S. Đurđića,
M. Pantelićb,
J. Trifkovića,
V. Vukojevića,
M. Natića,
Ž. Tešića and
J. Mutić*ac
aUniversity of Belgrade, Faculty of Chemistry, P.O. Box 51, 11158 Belgrade, Serbia
bInnovation Centre of Faculty of Chemistry Ltd, Studentski trg 12-16, 11000, Belgrade, Serbia
cGhent University Global Campus, Incheon, South Korea. E-mail: jelena.mutic@ghent.ac.kr
First published on 12th January 2017
The elemental profiles of 63 red and white wine samples from four different regions in Serbia were investigated. Twenty-one elements were analysed (Ca, Mg, Na, K, Fe, Mn, Cu, Zn, Co, Se, Cr, V, Ni, Cd, As, Al, Sb, Pb, Ba, Rb, and Be) by inductively coupled plasma quadrupole mass spectrometry (ICP-Q-MS) and inductively coupled plasma with optical emission spectrometry (ICP-OES). A pattern recognition method was applied in order to classify and differentiate type, seasonal variability, and geographical origin of the wine. Dietary mineral intake for elements was calculated in order to assess their contribution to daily intake. The most important descriptors for discrimination among red and white wine samples were Be, Al, Rb, Mg, K, Cu, Mn, and Na, in descending order. The variables Cd, Pb, As, Sb, V, Na, K, and Zn have the highest influence on vintage-to-vintage classification of red wines. Furthermore, the model revealed the existence of three groups of descriptors for different regions of production. All obtained statistical models confirmed that data from the elemental content of wine samples could be used for accurate prediction of wine type, seasonal variability, and regional origin.
Serbia has a very long tradition of wine production due to its favourable climate and good geographical conditions, but export capacity is very small due to the lack of competitiveness in price as a consequence of the high cost of production in relatively small wineries. The elevated prices are justified by the wine quality and by the relationship between the quality and the region of origin. The evaluation of the quality and authenticity of the wine is a key factor in establishing cost-effective wine, which would help numerous small Serbian wineries to enter this highly competitive market. In that sense, the aim of this work was, firstly, to obtain basic information about the elemental composition of genuine Serbian wines. The second objective was to apply chemometric techniques, such as principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA), in order to establish parameters that could be used for classification and differentiation of two types of wine (white and red), to determine which elements could be affected by seasonal variability between wines from one winery, and to try to establish the regional origin of red wine samples. A significant number of samples supplied by different wineries were collected from three different regions of Serbia. To the best of our knowledge, this is the first report on the influence of vintage on the elemental content of red wine samples analysed with chemometrics. The third objective was to evaluate the content of essential and toxic elements in wine samples in order to study their nutritional significance and to estimate the daily mineral intake provided by the consumption of this beverage.
Due to the lack of a reference material for wine, the accuracy of the measurements of elements was verified by CRM 1640a (trace elements in natural water) which contained all analysed elements. The results of the analysis showed good agreement with the certified levels of the standard (±10%).
All data were auto-scaled prior to any multivariate analysis to bring values to compatible units. Metal content served as the input data matrix. The PCA was performed using a singular value decomposition algorithm with a 0.95 confidence level for Q and T2 Hotelling limits for outliers. The analysis was based on a correlation matrix, and factors with eigenvalues greater than 1 were retained. The PLS-DA used the SIMPLS algorithm without forcing orthogonal conditions on the model in order to condense Y-block variance into the first latent variables. The models were validated using a venetian blinds validation procedure. The quality of the models was monitored with two different sets of parameters. The values of Rcal2, the cumulative sum of squares of the Ys explained by all extracted components, RCV2, the cumulative fraction of the total variation of the Ys that can be predicted by all extracted components, and RPRED2, the cumulative fraction of the total variation of the Ys that can be predicted by test components should be as high as possible. In contrast, RMSEC (Root Mean Square Errors of Calibration), RMSECV (Root Mean Square Errors of Cross-Validation), and RMSEP (Root Mean Square Errors of Prediction) should be as low as possible, with minimal differences between them.24
Wine type | K (mg L−1) | Mg (mg L−1) | Ca (mg L−1) | Na (mg L−1) | Fe (mg L−1) | Rb (mg L−1) | Mn (μg L−1) | Al (μg L−1) | Zn (μg L−1) | Cu (μg L−1) | V (μg L−1) | Ba (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
a Differences between two set of data is significant when P value is less or equal to P = 0.05. | |||||||||||||
Red wine (n = 46) | Mean | 690 | 94.9 | 83.1 | 8.48 | 1.32 | 2.58 | 1194 | 225 | 635 | 131 | 1.54 | 90.3 |
Stdev | 173 | 15.7 | 21.3 | 7.45 | 0.93 | 0.72 | 681 | 186 | 263 | 135 | 2.58 | 66.4 | |
Median | 656 | 97.3 | 80.4 | 5.70 | 1.23 | 2.44 | 994 | 175 | 598 | 81 | 0.34 | 73.5 | |
Min | 413 | 53.0 | 41.2 | 2.94 | 0.01 | 1.25 | 385 | 6 | 287 | 24 | 0.02 | 20.6 | |
Max | 1167 | 124.8 | 131.2 | 37.0 | 6.01 | 4.70 | 3310 | 800 | 1499 | 646 | 14.62 | 301.2 | |
White wine (n = 17) | Mean | 396 | 66.6 | 84.6 | 17.6 | 1.54 | 1.24 | 550 | 1145 | 977 | 558 | 3.07 | 79.9 |
Stdev | 71 | 14.5 | 16.2 | 12.0 | 0.68 | 0.28 | 178 | 469 | 693 | 515 | 1.75 | 33.7 | |
Median | 400 | 68.1 | 83.8 | 13.7 | 1.37 | 1.20 | 562 | 1004 | 837 | 501 | 3.40 | 70.5 | |
Min | 298 | 38.0 | 55.3 | 4.5 | 0.36 | 0.79 | 294 | 514 | 215 | 33 | 0.83 | 35.4 | |
Max | 560 | 89.4 | 115.7 | 34.3 | 2.80 | 1.71 | 885 | 2189 | 2585 | 1684 | 7.22 | 157.4 | |
Man–Whitney U-testa | P value | <0.0001 | <0.0001 | 0.5562 | 0.0013 | 0.1371 | <0.0001 | <0.0001 | <0.0001 | 0.0829 | 0.0008 | 0.0001 | 0.8892 |
H0 | Reject | Reject | Accept | Reject | Accept | Reject | Reject | Reject | Accept | Reject | Reject | Accept |
Wine type | Pb (μg L−1) | Ni (μg L−1) | Cr (μg L−1) | Sb (μg L−1) | As (μg L−1) | Co (μg L−1) | Be (μg L−1) | Cd (μg L−1) | Se (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|
Red wine (n = 46) | Mean | 47.8 | 37.5 | 5.49 | 8.78 | 16.1 | 3.89 | 0.129 | 1.99 | 3.40 |
Stdev | 40.4 | 28.4 | 4.90 | 4.76 | 10.9 | 2.28 | 0.196 | 2.53 | 3.02 | |
Median | 26.3 | 27.8 | 4.73 | 8.50 | 18.1 | 3.59 | 0.023 | 0.82 | 2.31 | |
Min | 9.3 | 8.5 | 0.10 | 0.66 | 1.2 | 0.40 | 0.001 | 0.07 | 0.06 | |
Max | 168.5 | 123.6 | 20.67 | 18.65 | 37.9 | 9.16 | 1.028 | 11.1 | 10.52 | |
White wine (n = 17) | Mean | 83.8 | 88.0 | 1.85 | 8.95 | 21.2 | 3.96 | 4.34 | 1.31 | 4.98 |
Stdev | 52.5 | 122.9 | 1.72 | 5.92 | 9.46 | 2.60 | 1.55 | 1.47 | 2.07 | |
Median | 68.8 | 41.4 | 1.34 | 10.54 | 23.0 | 3.65 | 4.00 | 0.71 | 5.39 | |
Min | 16.0 | 8.58 | 0.10 | 0.06 | 2.00 | 0.50 | 0.39 | 0.14 | 1.19 | |
Max | 188.2 | 420.5 | 6.59 | 17.77 | 35.7 | 9.60 | 7.01 | 4.75 | 8.00 | |
Man–Whitney U-test | P value | 0.0051 | 0.0492 | 0.0024 | 0.7217 | 0.0945 | 0.9938 | <0.0001 | 0.9629 | 0.0132 |
H0 | Reject | Reject | Reject | Accept | Accept | Accept | Reject | Accept | Reject |
Vintage | K (mg L−1) | Mg (mg L−1) | Ca (mg L−1) | Na (mg L−1) | Fe (mg L−1) | Rb (mg L−1) | Mn (μg L−1) | Al (μg L−1) | Zn (μg L−1) | Cu (μg L−1) | V (μg L−1) | Ba (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2012 (A) | Mean | 539 | 110.9 | 89.7 | 5.87 | 1.77 | 2.38 | 776 | 250 | 1013 | 230.2 | 0.82 | 66.4 |
Stdev | 155 | 9.7 | 14.7 | 0.92 | 1.77 | 0.72 | 225 | 127 | 271 | 208.5 | 1.20 | 15.5 | |
Median | 493 | 110.3 | 84.2 | 6.09 | 1.13 | 2.25 | 784 | 187 | 920 | 131.3 | 0.20 | 61.5 | |
Min | 413 | 97.0 | 73.3 | 4.66 | 0.70 | 1.58 | 496 | 124 | 738 | 80.9 | 0.03 | 48.1 | |
Max | 903 | 124.8 | 112.9 | 7.30 | 6.01 | 3.69 | 1169 | 488 | 1499 | 645.8 | 3.32 | 88.6 | |
2013 (B) | Mean | 635 | 94.5 | 78.0 | 4.41 | 1.42 | 2.53 | 966 | 112 | 420 | 63.1 | 0.88 | 43.5 |
Stdev | 188 | 13.6 | 14.0 | 1.22 | 0.37 | 0.72 | 240 | 112 | 134 | 21.5 | 0.87 | 17.9 | |
Median | 613 | 95.4 | 75.7 | 4.01 | 1.36 | 2.58 | 995 | 66 | 357 | 71.1 | 0.34 | 38.1 | |
Min | 430 | 59.3 | 55.0 | 2.94 | 0.86 | 1.25 | 614 | 6 | 292 | 24.4 | 0.09 | 20.6 | |
Max | 1167 | 109.9 | 100.7 | 7.08 | 2.26 | 4.22 | 1384 | 356 | 742 | 89.2 | 2.33 | 75.4 | |
2014 (C) | Mean | 897 | 99.1 | 120.1 | 3.98 | 1.23 | 3.37 | 1036 | 221 | 671 | 153.0 | 0.59 | 73.2 |
Stdev | 73 | 4.9 | 7.9 | 0.57 | 0.24 | 1.03 | 125 | 160 | 97 | 116.8 | 0.80 | 7.9 | |
Median | 915 | 99.3 | 118.3 | 3.80 | 1.23 | 3.12 | 1015 | 165 | 690 | 137.4 | 0.23 | 76.1 | |
Min | 797 | 93.2 | 112.7 | 3.55 | 0.99 | 2.51 | 907 | 99 | 549 | 49.1 | 0.10 | 61.9 | |
Max | 961 | 104.5 | 131.2 | 4.78 | 1.49 | 4.70 | 1207 | 455 | 757 | 287.9 | 1.79 | 78.8 | |
Kruskal–Wallis test | P value | 0.0120 | 0.0181 | 0.0047 | 0.0133 | 0.6873 | 0.1717 | 0.1144 | 0.0918 | 0.0002 | 0.0041 | 0.6565 | 0.0127 |
Z-Test | C(A, B) | A(B) | C(A, B) | B(A, C) | — | — | — | — | A(B) | A(B) | — | A(B, C) |
Vintage | Pb (μg L−1) | Ni (μg L−1) | Cr (μg L−1) | Sb (μg L−1) | As (μg L−1) | Co (μg L−1) | Be (μg L−1) | Cd (μg L−1) | Se (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|
2012 (A) | Mean | 44.1 | 34.5 | 3.63 | 13.56 | 19.5 | 3.65 | 0.049 | 1.07 | 2.88 |
Stdev | 39.2 | 28.7 | 1.76 | 2.57 | 4.8 | 1.98 | 0.052 | 1.43 | 1.70 | |
Median | 26.0 | 21.9 | 3.59 | 12.46 | 19.0 | 3.28 | 0.023 | 0.55 | 3.57 | |
Min | 15.8 | 16.9 | 0.22 | 10.42 | 12.6 | 0.50 | 0.011 | 0.07 | 0.42 | |
Max | 128.6 | 102.1 | 5.89 | 17.51 | 26.7 | 7.35 | 0.138 | 4.44 | 4.76 | |
2013 (B) | Mean | 49.4 | 18.6 | 2.33 | 6.18 | 23.2 | 3.12 | 0.042 | 2.84 | 3.40 |
Stdev | 34.7 | 8.9 | 1.92 | 3.13 | 4.9 | 2.35 | 0.040 | 2.39 | 2.61 | |
Median | 44.2 | 15.8 | 2.12 | 6.56 | 22.1 | 2.49 | 0.023 | 2.82 | 3.57 | |
Min | 9.3 | 8.8 | 0.10 | 0.66 | 15.2 | 0.50 | 0.023 | 0.07 | 0.42 | |
Max | 100.3 | 41.1 | 5.10 | 11.13 | 30.7 | 6.81 | 0.131 | 6.43 | 6.67 | |
2014 (C) | Mean | 55.7 | 26.2 | 4.27 | 6.89 | 20.2 | 4.03 | 0.103 | 3.33 | 6.03 |
Stdev | 75.4 | 3.4 | 2.21 | 5.86 | 3.4 | 2.90 | 0.096 | 5.19 | 2.46 | |
Median | 20.8 | 25.5 | 4.26 | 4.95 | 21.5 | 3.65 | 0.087 | 0.87 | 6.67 | |
Min | 12.6 | 23.3 | 2.15 | 2.20 | 15.2 | 1.01 | 0.023 | 0.49 | 2.82 | |
Max | 168.5 | 30.7 | 6.42 | 15.45 | 22.7 | 7.80 | 0.214 | 11.10 | 7.96 | |
Kruskal–Wallis test | P value | 0.8704 | 0.0249 | 0.1533 | 0.0022 | 0.2887 | 0.7373 | 0.4723 | 0.3384 | 0.1319 |
Z-Test | — | A(B, C) | — | B(A, C) | — | — | — | — | — |
Regional origin | K (mg L−1) | Mg (mg L−1) | Ca (mg L−1) | Na (mg L−1) | Fe (mg L−1) | Rb (mg L−1) | Mn (μg L−1) | Al (μg L−1) | Zn (μg L−1) | Cu (μg L−1) | V (μg L−1) | Ba (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Central Serbia (1) | Mean | 762 | 88.2 | 74.5 | 11.00 | 0.64 | 2.63 | 1632 | 325 | 617 | 176.4 | 2.73 | 175.2 |
Stdev | 104 | 14.9 | 12.4 | 5.82 | 0.54 | 0.64 | 1056 | 247 | 178 | 184.4 | 4.56 | 76.5 | |
Median | 783 | 92.9 | 74.6 | 8.98 | 0.50 | 2.32 | 1172 | 305 | 601 | 110.7 | 1.72 | 138.9 | |
Min | 573 | 53.1 | 53.8 | 3.07 | 0.01 | 1.99 | 385 | 9 | 287 | 44.7 | 0.20 | 94.3 | |
Max | 877 | 102.6 | 93.9 | 21.26 | 1.62 | 3.69 | 3310 | 800 | 948 | 619.2 | 14.62 | 301.2 | |
Vojvodina (2) | Mean | 678 | 101.9 | 78.0 | 23.02 | 2.01 | 2.13 | 1884 | 312 | 586 | 142.1 | 2.80 | 89.4 |
Stdev | 74 | 16.3 | 7.3 | 11.68 | 0.52 | 0.32 | 718 | 201 | 207 | 38.4 | 2.62 | 47.4 | |
Median | 677 | 107.6 | 78.9 | 23.44 | 1.83 | 2.19 | 1921 | 321 | 511 | 150.0 | 3.20 | 79.1 | |
Min | 607 | 75.6 | 68.7 | 5.65 | 1.42 | 1.77 | 774 | 9 | 375 | 92.6 | 0.20 | 27.9 | |
Max | 798 | 118.7 | 86.0 | 37.05 | 2.81 | 2.51 | 2609 | 574 | 834 | 182.3 | 6.28 | 155.1 | |
South Serbia (3) | Mean | 735 | 78.0 | 71.4 | 7.33 | 0.91 | 2.72 | 1141 | 148 | 567 | 68.3 | 0.67 | 99.6 |
Stdev | 179 | 11.4 | 29.3 | 5.81 | 0.45 | 0.74 | 645 | 138 | 132 | 15.0 | 0.70 | 81.3 | |
Median | 621 | 76.5 | 67.2 | 4.82 | 0.69 | 2.35 | 895 | 122 | 550 | 72.5 | 0.20 | 62.8 | |
Min | 550 | 66.9 | 41.2 | 4.02 | 0.52 | 2.02 | 594 | 10 | 403 | 40.1 | 0.19 | 30.9 | |
Max | 998 | 100.1 | 127.4 | 20.30 | 1.87 | 4.20 | 2456 | 437 | 766 | 80.2 | 1.90 | 260.0 | |
Kruskal–Wallis test | P value | 0.4768 | 0.0484 | 0.4668 | 0.0293 | 0.0058 | 0.1533 | 0.2232 | 0.2862 | 0.8237 | 0.0224 | 0.2558 | 0.0817 |
Z-Test | — | 2(3) | — | 2(3) | 2(1) | — | — | — | — | 2(3) | — | — |
Regional origin | Pb (μg L−1) | Ni (μg L−1) | Cr (μg L−1) | Sb (μg L−1) | As (μg L−1) | Co (μg L−1) | Be (μg L−1) | Cd (μg L−1) | Se (μg L−1) | |
---|---|---|---|---|---|---|---|---|---|---|
Central Serbia (1) | Mean | 61.9 | 67.7 | 10.44 | 9.78 | 5.1 | 4.75 | 0.292 | 2.38 | 2.02 |
Stdev | 50.6 | 39.2 | 6.25 | 3.27 | 8.7 | 2.05 | 0.313 | 2.80 | 3.29 | |
Median | 61.5 | 52.1 | 10.19 | 10.59 | 2.3 | 4.39 | 0.262 | 0.35 | 0.42 | |
Min | 11.5 | 8.5 | 0.10 | 4.02 | 1.2 | 2.44 | 0.023 | 0.15 | 0.06 | |
Max | 154.7 | 123.6 | 20.67 | 14.48 | 28.3 | 9.16 | 1.028 | 7.31 | 10.52 | |
Vojvodina (2) | Mean | 53.6 | 30.1 | 6.74 | 6.51 | 13.1 | 4.61 | 0.120 | 1.87 | 2.64 |
Stdev | 31.1 | 17.7 | 4.23 | 7.23 | 10.8 | 2.61 | 0.156 | 2.37 | 2.36 | |
Median | 43.9 | 18.4 | 6.15 | 2.89 | 9.5 | 5.00 | 0.023 | 1.13 | 1.62 | |
Min | 27.9 | 15.8 | 1.06 | 0.69 | 5.0 | 0.40 | 0.023 | 0.13 | 0.93 | |
Max | 107.0 | 52.7 | 12.88 | 18.65 | 32.1 | 7.59 | 0.382 | 5.85 | 6.67 | |
South Serbia (3) | Mean | 21.8 | 49.7 | 6.11 | 10.14 | 12.3 | 4.24 | 0.135 | 0.54 | 4.38 |
Stdev | 11.9 | 12.9 | 6.15 | 3.97 | 16.4 | 2.29 | 0.227 | 0.96 | 4.50 | |
Median | 16.8 | 49.0 | 7.70 | 9.38 | 3.5 | 4.53 | 0.023 | 0.13 | 1.64 | |
Min | 14.5 | 33.1 | 0.10 | 4.72 | 1.4 | 0.59 | 0.001 | 0.07 | 0.47 | |
Max | 47.4 | 70.5 | 15.15 | 16.91 | 37.9 | 7.84 | 0.609 | 2.68 | 10.52 | |
Kruskal–Wallis test | P value | 0.0916 | 0.1984 | 0.5340 | 0.3049 | 0.0802 | 0.8596 | 0.3375 | 0.1035 | 0.2027 |
Z-Test | — | — | — | — | — | — | — | — | — |
Regarding the content of toxic elements (As, Cd, and Pb) (Tables 1–3), the As content was higher than in published data (Table S3†)16,27,30,31,36 while the Pb content was lower than that in Czech30 and Romanian33 wines. The average content of Cd was slightly higher than in wines from Serbia's neighbouring countries Croatia16 and Macedonia.5 Only one sample contained elevated concentrations of this element. Analysing data from Table S3,† Serbian wines were concluded to contain higher amounts of essential elements (Cu, Mn, and Zn) and major elements (Ca and Mg) in comparison to Belgian32 or Argentinian31 wines. Additionally, our wines contained less Ni, Pb, and V than Turkish,36 Czech,30 and Romanian33 wines, respectively. Although Serbian wines contained higher amounts of arsenic compared to other wines, its concentrations were under the maximum acceptable limit. Due to significant deviations from the normal distribution for each of the studied variables, statistical evaluation of the differences in the elemental content in the two wine types was evaluated by the Mann–Whitney U-test, while the differences in the elemental content according to the vintage and regional origin were determined by the Kruskal–Wallis test. Tests were employed for each variable taking the appropriate variety as a single factor. The results are presented in Tables 1–3. Differences between two or three sets of data were considered significant when the P value was less than or equal to P = 0.05. In the cases where the Kruskal–Wallis test has indicated a statistically significant difference between the medians, the Kruskal–Wallis multiple-comparison Z-value test was also performed. Vintages and regions with different contents of a given metal are denoted in parentheses with a letter or number (Tables 2 and 3).
The Mann–Whitney U-test revealed that the variables K, Mg, Na, Rb, Be, Cr, Ni, Pb, Se, V, Al, Mn, and Cu governed the differences between white and red wines (Table 1) and suggest a totally different elemental profile for the two wine types. The Kruskal–Wallis test revealed that the macro elements K, Mg, Ca, and Na and trace elements Al, Cu, Ba, Ni, and Sb had statistically significant differences between vintages (Table 2). Additionally, the statistically significant differences between the metal content of samples from different geographical regions could only be observed for the elements Mg, Na, Fe, and Cu with the differentiation in samples arising from the Vojvodina region (Table 3). Pearson correlation coefficients between all elemental contents in the wine samples were determined separately for white and red wines due to the observed differences in relationships between elements (Tables S4 and S5, ESI†). Significant r values at the 99% confidence level37 are represented in bold. A significant correlation between two or more elements indicates a similar ability to penetrate the grape or the existence of the same source.16 Correlation matrices (Tables S4 and S5†) showed that the different elements were mutually correlated for the two wine types indicating different sources of elements in red and white wines. Due to the high number of samples and the fact that t value for the correlation is directly proportional to the number of samples, a significant r value was very low in both cases. For that reason, herein we will discuss only dependencies with r > 0.700. For white wine samples, correlations were obtained between the following elements: Pb and Zn (r = 0.9177), Pb and Cu (r = 0.8824), Zn and Cu (r = 0.9247), and Mn with Ca (r = 0.7401) (Table S4†). The correlations between Pb and these two essential metals could be explained by their similarity as micronutrients or by anthropogenic influences. Furthermore, the existence of a correlation between Mn and Ca could be attributed to the type of soil, which is rich in manganese, iron oxide minerals, and hydrated silicate of Na, Ca, Al, and Mg. In red wine samples, significant correlations were observed between Ba and Mn (r = 0.7519), Ba and Ni (r = 0.7054), Cd and Pb (r = 0.7527), and Mn and Na (r = 0.7001) (Table S5†). Mn, Ba, and Ni together with alkali and alkaline earth metals belong to the lithophile elements which form the main components of the corresponding soil. Therefore, their origin in wine was likely mostly from the soil. On the contrary, Cd and Pb are toxic elements with similar chemical properties, and they predominately come from anthropogenic sources.
Samples of different wine types were also modelled simultaneously using PLS-DA. The number of the latent variables (LVs) was selected based on the minimum value of RMSECV, which was achieved with one LV. Classification and validation results were expressed as Rcal2, RCV2, Rpred2, RMSEC, RMSECV, and RMSEP values. Two models were statistically significant, with relatively high values of Rcal2, RCV2, and Rpred2 (0.8484, 0.8094 and 0.8382, respectively), as well as low differences between RMSEC, RMSECV, and RMSEP values (0.169, 0.189, and 0.193, respectively). Hence, data from the elemental content of wine samples allows accurate prediction of wine type and provides a useful tool for their determination (Fig. 2c). The contribution of variables that had the strongest influence on differentiation between types was analysed using the variable importance in projection (VIP) value (Fig. 2d). The variables with a VIP score higher than one were considered the most relevant for classification of the wine samples by type. The most important descriptors for discrimination among red and white wine samples were Be, Al, Rb, Mg, K, Cu, Mn, and Na, in descending order. The standardized regression coefficients that indicate in which direction (i.e. positively or negatively) the element is contributing to the model are shown on Fig. 2e and f. The regression coefficients indicated identical but opposite influence of the variables in the two models. The highest positive impacts on classification of red wines arose from the elements Rb, K, Mg, Mn, and Cr. The model for white wines indicated high positive regression coefficients for Be, Al, Cu, Na, Ni, Pb, Zn, V, and Se. These differences could be attributed to differences in winemaking techniques between red and white grapes such as long maceration, length of time in contact with the grape skins, and the filtration process typically undergone by white wines.1,38–40,44
Annual reports for 2012, 2013, and 2014 about average air temperature, total rainfall, and total sunshine hours in Serbia were obtained from the Republic Hydrometeorological Service of Serbia (RHSS).46 According to these data, the three chosen vintages differed by rainfall and number of sunny days. The year 2012 had a high number of sunny days, an average temperature of 14.0 °C, and low rainfall (563 mm); in 2013, the rainfall was within normal limits (680.0 mm) and the average temperature was 11.6 °C. The average rainfall in the Serbia in 2014 was 1017.9 mm, which was the rainiest year in 60 years. The effect of vintage on the average content of some selected elements is presented in Fig. 3. Significant differences were found between the average content of Zn and Fe in the wine samples in 2012 where they were higher in comparison with other two years. The same trend was observed for the trace elements Sb and Ni. The content of Cr, Co, Cd, and Se observed in samples from 2014 was higher than in those from the other two other vintages. Similar results were also found for the content of Pb and Ba (Fig. 3). The variation of soil pH is known to be highly dependent on precipitation amount,14 and the (bio)availability of some elements from soil is influenced by water (in this case from rain). Therefore, a high amount of precipitation changed soil conditions and led to leaching of some elements dependent on their bound in soil substrate to make them available for plants. The changes in ionic strength of water and pH affect leaching of elements which are usually in ion-exchangeable forms in soil such as: Cu, Pb, Ni, As, Co, Cd, and Mg.47
Fig. 3 Results from different vintages represented as an average of selected measured variables of the wines. |
The influence of seasonal variability on the elemental content of red wine samples obtained from the same vineyard and with the same production procedure in the winery Radmilovac (Smederevo, Belgrade region) was analysed by PCA. To our knowledge, this is the first report of vintage-to-vintage influence on the elemental profile of wine. The PCA model resulted in a seven-component model that explained 78.99% of the total variance. The parameters of the PCA model, eigenvalues, and percentage variance captured by each PC are presented in Table S6 (ESI†), and mutual projections of factor scores and their loadings are presented in Fig. 4. Score plots (Fig. 4a and b, seasonal classification) of the model revealed the existence of three distinctive groups of characteristics belonging to different vintages with some overlap of samples from 2013 and 2014. The loading plot (Fig. 4c) implies that variables Cd, Pb, As, Sb, V, Na, K, and Zn have the highest influence on the classification. Interestingly, Cd, Pb, As, and Sb are most probably derived from anthropogenic sources rather than from natural sources. Pb and Cd come predominately from atmospheric pollution or from fungicide treatment. Among these elements, Na, K and Zn are elements that derived from both natural and artificial sources with their concentration determined by many different factors. The results of the PCA confirm that seasonal variability does have a significant influence on the elemental profile of red wines.
Fig. 4 PCA for seasonal and regional classifications: (a) 3D score plots, (b) 2D score plots, and (c) loading plots. |
We tried to determine regional origin of red wine samples collected from three statistical regions of Serbia (Vojvodina, Central region, and South region) and from different wineries within each region, by using the elemental content and chemometrics. The PCA model resulted in a seven-component model that explained 82.92% of the total variance. Parameters of the PCA model, eigenvalues, and percentage variance captured by each PC are presented in Table S6.† The score plots (Fig. 4a and b, regional classification) of the model revealed the existence of three groups of characteristics associated with different regions of production. Using this classification, we could assume that the regional origin of red wine from Serbia could be determined. However, we should emphasize that the three samples in the left upper part of the score plot (circled, Fig. 4b) represented samples whose grapes originated from the observed regions but were produced in Radmilovac, a winery that did not belong to the regions included in this analysis. Since each of the observed regions includes several different wineries whose production procedure does not have influence on classification, we assumed that the winery Radmilovac differed from all other wineries in Serbia. A similar conclusion was found by Atanacković et al.49 These authors studied the influence of winemaking techniques on the resveratrol content, total phenolic content, and antioxidant potential of red wines made at the oenological station Radmilovac from the grape varietals Merlot (M), Cabernet Sauvignon (KS), Pinot noir (PN) and Prokupac (P). Their results implied that the use of thermovinification procedures could enhance the phenolic composition. Obviously, besides the composition of soil, vintage, and grape varietal, the winery also contributes to the fingerprint of wine. The existence of three groups of characteristics associated with different regions of production could be explained by different types of soil in Serbian regions. Samples from the Central region form a cluster in the right lower part of score plot and showed higher content of the elements Al, Mn, Be, Ba, Cr, and Ni (Fig. 4b and c). In central Serbia, high concentrations of Cr and Ni in the soil on serpentine rocks was recorded.50 Compared to other soil types, the Rankers on serpentine stands out for having significantly higher concentrations of Ni and Cr, which is characteristic for soil on this substrate. Wines from Vojvodina, grouped in the right upper part of score plot, were characterized by higher content of Ca, Na, and Mg (Fig. 4b and c). The dominant soil type in Vojvodina is chernozem that is rich in humus and minerals such as hydrated silicate of Ca, Na, and Mg. Samples from the South region were positioned in the left lower part of the score plot and showed higher content of K and Rb (Fig. 4b and c). Limestone type soil in Southeast Serbia is rich in alkaline metals such as Rb and K. Rubidium content in soils is largely inherited from the parent rocks, as indicated by the highest mean Rb content in soils over granites, limestones, and calcareous rocks.51 Rubidium is very closely linked with K as their properties are similar, and it may partially substitute for K sites in plants.51
Essential element | Average content in 100 mL wine (mg) | RDA (mg per day) | DMI (%) | ||
---|---|---|---|---|---|
Red | White | Red | White | ||
K | 67.5 | 39.6 | 2000 | 3.38 | 1.98 |
Mg | 9.74 | 6.66 | 375 | 2.60 | 1.78 |
Ca | 8.18 | 8.46 | 800 | 1.02 | 1.06 |
Na | 0.88 | 1.76 | 1500 | 0.06 | 0.12 |
Fe | 0.131 | 0.15 | 14 | 0.94 | 1.10 |
Cu | 0.0014 | 0.068 | 1 | 0.14 | 6.75 |
Zn | 0.0065 | 0.098 | 10 | 0.07 | 0.98 |
Se | 0.0003 | 0.0005 | 0.055 | 0.49 | 0.71 |
Mn | 0.014 | 0.087 | 2 | 0.71 | 4.37 |
Cr | 0.0010 | 0.0008 | 0.04 | 2.60 | 1.96 |
Toxic element | Average content in 100 mL wine (μg) | MDI (μg per day) | DI (%) | ||
---|---|---|---|---|---|
Red | White | Red | White | ||
Pb | 5.15 | 8.97 | 250 | 2.06 | 3.59 |
As | 1.40 | 2.12 | 150 | 0.96 | 1.41 |
Cd | 0.19 | 0.13 | 25 | 0.78 | 0.52 |
A 100 mL glass of analysed wine will provide significant amounts of essential elements including Cu (white wine), Mn (white wine), K (red wine) and seem to be good source of these metals. These results highlight that no element could be labelled as significant in a single dose, since none achieves 15% of the RDA, as recommended for labelling purposes.26 Importantly, the presence of other bioactive compounds in the beverages can influence the bioavailability of the minerals.
For Cd, the EFSA's Panel on Contaminants in the Food Chain (CONTAM Panel) set the Tolerable Weekly Intake (TWI) of 2.5 μg per kg b.w.60 in order to ensure a high level of protection for all consumers, which corresponds 25 μg per day for an adult person of 70 kg. The provisional tolerable weekly intake for Pb of 25 μg per kg b.w. is no longer appropriate.61 The respective benchmark dose lower confidence limit (BMDL) derived from blood lead levels in μg L−1 (corresponding to dietary intake values in μg per kg b.w. per day) were the following: developmental neurotoxicity BMDL01 12 μg L−1 (0.50 μg per kg b.w. per day), effects on systolic blood pressure BMDL01 36 μg L−1 (1.50 μg per kg b.w. per day), and effects on kidney in adults BMDL10 15 μg L−1 (0.63 μg per kg b.w. per day).55 Lead has the greatest concentration in wine of the toxic elements (3.59% and 2.06% of DI for white and red wines, respectively). In the studied wines, the contribution to the daily intake decreased in the order Pb > As > Cd. Estimated daily intake of toxic elements from consumption of 100 mL wine hence poses no toxicological risk.
PCA and PLS-DA were applied in order to establish parameters for classification and differentiation of two types of wine (white and red). Data from the elemental content of wine samples allows accurate prediction of wine type and provides a useful tool for their determination. The most important descriptors for discrimination among red and white wine samples were Be, Al, Rb, Mg, K, Cu, Mn, and Na, in descending order. In addition, PCA was applied to determine which elements affected seasonal variability between wines from one winery. The obtained model revealed the existence of three distinctive groups of characteristics associated with three vintages with the highest influence from the variables Cd, Pb, As, Sb, V, Na, K and Zn. The entire elemental profile of wine is influenced by rain and number of sunny days during the vintage. A pattern recognition technique was also applied for the determination of regional origin of red wine samples. A significant number of samples were supplied by different wineries from three different regions of Serbia. The PCA model revealed the existence of three groups of characteristics associated with different regions of production which could be explained by different soil types in the selected Serbian regions.
A 100 mL glass of the analysed wines will provide significant amount of Cu (white wine, 6.7% of RDA), Mn (white wine, 4.4% of RDA), and K (red wine, 3.4% of RDA). These results highlight that no element could be labelled as significant in a single dose and that the estimated daily intake of toxic elements hence posed no toxicological risk.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra25105f |
This journal is © The Royal Society of Chemistry 2017 |