Characterization and authentication of Spanish PDO wine vinegars using multidimensional fluorescence and chemometrics

1 Postprint of Food Chemistry Volume 230, 1 September 2017, Pages 108–116 DOI 10.1016/j.foodchem.2017.02.118 CHARACTERIZATION AND AUTHENTICATION OF SPANISH PDO WINE VINEGARS USING MULTIDIMENSIONAL FLUORESCENCE AND CHEMOMETRICS Rocío Ríos-Reinaa, Saioa Elcoroaristizabalb, Juan A Ocaña-Gonzálezc, Diego L. García-Gonzálezd, José M. Amigoe, Raquel M. Callejóna,* aÁrea de Nutrición y Bromatología, Facultad de Farmacia, Universidad de Sevilla, C/P. García González n°2, E- 41012, Sevilla, Spain b Chemical and Environmental Engineering Department, Faculty of Engineering, University of the Basque Country, Alameda de Urquijo s/n, E-48013 Bilbao, Spain c Departamento Química Analítica, Facultad de Química, Universidad de Sevilla, C/P. García González s/n, E-41012 Sevilla, Spain d Instituto de la Grasa (CSIC), Campus University Pablo de Olavide - Building 46, Ctra. de Utrera, km. 1 E– 41013, Sevilla, Spain eChemometric Analytical Technologies, Department of Food Sciences, Faculty of Science, University of Copenhagen, Rolighedsvej 30, Frederiksberg CDK-1958, Denmark Running Title: Characterization and authentication of Spanish PDO vinegars by fluorescence and chemometrics. *Author to whom correspondence should be sent. E-mail: [email protected] 2 Abstract 1 This work assesses the potential of multidimensional fluorescence spectroscopy 2 combined with chemometrics for characterization and authentication of Spanish 3 Protected Designation of Origin wine vinegars. Seventy-nine vinegars of different 4 categories (aged and sweet) belonging to the Spanish PDOs “Vinagre de Jerez”, 5 “Vinagre de Montilla-Moriles” and “Vinagre de Condado de Huelva”, were analyzed by 6 excitation-emission fluorescence spectroscopy. A visual assessment of fluorescence 7 landscapes pointed out different trends with vinegar categories. PARAllel FACtor 8 analysis (PARAFAC) extracted the potential fluorophores and their values in the PDO 9 vinegars. This information, coupled with different classification methods (Partial Least 10 Square Discrimination Analysis “PLS-DA” and Support Vectors Machines “SVM”), was 11 able to discriminate the wine vinegar category within each PDO, for which SVM models 12 obtained better results (>92% of classification). In each category, SVM also allows the 13 differentiation between PDOs. The proposed methodology could be used as an 14 analysis method for the authentication of Spanish PDO wine vinegars. 15 3 4 1. Introduction 16 Vinegar is a product used worldwide as a condiment and food preserving agent, 17 obtained by a double fermentation process (alcoholic and acetic fermentation) of 18 sugary and starchy substrates (FAO, 1998). Vinegar can be produced by different 19 methods and raw materials (such as malt, apple, rice, etc.), among which, wine vinegar 20 is the most commonly produced and consumed vinegar in Mediterranean countries and 21 Central Europe (Polo & Sanchez-Luengo, 1991). 22 For many years, wine vinegar has been considered as a low-cost secondary 23 product spontaneously derived from wine production. However, in recent years wine 24 vinegar has become a valued food product much appreciated in gastronomy. As a 25 result, the demand for high-quality wine vinegars has significantly increased over the 26 last years. In this framework, Spain is one of the major producers of high-quality wine 27 vinegars, including three of the five types of vinegar registered in Europe (Council 28 Regulation (EC) No 510/2006) with a “Protected Designation of Origin” (PDO): “Vinagre 29 de Jerez”, “Vinagre de Montilla-Moriles” and “Vinagre de Condado de Huelva” (Table I 30 Supplemental Material). The production of these high-quality PDO wine vinegars in 31 Spain is centered in Andalusia, each of them made from the corresponding protected 32 wines (Jerez, Montilla-Moriles and Condado de Huelva), which provides singular and 33 specific characteristics to each vinegar. In addition, the production of high-quality 34 vinegars requires an ageing period in wooden butts. During the period of aging, some 35 chemical modifications take place providing them with unique organoleptic properties 36 and higher sensory quality (Morales, Tesfaye, García-Parrilla, Casas, & Troncoso, 37 2002). According to the sweetness, time and method of ageing (“criaderas and solera” 38 and “añada” system), different categories are considered within each Spanish PDO 39 (Table 1). 40 The longer aging time is directly related to both the higher quality and production 41 costs of these wine vinegars. This fact increases the final market price and makes the 42 quality assurance and authentication of the Spanish PDOs wine vinegars an important 43 issue. For this reason, objective analytical methodologies are required to guarantee the 44 wine vinegar authenticity and fight against frauds. However, the most common 45 analytical techniques used for the characterization and authentication of these vinegars 46 rely on chromatographic methods that are often expensive and time-consuming 47 (Aceña, Vera, Guasch, Busto, & Mestres, 2011; Cirlini, Caligiani, Palla, & Palla, 2011). 48 Thus, in recent years there has been a growing interest in developing rapid, 49 5 inexpensive, non-destructive and direct methodologies based on non-targeted 50 techniques for food authentication. Fluorescence spectroscopy has been increasingly 51 applied as a competitive, high sensitivity, fast and non-destructive technique in food 52 analysis (Karoui & Blecker, 2011). This spectroscopic technique has been more 53 commonly used in wine (Airado-Rodŕiguez, Galeano-D iaz, Durán-Merás, & Wold, 54 2009; Azcarate et al., 2015), but rarely adopted for wine vinegar samples (Callejón et 55 al., 2012) and hence, there is still scarce information about vinegar fluorescent 56 components. 57 In this sense, wine vinegar is a very complex multi-component mixture of chemical 58 compounds due to its traditional making procedure, the raw material used and the 59 ageing period and method employed. Some of these chemical compounds are 60 polyphenols, amino acids and vitamins (Airado-Rodríguez, Durán-Merás, Galeano-61 Díaz, & Wold, 2011), whose presence is related to the wine chemical basis. To handle 62 this complexity, fluorescence multidimensional measurements, such as excitation-63 emission fluorescence spectroscopy, combined with adequate multi-way methods 64 (Andersen & Bro, 2003; Sádecká & Tóthová, 2007) have been proven to be useful for 65 characterization of complex food matrices (Callejón et al., 2012; Christensen, Becker, 66 & Frederiksen, 2005; Elcoroaristizabal et al., 2016; Lenhardt, Bro, Zeković, Dramićanin, 67 & Dramićanin, 2015). Measuring the emission spectra at different excitation 68 wavelengths results into a bi-dimensional Excitation-Emission Matrix (EEM), which 69 contains unique information of each measured sample. Therefore, a three dimensional 70 array is obtained when all the samples are gathered together, so requiring an 71 appropriate data processing for its interpretation. 72 An adequate multiway method, such as PARAllel FACtor Analysis (PARAFAC), can 73 be used to decompose fluorescence EEMs into different independent groups of 74 fluorescence components (fluorophores), as well as their relative concentration 75 (scores) in each sample (Bro, 1997). The information provided by the resolved 76 fluorophores has been successfully applied in food quality control, since it can reveal 77 clearer insights into the relationships between the intrinsic food properties and the 78 quality of the product. For instance, EEM-PARAFAC has been applied for monitoring 79 the changes occurring during the storage and production of different food samples 80 (Christensen, Becker, & Frederiksen, 2005; Elcoroaristizabal et al., 2016) and their 81 characterization (Lenhardt et al., 2015; Tena, Aparicio, & García-González, 2012). 82 Furthermore, the information obtained after EEM data decomposition by PARAFAC 83 6 modeling could be coupled with different classification methods in order to characterize 84 and classify different food products or detect fraudulent samples (Callejón et al., 2012). 85 There are numerous classification algorithms such as Partial Least Square 86 Discrimination Analysis (PLS-DA), K-Nearest Neighbors (KNN), Support Vector 87 Machines (SVM) and Soft Independent Modelling of Class Analogy (SIMCA) (Cover & 88 Hart, 1967; Vapnik, 1999; Wold, 1966; Wold, 1976). Among them, Partial Least 89 Squares-Discriminant Analysis (PLS-DA) and Support Vectors Machines (SVM) are 90 two of the most common used ones. PLS-DA is a supervised class-modelling method 91 used for building linear discriminant models (Nocairi, Qannari, Vigneau, & Bertrand, 92 2005), which has been successfully applied to a wide variety of food matrices for 93 classification purposes (Azcarate, Cantarelli, Pellerano, Marchevsky, & Camiña, 2013; 94 Lenhardt et al., 2015; Liu, He, & Wang, 2008). The main advantage of the PLS-DA 95 approach is the ability of handling highly collinear and noisy data. However, one of the 96 main issues is that PLS-DA models need a sufficient and balanced amount of samples 97 for each class; and sometimes it is difficult to acquire sufficient samples of some 98 classes, due to their cost of production or their non-availability in the market. Moreover, 99 classes that are not effectively separated linearly are common in food products. 100 Support Vector Machines (SVM) is an effective non-linear machine learning technique 101 suitable for both classification and regression analysis (Xu, Zomer, & Brereton, 2006). 102 In comparison to PLS-DA, the main advantage of SVM is its flexibility in modelling 103 complex classification problems that are non-linear. A common disadvantage of SVM is 104 the lack of transparency of the results, since there are no statistics such as scores and 105 loadings available for easy visualization. 106 Several researchers have tested the SVM’s performance in different food 107 authentication problems obtaining better results than other traditional classification 108 methods. For instance, Acevedo et al. (2007) (Acevedo, Jiménez, Maldonado, 109 Domínguez, & Narváez, 2007) observed that SVM performed better than SIMCA, k-110 NN, and PLS-DA for discrimination of wines according to their PDO, which also 111 enabled the selection of the most relevant UV-Vis wavelengths for samples 112 classification. In the same way Callejón et al. (2012) (Callejón et al., 2012) proved that 113 SVM in conjunction with excitation-emission fluorescence spectroscopy was a more 114 adequate methodology than PLS-DA for the classification of sherry vinegars according 115 to their ageing time. However, the aforementioned study was only focused on the 116 classification of a limited number of wine vinegar categories (aged vinegars) belonging 117 to one Spanish PDO (“Vinagre de Jerez”). 118 119 7 In this context, the aim of this work was to investigate the feasibility of using 120 excitation-emission fluorescence spectroscopy combined with several chemometric 121 techniques for characterization and classification of the three Spanish PDOs wine 122 vinegars and their commercialized categories. First, EEM data will be analyzed by 123 PARAFAC in order to characterize spectroscopically and chemically different 124 commercialized wine vinegar categories (aged and sweet) according to each Spanish 125 PDO. Then, these results will be used to build reliable classification models able to 126 differentiate between the wine vinegar categories corresponding to each Spanish PDO, 127 and each PDO within the same wine vinegar category. 128 129 2. Materials and Methods 130 131 2.1 Wine vinegar samples 132 Seventy-nine wine vinegar samples from the three Spanish PDOs coming from 133 several producers were analyzed in this study (Table 1): 30 “Vinagre de Jerez”, 18 134 “Vinagre de Montilla-Moriles” and 21 “Vinagre de Condado de Huelva” samples. 135 Among the aged categories, these vinegars are aged by the traditional system called 136 “criaderas and solera”, except for the “Vinagre de Condado de Huelva Añada” which is 137 aged by using the static aging system called “añada”. Regarding the “Pedro Ximenez” 138 category, it should be highlighted that this sweet category differs from the aged 139 category not only by the aging time but also by other factors such as their different 140 production process. Thus, they are produced by the addition of must of raisined “Pedro 141 Ximenez” grapes (in the case of “Vinagre de Montilla-Moriles”) or the addition of “Pedro 142 Ximenez” wine to the vinegar. All the samples were purchased from local wineries 143 working in compliance with current regulations of each Spanish PDO. The samples 144 were collected in triplicate and stored in amber vials at room temperature until the 145 analysis. 146 Within each PDO, a different number of samples were collected for the established 147 categories (aged and sweet) according to the production/sale rates of each category 148 during the last years (2014-2015). In these years, the general trend for the three 149 Spanish PDOs reveals a higher production of the categories with less ageing time due 150 to market trends. For instance, “Vinagre de Jerez” (JCR) represented approximately 151 60% of total sales in the PDO “Vinagre de Jerez”, while sales of “Vinagre de Jerez 152 Reserva” (JRE) and “Vinagre de Jerez Gran Reserva” (JGR) accounted for 40% and 153 1% of the total, respectively. Similarly, “Vinagre de Condado de Huelva” (CSC) 154 category had the highest sales growth up 38% of the total. Meanwhile among the aged 155 8 categories of “Vinagre de Condado de Huelva” PDO, the most commercialized vinegar 156 categories were, in decreasing order: “Solera” (CSO), “Reserva” (CRE) and “Añada” 157 (CAN). In the same way, “Vinagre de Montilla-Moriles Crianza” (MCR) was the most 158 commercialized one of the “Vinagre de Montilla-Moriles” PDO due to the recent 159 incorporation to the Spanish PDOs. 160 2.2 Fluorescence analysis 161 Fluorescence measurements were recorded using a Varian Cary-Eclipse 162 fluorescence spectrophotometer (Varian Iberica, Madrid, Spain), equipped with two 163 Czerny-Turner monochromators, and a Xenon discharge lamp pulsed at 80 Hz with a 164 half peak height of ∼2 µs (peak power equivalent to 75 kW). A high-performance R298 165 photomultiplier tube detector was used for collection of the fluorescence spectra. Wine 166 vinegar samples were directly analysed without sample pre-treatment by pipetting them 167 into 3.5 mL quartz cuvettes before measurement. Standard quartz cells (Hellma 168 Analytics, Müllheim, Germany) of 1 cm path length were used to carry out the 169 measurements in a Peltier thermostatted cuvette holder (25.00 ± 0.05 ◦C). The 170 spectrometer was interfaced to a computer with Cary-Eclipse software for spectral 171 acquisition and exportation. 172 The fluorescence Excitation-Emission Matrices (EEMs) were obtained by varying 173 the excitation wavelength (λex) ranging between 250 and 700 nm (every 5 nm), and 174 recording the emission spectra (λem) from 300 to 800 (every 2 nm). For these 175 measurements, excitation and emission slits were both set at 5 nm, and the scan rate 176 was fixed to 1200 nm min-1. The system was wavelength calibrated every day by 177 means of the water Raman peak to account for a possible wavelength drift of the 178 instrument. EEMs were registered by triplicate for each sample and preprocessed in 179 order to avoid noisy and non-informative areas by selecting shorter spectral ranges (λex 180 from 250 to 680 nm, and λem from 310 to 800 nm). 181 182 2.3 Software and data analysis 183 EEM data analysis was performed by using the PLS_Toolbox 7.9.5 (Eigenvector 184 Research Inc., Wenatchee, WA) working under Matlab v.8.5.0 environment (The 185 Mathworks Inc., Natick, MA). Before the analysis, EEMs data were corrected for –186 Rayleigh and Raman scattering (Elcoroaristizabal, Bro, García, & Alonso, 2015) – by 187 removing and replacing the scattering areas with interpolated values by using the 188 FLUCUT function included in the PLS_Toolbox. FLUCUT Removes Rayleigh scattering 189 (and possibly Raman) by inserting NaN and 0 values in Excitation-Emission Matrices 190 9 (EEMs) where the Rayleigh bands are. Alternatively, FLUCUT may also be used to 191 generate weights that can be used for deweighting (instead of eliminating) these 192 regions. 193 2.3.1 Parallel Factor Analysis (PARAFAC) 194 PARAllel FACtor models were performed on the corrected EEM data in order to 195 extract the relevant information and develop models for: (1) different wine vinegar 196 categories belonging to the same Spanish PDO, and (2) similar wine vinegar 197 categories belonging to different Spanish PDOs. 198 Before modelling, the EEM landscapes corresponding to the same Spanish 199 PDO (1) were rearranged into a three-dimensional structure (X) of size (3 replicated 200 samples x λem x λex): 90 x 246 x 87 for the PDO “Vinagre de Jerez”; 54 x 246 x 87 for 201 the PDO “Vinagre de Montilla-Moriles”, and 63 x 246 x 87 for the PDO “Vinagre de 202 Condado de Huelva”. In a similar way, the EEM landscapes corresponding to similar 203 wine vinegar categories but different Spanish PDOs (2) were organized into a three-204 way array (X) of size (3 replicated samples x λem x λex): 78 x 246 x 87 for “Crianza”; 66 205 x 246 x 87 for “Reserva” category, and 27 x 246 x 87 for “Pedro Ximenez” categories. 206 No PARAFAC analysis was carried out for the “Gran Reserva” category due to the 207 limited number of samples. 208 Then, each three-way dataset (X) was decomposed by PARAFAC (Bro, 1998). 209 The proper number of factors for each model was determined by using the CORe 210 CONsistency DIAgnostic test (CORCONDIA) (Bro & Kiers, 2003), the percentage of 211 variance explained by the model and the visual inspection of the recovered spectral 212 profiles and residuals. Non-negative constraints for all modes (concentrations and both 213 spectral profiles) were applied to obtain meaningful chemical solutions. 214 215 2.3.2 Classification methods 216 Partial Least Squares-Discriminant Analysis (PLS-DA) (Nocairi et al., 2005) and 217 Support Vectors Machines (SVM) (Vapnik, 1999) algorithms were used to build 218 classification models for discrimination the wine vinegar category within each Spanish 219 PDO. On the one hand, PLS-DA is a classification method based on partial least 220 squares regression (PLS) that transforms the data into a set of linear latent variables 221 for predicting the dependent or class variable, making models that allow the maximum 222 separation among classes. The class variable forms a so-called dummy matrix that 223 indicates whether a sample belongs to a certain class or category. In our study, three 224 16 aged category (CSC) of “Vinagre de Condado de Huelva” PDO, the different 435 fluorescence profile observed in the landscapes and the higher mean values of F1 and 436 F3 with respect to the rest of categories (Table V Supplementary Material), could be 437 related to the lowest errors of classification obtained for this category (25.0% of the 438 samples misclassified). However, unsatisfactory results were obtained for the rest of 439 aged categories within each Spanish PDO: JCR, JRE, MCR, CSO and CRE. The low 440 sensitivity and specificity values (mainly under 65.0%) and high classification errors 441 obtained in terms of prediction (between 25.0 and 63.0%), confirm the difficulty in 442 correctly classifying these aged vinegar categories (≥ 6 and ≥ 24 months) by using 443 linear classification models. This could be related to the similar score values followed 444 by the modelled factors of these categories within each Spanish PDOs (Table V 445 Supplementary Material). Among them, the ≥ 6 months aged samples (JCR, MCR and 446 CSO) were the worst classified ones in all the Spanish PDO vinegar models, showing 447 classification errors until 63.0%. This may be explained due to these wine vinegars are 448 aged over a wide range of time (from 6 to 24 months). Thus, those samples aged until 449 24 months are expected to be spectroscopically and chemically quite similar to the 450 vinegars of the following category (≥ 24 month). Similar results were obtained by 451 Callejón et al., (2012). 452 In contrast, higher sensitivity and specificity levels were obtained for all Spanish 453 PDO vinegars using SVM models (Table 2). The optimal parameters for the 454 optimization of SVM models ,log10(C) and log10 (γ), were optimized in the traditional 455 way by using an independent test set (Cristianini & Shawe-Taylor, 2000). Between 456 92% and 100% of the samples were correctly classified in all categories. Even more, all 457 samples belonging to the Spanish PDO “Vinagre de Montilla-Moriles” were perfectly 458 classified. Further information about the misclassified category samples within each 459 Spanish PDO is summarized in the confusion matrices shown in Table VI 460 (Supplementary Material). These results also point out that SVM does not need a large 461 number of samples to make a good model, as occurs in our study with some 462 categories, and further, it is not affected by the presence of outliers. These results 463 demonstrated that this methodology could be successfully used for the authentication 464 of the vinegar category belonging to each Spanish PDO. 465 466 3.4 Spanish PDO classification within similar vinegar categories (“Crianza”, 467 “Reserva” and “Pedro Ximenez”) 468 17 In this classification task, the objective was to classify the samples by their PDO for 469 a single category. Three PARAFAC models were built according to the vinegar 470 categories under study, i.e. “Crianza”, “Reserva” and “Pedro Ximenez”, in order to 471 discriminate their corresponding Spanish PDO. In a similar way to the previous 472 sections, the best PARAFAC models obtained for each vinegar category were selected 473 comparing the quality parameters of the models, which are shown in Table VII 474 (Supplementary Material). In this case, a 4-factor PARAFAC model was obtained for 475 “Crianza”, while a 5-factor PARAFAC model was built for “Reserva”, and a 3-factor 476 PARAFAC model was constructed for “Pedro Ximenez” category. The obtained models 477 explained more than 97.0% of the variance with a core consistency over zero. The 478 related PARAFAC loadings (excitation and emission spectra) obtained for the models 479 corresponding to each vinegar category are illustrated in Figure 3. 480 The maxima wavelengths (λex and λem) of the different factors obtained for the 481 “Crianza”, “Reserva” and “Pedro Ximenez” PARAFAC models match with the different 482 fluorophores described in detail in section 3.2. As shown in Figure 3, the “Crianza” 483 category is characterized by factors with excitation and emission ranges between 340-484 500 nm and 430-580 nm, respectively. These factors are mainly related to fluorescent 485 compounds naturally presented in high concentration in wine such as phenols, 486 flavonols and vitamins as previously described (section 3.2). These compounds have a 487 higher contribution in this category (Table V Supplementary Material) since the less 488 aged wine vinegars retain more compounds coming from the raw materials. Regarding 489 the “Reserva” samples, a higher number of factors were required to model this 490 category, i.e., more fluorescent compounds with a wide range of excitation/emission 491 spectra were needed to describe its underlying chemical composition. In this case, the 492 longer aging period undergone by vinegars of the “Reserva” category plays a crucial 493 role in this chemical complexity. There is an enrichment of these vinegars with more 494 phenolic compounds (released by the wood barrels) and oxidation products (derived 495 from the development of certain chemical reactions among vinegar components), 496 whose concentration levels have been proven to increase during the aging process 497 (Callejón, Morales, Silva Ferreira, & Troncoso, 2008). The wavelengths of the factors 498 modelled for “Pedro Ximenez” category are associated to fluorophores emitting at the 499 highest excitation and emission wavelengths (upper than 475 and 550 nm, 500 respectively). However, more information, not available in the literature, is needed to 501 identify these fluorophores. 502 For all the categories under study, the relative values of these factors (scores) vary 503 as a function of the Spanish PDO, which highlights that the composition of the vinegar 504 18 categories depends also on the raw material used (wine) and on the different 505 production methods to which the vinegars have been subjected in each PDO. Thus, 506 these particular characteristics reflected by the scores provide the chance to 507 discriminate the Spanish PDO corresponding to similar wine vinegar categories. In this 508 case, related to the proven higher ability of prediction previously obtained (Table 2), 509 only SVM classification models were built. The parameters for the optimization of the 510 SVM models, log10(C) and log10 (γ), were found to be 2 and between -1 and 0, 511 respectively. Table 3 summarizes the statistical results (sensitivity, specificity and 512 errors) of the performance of the SVM models. 513 Regarding the “Crianza” category, high sensitivity and specificity values were 514 obtained for these samples according to their origin (Spanish PDO) with classification 515 errors of prediction lower than 3.5% (Table 3). These results demonstrate that it is 516 possible to successful differentiate the Spanish PDO of “Crianza” vinegars according to 517 their fluorescent composition that is highly related to the raw material (wine) used. 518 Furthermore, all the samples belonging to the “Pedro Ximenez” category were correctly 519 classified in the “Vinagre de Jerez” and “Vinagre de Montilla-Moriles” PDOs. The high 520 levels of sensitivity and specificity and the good classification rates obtained were 521 explained by the different production process employed by each Spanish PDO. 522 “Reserva” was the worst classified category according to the PDOs, showing sensitivity 523 and specificity values higher than 70% and predicted errors lower than 15%. In the 524 case of samples belonging to “Vinagre de Montilla-Moriles”, 100% of the “Reserva” 525 vinegars were correctly classified, and only some samples were misclassified between 526 “Vinagre de Jerez” and “Vinagre de Condado de Huelva” PDOs (Table VIII 527 Supplementary Material). These results are considered acceptable considering the 528 high variability of these samples due to the wide range of ageing periods that are 529 reflected by their complex chemical composition (Casale et al., 2006; García Parrilla et 530 al., 1999). 531 532 4. Conclusions 533 The analytical methodology proposed in this study, namely fluorescence 534 excitation–emission spectroscopy coupled to PARAFAC modeling and SVM 535 classification method, has demonstrated to be able to characterize and classify the 536 three Spanish PDOs wine vinegars according to their “Protected Denomination of 537 Origin” as well as their categories (aged and sweet). As a simple preliminary 538 characterization, a visual assessment of the fluorescence Excitation-Emission Matrices 539 (EEMs) of the aged categories pointed out similarities in the fluorescence landscapes 540 19 for the three Spanish PDOs wine vinegars: the spectral maxima were shifted towards 541 longer wavelengths with the aging of these vinegars. Moreover, the sweet category 542 “Pedro Ximenez” showed its excitation and emission maxima even at longer 543 wavelengths than the aged categories, probably due to the different production process 544 to which these vinegars are subjected. PARAFAC was carried out to spectroscopically 545 and chemically characterize the different wine vinegars. It gave information about the 546 potential fluorescent compounds present in the wine vinegars as well as their 547 contribution in each Spanish PDO and category. These dissimilar spectroscopic and 548 chemical features allowed us their differentiation according to their category and origin 549 (Spanish PDO) using suitable classification methods. The feasibility of SVM 550 methodology to classify the different categories of wine vinegars within each PDO was 551 successfully demonstrated. The built SVM classification models proved a higher ability 552 of prediction (between 92% and 100% correctly classified samples) than PLS-DA 553 models, especially for classifying aged vinegar categories with similar spectroscopic 554 characteristics. Furthermore, SVM models were also able to differentiate the Spanish 555 PDOs even for similar vinegar categories due to their spectral differences. 556 The advantages of this methodology, e.g. fast, non-destructive and nonsample 557 preparation, would allow implementing this method as an alternative tool for PDO 558 regulatory councils and producers to be implemented in routine analysis. It could be 559 applied for assessing the authenticity of the Spanish PDO and the vinegar category. 560 Finally, it is expected that further information about the specific ageing periods to which 561 these vinegars are subjected will improve the performance of some classification 562 models. 563 564 5. Acknowledgments 565 The authors want to thank “Consejería de Economía, Innovación y Ciencia” of the 566 “Junta de Andalucía” for the financial support provided through the project P12-AGR-567 1601, and the Spanish Regulatory Councils of the wine vinegar PDOs for their 568 invaluable help with the acquisition of the samples for this study. In addition, authors 569 want to thanks “Ministerio de Educación, Cultura y Deporte” (MECD, Spanish 570 Government) for the FPU pre-doctoral fellowship and UPV/EHU (University of the 571 Basque Country) for the grant received from the program ‘‘Contratación de Doctores 572 Recientes”. 573 574 6. References 575 Aceña, L., Vera, L., Guasch, J., Busto, O., & Mestres, M. (2011). Chemical 576 20 characterization of commercial sherry vinegar aroma by headspace solid-phase 577 microextraction and gas chromatography-olfactometry. Journal of Agricultural and 578 Food Chemistry, 59(8), 4062–4070. http://doi.org/10.1021/jf104763u 579 Acevedo, F. J., Jiménez, J., Maldonado, S., Domínguez, E., & Narváez, A. (2007). 580 Classification of wines produced in specific regions by UV-visible spectroscopy 581 combined with support vector machines. Journal of Agricultural and Food 582 Chemistry, 55(17), 6842–6849. http://doi.org/10.1021/jf070634q 583 Airado-Rodríguez, D., Durán-Merás, I., Galeano-Díaz, T., & Wold, J. P. (2011). Front-584 face fluorescence spectroscopy: A new tool for control in the wine industry. 585 Journal of Food Composition and Analysis, 24(2), 257–264. 586 http://doi.org/10.1016/j.jfca.2010.10.005 587 Airado-Rodríguez, D., Galeano-D iaz, T., Durán-Merás, I., & Wold, J. P. (2009). 588 Usefulness of fluorescence excitation-emission matrices in combination with 589 parafac, as fingerprints of red wines. Journal of Agricultural and Food Chemistry, 590 57(5), 1711–1720. http://doi.org/10.1021/jf8033623 591 Andersen, C. M., & Bro, R. (2003). Practical aspects of PARAFAC modeling of 592 fluorescence excitation-emission data. Journal of Chemometrics, 17(4), 200–215. 593 http://doi.org/10.1002/cem.790 594 Azcarate, S. M., Araújo, A. De, Alcaraz, M. R., Ugulino, M. C., Araújo, D., Camiña, J. 595 M., & Goicoechea, H. C. (2015). Modeling excitationemission fluorescence 596 matrices with pattern recognition algorithms for classification of Argentine white 597 wines according, 184, 214–219. 598 Azcarate, S. M., Cantarelli, M. A., Pellerano, R. G., Marchevsky, E. J., & Camiña, J. M. 599 (2013). Classification of argentinean sauvignon blanc wines by UV spectroscopy 600 and chemometric methods. Journal of Food Science, 78(3), 432–436. 601 http://doi.org/10.1111/1750-3841.12060 602 Bro, R. (1997). PARAFAC. Tutorial and applications. Chemometrics and Intelligent 603 Laboratory Systems, 38(2), 149–171. http://doi.org/10.1016/S0169-604 7439(97)00032-4 605 Bro, R. (1998). Multi-way analysis in the food industry. Models, algorithms, and 606 applications., (1998). 607 Bro, R., & Kiers, H. A. L. (2003). A new efficient method for determining the number of 608 components in PARAFAC models. Journal of Chemometrics, 17(5), 274–286. 609 21 http://doi.org/10.1002/cem.801 610 Callejón, R. M., Amigo, J. M., Pairo, E., Garmón, S., Ocaña, J. A., & Morales, M. L. 611 (2012). Classification of Sherry vinegars by combining multidimensional 612 fluorescence, parafac and different classification approaches. Talanta, 88, 456–613 462. http://doi.org/10.1016/j.talanta.2011.11.014 614 Callejón, R. M., Morales, M. L., Silva Ferreira, A. C., & Troncoso, A. M. (2008). 615 Defining the typical aroma of Sherry vinegar: Sensory and chemical approach. 616 Journal of Agricultural and Food Chemistry, 56(17), 8086–8095. 617 http://doi.org/10.1021/jf800903n 618 Casale, M., Sáiz Abajo, M. J., González Sáiz, J. M., Pizarro, C., & Forina, M. (2006). 619 Study of the aging and oxidation processes of vinegar samples from different 620 origins during storage by near-infrared spectroscopy. Analytica Chimica Acta, 621 557(1–2), 360–366. http://doi.org/10.1016/j.aca.2005.10.063 622 Christensen, J., Becker, E. M., & Frederiksen, C. S. (2005). Fluorescence 623 spectroscopy and PARAFAC in the analysis of yogurt. Chemometrics and 624 Intelligent Laboratory Systems, 75(2), 201–208. 625 http://doi.org/10.1016/j.chemolab.2004.07.007 626 Christiani, N., & Shawe-Taylor, J. (2000). An Introduction to Support Vector Machines. 627 Cambridge Univ. Press, Cambridge, UK. 628 Cirlini, M., Caligiani, a., Palla, L., & Palla, G. (2011). HS-SPME/GC-MS and 629 chemometrics for the classification of Balsamic Vinegars of Modena of different 630 maturation and ageing. Food Chemistry, 124(4), 1678–1683. 631 http://doi.org/10.1016/j.foodchem.2010.07.065 632 Council Regulation (EC) No 510/2006 of 20 March 2006 on the protection of 633 geographical indications and designations of origin for agricultural products and 634 foodstuffs. 635 Cover, T., & Hart, P. (1967). Nearest neighbor pattern classification. IEEE Transactions 636 on Information Theory, 13, 21 – 27. 637 Dufour, É., Letort, A., Laguet, A., Lebecque, A., & Serra, J. N. (2006). Investigation of 638 variety, typicality and vintage of French and German wines using front-face 639 fluorescence spectroscopy. Analytica Chimica Acta, 563(1–2 SPEC. ISS.), 292–640 299. http://doi.org/10.1016/j.aca.2005.11.005 641 Elcoroaristizabal, S., Bro, R., García, J. A., & Alonso, L. (2015). PARAFAC models of 642 22 fluorescence data with scattering: A comparative study. Chemometrics and 643 Intelligent Laboratory Systems, 142, 124–130. 644 http://doi.org/10.1016/j.chemolab.2015.01.017 645 Elcoroaristizabal, S., Callejón, R. M., Amigo, J. M., Ocaña-González, J. A., Morales, M. 646 L., & Ubeda, C. (2016). Fluorescence excitation-emission matrix spectroscopy as 647 a tool for determining quality of sparkling wines. Food Chemistry, 206, 284–290. 648 http://doi.org/10.1016/j.foodchem.2016.03.037 649 García Parrilla, M. C., Heredia, F. J., & Troncoso, A. M. (1999). Sherry wine vinegars: 650 Phenolic composition changes during aging. Food Research International, 32(6), 651 433–440. http://doi.org/10.1016/S0963-9969(99)00105-2 652 Joint FAO/WHO Food Standards Programme.Report No: FAO-ESN-CX/PFV-98/8. 653 Codex Committee on Processed Fruits and Vegetables. Sess 19. Whashington, 654 DC, 16-18, March 1998. FAO, R. Conversion of European Regional Standard for 655 Vinegar into World-wide Standard. (1998). 656 Karoui, R., & Blecker, C. (2011). Fluorescence Spectroscopy Measurement for Quality 657 Assessment of Food Systems-a Review. Food and Bioprocess Technology, 4(3), 658 364–386. http://doi.org/10.1007/s11947-010-0370-0 659 Karoui, R., Dufour, É., & De Baerdemaeker, J. (2006). Front face fluorescence 660 spectroscopy coupled with chemometric tools for monitoring the oxidation of semi-661 hard cheeses throughout ripening. Food Chemistry, 101(3), 1305–1314. 662 http://doi.org/10.1016/j.foodchem.2006.01.028 663 Lenhardt, L., Bro, R., Zeković, I., Dramićanin, T., & Dramićanin, M. D. (2015). 664 Fluorescence spectroscopy coupled with PARAFAC and PLS DA for 665 characterization and classification of honey. Food Chemistry, 175, 284–291. 666 http://doi.org/10.1016/j.foodchem.2014.11.162 667 Liu, F., He, Y., & Wang, L. (2008). Determination of effective wavelengths for 668 discrimination of fruit vinegars using near infrared spectroscopy and multivariate 669 analysis. Analytica Chimica Acta, 615(1), 10–17. 670 http://doi.org/10.1016/j.aca.2008.03.030 671 Margraf, T., Santos, É. N. T., de Andrade, E. F., van Ruth, S. M., & Granato, D. (2016). 672 Effects of geographical origin, variety and farming system on the chemical 673 markers and in vitro antioxidant capacity of Brazilian purple grape juices. Food 674 Research International, 82, 145–155. http://doi.org/10.1016/j.foodres.2016.02.003 675 23 Morales, M. L., Tesfaye, W., García-Parrilla, M. C., Casas, J. a, & Troncoso, A. M. 676 (2002). Evolution of the aroma profile of sherry wine vinegars during an 677 experimental aging in wood. Journal of Agricultural and Food Chemistry, 50(11), 678 3173–8. http://doi.org/10.1021/jf011313w 679 Nocairi, H., Qannari, E. M., Vigneau, E., & Bertrand, D. (2005). Discrimination on latent 680 components with respect to patterns. Application to multicollinear data. 681 Computational Statistics and Data Analysis, 48(1), 139–147. 682 http://doi.org/10.1016/j.csda.2003.09.008 683 Polo M.C., & Sanchez-Luengo, A. (1991). El vinagre de vino. In in: C. Llaguno, M.C. 684 Polo (Eds.), El vinagre de vino, C.S.I.C. Espana, Madrid (pp. 23–47). 685 Rodríguez-Delgado, M. A., Malovaná, S., Pérez, J. P., Borges, T., & García 686 Montelongo, F. J. (2001). Separation of phenolic compounds by high-performance 687 liquid chromatography with absorbance and fluorimetric detection. Journal of 688 Chromatography A, 912(2), 249–257. http://doi.org/10.1016/S0021-689 9673(01)00598-2 690 Sádecká, J., & Tóthová, J. (2007). Fluorescence Spectroscopy and Chemometrics in 691 the Food Classification− a Review. Czech J. Food Sci. Vol, 25(4), 159–692 173. Retrieved from http://www.journals.uzpi.cz/publicFiles/00302.pdf 693 Sádecká, J., Tóthová, J., & Májek, P. (2009). Classification of brandies and wine 694 distillates using front face fluorescence spectroscopy. Food Chemistry, 117(3), 695 491–498. http://doi.org/10.1016/j.foodchem.2009.04.053 696 Tena, N., Aparicio, R., & García-González, D. L. (2012). Chemical changes of 697 thermoxidized virgin olive oil determined by excitation-emission fluorescence 698 spectroscopy (EEFS). Food Research International, 45(1), 103–108. 699 http://doi.org/10.1016/j.foodres.2011.10.015 700 Tóthová, J., & Sádecká, Ľ. Ž. J. (2008). Characterization and Classification of Distilled 701 Drinks Using Total Luminescence and Synchronous Fluorescence Spectroscopy. 702 Acta Chimica Slovaca, 1(1), 265–275. 703 Vapnik, V. (1999). The Nature of Statistical Learning Theory . In New York : Springer-704 Verlag. 705 Wold, H. (1966). Estimation of principal components and related models by iterative 706 least squares . In In: Multivariate Analysis ( Krishnaiah PR , ed.). New York : 707 Academic Press . 708 24 Wold, S. (1976). Pattern recognition by means of disjoint principal components models 709 . Pattern Recognition, 8, 127 – 139. 710 Xu, Y., Zomer, S., & Brereton, R. G. (2006). Support Vector Machines: A Recent 711 Method for Classification in Chemometrics. Critical Reviews in Analytical 712 Chemistry, 36(3–4), 177–188. http://doi.org/10.1080/10408340600969486 713 Zhu, D., Ji, B., Eum, H. L., & Zude, M. (2009). Evaluation of the non-enzymatic 714 browning in thermally processed apple juice by front-face fluorescence 715 spectroscopy. Food Chemistry, 113(1), 272–279. 716 http://doi.org/10.1016/j.foodchem.2008.07.009 717 718 FIGURE CAPTIONS 719 Figure 1. Excitation-Emission fluorescence landscapes obtained for different 720 categories of Spanish PDO vinegars. The color scale of fluorescence intensity (in 721 arbitrary units) varies from dark blue (lowest signal intensity) to yellow (highest signal 722 intensity).The acronyms for the different wine vinegar categories are defined in Table 1. 723 724 Figure 2. Excitation and Emission spectra (PARAFAC loadings) of the main 725 fluorophores present in the vinegars of the three Spanish PDOs. 726 727 Figure 3. Excitation and Emission spectra (PARAFAC loadings) of the main 728 fluorophores present in different vinegars categories of the Spanish PDOs. 729 730 25 Table 1. Wine vinegar samples analyzed according to the Spanish PDOs 731 732 *Note: each samplecorresponds to different producers 733 734 Protected Designation of Origin (PDO) Vinegar Category Category Name Acronym Aging time (months) Number of samples* “Vinagre de Jerez” Aged “Vinagre de Jerez” JCR ≥6 13 “Reserva” JRE ≥24 11 “Gran Reserva” JGR ≥ 120 2 Sweet “Pedro Ximenez” JPX - 4 “Vinagre de Montilla- Moriles” Aged “Crianza” MCR ≥6 7 “Reserva” MRE ≥24 4 “Gran Reserva” MGR ≥120 2 Sweet “Pedro Ximenez” MPX - 5 “Vinagre de Condado de Huelva” Non – aged “Vinagre Condado de Huelva” CSC 0 6 Aged “Viejo Solera” CSO ≥6 6 “Viejo Reserva” CRE ≥24 7 “Viejo Añada” CAN ≥36 2