ULTRAVIOLET – VISIBLE - NEAR INFRARED SPECTROSCOPY FOR RAPID DETERMINATION OF VOLATILE COMPOUNDS IN WHITE GRAPES DURING RIPENING

Near Infrared (NIR) spectroscopy is increasingly used in food analysis due to its speed and ease of use. Ultraviolet-visible (UV-VIS) spectroscopy is commonly used in any laboratory worldwide. The objective of this work was to develop a rapid method based on UV-VIS-NIR spectroscopy for the determination of volatile compounds in white ‘Albariño’ grapes from DO Rias Baixas (Spain). A total of 52 samples of white ‘Albariño’ must were analysed. Gas chromatography-mass spectrometry (GC-MS) was used as reference method. Partial Least Squares regression was used to fit mathematical models to relate the UV-VIS-NIR spectra with the volatile compounds determined by GC-MS. Reliable models for predicting the following compounds were obtained: (E)-2-hexenal, 1-hexanol, (Z)-2-hexanol, benzaldehyde, phenylethanal, cis pyran linalool oxide


INTRODUCTION
Grape composition at harvest is one of the most important factors determining the future quality of wine. Volatiles, important flavour components in white wine, are formed during grape berry metabolism and they are very influenced by the ripening stage. Measurement of grape volatile composition in the last ripening stages is an important requirement for an optimum production of white wines. Therefore a rapid method is necessary to know the aromatic ripening of grape.
'Albariño' is one of the most important white grape varieties in Galicia (NW Spain). The aim of this work was to apply of UV-Vis-NIR spectroscopy to predict the volatile composition of grapes during ripening with the objective of optimizing the wine aroma.

Vineyard locations and grape samples
This study was tested on 'Albariño' grapes grown on the Controlled Designation of Origin Rías Baixas (Galicia, Spain). A total of 52 grape samples of Vitis vinifera 'Albariño' were studied at different maturity stages, 13 from M-1 (16.3 ºBrix), 19 from M-2 (17.8 ºBrix) and 20 from M-3 (18.7 ºBrix). The grape samples (1 kg) were collected during 2014 vintage. After harvest, ºBrix was analyzed by refractometry and the samples were immediately frozen and stored at -20°C until chemical analyses. Grape samples of 500 g were used to analyze free volatile composition of 'Albariño' musts.

Extraction, identification and quantification of free volatile compounds
About 500 g of frozen berries were thawed at 4 ºC overnight, and then manually crushed, centrifuged (9287 rpm, 20 min, 4 ºC) and filtered through a glass wool bed. To 75 mL of juice 4-nonanol (Merck, ref. 818773) was added as internal standard (10 µL of 40 µg/L solution in 10 %, v/v ethanol) and passed through a LiChrolut EN cartridge (Merck, 500 mg, 40-120 µm) according to Oliveira et al. (2000). The resin was previously pre-conditioned with 10 mL of dichloromethane, 5 mL of methanol and 10 mL of aqueous alcoholic solution (10 %, v/v). Free volatile compounds were eluted with 5 mL of pentanedichloromethane. The pentane-dichloromethane elute was dried over anhydrous sodium sulphate and concentrated to 200 µL by solvent evaporation under a nitrogen stream prior to analysis.
Gas chromatographic analysis of volatile compounds was performed using a GC-MS system constituted by an Agilent Chromatograph 6890N and an ion-trap mass spectrometer 5975C. A 1 μL injection was made into a capillary column, coated with CP-Wax 52 CB (50 m × 0.25 mm i.d., 0.2 μm film thickness, Chrompack). The temperature of the injector (SPIseptum-equipped programmable temperature) was programmed from 20 ºC to 250 ºC, at 180 ºC/min. The oven temperature was held at 40 ºC for 5 min, then programmed to rise from 40 ºC to 250 ºC at 3 ºC/min, then held 20 min at 250 ºC and finally programmed to go from 250 ºC to 255 ºC at 1 ºC/min. The carrier gas was helium N60 (Air Liquide) at 103 kPa, which corresponds to a linear speed of 180 cm/s at 150 ºC. The detector was set to electronic impact mode (70 eV), with an acquisition range from 29 to 360 m/z, and an acquisition rate of 610 ms.
Identification was performed using the software Saturn version 5.2 (Varian), by comparing mass spectra and retention indices with those of pure standard compounds. All of the compounds were quantified as 4-nonanol equivalents.

Spectral analysis
Samples of musts (5 mL) were analysed in a spectrophotometer V-670 (Jasco Inc, Japan) using transmittance mode at 2 nm intervals in UV-VIS-NIR regions (190 nm-2500 nm). Prior spectral analysis, samples were equilibrated at 33 ºC for 10 min before scanning and filtered through 0.45 μm filter (Cozzolino et al., 2007). Cell quartz with 1 mm path length was used to scan samples. Data were collected using Spectra Manager™ II software (Jasco Inc, Japan). Samples were scanned in duplicate obtaining 104 spectra.

Chemometric analysis
The chemometric analysis was performed according to Martelo-Vidal and Vazquez (2014). Spectral data were exported from Spectra Manager™II software into Uncrambler software (version X 10.2; CAMO, Oslo, Norway) for pre-treatment and obtain calibration models. Two replicates of each sample (104 spectra) were analysed in Unscrambler software.
Calibration models for measurement of volatile compounds were performed using partial least square regression (PLS). Calibration models were developed using full-cross validations. Spectral data were pretreated before PLS modeling. The pre-treatments tested were standard normal variate (SNV), first derived Savitzky-Golay (1 st derived), second derived Savitzky-Golay (2 nd derived). The statistical parameters Correlation Coefficient-squared (r 2 ), Root Mean Square of Calibration (RMSEC), Root Mean Square Error of Cross Validation (RMSECV) and Residual Predictive Deviation (RPD) was used to evaluate how well the calibration model of spectra could predict volatile compounds (Cozzolino et al., 2004(Cozzolino et al., , 2011Lorenzo et al., 2009;Garde-Cerdan et al., 2012).

Chemical analysis
'Albariño' cultivar from controlled designation of origin Rías Baixas was sampled at different ripening dates, which influences in the volatile composition.
Table I and Figure 1 shows the variation of concentration volatile compounds identified and quantified in 52 samples of 'Albariño' musts during 3 stages of ripening (16.3, 17.8 and 18.7 ºBrix). Data in Table I have been arranged into the nine chemical families where were identified and quantified 26 free volatile compounds: 6 alcohols, 7 C 6-compounds, 3 volatile acids, 2 terpenes, 2 ethyl esters and acetates, 2 aldehydes, 1 lactone, 2 volatile phenols and 1 carbonyl compound.

M-1 (16,3 ° Brix), H-2 (17,8 ° Brix) e M-3 (18,7 ° Brix).
Analytical results showed differences in the ripening of 'Albariño' cultivar where C 6 -compounds (represented by six compounds) were quantitatively the largest group of free volatile compounds quantified in 'Albariño' grape cultivar. C 6compounds are related to varietal origin because they can be formed, via C 6 -aldehydes, through lipoxygenase activity, from linoleic and linolenic acids present in grapes and supply vegetal and herbaceous nuances to the wine (Oliveira et al., 2006;Kalua and Boss, 2009). 'Albariño' was described as a terpenic wine by several authors because the bound terpenic content of the must (Oliveira et al., 2000;Vilanova et al., 2007). However in our work only volatiles (free fraction) were analysed where only cis pyran linalool oxide and diendiol I was identified.
Among ripening data, the highest total value of volatile composition was showed in M-2 (17.8 ºBrix) with 5.939 μg/L. Different behaviour was shown among compounds ( Figure 2). Ripening data M-2 showed the highest values of C 6 -compouds, volatile esters and acetates and terpenes. All volatile families have shown a decreased in the last ripening data (M-3) with exception of aldehydes and carbonyl compounds. Volatile phenols and lactones concentration decreased during ripening. The evolution of volatiles during ripening of grape juice was not proportional to the changes in sugar content, which shows that the technological and aromatic maturities did not occur at the same time (Vilanova et al., 2009).  Figure 2 shows raw UV-VIS-NIR spectral data of 'Albariño' musts samples. UV-VIS zones (234 a 850 nm) showed big differences among samples meanwhile NIR zone (850 a 2500 nm) showed more homogeneity. The variability observed in UV-VIS-NIR could be useful to obtain mathematical models that relate the volatile compounds with the spectra. PLS was performed on raw or pre-treated spectral data with the aim to correlate the volatile compounds and the spectral data. Results showed significant correlations between spectral data and some volatile compounds.

Spectral analysis
With the aim of showing the variability in volatile composition and spectral data, a score plot of the PCA of UV-VIS-NIR spectra from must samples is shown in Figure 3. This PCA explains 81 % of total variance with the first two PC. Separation among ripening samples was observed in the score plots. Some of the most ripening samples (18.7 ºBrix, samples 1-19) were located in the positive side of PC1 and negative side of PC2.

Análise em componentes principais de dados brutos UV-VIS-NIR de mostos de 'Albariño' (M1-M52) da DO Rías Baixas.
The PLS analysis showed that UV-VIS-NIR spectra allowed to obtain accurate mathematic models to predict some volatile compounds concentrations from spectral data. Statistical results of the models for volatile compounds and UV-VIS-NIR spectral data (raw and pre-treatments) are shown in Table II. Only statistical results for the model of the best pretreatment are showed.
Results showed variations based on the treatments (raw, 1 st derived, 2 nd derived or SNV). For many volatile compounds, the best models were obtained with the pre-treatments 1 st derived and 2 nd derived. For only one compound the best pre-treatment was SNV (vanillin) and for only one no treatment was the best (2-phenylethylacetate). Models for one alcohol and three C 6 -compounds showed a r 2 > 0.9 using the 2 nd derived pre-treatment while two aldehydes and one terpenes showed values of r 2 > 0.8 using the 1 st derived. Other studies performed in 'Tannat' grapes showed lower values of r 2 in glycosylated aroma compounds (Boido et al., 2013).  Figure 4 shows the predicted vs real values of the three best models for the volatile compounds 2phenylethanol, 1-hexanol and phenylethanal by UV-VIS-NIR spectroscopy. A good correlation can be observed.
Considering that C 6 compounds in grape are a ripening marker, which could be a good indicator of optimal timing of harvest, the application of UV-VIS-NIR spectroscopy for the rapid analysis of these compounds can be new tools for winemakers to define the harvest time. High concentrations of these compounds indicate lower ripening of the grapes, providing an herbaceous aroma in future wines.
Residual predictive deviation (RPD) was used to evaluate the prediction capacity of models. RPD is the most commonly used statistical index to account for model reliability. In our study a good capacity of prediction for PLS calibrations was achieved for any volatile compounds quantified. Values around 1 were considered good for the prediction models. A total of 25 free volatile compounds from 26 showed RPD values > 0.9. Similar results were shown by Boido et al. (2013) when they analyzed bound glicosidically compounds in 'Tannat' juice.

CONCLUSIONS
PLS regression models showed good results for some compounds related with grape ripening. UV-VIS-NIR spectroscopy could be a rapid and nondestructive method for evaluating grape aroma during ripening and establish the harvest data. UV-VIS-NIR spectroscopy could be a good tool for viticulture decisions in the vineyard. In conclusion, UV-VIS-NIR spectroscopy is a fast and feasible method for the determination of some volatile compounds in white 'Albariño' musts from DO Rías Baixas.