Wine samples

The study included 130 bottles of commercial wines representing six varieties, two viticultural regions, six subregions, four zones/areas (Figure 1), and three different vintages (Table 1). These samples were donated by different wineries or purchased from local stores and were selected based on a similar price range (approximately US$ 5 to 10). The alcohol strength of the samples varied between 12.5 and 14% for white wines and 13 to 14.5% v/v for reds. Their pH ranged from 2.99 to 3.44 for white wines and 3.04 to 3.68 for reds (Table a, supplemental information).

Figure 1 Wine regions of Chile. 

Reagents and solutions

All solutions were prepared using ultra-pure water, obtained from a Millipore system (Bedford, MA, USA). Nitric acid (HNO₃, 65%) and hydrogen peroxide (H₂O₂, 30%) were purchased from Sigma Aldrich (Darmstadt, Germany), and Lanthanum oxide (La₂O₃) was bought from Merck (Darmstadt, Germany). The calibration curves for K, Na, Mg, Ca, Fe, Zn, Cu and Mn were prepared from certified individual commercial solutions of 1000 mg L^-1 as explained below (Spex Industries Inc., Edison, NJ, USA). Additionally, a Certipur® ICP multi-element standard solution of 1000 mg L^-1 obtained from Merck was used to develop the calibration curves for the analyses of microelements.

Glassware was cleaned by soaking in 10% v/v HNO₃ solution for 24 h and rinsed at least three times with ultra-pure water. Then, the material was dried and stored in a class-100 laminar flow hood.

Calibration curves

Five-point calibration curves were prepared for each of the seventeen metals studied (i.e., K, Na, Ca, Zn, Cu, Fe, Mg, Mn, Ni, Cr, Al, Pb, Cd, Hg, Se, Co and Sn). For the atomic absorption spectroscopy (AAS) analyses, they were prepared from certified standard solutions of 1000 mg L^-1 of each metal. The concentration ranges covered for each calibration curve were as follows: Ca, 1-25 mg L^-1; Mg, 0.25-5 mg L^-1; K, 2-20 mg L^-1; Na, 1-10 mg L^-1, Mn, 1-6 mg L^-1, Cu, 0.5-5 mg L^-1, Fe, 1-10 mg L^-1, and Zn, 0.5-3 mg L^-1. Before measuring the macroelements (Ca, Mg, K and Na) in the AAS, a 1:10 v/v dilution of the samples was performed with La₂O₃ (1.1 g L^-1 solution). Conversely, Mn, Cu, Fe and Zn were determined with no dilution because of their low concentration.

The inductively coupled plasma-mass spectrometry (ICP-MS) analyses were based on the methodology reported by Aceto and collaborators (^{Aceto et al. 2013}). The certified standard solution was a mixture of 5% HNO₃ solution (w/v) with the following additions: 10.0 mg L^-1 for Ca, Cd, Cr, Fe, Na, Ni, Pb, Se, and Zn; 10.1 mg L^-1 for Al, Co, Cu, Mn, Rb, and Sr; and 9.9 mg L^-1 for Hg, K, and Mg. From this solution, a calibration curve was prepared with the following concentration points: 0.0002, 0.0005, 0.001, 0.003, 0.007, 0.01, 0.03, 0.05, 0.07, and 0.1 mg L^-1, using nitric acid (Suprapur®) for dilution (5% w/v). The calibration curves obtained had coefficients of determination, R², ranging between 0.997 and 0.999.

Wine analyses

Once opened, the wine from each bottle was subjected to microwave-assisted acid digestion in order to eliminate organic interfering compounds (^{Alkıs, et al., 2014}; ^{Grindlay, et al., 2009}). Five mL of wine, 3 mL of 65% HNO₃ and 1.5 mL of H₂O₂ 30% were placed in 55 mL Teflon® tubes (MARS Xpress, Matthews, NC, USA). The samples were manually shaken for approximately 10 seconds and left to rest for 24 h before proceeding with the microwave-assisted heating program, which included a ramp time of 5 minutes going from 20 to 175 °C with a 3 min hold time, followed by a 5 min ramp from 175 to 30 °C with a hold time of 3 min. This cycle was repeated twice at 800 W of applied power.

Once the samples were digested, the content of each Teflon tube was transferred into a 50 mL volumetric flask and filled with ultra-pure water before the analyses in the AAS or ICP-MS were performed.

The analysis for the most concentrated elements (i.e., Ca, K, Na, Mg, Cu, Mn, Fe and Zn) was completed in an AAS (280FS AA, Agilent Technologies), whereas trace metal ions (i.e., Al, Cd, Cr, Hg, Ni, Pb, Se, Sn and Co) were analyzed with an ICP-MS (7700x, Agilent Technologies), using argon gas to form the plasma for the instrument. The ICP-MS was equipped with an ASX 500-CETAC auto-sampler, a micro-concentric nebulizer, nickel cones, and a peristaltic sample delivery pump. The operation conditions of the AAS and ICP-MS are detailed in Table 2.

Each wine bottle was sampled, pre-treated, and analyzed three times. Each of the analytical readings was also repeated three times.

Data analysis

The metal content of the samples was analyzed using descriptive statistics (i.e., mean, range, standard deviation, and median) and presented as the average of all the wine samples, or separated into white and red wine subgroups.

One-way analysis of variance (ANOVA), followed by a post hoc Tukey test, was conducted for each element, looking for metals that could allow for the grouping of samples according to their place of origin (p≤0.05). Additionally, principal component analysis (PCA) was used to identify the main elements that explain the data variance, whereas linear discriminant analysis (LDA) was used for sample discrimination (^{Angus et al. 2006}; ^{Laurie et al., 2010}; ^{Coetzee et al. 2014}). These latter analyses (i.e., PCA and LDA) were performed separately for white and red wines, excluding the samples representing the Central Valley (n=30 white and 70 red wine samples). These analyses were done at a subregional level to avoid collinearity between samples of different hierarchical classification.

To avoid attributing subregional differences to wines whose metal profiles could have been influenced by varietal disparities, multivariate analyses were conducted only with metals that did not show significant differences according to varieties (e.g., Zn, Ca, Na, Mn, Fe, Cu, K, Al, Sn, Pb, Ni, and Cr in white wine and Zn, Ca, Na, Mn, Mg, Fe, Cu, K, Al, Sn, Ni, and Cr in red wine) (Table b, supplemental information).

In all cases, the statistical analyses were performed using Statgraphics Centurion XV (Statpoint technologies, Warrenton, VA, USA), and graphical representations were made using Sigma-Plot version 12 (Systat Software Inc. Chicago, IL, USA).

Content of metals and wine discrimination according to geographic area

Statistical differences were observed for several of the metals analyzed (p≤0.05) when the samples were grouped and compared based on the location declared on their labels. For example, Maipo wines (predominantly producing Cabernet Sauvignon) had a higher concentration of K when compared with Aconcagua, Casablanca (predominantly producing white wines), and Curicó valleys. Colchagua and Maipo wines had more Mn than Curicó samples. Casablanca wines had more Na than most of the other wines. Colchagua wines had more Zn than Lontué and Loncomilla samples. Metal elements such as Ca, Cu, Fe, Al, Ni, Sn and Pb did not show significant differences among the wines compared from different geographic areas (Figure a, supplemental information).

Geological and climatic differences seemed to be decisive in the concentration of some of these elements (^{Zou et al., 2012}; ^{Coetzee et al., 2014}; ^{Paneque et al., 2017}). For example, a high Na content in Casablanca could be due to the presence of saline ground water or sea winds that contain salty water could reach some of these vineyards (^{Pohl, 2007}; ^{Blume et al., 2015}). This could be exacerbated by the limited rainfall experienced in recent years; this is similar to semi-arid areas where these types of cations accumulate in the soil due to the scarcity of rainfall (^{González et al., 1996}; ^{Sierra et al., 2001}).

Principal component analysis

Given the differences observed in the concentration of metals among the geographic areas studied, PCA was performed to identify the main vectors that explain the variability for these samples. An initial multivariate analysis was carried out for data reduction, allowing for the removal of elements that had strong linear correlations (^{Hopfer et al., 2015}). Metal elements that showed significant differences in their concentration between white and red varieties were also excluded from the multivariate analyses (Mg in case of white varieties, and Pb in case of red varieties), thus limiting the influence of varietal differences on the geographic area discrimination.

When a PCA was performed comparing red and white wines, 84.50% of the variability was explained by the first two principal components. In this case, the variability was mainly explained by the content of Na and the concentration of Mg and K (Figure 2a). Consequently, further analyses were performed separately for white and red wines; values for the first two principal components reached 95.28% for white wines and 77.69% for reds (Figure 2b and c). The main components that explained the variability among groups of samples were clearly identified in white wines but not as much in reds, where a higher degree of overlapping among neighboring sub-regions and zones (and even areas) was observed (e.g., Rapel, Maule, San Javier, Loncomilla). This finding was probably because Maule is the largest viticultural area in Chile (Figure 1) with approximately 30,296.1 Km² that spans from the coast to the Andes (^{ODEPA 2016}).

Figure 2 Score plot of the first two principal components (principal component analysis, PCA) of (a) 130 Chilean wine samples grouped by the type of wine, (b) 39 white wines grouped by geographic origin, and (c) 91 red wines grouped by geographic origin. 

Linear Discriminant Analysis

Given the results obtained by PCA, linear discriminant analysis (LDA) was used to divide the sample space, to separate the best possible subgroups, and to discriminate wines according to geographic area. As before, red and white samples were treated separately.

In the case of white wines, the magnitudes of twelve independent variables (i.e., Ca, K, Cu, Na, Mn, Fe, Zn, Al, Cr, Ni, Sn and Pb) were used as predictors to discriminate thirty wine samples according to their places of origin. The model allowed for a 96.9% discrimination of white wines according to their places of origin (Figure 3a). As shown in Figure 3a, the samples representing the five subregions studied were well grouped. As indicated before, the wines labeled under the Central Valley denomination were eliminated from these analyses because they include wines containing grapes may have come from different geographic origins. A minimum of 75% of the grapes should come from a specific region to label a wine from a specific place in Chile (^{Ministerio de Agricultura, 1995}).

Figure 3 Scatter plot of the linear discriminant functions for (a) 39 samples of white wine analyzed according the region of origin, and (b) 91 samples of red wine analyzed according the region of origin. 

In the case of red wines, twelve predictor variables (Ca, K, Cu, Na, Mn, Fe, Zn, Al, Cr, Ni, Sn and Mg) were used to discriminate seventy samples according to their place of origin. In this case, the model allowed for 79.43% sample discrimination (Figure 3b). The wine samples representing the Maule subregion had the highest variability, as they comprise a high diversity of mesoclimates and soils. Maule valley is the largest viticultural origin in Chile with approximately 53,497 ha, representing 36.7% of the national surface area of vineyards dedicated to winemaking (^{ODEPA, 2016}).

As indicated elsewhere (^{Zou et al., 2012}; ^{Versari et al., 2014}), the metal composition of wines is a useful way to characterize and discriminate wines according to their geographic area; they largely reflect the composition of metals in the soil and water of the places in which the grapes are grown (^{Pohl, 2007}). The limitations of this technique include the purity of the wines analyzed, variations in winemaking style, the use of winemaking aids, yeast culture and vinification practices, etc. (^{Pohl, 2007}; ^{Rossano et al., 2007}; ^{Álvarez et al., 2007}; ^{Hopfer et al., 2015}). In this case, the samples selected had a limited range of prices and winemaking styles, thus allowing for a good degree of geographic area discrimination. A more complete study that allows for the determination of the origin of a larger group of wines, the influence of their purity (i.e., unblended vs. blended), production and storage practices, and a greater number of samples per zone is necessary.

The main conclusions are as follows. All metal concentrations were within the ranges previously reported for wines from other regions. As expected, red wines had a higher concentration of K, Mg and Fe compared to white wines, in which Na, Ca, and Zn were particularly concentrated. The exploratory data analysis conducted with PCA allowed for the identification of the main components explaining the variability among samples. In the case of white wines, the first two main components explained 95.28% of the variability, whereas in red wines only 77.69% of the variability was explained by the first two principal components. On the other hand, LDA analysis allowed for a good degree of discrimination by geographic area for white (i.e., 96.9%) and red wines (i.e., 79.43%). To the best of our knowledge, this is the first time that the content of a large number of metallic elements in Chilean wines has been described and used to discriminate wine origins. No previous attempts to discriminate Chilean wine areas, based on a broad range of multi-elemental compositional data, have been published.