

non-destructive, spectral technology could potentially be considered as an alternative to

400

widely adopted HPLC analysis. This could be beneficial in terms of reducing the

401

analysis time and labour costs and avoid the use of solvents, while providing reliable

402

and robust information of the anthocyanin composition of grape berries.

403 404

ABBREVIATIONS

405

MvGl: Malvidin 3-O-glucoside

406

MvGlAc: Malvidin 3-O-(6-acetyl)glucoside

407

MvGlCm: Malvidin 3-O-(6-p-coumaroyl)glucoside

408

PtGl: Petunidin 3-O-glucoside

409

PtGlAc: Petunidin 3-O-(6-acetyl)glucoside

410

PtGlCm: Petunidin 3-O-(6-p-coumaroyl)glucoside

411

DpGl: Delphinidin 3-O-glucoside

412

DpGlAc: Delphinidin 3-O-(6-acetyl)glucoside

413

DpGlCm: Delphinidin 3-O-(6-p-coumaroyl)glucoside

414

PnGl: Peonidin 3-O-glucoside

415

PnGlAc: Peonidin 3-O-(6-acetyl)glucoside

416

PnGlCm: Peonidin 3-O-(6-p-coumaroyl)glucoside

17 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

417

CyGl: Cyanidin 3-O-glucoside

418

CyGlAc: Cyanidin 3-O-(6-acetyl)glucoside

419

CyGlCm: Cyanidin 3-O-(6-p-coumaroyl)glucoside

420

Total Anth: Total anthocyanins

421

Total Gl Anth: Total 3-O-glucoside anthocyanins

422

Total GlAc Anth: Total 3-O-(6-acetyl)glucoside anthocyanins

423

Total GlCm Anth: Total 3-O-(6-p-coumaroyl)glucoside anthocyanins

424

DAD: Diode array detector

425

HPLC: High performance liquid chromatography

426

HSI: Hyperspectral imaging

427

MPLS: Modified partial least squares

428

MSC: Multiplicative scatter correction

429

NIR: Near infrared radiation

430

PCA: Principal component analysis

431

PTFE: Polytetrafluoroethylene

432

R: Reflectance

433

R2CV: Determination coefficient of cross validation

434

R2P: Determination coefficient of prediction

435

RPD: Residual predictive deviation

436

SEC: Standard error of calibration

437

SECV: Standard error of cross validation

438

SNV: Standard normal variate

439

UV: Ultra violet radiation

440

VIS: Visible radiation

441

18 ACS Paragon Plus Environment

Page 18 of 35

Page 19 of 35

Journal of Agricultural and Food Chemistry

442

Acknowledgments

443

Special thanks to Borja Millan and Salvador Gutiérrez for their assistance during image

444

and acquisition and image processing, and to Victor Sicilia and Teresa Díaz for their

445

help in the extraction of anthocyanins and sample preparation.

446 447

This work is partially supported by: European Investment Funds by

448

FEDER/COMPETE/POCI– Operacional Competitiveness and Internacionalization

449

Programme, under Project POCI-01-0145-FEDER-006958 and National Funds by FCT

450

-

451

UID/AGR/04033/2013 and by I&D project NORTE-01-0145-FEDER-000017, Interact

452

- Integrative Research in Environment, Agro-Chain and Technology, co-funded by

453

Fundo Europeu de Desenvolvimento Regional (FEDER) through NORTE 2020

454

(Programa Operacional Regional do Norte 2014/2020).

Portuguese

Foundation

for

Science

and

Technology,

under the

project

455 456

Supporting Information Available: The correlation coefficients between the

457

concentrations of all individual and groups of anthocyanins are shown as supporting

458

information in Table S1. The loadings of all calibration models for individual and

459

groups of anthocyanins are plotted in Figure S1.

460 461 462 463 464 465 466

Literature Cited 1. Kennedy, J.A.; Saucier, C.; Glories, Y. Grape and wine phenolics: History and perspective. Am. J. Enol. Vitic. 2006. 57, 239-248. 2. Mazza, G. Anthocyanins in grapes and grape products. Crit. Rev. Food Sci. Nut. 1995, 35, 341-371. 3. Moskowitz, A.H.; Hrazdina, G. Vacuolar contents of fruit subepidermal cells

19 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

467

from Vitis species. Plant Physiol. 1981, 68, 686-692.

468

4. Fraige, K.; Pereira-Filho, E.R.; Carrilho, E. Fingerprinting of anthocyanins from

469

grapes produced in Brazil using HPLC-DAD-MS and exploratory analysis by

470

principal component analysis. Food Chem. 2014, 145, 395-403.

471 472 473 474

5. Singleton, V. L.; Esau, P. Phenolic substances in grapes and wine, and their significance. Adv. Food Rees. Suppl. 1. Academic Press, New York. 1969. 6. Mazza, G.; Miniati, E. Anthocyanins in fruits, vegetables and grains. Boca Raton, CRC Press. 1993.

475

7. Fong, R.A.; Kepner, R.R.; Webb, A.D. Acetic acid-acylated anthocyanin

476

pigments in the grape skins of a number of varieties of Vitis vinifera. Am. J.

477

Enol. Vitic. 1971, 22, 150-155.

478

8. De Rosso, M.; Tonidandel, L.; Larcher, R.; Nicolini, G.; Ruggeri, V.; Dalla

479

Vedova, A.; De Marchi, F.; Gardiman, M.; Flamini, R. Study of anthocyanic

480

profiles of twenty-one hybrid grape varieties by liquid chromatography and

481

precursor-ion mass spectrometry. Anal. Chim. Acta 2012, 30, 120-129.

482

9. García-Beneytez, E.; Revilla, E.; F. Cabello. Anthocyanin pattern of several red

483

grape cultivars and wines made from them. Eur. Food Res. Tech. 2002, 215, 32-

484

37.

485

10. Lorente, D., Aleixos, N., Gómez-Sanchís, J., Cubero, S., García-Navarrete, O.,

486

Blasco, J. Recent advances and applications of hyperspectral imaging for fruit

487

and vegetable quality assessment. Food Bioprocess Technol. 2012, 5, 1121-

488

1142.

489

11. Keresztes, J.C.; Goodarzi, M.; Saeys, W. Real-time pixel based early apple

490

bruise detection using short wave infrared hyperspectral imaging in combination

491

with calibration and glare correction techniques. Food Control 2016, 66, 215-

20 ACS Paragon Plus Environment

Page 20 of 35

Page 21 of 35

Journal of Agricultural and Food Chemistry

492

226.

493

12. Pan, L.; Zhang, Q.; Zhang, W.; Sun, Y.; Hu, P.; Tu, K. Detection of cold injury

494

in peaches by hyperspectral reflectance imaging and artificial neural network.

495

Food Chem. 2016, 192, 134-141.

496

13. Gowen, A.; O’Donnell, C.P.; Cullen, P.J.; Downey, G.; Frias, J.M.

497

Hyperspectral imaging-an emerging process analytical tool for food quality and

498

safety control. Trends Food Sci. Technol. 2007, 18, 590-598.

499 500

14. Sun, D.W. Hyperspectral Imaging for Food Quality Analysis and Control. Elsevier Science and Technology: San Diego, CA, USA, 2010.

501

15. Siche, R.; Vejarano, R.; Aredo, V.; Velasquez, L.; Saldaña, E.; Quevedo, R.

502

Evaluation of food quality and safety with hyperspectral imaging (HSI). Food

503

Eng. Rev. 2015, 1-17.

504

16. Cheng, J.-H.; Sun D.-W; Pu H.-B.; Chen, X.; Liu, Y.; Zhang, H.; Li, J.-L.

505

Integration of classifiers analysis and hyperspectral imaging for rapid

506

discrimination of fresh from col-stored and frozen-thawed fish fillets. J. Food

507

Eng. 2015, 161, 33-39.

508

17. Calvini, R.; Ulrici, A.; Amigo, J.M. Practical comparison of sparse methods for

509

classification of Arabica and Robusta coffee species using near infrared

510

hyperspectral imaging. Chemom. Intell. Lab. Syst. 2015, 146, 503-511.

511

18. Diago, M.P.; Fernandes, A.M.; Millan, B.; Tardaguila, J.; Melo-Pinto, P.

512

Identification of grapevine varieties using leaf spectroscopy and partial least

513

squares. Comput. Electron. Agr. 2013, 99, 7-13.

514

19. Nogales-Bueno, J.; Rodríguez-Pulido, F.J.; Heredia, F.J.; Henández-Hierro, J.M.

515

Comparative study on the use of anthocyanin profile, color image analysis and

516

near-infrarred hyperspectral imaging as tools to discriminate between four

21 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

517

autochthonous red grape cultivars from La Rioja (Spain). Talanta, 2015, 131,

518

412-416.

519

20. Fernandes, A.M.; Melo-Pinto, P., Millan, B.; Tardaguila, J.; Diago, M.P.

520

Automatic discrimination of grapevine (Vitis vinifera L.) clones using leaf

521

hyperspectral imaging and partial least squares. J. Agric. Sci. 2015, 153, 455-

522

465.

523

21. Fernandes, A.M.; Oliveira, P.; Moura, J.P.; Oliveira, A. A.; Falco, V.; Correia,

524

M.J.; Melo-Pinto, P. Determination of anthocyanin concentration in whole grape

525

skins using hyperspectral imaging and adaptive boosting neural networks. J.

526

Food Eng. 2011, 105, 216-226.

527

22. Hernández-Hierro, J.M.; Nogales-Bueno, J.; Rodríguez-Pulido, F.J.; Heredia,

528

F.J. Feasibility study on the use of near-infrared hyperspectral imaging for the

529

screening of anthocyanins in intact grapes during ripening. J. Agric. Food Chem.

530

2013, 61, 9804-9809.

531

23. Chen, S.: Zhang, F.; Ning, J.; Liu, X.; Zhang, Z.; Yang S. Predicting the

532

anthocyanin content of wine grapes by NIR hyperspectral imaging. Food Chem.

533

2015, 172, 788-793.

534

24. Martínez-Sandoval, J.R.; Nogales-Bueno, J.; Rodríguez-Pulido, F.J.; Hernández-

535

Hierro, J.M.; Segovia-Quintero, M.A.; Martínez-Rosas, M.E.; Heredia, F.J.

536

Screening of anthocyanins in single red grapes using non-destructive method

537

based on the near infrared hyperspectral technology and chemometrics. J. Sci.

538

Food Agric. 2016, 96, 1643-1647.

539

25. Piazzolla, F.; Amodio, M.L.; Colelli, G. The use of hyperspectral imaging in the

540

visible and near infrared region to discriminate between table grapes harvested

541

at different times. J. Agric. Eng. 2013, 44, 49-55.

22 ACS Paragon Plus Environment

Page 22 of 35

Page 23 of 35

Journal of Agricultural and Food Chemistry

542

26. Nogales-Bueno, J.; Hernández-Hierro, J.M.; Rodríguez-Pulido, F.J.; Heredia,

543

F.J. Determination of technological maturity of grapes and total phenolic

544

compounds of grape skins in red and white cultivars during ripening by near

545

infrared hyperspectral image: A preliminary approach. Food Chem. 2014, 152,

546

586-591.

547

27. Rodríguez-Pulido, F.J.; Hernández-Hierro, J.M.; Nogales-Bueno, J.; Gordillo,

548

B.; González-Miret, M.L.; Heredia, F.J. A novel method for evaluating flavanols

549

in grape seeds by near infrared hyperspectral imaging. Talanta 2014, 122, 145-

550

150.

551

28. Jara-Palacios, M.J.; Rodríguez-Pulido, F.J.; Hernanz, D.; Escudero-Gilete, M.L.;

552

Heredia, F.J. Determination of phenolic substances of seeds, skins and strems

553

from white grape marc by near-infrared hyperspectral imaging. Aust. J. Grape

554

Wine Res. 2016, 22, 11-15

555

29. Noguerol-Pato, R.; González-Barreiro, C.; Cancho-Grande, B.; Martínez, M.C.;

556

Santiago, J.L.; Simal-Gándara, J. Floral, spicy and herbaceous active odorants in

557

Gran Negro grapes from shoulders and tips into the cluster, and comparison with

558

Brancellao and Mouratón varieties. Food Chem. 2012, 135, 2771-2782.

559

30. Noguerol-Pato, R.; González-Barreiro, C.; Cancho-Grande, B.; Santiago, J.L.;

560

Martínez, M.C.; Simal-Gándara, J. Aroma potential of Brancellao grapes from

561

different cluster positions. Food Chem. 2012, 132, 112-124.

562

31. Goldfarb,

A.

Don't

Call'Em

Crushers.

Wines

&

Vines.

563

http://www.winesandvines.com/template.cfm?section=features&content=55730.

564

(Last access: 15th September 2016)

565

32. Ferrer-Gallego, R.; Hernández-Hierro, J.M.; Rivas-Gonzalo, J.C.; Escribano-

566

Bailón, M.T. Determination of phenolic compounds of grape skins during

23 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

567 568 569 570 571 572 573

ripenin by NIR spectroscopy. LWT-Food Sci. Technol. 2011, 44, 847-853. 33. Massart, D. L.; Vandeginste, B. G. M.; Deming, S. N.; Michotte, Y.; Kaufman, L. Chemometrics: a textbook. 1988. 34. Naes, T.; Isaksson, T.; Fearn, T.; Davies, T. A user friendly guide to multivariate calibration and classification; NIR publications, 2002. 35. Shenk, J. S.; WesterHaus, M. O. Near infrared spectroscopy: The future waves; C, D. A. M., P, W., Eds.; NIR Publications: UK, 1996.

574

36. Geladi, P.; MacDougall, D.; Martens, H. Linearization and scatter-correction for

575

near-infrared reflectance spectra of meat. Appl. Spectrosc. 1985, 39 (3), 491–

576

500.

577 578

37. Dhanoa, M. S.; Lister, S. J.; Barnes, R. J. On the scales associated with nearinfrared reflectance difference spectra. Appl. Spectrosc. 1995, 49 (6), 765–772.

579

38. Shenk, J. S.; Westerhaus, M. O. Routine operation, calibration, development and

580

network system management manual. NIRSystems Inc., Silver Spring, MD, USA

581

1995.

582

39. Mark, H.; Workman Jr, J. Statistics in spectroscopy; Academic Press, 2003.

583

40. Williams, P. C.; Sobering, D. C. How do we do it: A brief summary of the

584

methods we use in developing near infrared calibrations. In A. M. C. Davies; P.

585

Williams (Eds.), Near infrared spectroscopy: The future waves (pp. 185–

586

188).Chichester: NIR Publications.1996

587

41. Giusti, M. M.; Wrolstad, R.E. Characterization and measurement of

588

anthocyanins by UV-Visible spectroscopy. In Current Protocols in Food

589

Analytical Chemistry. 2001. John Wiley, New York. F:F1:F1.2

590

42. Dambergs, R. G.; Cozzolino, D.; Cynkar, W. U.; Janik, L.; Gishen, M. The

591

determination of red grape quality parameters using the LOCAL algorithm. J.

24 ACS Paragon Plus Environment

Page 24 of 35

Page 25 of 35

Journal of Agricultural and Food Chemistry

592

Near Infrared Spectrosc. 2006, 14 (2), 71.

593

43. Cozzolino, D.; Kwiatkowski, M. J.; Parker, M.; Cynkar, W. U.; Dambergs, R.

594

G.; Gishen, M.; Herderich, M. J. Prediction of phenolic compounds in red wine

595

fermentations by visible and near infrared spectroscopy. Anal. Chim. Acta 2004,

596

513 (1), 73–80.

597

44. Lichtenthaler, C.; Buschmann, C. Chlorophylls and carotenoids-Measurement

598

and characterisation by UV-VIS. Current Protocols in Food Analytical

599

Chemistry (CPFA). 2001. John Wiley, New York. Supplement 1. F4.3.1-F4.3.8.

600

45. Williams, P.C. Implementation of near-infrared technology. Near Infrared

601

Technology in the Agricultural and Food Industries (pp. 145-169). American

602

Association of Cereal Chemists. 2001.

603

46. Nogales-Bueno, J.; Ayala, F.; Hernández-Hierro, J.M.; Rodríguez-Pulido, F.J.;

604

Echávarri, J.F.; Heredia, F.J. Simplified method for the screening of

605

technological maturity of red grape and total pehnolic compouns of red grape

606

skin: Application of the characteristic vector method to near-infrared spectra. J.

607

Agric. Food Chem. 2015, 63, 4284-4290.

608 609 610 611

Figure captions

612

Figure 1. (A) Scheme of the 4-step grape berry rotation (90º each rotation step) for the

613

acquisition of hyperspectral images of four sides of the berry; (B) Average absorbance

614

spectra corresponding to the four sides of an imaged Cabernet Sauvignon grape berry.

615

25 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

616

Figure 2. (A) Absorbance (Log (1/R)) (A) and (B) first derivative of the absorbance

617

(D1log (1/R)) spectra of the whole set of grape berries of eight different varieties (80

618

samples) in the VIS-NIR region studied.

619 620

Figure 3. Correlation of the values obtained of HPLC with respect to those predicted by

621

the hyperspectral imaging models and external validation. (A) Total Anthocyanins

622

(Total Anths); (B) Total 3-glucoside anthocyanins (Total Gl Anth). Correlation lines:

623

solid line and predictions intervals: dashed line for cross-validation models,

624

respectively. Dotted line represents the 1:1 line.

26 ACS Paragon Plus Environment

Page 26 of 35

Page 27 of 35

625 626 627

Journal of Agricultural and Food Chemistry

Table 1. Overview of the mean and standard deviation anthocyanin concentration in grape skin extracts of each grapevine variety. Results are expressed in mg L-1 malvidin-3-O-glucoside. Grapevine variety

Compound Cabernet Franc

Cabernet Sauvignon

Grenache

Merlot

Petit Verdot

Pignolo

Rebo

Sangiovese

Total Anth.

919.49 ± 223.52

877.86 ± 172.80

595.01 ± 149.11

1358.35 ± 299.72

505.82 ±152.68

997.13 ± 265.44

2463.18 ± 379.25

1736.62 ± 378.87

Total Gl Anth.

801.45 ± 186.56

827.33 ± 157.30

544.71 ± 135.48

1197.84 ± 238.06

419.09 ± 128.50

941.13 ± 255.45

2262.66 ± 324.61

1723.88 ± 375.84

4.14 ± 1.35

6.22 ± 3.30

3.55 ± 1.43

24.72 ± 9.37

8.97 ± 3.90

5.74 ± 3.77

46.88 ± 22.64

2.72 ± 1.36

Total GlCm Anth.

113.90 ± 47.53

44.31 ± 22.93

46.74 ± 14.68

135.80 ± 61.57

77.76 ± 26.34

50.26 ± 17.33

153.63 ± 52.00

10.02 3.65

MvGl

554.94 ± 143.38

472.75 ± 100.70

433.10 ± 100.08

605.40 ± 99.56

339.44 ± 91.39

390.56 ± 117.40

948.86 ± 120.03

602.44 ± 157.13

MvGlAc

0.00 ± 0.00

1.87 ± 3.04

2.03 ± 1.24

21.42 ± 8.99

3.49 ± 2.87

1.29 ± 2.82

38.57 ± 22.17

0.62 ± 0.81

MvGlCm

77.14 ± 35.40

23.58 ± 8.59

33.82 ± 10.76

66.00 ± 28.15

61.07 ± 19.93

24.93 ± 9.59

60.34 ± 20.88

2.61 ± 1.08

PtGl

85.40 ± 23.52

104.82 ± 23.17

36.46 ± 12.66

173.98 ± 35.28

36.14 ± 17.43

140.98 ± 46.72

381.52 ± 56.14

265.59 ± 60.62

PtGlAc

3.30 ± 1.38

3.83 ± 0.88

1.42 ± 0.45

1.98 ± 0.73

5.36 ± 2.69

3.77 ± 0.95

4.45 ± 1.24

0.58 ± 0.39

PtGlCm

9.19 ± 3.01

3.63 ± 1.44

2.43 ± 0.87

15.53 ± 6.82

5.51 ± 2.72

6.03 ± 2.00

23.47 ± 7.67

1.00 ± 0.40

80.23 ± 30.54

157.67 ± 40.98

25.20 ± 10.00

202.90 ± 44.04

24.52 ± 14.01

168.59 ± 57.17

611.56 ± 100.52

203.50 47.09

DpGlAc

nd

nd

nd

0.25 ± 0.29

nd

nd

0.01 ± 0.05

0.16 ± 0.30

DpGlCm

8.50 ± 2.67

5.25 ± 1.63

1.80 ± 0.77

20.67 ± 9.91

4.72 ± 2.51

8.00 ± 2.38

43.78 ± 14.55

1.10 ± 0.51

67.48 ± 23.76

66.68 ± 20.30

45.41 ± 29.00

164.52 ± 62.50

17.51 ± 5.96

151.25 ± 39.39

167.94 ± 32.35

338.14 ± 77.89

PnGlAc

0.62 ± 0.41

0.37 ± 0.12

0.09 ± 0.11

0.71 ± 0.51

0.12 ± 0.11

0.47 ± 0.30

2.46 ± 0.43

0.90 ± 0.68

PnGlCm

12.04 ± 7.01

7.64 ± 12.44

5.28 ± 2.30

22.50 ± 12.89

3.81 ± 1.30

7.58 ± 3.21

9.77 ± 3.67

1.93 ± 0.85

CyGl

13.40 ± 4.96

25.41 ± 14.10

4.55 ± 3.26

51.04 ± 18.41

1.47 ± 0.88

89.75 ± 25.58

152.79 ± 38.17

314.21 ± 91.29

CyGlAc

0.22 ± 0.09

0.15 ± 0.13

0.01 ± 0.03

0.35 ± 0.33

0.01 ± 0.01

0.21 ± 0.34

1.39 ± 1.01

0.47 ± 0.41

CyGlCm

7.03 ± 3.60

4.21 ± 1.04

3.42 ± 1.32

11.10 ± 5.32

2.64 ± 0.53

3.71 ± 0.97

16.26 ± 6.16

3.37 ± 1.34

Total GlAc Anth

DpGI

PnGl

1

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

Page 28 of 35

628 629 630 631 632 633

Total Anthocyanins (Total Anth.); Total 3-O-glucoside anthocyanins (Total Gl Anth.); Total 3-O-(6-acetyl)glucoside anthocyanins (Total GlAc Anth.); Total 3-O-(6-p-coumaroyl)glucoside anthocyanins (Total GlCm Anth.); Malvidin 3-O-glucoside (MvGl); Malvidin 3-O-(6-acetyl)glucoside (MvGlAc); Malvidin 3-O-(6-p-coumaroyl)glucoside (MvGlCm); Petunidin 3-O-glucoside (PtGl); Petunidin 3-O-(6-acetyl)glucoside (PtGlAc); Petunidin 3-O-(6-p-coumaroyl)glucoside (PtGlCm); Delphinidin 3-O-glucoside (DpGl); Delphinidin 3-O-(6-acetyl)glucoside (DpGlAc); Delphinidin 3-O-(6-p-coumaroyl)glucoside (DpGlCm); Peonidin 3-Oglucoside (PnGl); Peonidin 3-O-(6-acetyl)glucoside (PnGlAc); Peonidin 3-O-(6-p-coumaroyl)glucoside (PnGlCm); Cyanidin 3-O-glucoside (CyGl); Cyanidin 3-O-(6-acetyl)glucoside (CyGlAc); Cyanidin 3-O-(6-pcoumaroyl)glucoside (CyGlCm). nd: not detected.

634 635

2

ACS Paragon Plus Environment

Page 29 of 35

Journal of Agricultural and Food Chemistry

636 637

638 639 640 641 642 643 644 645 646

Table 2. Statistics overview of the anthocyanin concentration of grape skin extracts of all grapevine varieties, expressed in mg L-1 of malvidin-3-O-glucoside. Compound

N

Minimum

Maximum

Mean

SD

Total Anth.

80

315.87

3250.33

1181.68

664.96

Total Gl Anth.

80

280.73

2942.01

1089.76

629.16

Total GlAc Anth.

80

1.16

75.88

12.87

16.89

Total GlCm Anth.

80

4.02

232.45

79.05

58.78

MvGl

80

226.56

1153.86

543.44

211.88

MvGlAc

52

nd

65.69

8.66

15.54

MvGlCm

80

0.73

137.64

43.69

30.98

PtGl

80

16.25

480.79

153.11

118.13

PtGlAc

78

nd

9.42

3.09

1.96

PtGlCm

80

0.28

34.65

8.35

8.13

DpGI

80

9.67

807.71

184.27

182.90

DpGlAc

15

nd

0.98

0.05

0.17

DpGlCm

80

0.50

65.69

11.73

14.81

PnGl

80

10.06

461.56

127.37

104.98

PnGlAc

74

nd

3.25

0.72

0.80

PnGlCm

80

0.34

42.85

8.82

9.06

CyGl

80

0.58

476.94

81.58

106.89

CyGlAc

62

nd

4.13

0.35

0.58

CyGlCm

80

0.74

28.41

6.47

5.52

N: Number of samples; SD: standard deviation. Total Anthocyanins (Total Anth.); Total 3-O-glucoside anthocyanins (Total Gl Anth.); Total 3-O-(6-acetyl)glucoside anthocyanins (Total GlAc Anth.); Total 3-O-(6-p-coumaroyl)glucoside anthocyanins (Total GlCm Anth.); Malvidin 3-O-glucoside (MvGl); Malvidin 3-O-(6-acetyl)glucoside (MvGlAc); Malvidin 3-O-(6-pcoumaroyl)glucoside (MvGlCm); Petunidin 3-O-glucoside (PtGl); Petunidin 3-O-(6-acetyl)glucoside (PtGlAc); Petunidin 3-O-(6-pcoumaroyl)glucoside (PtGlCm); Delphinidin 3-O-glucoside (DpGl); Delphinidin 3-O-(6-acetyl)glucoside (DpGlAc); Delphinidin 3O-(6-p-coumaroyl)glucoside (DpGlCm); Peonidin 3-O-glucoside (PnGl); Peonidin 3-O-(6-acetyl)glucoside (PnGlAc); Peonidin 3O-(6-p-coumaroyl)glucoside (PnGlCm); Cyanidin 3-O-glucoside (CyGl); Cyanidin 3-O-(6-acetyl)glucoside (CyGlAc); Cyanidin 3O-(6-p-coumaroyl)glucoside (CyGlCm). nd: not detected.

647 648 649 650

1 ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

651 652

Page 30 of 35

Table 3. Calibration statistical descriptors and internal and external validation results for the VIS-NIR spectral models of anthocyanin concentration in grapes skin extracts (values are expressed as mg L-1 malvidin-3-O-glucoside) Compound

Spectral Treatment

N

SD

Minimum

Maximum

PLS Factor

Calibration

Cross Validation

SEC

R2

C

SECV

R2

cv

RPD

External1 Validation SEP R2

p

Snv-DT 1.5.5.1

76

629.46

315.87

2839.96

4

148.25

0.94

189.05

0.91

3.51

281.58

0.86

Total Gl Anth.

1.5.5.1

74

562.69

280.73

2521.24

4

125.97

0.95

155.94

0.92

4.03

272.60

0.86

Total GlAc Anth.

2.5.5.1

73

12.80

1.16

54.26

5

1.98

0.98

4.12

0.90

4.10

4.82

0.51

Total GlCm Anth.

1.5.5.1

73

56.49

4.02

221.37

6

13.42

0.94

23.09

0.83

2.55

42.72

0.40

Snv-DT 1.5.5.1

76

201.85

226.56

1100.02

4

56.014

0.92

73.93

0.87

2.87

109.10

0.83

MvGlAc

2.5.5.1

48

13.39

0.09

46.49

6

1.32

0.99

4.13

0.90

4.27

3.30

0.90

MvGlCm

Snv-DT 1.5.5.1

75

29.78

0.73

119.94

7

6.99

0.94

13.32

0.80

2.33

23.33

0.57

PtGl

Snv-DT 1.5.5.1

75

108.00

16.25

426.94

4

23.29

0.95

29.44

0.93

4.01

49.66

0.87

PtGlAc

1.5.5.1

72

1.50

0.23

6.09

3

0.73

0.76

0.98

0.57

2.00

1.18

0.35

PtGlCm

1.5.5.1

71

6.72

0.28

26.06

7

1.12

0.97

2.04

0.91

3.98

4.44

0.84

Snv-DT 1.5.5.1

78

158.65

9.67

657.71

4

36.89

0.95

48.41

0.91

3.78

71.39

0.88

DpGlAc

--

--

--

--

--

--

--

--

--

--

--

--

--

DpGlCm

Snv-DT 1.5.5.1

72

11.59

0.50

46.49

7

1.83

0.97

3.28

0.92

4.51

2.34

0.48

PnGl

Snv-DT 1.5.5.1

73

86.47

10.06

383.09

7

15.93

0.97

30.35

0.88

3.46

49.10

0.81

PnGlAc

Snv-DT 2.5.5.1

69

0.69

0.02

2.82

2

0.26

0.86

0.31

0.80

2.61

0.35

0.88

PnGlCm

Snv-DT 2.5.5.1

71

7.56

0.34

37.53

4

2.14

0.92

3.81

0.75

2.38

6.50

0.54

CyGl

1.5.5.1

67

47.50

0.58

159.63

6

8.59

0.97

16.43

0.88

6.51

48.85

0.69

CyGlAc

1.5.5.1

54

0.33

0.01

1.26

2

0.14

0.81

0.16

0.77

3.94

0.16

0.85

CyGlCm

1.5.5.1

70

4.80

1.46

19.14

5

1.21

0.94

1.78

0.86

3.10

3.82

0.63

Total Anth.

MvGl

DpGI

1

ACS Paragon Plus Environment

Page 31 of 35

Journal of Agricultural and Food Chemistry

653 654 655 656 657 658 659

Snv-DT: standard normal variate plus Detrending; N: number of samples were the ones used for calibration and cross validation models after outlier detection; SD: standard deviation; SEC: standard error of calibration; R2c: determination coefficient of calibration; SECV: standard error of cross-validation; R2cv: determination coefficient of cross-validation; RPD: residual predictive deviation. Total Anthocyanins (Total Anth.); Total 3O-glucoside anthocyanins (Total Gl Anth.); Total 3-O-(6-acetyl)glucoside anthocyanins (Total GlAc Anth.); Total 3-O-(6-p-coumaroyl)glucoside anthocyanins (Total GlCm Anth.); Malvidin 3-O-glucoside (MvGl); Malvidin 3-O-(6-acetyl)glucoside (MvGlAc); Malvidin 3-O-(6-p-coumaroyl)glucoside (MvGlCm); Petunidin 3-O-glucoside (PtGl); Petunidin 3-O-(6-acetyl)glucoside (PtGlAc); Petunidin 3-O-(6-pcoumaroyl)glucoside (PtGlCm); Delphinidin 3-O-glucoside (DpGl); Delphinidin 3-O-(6-acetyl)glucoside (DpGlAc); Delphinidin 3-O-(6-p-coumaroyl)glucoside (DpGlCm); Peonidin 3-O-glucoside (PnGl); Peonidin 3O-(6-acetyl)glucoside (PnGlAc); Peonidin 3-O-(6-p-coumaroyl)glucoside (PnGlCm); Cyanidin 3-O-glucoside (CyGl); Cyanidin 3-O-(6-acetyl)glucoside (CyGlAc); Cyanidin 3-O-(6-p-coumaroyl)glucoside (CyGlCm).

660

1

For the predictive models, internal validation was carried out with N=60 samples while external validation used N=20 samples.

661 662 663

2

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

Figure 1. (A) Scheme of the 4-step grape berry rotation (90º each rotation step) for the acquisition of hyperspectral images of four sides of the berry; (B) Average absorbance spectra corresponding to the four sides of an imaged Cabernet Sauvignon grape berry. 248x92mm (96 x 96 DPI)

ACS Paragon Plus Environment

Page 32 of 35

Page 33 of 35

Journal of Agricultural and Food Chemistry

Figure 2. (A) Absorbance (Log (1/R)) (A) and (B) first derivative of the absorbance (D1log (1/R)) spectra of the whole set of grape berries of eight different varieties (80 samples) in the VIS-NIR region studied. 145x179mm (96 x 96 DPI)

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

Figure 3. Correlation of the values obtained of HPLC with respect to those predicted by the hyperspectral imaging models and external validation. (A) Total Anthocyanins (Total Anths); (B) Total 3-glucoside anthocyanins (Total Gl Anth). Correlation lines: solid line and predictions intervals: dashed line for crossvalidation models, respectively. Dotted line represents the 1:1 line. 96x190mm (96 x 96 DPI)

ACS Paragon Plus Environment

Page 34 of 35

Page 35 of 35

Journal of Agricultural and Food Chemistry

TOC graphic 44x23mm (300 x 300 DPI)

ACS Paragon Plus Environment