Abstract
A rapid method for the simultaneous measurement of gluconic acid and glycerol, two important chemical markers of grape infection, was developed using Fourier-transform mid-infrared (FT-MIR) spectroscopy. The method used a combination of FT-MIR with partial-least squares (PLS) regression and compositional data of Trebbiano and Sangiovese grapes obtained with high performance ion-exchange chromatography (HPLC). A total of 320 grape samples were analyzed, including grapes hand picked in the vineyards (200 samples) and mechanically sampled from trucks upon arrival at the wineries (120 samples). Gluconic acid and glycerol increased with increasing percentage of grape infection assessed by visual inspection (r = 0.928). The best result was obtained for gluconic acid (R2 = 0.979; RMSECV = 0.63; RPD = 7.0), indicating that FT-MIR was suitable for process control.
The quality of grapes diminishes when infections occur (Ribereau-Gayon 1983), thus it is important to determine the sanitary status of grapes before crushing the fruit. Among the numerous parasites of the grapevine, Botrytis cinerea Pers., the causal agent of grey mold, is highly problematic in the Emilia-Romagna region (northern Italy). Development of the fungi Botrytis cinerea in the vineyard modifies the chemical composition of grape berry, including the accumulation of glycerol, gluconic acid, galacturonic acid, polyols, and laccase (Donèche 1993). Although specific methods are available for the detection and quantification of Botrytis cinerea infection, such as microscopy, immunological techniques, and molecular biology (Dewey et al. 2000, Dewey and Yohalem 2004), the use of these methods for routine quality control of grapes is limited. In many wineries the sanitary state of grapes is evaluated solely by visual inspection, which can result in much variability when there are multiple inspectors. The problem is even greater with machine-harvested grapes, which usually lose their structural integrity, making visual inspection and evaluation more difficult (Kupina 1984). A more reliable and objective method is needed to assess the sanitary condition of grapes.
Spectroscopic techniques, combined with multivariate data analysis, are well suited for correlating the spectral response of a sample to its compositional data. Mid-infrared (MIR) spectroscopy technology measures the absorption of radiation in the spectral region, ranging from 4000 to 400 cm−1 or from 2500 to 2.5 × 10−4 nm (Smith 1999). Fourier transform MIR (FT-MIR) spectroscopy is based on interferometry, which allows the measurement of all wavelengths simultaneously. The interferogram is then converted into a conventional spectrum using a Fourier transform algorithm. FT-MIR spectra contain much information that can be easily calibrated for many types of constituents. When multiple analytes are to be determined and when the absorbing peaks overlap, a multivariate method such as partial least squares (PLS) regression is used in conjunction with the spectra to develop multicomponent calibration models (Esbensen et al. 1998).
FT-MIR spectrometry is a suitable technique for simultaneous multicomponent analysis of many enological parameters, including alcohol, organic acids, volatile acidity, reducing sugars (Schindler et al. 1998, Patz et al. 1999, Gishen and Holdstock 2000, Kupina and Shrikhande 2003, Urbano Cuadrado et al. 2005), glycerol (Nieuwoudt et al. 2004), anthocyanins (Versari et al. 2006, Soriano et al. 2007), aroma precursors (Schneider et al. 2004), and polysaccharides (Boulet et al. 2007). Gluconic acid exhibits several IR absorption bands located ~1100 cm−1 that reflect C–O and C–C stretching and possesses a carboxylic acid group that absorbs IR radiation at ~1723 cm−1, which is likely caused by the carboxylic acid group in the protonated form (COOH). The IR region 1500 to 900 cm−1 corresponds to the absorbance zone of major constituents of wine, such as ethyl alcohol, sugars, organic acids, and glycerol (Vonach et al. 1998). The FT-IR spectrum of the aqueous glycerol solution shows prominent absorbance peaks in the 1600 to 929 cm−1 and 3000 to 2525 cm−1 regions (Patz et al. 1999). In particular, more than 85% of the variation in the glycerol concentration of the wine samples could be correlated to the absorbance in the region from 1229 to 929 cm−1 (Nieuwoudt et al. 2004). Some characteristic features of the FT-IR spectrum of glycerol include the peak at 1462 to 1458 cm−1 (representing the H-C bend) and the peak at 1100 to 1075 cm−1 (representing the C-O stretch) (Pavia et al. 2001). Most studies on FT-MIR focus on wine composition and little information is available on its application to grape juice analysis. The FT-MIR spectra represent the whole grape juice matrix, and the chemical modification produced by B. cinerea represents an interesting application of this technique. However, limited information is currently available (Dubernet et al. 2001, Rousseau et al. 2002).
The aim of this work was to show that FT-MIR, using chemometrics, can rapidly measure levels of gluconic acid and glycerol in the grape juice and that these measurements could be used to indicate the quality of botrytized grape berries. The procedure could serve as an alternative to conventional visual inspection of grapes as a quality control tool in wineries.
Materials and Methods
Samples.
Two sets of Sangiovese and Trebbiano grapes were collected at maturity with total soluble solids between 17 and 21 Brix. Grapes were collected during two vintages, 2005 and 2006. The first set of 200 grape bunches (~2 kg each) was hand picked from the field within each vine and along the row of the vineyards at the University of Bologna and Extension Center (Faenza, RA, Italy). The second set of 120 bunches (~5 kg each) was collected by a mechanical core sampler from the harvest load of grapes upon their arrival at five local wineries. Both sets were designed for analyzing the sanitary quality of the incoming grapes, thus representing grapes from healthy (0% infection) to extremely poor quality (>80% B. cinerea infection) estimated by visual inspection, using a 1% interval scale. The hand-picked grapes (set 1) were specifically sampled to cover the widest range of infection. Grape berries from both sets were placed in polyethylene bags, pressed manually as described in the literature (Iland et al. 2004), then filtered with a filtration unit (model 79500; Foss, Hillerød, Denmark) using filter paper circles graded at 20–25 μm, 185 mm diam (model 10312714; Whatman, Milan, Italy) until 100 mL of juice was collected and used for FT-MIR analysis, or stored at −18°C for later reference analysis.
Infrared spectroscopy equipment.
FT-MIR analyses were carried out using a GrapeScan FT120 instrument (Foss) equipped with a pyroelectric DTGS detector and a Michelson interferometer. Samples were pumped through a flow-through cuvette equipped with CaF2 windows and a 37-μm optical path length that is housed in the heater unit of the instrument set at 40°C. Background absorbance in the sample (which includes the absorbance of water) was corrected through the use of a Foss Zero Liquid S-6060 (WineScan FT120, model 77110 and 77310), which is scanned prior to the grape sample. The absorbance spectra were recorded in duplicate for each sample from 5012–926 cm−1, with a 4 cm−1 spectral resolution. Although the whole spectral range (5012–926 cm−1) was stored for each sample, only the following three areas were used for calculation: 1543–965cm−1, 2280–1717 cm−1, and 2971–2435 cm−1. The other ranges of frequencies were eliminated to prevent noise in the calculation. In particular, the two regions 1717–1543 cm−1 and 3627–2971 cm−1 are water absorption bands preventing any energy from passing through the cell, whereas the region 5012–3627 cm−1 contains very little useful information (Figure 1⇓).
Reference analysis.
The sanitary condition of grapes was assessed by visual inspection and by high-performance ion-exchange chromatography of gluconic acid (CH2OH-CHOH-CHOH-CHOH-CHOH-COOH) and glycerol (CH2OH-CHOH-CH2OH). HPLC was used as the reference method to calibrate the FT-MIR instrument. A DX600 system (Dionex, Sunnyvale, CA) was used equipped with a quaternary pump module, electrochemical detection (ED50), automated sampler (AS40), and Peak-Net 6.20 software for data acquisition and processing. Analysis of gluconic acid was conducted on an Ion-Pac AS11–HC column (4 × 250 mm; Dionex) kept at 30°C, and by conductivity detection following suppression of background noise with an ASRS-Ultra, 4-mm self-regenerating suppressor (Dionex). The column operated at flow rate of 1.2 mL/min using HPLC grade water as eluent A and NaOH 100 mM as eluent B, following the linear gradient elution conditions (min/A%): 0/98, 10/98, 20/75, 30/50.
Analysis of glycerol was carried out on a CarboPac PA10 column (4 × 250 mm; Dionex) kept at 30°C and by pulsed amperometric detection using a gold electrode. The column operated at a flow rate of 1.0 mL/min, following isocratic gradient elution conditions with HPLC grade water as eluent A (30%) and NaOH 100 mM as eluent B (70%) for 15 min. Mobile phases were degassed with helium for 30 min and kept anaerobic under pressure with helium.
Standards of gluconic acid (code 49120) and glycerol were from a commercial source (Sigma, Milan, Italy). Peak identification was based on the retention times (tR) and spiking technique, whereas peak quantification was based on the external standard calibration method using linear regression analysis (r > 0.999).
Statistical analysis.
The experimental approach included a preliminary check for outliers, set up of a calibration model for gluconic acid and glycerol by HPLC, visual inspection for grape quality, and cross-validation of the models. For the calibration of the selected parameters, partial least squares (PLS) regression was chosen among the most commonly used multivariate calibration method for the evaluation of MIR spectra (Martens and Martens 2001). Each raw spectrum was mean centered before statistical analyses. The cross-validation method, following the leave-one-out procedure, was used to determine the maximum number of significant factors (principal components; PCs) to ensure the predictive ability and to avoid over-fitting of the data (Martens and Naes 1989). With cross-validation, the same samples are used both for model estimation and testing. A few samples are omitted from the calibration data set and the model is calibrated on the remaining data points. The values for the omitted samples are predicted and the prediction residuals are computed. The process is repeated with another subset of the calibration set, and so on until every object has been omitted once; then all prediction residuals are combined to compute the validation residual variance and root mean squared error of prediction (RMSEP). Full cross-validation omits only one sample at a time; it is the original version of the method. In spectroscopy applications it is often appropriate to refine the model by deleting wavelengths. In practice, the B coefficients (regression coefficients) give the accumulated picture of the most important wavelengths. The optimum wave numbers for inclusion in the calibration equations was determined by comparing the regression results in terms of correlation coefficient (r) and the RMSEP, which should be minimized (Esbensen et al. 1998). The relative error (RMSEP%) corresponds to the formula: RMSEP% = (RMSEP/mean)*100. The residual predictive deviation (RPD = standard deviation of the data/standard error of prediction) was also used to evaluate the predictive ability of the calibration models. An RPD value <3 indicates the calibration model is unsuitable for quantification, a value between 3 and 5 is suitable for screening, while a value >5 is suitable for quantification (Pink et al. 1998, Williams 1995). Statistical analyses were performed using two commercially available chemometric software packages (Opus 5.5, Bruker Optics, Milan, Italy; Unscrambler 7.6, CAMO, Oslo, Norway).
Results and Discussion
Analysis of gluconic acid and glycerol.
After preliminary data plotting, five outliers were removed from the first set of data, resulting in 99 samples of Trebbiano white grapes and 96 samples of Sangiovese red grapes (total 195 samples). The second set included 60 samples for each variety (total 120 samples). Taking into account the time and cost required for HPLC analysis and the information available in the literature (mainly on glycerol), all samples were assessed by visual inspection, whereas gluconic acid and glycerol were analyzed based on 120 and 60 samples, respectively.
The performance of FT-MIR calibrations as compared with reference analysis is shown in Table 1⇓). The frequency regions of the IR spectra primarily related to each parameter were between 1563 and 1065 cm−1, using 7 to 10 PCs. The calibration results of hand-picked grapes (set one) were better than the results of mechanically sampled grapes (set two). The only differences between the two data sets were sampling procedure (i.e., hand picking and mechanical sampling) and range of values for each parameter, particularly gluconic acid and glycerol. Hand picking of grapes fully respected the physical integrity of the berries, whereas mechanical sampling was handled automatically by a core-type sampler that also separates stems from grape must. The latter is a standard procedure used in many wineries and the sample collected is used for internal quality control of grapes (such as sugar content, pH, and yeast assimilable nitrogen).
FT-MIR was suitable in process control for the analysis of gluconic acid in hand-picked samples (RPD = 7.0) from Trebbiano and Sangiovese grapes (Figure 2⇓). Data from Trebbiano and Sangiovese grapes were pooled in a single group taking into account that no maceration occurred, and thus no color extraction from Sangiovese red grapes. The fitting of data on the plot confirmed that gluconic acid can be well predicted on a combined set of white and red grapes. Samples from hand-picked and mechanically sampled grapes clearly overlapped, however, the latter covered only part of the range value (SD = 0.8), thus resulting in a RPD value of 1.8, which is not recommended for application. Even though glycerol showed a high value of coefficient of determination (R2 = 0.959), the FT-MIR was appropriate for screening only (RPD = 4.9), being close to the value for quality-control application (RPD > 5.0).
The predictive ability of the FT-MIR spectrometry is highly dependent on the composition of the sample set. The application of FT-MIR for quantifying glycerol in dry wine has provided a relative standard error of cross-validation (%) of 7.2 g/L (Patz et al. 1999) and 8.7 g/L (Urbano Cuadrado et al. 2005). The better result found in the literature for glycerol determination in dry wine can be explained taking into account the role of sugar content. Researchers using the FT-MIR analysis found a higher relative standard error of cross-validation (%) for sweet wines with 31–147 g/L reducing sugars compared with dry wines, 7.5 g/L and 4.3 g/L, respectively (Nieuwoudt et al. 2004).
Sampling procedure, sample storage, and timing of each analysis can each play a critical role in FT-MIR calibration. A recent study reported that frozen storage longer than one month does impact significantly on the predictive ability of NIR calibrations when attempting to predict fresh samples in terms of total anthocyanins, total soluble solids, and pH (Cozzolino et al. 2005).
Analysis of grape infection.
When designing a data set, it is important to consider the number of samples, the concentration range covered by the samples, and the distribution of samples within this range. The distribution of grape infection (%) assessed by visual inspection for the hand-picked and the mechanically sampled grapes is shown in Figure 3⇓. The hand-picked samples covered a wide range of grape infection, representing grapes from healthy (0% infection) to extremely poor quality (>80% Botrytis cinerea infection). As expected, the distribution of mechanically sampled grapes is left-skewed toward the lowest values of grape infection. In fact, when the degradation of the sanitary conditions of grapes begins, grape-growers may harvest the product without close evaluation of the value of other quality parameters of berries (such as sugar content, pH, and color density).
Visual inspection of hand-picked (data set 1) and mechanically sampled (data set 2) grapes was carried out by the same two operators who identified the symptoms of Botrytis cinerea as the main infection present on grapes. Botrytis symptoms are characterized by slipskin lesions and the development of grey-white mycelium on the berries, which may progress to infect whole clusters. Although the occurrence of secondary microorganisms in the grapes cannot be excluded, the microbiological characterization of the pathogen populations is beyond the scope of the present investigation.
The model development for assessing grape quality at harvest involved FT-MIR analysis of hand-picked grapes and mechanically sampled grapes. A preliminary attempt to correlate the visual inspection with the FT-MIR was conducted. The FT-MIR showed a poor ability (R2 = 0.777) to predict grape quality on the basis of visual inspection of hand-picked grapes (Figure 4⇓), and a total lack of fitting was found in mechanically sampled grapes (R2 = 0.1). This result confirmed the difficulty of visual inspection of grapes at the winery and stressed the need for a suitable reference method (such as HPLC of chemical markers).
The correlation between methods for measuring grape quality is shown in Table 2⇓. The fit correlation found between the visual inspection of the sanitary status of hand-picked grapes (set 1) with glycerol (r = 0.892) and gluconic acid (r = 0.827) confirmed the suitability of these compounds as chemical markers of grape infection and underlined the appropriateness of visual inspection of hand-picked grapes. However, the promising results obtained on set 1 were only partially confirmed on grapes mechanically sampled from the truck upon arrival at the commercial wineries (set 2) (glycerol with grape infection: r = 0.41). This result was mostly related to the poor visual inspection of set 2. Further evidence of the difficulty of visual inspection of set 2 is indicated by the positive intercept values (a) found between visual inspection with gluconic acid (r = 0.586; a = 3.2 g/L) and with glycerol (r = 0.225; a = 1.4 g/L), suggesting that grapes with infection value 0% showed an average of gluconic acid and glycerol of 3.2 g/L and 1.4 g/L, respectively. These findings can be explained by the presence of latent (not visible) Botrytis that under some conditions may account for a high percentage of berry infections. A lack of correlation has also been found between visual inspection of grape infection and FT-MIR in Grenache, Syrah, Carignane, Merlot, and Marselan grapes (Rousseau et al. 2002).
The good correlation between gluconic acid and glycerol found in hand-picked grapes (r = 0.928) supported the correct identification of Botrytis cinerea as the main infection agent in grapes (Table 2⇑). On the other hand, the low correlation between gluconic acid and glycerol in mechanically sampled grapes (r = 0.758) may indicate the presence on infected grapes of additional microorganisms with a metabolism different from Botrytis cinerea. Data from the literature confirms that wines from healthy grapes have low glycerol, whereas in wines from grapes infected by Botrytis cinerea the concentration of glycerol and gluconic acid increased proportionately (Slobodan 1980). In the case of Botrytis-infected berries, the content of both compounds increases considerably, depending on the level of infection, with gluconic acid up to 6.5 g/L and glycerol up to 31.7 g/L (Holbach and Woller 1976). Research on the role of glycerol (range 0–6 g/L) and gluconic acid (range 0–15 g/L) as suitable chemical markers for Botrytis infection considers gluconic acid and glycerol as markers for Botrytis cinerea, acetic acid and sorbitol for acetic bacteria, mannitol and butanediol for yeasts, and lactic acid for lactic bacteria (Dubernet and Dubernet 2000).
Conclusions
Visual inspection for quality of botrytized grapes is problematic. However, gluconic acid and glycerol are suitable chemical markers of grape infection and they can be rapidly measured by FT-MIR for daily routine analysis. Sampling procedure, timing of the analysis, and reliability of the data used for calibration are among the critical factors in the application of this technique. Taking into account that the development and metabolism of Botrytis infection can depend on the grape type, the prediction model can be improved using selected sample sets with common characteristics. For optimal results with FT-MIR, calibration and reference analysis should be carried out rapidly on a high number (>75) of fresh grapes.
Footnotes
Acknowledgments: The authors acknowledge Regione Emilia-Romagna, Italy, for financial support (L.R. n. 28, project 11653-11052) and the staff of CRPV (Faenza, RA, Italy) for technical support.
- Received July 2007.
- Revision received October 2007.
- Revision received February 2008.
- Copyright © 2008 by the American Society for Enology and Viticulture