Wine Authenticity by Quantitative 1H NMR Versus Multitechnique Analysis: a Case Study

Louis Gougeon, Gregory da Costa, Tristan Richard & François Guyon

Food Analytical Methods

ISSN 1936-9751 Volume 12 Number 4

Food Anal. Methods (2019) 12:956-965 DOI 10.1007/s12161-018-01425-z

1 23 Your article is protected by copyright and all rights are held exclusively by Springer Science+Business Media, LLC, part of Springer Nature. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self- archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.

1 23 Author's personal copy

Food Analytical Methods (2019) 12:956–965 https://doi.org/10.1007/s12161-018-01425-z

Wine Authenticity by Quantitative 1H NMR Versus Multitechnique Analysis: a Case Study

Louis Gougeon 1 & Gregory da Costa1 & Tristan Richard1 & François Guyon2

Received: 27 September 2018 /Accepted: 26 December 2018 /Published online: 15 January 2019 # Springer Science+Business Media, LLC, part of Springer Nature 2019

Abstract Food counterfeit is a major issue for food industries. In the case of , there can be multiple forgeries such as geographical, , or mislabeling. For the last 10 years, the use of quantitative 1H NMR (q-NMR) for food authentication has experienced an enormous increase. This study was based on a comparative evaluation of the results obtained for three sets of authentic high-valued wines and suspected wines. Two methodologies have been followed: (i) the usual wine analysis, based on the utilization of multiple techniques, which is the traditional way of analysis for wine authentication and (ii) q-NMR profiling, the alternative proposed method. For wine comparison, an original approach based on similarity score (S-score) by analogy to the Z-score calculation was developed. A limit of four S-scores outside the range − 2/2 has been, arbitrary, defined to reveal that wines are different using q-NMR analysis. Data treatments allowed the extraction of discriminating parameters, some of which were common to both approaches while others could be linked to the techniques’ capabilities showing the complementarity of the two approaches. This study demonstrated the potential of q-NMR in wine authentication by comparative analysis with authentic samples. The q-NMR alternative method has also the advantages of rapidity (around 20 min including sample preparation, analysis, and data treatment) and low volume (0.5 mL), which is a prerequisite for analyzing priceless wines.

Keywords Food authenticity . Food quality control . Wine . Classical wine analysis . 1HNMR . q-NMR . Chemometric

Introduction using various analytical (Schlesier et al. 2009;Medina et al. 2013) and statistical methods (Smeyers-Verbeke et Food authenticity is a major concern in the global food al. 2009). Multi-instrumental approaches have been used system for many reasons such as safety, consumer trust, including liquid and gas chromatography, UV-visible spec- and producer concurrency. Falsifications occurring in wine trometry, near-infrared spectrometry, NMR, mass spec- can belong to multiple types including geographical, vin- trometry (MS), isotope ratio mass spectrometry (IRMS), tage, and varietal mislabeling. The detection of such falsi- and coupled plasma mass spectroscopy (ICP-MS). All fications is usually performed using the confrontation of these techniques are widely employed in the investigation analytical results between authentic and suspicious wines of wine authenticity allowing the measurement of several wine parameters such as and alcohols, sulfur diox- ide, reducing , polyphenols, organic acids, total and Electronic supplementary material The online version of this article volatile acidity, minerals, metals, and isotopic ratios (D/H, 18 16 13 12 (https://doi.org/10.1007/s12161-018-01425-z) contains supplementary O/ O, C/ C). Until now, all these multidisciplinary material, which is available to authorized users. approaches are necessary to obtain discriminant results (so-called in this paper usual wine analysis) (Medina * François Guyon et al. 2013). As a result, wine authenticity study requires [email protected] highly skilled technicians, numerous analytical systems, and a large volume of wine. The overall analysis can take 1 Univ. Bordeaux, ISVV, EA 4577, USC 1366 INRA, Unité de Recherche, Œnologie, 210 Chemin de Leysotte, 33882 Villenave time and is costly. The NMR quantitative analysis (q- d’Ornon, France NMR) could be a complementary alternative to the usual 2 Service Commun des Laboratoires, 3 Avenue du Dr. Albert wine analysis because it can be considered as a universal Schweitzer, 33600 Pessac, France method that allows the measurement of several parameters Author's personal copy

Food Anal. Methods (2019) 12:956–965 957 on a small volume of wine without specific sample prepa- treatment will compare analytical information provided by ration procedure (Pauli et al. 2012). these two different approaches in order to evaluate q-NMR Since 10 years, 1H NMR analysis has been more and more potential in wine authentication by comparative data analysis widely applied to food control and, as a result, to wine (Larive of suspicious and authentic wines. et al. 2014; Simmler et al. 2014; Amargianitaki and Spyros 2017). The application of proton profiling or compound quan- tification obtained by q-NMR along with chemometrics has Material and Methods been successfully used for recognizing grape varieties (Son et al. 2009a, b; Anastasiadi et al. 2009; Godelmann et al. Samples 2013), geographical origin (Caruso et al. 2012; Viggiani and Castiglione Morelli 2008; Son et al. 2008; Pereira et al. 2007) In order to respect the anonymity, three sets of samples were and vintage (Consonni et al. 2011). The q-NMR interest identified as A, B, and C, corresponding to three different comes from the global analysis of the food matrix, the repeat- , one old (A, prior to the 2000 vintage) and two youn- ability and reproducibility of the measurements as well as its ger ones (B and C). The first set of samples was composed of rapidity in comparison to classical methods (Godelmann et al. two bottles (As1 and As2) which are expected to be used for top 2016). up leveling. The second set of samples, corresponding to sus- High-value wines are to be kept for aging, and, sometimes, picious bottles found on the market, was based on a single some of them become wines of collection. Authenticity as- bottle for each of the two vintages (Bs and Cs). For each sessment and good conservation of these wines are thus crit- vintage, Château Mouton Rothschild provided a bottle of the ical questions. On one hand, greed and profit could lead un- authentic wine of their own cellar identified as Aa,Ba,andCa. scrupulous resellers to practice wine forgeries (geographical, In addition, Château Mouton Rothschild provided three sets of vintage, or varietal mislabeling) on such priceless wines. On authentic bottles from different old vintages (prior to the 1980 the other hand, during long-aging period, corks can dry out vintage) identified as vintages N, N +2,andN + 4 (5, 4, and 4 due to a lack of humidity which, in addition to wine leak, bottles, respectively). induces oxygen exchanges causing wine spoilage or oxida- To validate the 1H NMR wine analysis method, 20 French tion. As a result, these two aspects become critical for driving wines were purchased commercially: a set of 13 wines from an early wine degradation. The unique solution is to replace South-West region (6 Médoc, 6 Libournais, and 1 Gascogne), the and to top the bottle up with a similar wine, i.e., same a set of 4 wines from Burgundy region (2 Burgundy and 2 vintage and same production estate. The wine producer is the ), 2 wines, and 1 Roussillon wine only one licensed to perform this top-up level step which is (Table S1, supplementary data). realized after a control of wine authenticity. For both applica- tions, estates are searching for reliable authentication Usual Wine Analysis Methods techniques requiring the lowest sample volume to be used for analysis. Among q-NMR advantages, a valuable one is the All these analyses were performed strictly following the offi- low quantity of sample required for the analysis (Gougeon cial methods internationally recognized and referenced in the et al. 2018). As a result, this tool can be interesting in old Organisation Internationale de la Vigne et du Vin (OIV) com- vintage wine authentication using proton profile comparison pendium: alcoholic strength (ethanol) [OIV a], sugars (glu- between the authentic estate wine and the commercial one. cose, , , glycerol) [OIV b], acidity (total and But until now, q-NMR has not been applied for authentication volatile acidity) [OIV c], volatile compounds (ethanal, ethyl applications using the wine proton profile comparison. acetate, methanol, butanol 2, propanol 1) [OIV d], organic Recently, in collaboration with Château Mouton acid concentration (tartaric, malic, succinic, lactic and Rothschild, a prestigious winemaker, the authenticity of dif- gluconic acids) by high pressure capillary electrophoresis ferent wines samples was checked using the q-NMR method (HPCE) [OIV e], metal and trace metal analysis by inductive and the usual analysis (multitechnique approach) applied for plasma atomic emission spectrometry (ICP-AES) [OIV f], wine authentication (Medina et al. 2013). The first case was ethanol deuterium analysis by 2H NMR ((D/H)1 and on the authentication of the wine contained in two bottles, half (D/H)2) [OIV g], ethanol and wine carbone-13 (written, re- filled, that would be used to top-up the level of one bottle. The spectively, δ13Candδ13C wine) [OIV h], and wine oxygen-18 second case concerned two suspected bottles of different vin- (δ18O) [OIV i] by IRMS. This isotope value corresponds to tages, fully filed and sampled on foreign market. Château the ratio heavy/light element against an international standard A/A′ A/A′ A/A′ Mouton Rothschild’s requested the authentication of the wine computed according to δ X =([( X)sample/( X)reference] contained in the four bottles by comparative analysis with − 1) where X is the studied element and A and A′ denote higher authentic samples coming from their own cellar. This work and lower atomic mass number, respectively, of element X. presents the results obtained by the two methods. Data Stable isotope ratios are expressed as multiple of 1000 (unit Author's personal copy

958 Food Anal. Methods (2019) 12:956–965

per mil ‰) because of the smallness of the measured values, where CS and CA are, respectively, the concentration of a spe- total and free sulfur dioxide (respectively, SO2TandSO2L) cific analyte of the suspicious sample and the authentic sam- [OIV j]. These analyses were performed following quality ple. σ, the measurement uncertainty, is computed applying the control requirement (control cards, analytical standards, and method uncertainty to the highest analyte concentration, CA reference samples) to ensure the results’ accuracy. and CS. Unsupervised principal component analysis (PCA) was 1H NMR Experiments carried out using the FactoMineR plugin of R® software (ver- sion 3.5.0). The matrix was created with the normalized quan- tified data. The centered-reduced data were obtained in order Themethodusedfor1H NMR analysis was reported in a pre- to make the variables comparable. vious article (Gougeon et al. 2018), with a slight modification concerning pH adjustment. Briefly, for each sample, 120 μLof μ phosphate solution (1 M, pH 2.6), 60 L calibration solution Results and Discussion (77 mM calcium formate and 5 mM trimethylsilylpropanoic acid in D O) were added to 420 μL of previously centrifuged 2 As an introduction of this results and discussion part, a brief wine. The pH was fine-adjusted to 3.1 using a semi-automatic synopsis of the context of this study needs to be provided for small-scale system (BTpH, Bruker BioSpin, Germany) using a the reader comprehension: Château Mouton Rothschild re- 1 M HCl solution. Finally, 550 μL of the solution was trans- quested usual wine analysis for the authentication of four ferred in a 5-mm NMR tube. wines of their estate. Additionally, all these samples were an- 1H NMR spectra were manually recorded at 293 K on a alyzed using 1H NMR experiments. Face to the interest of this 600-MHz Advance spectrometer (Bruker Biospin, France) study, Château Mouton Rothschild gave us its authorization to operating at 600.27 MHz using a 5-mm TXI probe with Z- publish this work with the restriction of vintage anonymity gradient coils. Two successive 1H NMR experiments were and analytical data confidentiality. For this reason, statistical used for the acquisition; the first one was a classical pulse data treatments have been presented for each set of sample program using presaturation to suppress water signal identified as wine sets A, B, and C (three different vintages). (ZGPR, TD = 65,536, SW = 12,019.23 Hz, RD = 5 s, AQ = 2.726 s, NS = 8, RG = 5.6, DS = 2). The second one Usual Wine Analysis was a 1H NMR pulse sequence with water and ethanol signal suppression (NOESYGPPS1D, TD = 65,536, The usual wine analyses were performed according to OIV SW = 12,019.23 Hz, RD = 5 s, AQ = 2.726 s, NS = 32, official methods to measure the following parameters: alco- RG = 64, DS = 4). NMR spectra were processed by using holic strength, sugars, volatile acidity, volatile compounds, both Topspin® software (version 3.2, Bruker Biospin, organic acids, metals concentration, deuterium, carbone-13, Germany) and MestReNova® NMR software (version and oxygen-18 stable isotope ratios. All these analyses were 11.0.3, Mestrelab Research, Spain). FIDs were multiplied performed under quality control conditions, and all the com- with an exponential function (LB = 0.3 Hz) followed by pounds were analyzed in a row with duplicates. Beside the Fourier transformation. Spectra were manually phased accuracy of the measurements, the critical step is the compar- and aligned with the TMSP signal shift to zero, and base- ison between two measurements and the limit to establish that lines were corrected automatically using MestReNova®. two samples are different. In order to define a rational decision Semi-automatic quantification of 38 molecules was per- that can be used in any condition, a similarity score (S-score) formed with MestReNova® Simple Mixture Analysis computation (Eq. 1) was envisaged: the analytical results of (SMA) plugin. Global Spectral Deconvolution (GSD) the authentic wine are considered as the target value and the (Cobas et al. 2011) was applied for peak deconvolution suspected wine as the variable; finally, the analytical uncer- to ensure a good integration in a complex mixture. tainty applied to the higher concentration among authentic and suspected analytical measurements provided a value assimi- Data Treatment lated to the laboratories standard deviation. A S-score value higher than 2, which corresponds to a confidence interval of Comparisons between authentic and suspicious wines were 95%, indicates a significant difference between the authentic performed using an adapted Z-score value approach, and a and suspected wine. The graphical representations of these similarity score (S-score) was computed, for each analyte, data treatments are grouped in Fig. 1. according to The results concerning the wines planned for top up level- − ing (As1 and As2 samples) are presented in Fig. 1a. For the − ¼ CS CA ð Þ S score σ 1 analyses, only 30 mL of each wine was provided by the estate. Because usual analyses require large volume of wines, only a Author's personal copy

Food Anal. Methods (2019) 12:956–965 959 limited number of parameters were determined for these sam- samples, significant differences were observed for dry extract ples: organic acids, sugars, metal concentration, deuterium, values, sulfur dioxide concentrations (SO2LforBandSO2T carbone-13, and oxygen-18 stable isotope ratios. No signifi- for C), ethanol δ13C and (D/H)I ratios, total wine δ13C, and Fe cant differences were detected between suspicious and authen- contents. Additionally, oxygen-18 ratios were significantly tic wines. All the S-scores appeared to be lower than 1 indi- different for B samples, just as (D/H)II ratio, Zn, gluconic cating that the suspicious wines were effectively from the acid, and methyl-2-butan-1-ol contents were for C samples. estate and the announced vintage. Some of these discriminant parameters can be directly related The data treatments of samples B and C are presented on to aging process (volatile compounds, sulfite concentration), Fig. 1b, c, respectively. One bottle of each wine was provided but others, like stable isotope ratios, organic acid (Suarez et al. allowing a full analysis using these usual methods. For both 1979), and metal concentration do not evolve significantly

Fig. 1 Graphical representation of the similarity score (S-score) values for the three wine sets calculated using usual analysis. Dashed lines correspond to the limit used for a discriminating factor. The ordinate scale has been voluntary limited to 3, and for this reason, open bars result of higher S-score values Author's personal copy

960 Food Anal. Methods (2019) 12:956–965 with wine aging. Thus, it appears that suspicious wines do not allows the identification of many metabolites from different correspond to the authentic samples provided by the vineyard families such sugars, organic acids, amino acids, higher alco- estate. hols, or polyphenols (Consonni et al. 2011; Godelmann et al. 2016; Fotakis et al. 2013; López-Rituerto et al. 2012; Son 1 H NMR Wine Analysis et al. 2009a, b). For quantification, the 1H NMR spectra were directly inserted in the data treatment software Usual wine analysis can often be time-consuming, laborious, (MestReNova®) for molecule identification and quantifica- and require a large volume of wine. Therefore, q-NMR meth- tion using GSD method (Cobas et al. 2011), allowing the od was applied to evaluate its ability to distinguish between semi-automatized quantification of 38 molecules listed in authentic and counterfeit wines. The three wine sets were Table 1 (Gougeon et al. 2018). In Fig. 2, zooms on six partic- analyzed by q-NMR (Gougeon et al. 2018). The pH of the ular proton resonances are shown highlighting possible dis- 1 analyzed solution has a key role in H NMR as it can be source criminant signals between the both B wines: proline (δH of chemical shift variation (Bharti and Roy 2012). When the 1.99 ppm), succinic acid (δH 2.62 ppm), myo-inositol (δH quantification is performed manually, fixing the pH is not a 4.05 ppm), shikimic acid (δH 6.81 ppm), and ethanal (δH prerequisite; but, as far as the quantification is automatized, a 9.7 ppm). chemical shift could make the quantification not be accurate. Because of the high sensitivity of 1H NMR analysis In this study, the pH was adjusted to 3.1 ± 0.1 using a semi- (Gougeon et al. 2018), the question of wine evolution in bot- automatized system. After 10 min of acquisition, the 1HNMR tle, particularly for old wines, is critical as two wines coming spectra were obtained corresponding to the proton fingerprint from the same production batch can evolve slightly differently of each wine sample. To illustrate the comparison between even in similar conditions of aging. However, the precision of wine samples, a stack of the two 1H NMR spectra of B sam- 1H NMR measurements, along with the S-score criteria, can ples is presented in Fig. 2. The analysis of 1H NMR spectra lead to consider significantly different two similar wines using

Fig. 2 Differences observed in 1H NMR spectra between suspicious (red) and authentic (blue) B wines. Proline (a), succinic acid (b), myo-inositol (c), shikimic acid (d), and acetaldehyde (e). Water and ethanol regions were masked Author's personal copy

Food Anal. Methods (2019) 12:956–965 961 only the measurement uncertainty. In order to consider a glob- Table 1 Computed uncertainty applied to q-NMR results to determine al aging uncertainty, Château Mouton Rothschild provided the S-score between an authentic and a suspected wine samples from the same batch and for three different old vin- Compound Uncertainty (%) tages allowing the estimation of aging impact on wine param- eters. For each vintage, the uncertainties were calculated for 1 2,3-Butanediol 20 all the quantified compounds. The highest value for each mol- 2 Acetic acid 20 ecule was considered as the uncertainty value due to wine 3 Acetoin 20 evolution during aging. The results, listed on Table 1,show 4 Alanine 20 that the uncertainty among similar wines is way higher than 5 Ethanol 15 the measurement uncertainty (Gougeon et al. 2018). This un- 6 20 certainty can be as high as 30% for the so-called degradation 7 Arginine 15 products such as ethyl acetate. These values have been con- 8 Catechin 20 sidered as the method uncertainty values in the S-score 9Choline10 computation. 10 Citric acid 10 The S-score values using q-NMR results of the three sets of 11 Epicatechin 20 suspicious wines are presented on Fig. 3. Concerning the 12 Ethanal 30 wines planned for top up leveling (As1 and As2 samples), the 13 Ethyl acetate 30 spread of the S-score obtained by q-NMR method (Fig. 3a) is 14 Ethyl lactate 20 greater than that observed in usual wine analysis even for 15 Fructose 15 similar compounds such as organic acids (Fig. 1a). This ob- 16 Gallic acid 20 servation indicates that q-NMR is, in a way, too sensitive for 17 γ-Aminobutyric acid 30 wine comparison. Moreover, it could be also estimated that 18 Galacturonic acid 10 the proposed uncertainties (Table 1) are still too low to 19 15 compensate q-NMR precision. Nonetheless, it appears that 20 Glycerol 15 the S-scores of most of the quantified molecules are lower 21 Methyl-2 propanol 1 15 than 2, except for acetoin and acetic acid which are aging 22 Methyl-3 butanol-1 20 markers with an important variability. These results are in 23 Lactic acid 30 agreement with the previous conclusion established from clas- 24 Malic acid 20 sical analysis, i.e., that the two suspected wines are similar to 25 Methanol 20 the authentic wine. Comparison between S-score values ob- 26 Myo-inositol 10 tained by usual methods and q-NMR shows some differences 27 Phenylethyl alcohol 15 for some compounds quantified by the two methods. These 28 Proline 15 differences are only due to the method sensitivity, particularly 29 Shikimic acid 15 when the concentration is at the limit of quantification: as an 30 Succinic acid 10 example for glucose, the HPLC quantification limit is 0.5 g/L, 31 Syringic acid 20 while for q-NMR, it is 0.1 g/L (Gougeon et al. 2018). Figure 4 32 Tartaric acid 20 presents the cross-validation analysis based on the total 33 Trans-caffeic acid 20 number of S-scores higher than 2 between each analyzed 34 Threonine 20 wines. Only the similar wines, i.e., A sample ones, ex- 35 Trigonelline 20 hibitatotalS-scorelowerthan4. 36 Tyrosine 30 Figure 3b, c present the S-scores for the wine sample sets B 37 Valine 30 and C. In contrast to the A wine samples, many molecules 38 30 (13 and 10 molecules for wines B and C, respectively) present a S-score higher than 2, which represents around 30% of the Met2Prop1 methyl-2 propanol-1, Met3but 1 methyl-3 butanol-1, GABA molecules identified. On this basis, the two suspected wines γ-aminobutyric acid cannot be considered as authentic, which is in a good agreement with the interpretation based on the classical analysis. More specifically, contents in succinic acid, ethanal, myo-inositol, are well known as compounds whose contents are specific to and isopentanol were found to be different in the B samples, each grape variety (Huang and Ough 1989;Mardonesetal. whereas methanol, gallic acid, glucose, and threonine were 2005). In this study, their concentrations are quite similar in found as different in the C samples. Additionally, proline, the two authentic wines of lots B and C, and in the two suspect shikimic, and succinic acids and phenyl ethanol showed dif- wines of these two lots. It would seem that fraudulent wines ferent values in the two wine sets. Shikimic acid and proline and authentic wines are not from the same grape variety. Author's personal copy

962 Food Anal. Methods (2019) 12:956–965

Fig. 3 Graphical representation of the S-score values for the three wine sets calculated using q-NMR analysis. Dashed lines correspond to the limit used for a discriminating factor

As a result and for coherence between data interpretation of from two noted exceptions, all the wines are properly discrim- usual wine analysis results and q-NMR results, a 4 S-score inated. The total S-scores were higher than five except for two output has been arbitrary fixed as the limit indicating that Médoc wines (Medoc 5 and 6) and two Libournais ones two wines are significantly different in terms of q-NMR (Libourne 1 and 2). It is not surprising that in the same vine- analyses. yard, some wines bear similar chemical profiles. These wines In order to validate this S-score assumption, associated to (the two Médoc and the two Libournais, respectively) were q-NMR analysis for wine discrimination, 20 commercial produced by close wineries from the same and the same wines (12 Bordeaux, 2 Burgundy, 2 Beaujolais, 2 Loire vintage (Table S1). These results illustrate the potential and Valley, 1 Gascogne, 1 Roussillon) were analyzed using the the limitations of the q-NMR method combined to the total q-NMR method. A validation analysis based on the total num- S-score calculation. Secondly, the total S-scores between wines ber of S-scores higher than 2 between these 20 wines was from the same region (Bordeaux, Burgundy, or Beaujolais) realized. The results are reported in Fig. 4. First of all, aside appear lower than those between wines from different area. Author's personal copy

Food Anal. Methods (2019) 12:956–965 963

Fig. 4 Cross-validation of q- Aa As1 As2 Bs Ba Cs Ca Number of z-scores > 2 NMR method for wine A discrimination based on total S- a 0 score. Top: Château Mouton As1 1-2 Rothschild’s wines; bottom: set of 3-4 As2 20 commercial wines (6 Medoc, 6 5-7 Libournais, 1 Gascogne, 2 Bs 8-10

Burgundy, 2 Beaujolais, 2 Loire Ba 11-14 Valley,and1Roussillon) 15-20 Cs > 20 Ca

1 2 3 4 5 6 7 8 9 1011121314151617181920 1 Medoc 1 2 Medoc 2 3 Medoc 3 4 Medoc 4 5 Medoc 5 6 Medoc 6 7 Libourne 1 8 Libourne 2 9 Libourne 3 10 Libourne 4 11 Libourne 5 12 Libourne 6 13 Loire 1 14 Loire 2 15 Bourgogne 1 16 Bourgogne 2 17 Roussillon 18 Gascogne 19 Beaujolais 1 20 Beaujolais 2

Even if it is possible to distinguish wines from the same area in which represents the S-score values computed between the 13 most cases, the wines from the same terroir have similar chem- Château Mouton Rothschild’s wines used for the uncertainty ical signatures. This observation is illustrated in Fig. S-1 estimation taking into account the wine aging. The vintage

Usual analyses q-NMR

Fig. 5 Scatter-plot of the scores generated from the two principal components, PC1 and PC2, obtained using analytical data from usual analysis in the left side and 1H NMR wine spectra in the right side for the three wine sets A, B, and C Author's personal copy

964 Food Anal. Methods (2019) 12:956–965 interval between the three sets of samples is 2 years. Similarities Acknowledgments Château Mouton Rothschild is warmly thanked for ’ are observable in-between the samples coming from the same their authorization to publish these results. The authors thank the SCL s team of Bordeaux-Pessac for performing usual analysis: Laetitia Gaillard vineyard, but discrimination occurs for vintage gaps of 4 years. for stable isotope analyses; Ludovic Pottier, François Auger, Claude Roux, and Nathalie Diet for ICP-AES; Carole Lagrèze and André Domec for HPCE; and Rodolphe Robin, Maryse Viateau, and Sebastien Complementarity of Techniques Raud for GC and standard wine analysis. Josep Valls-Fonayet (ISVV) is greatly acknowledged for his comments and advises.

From these results, the two ways to approach the wine authen- Funding Information This study is financially supported by the Conseil ticity appear to be complementary: q-NMR brings results on Régional d’Aquitaine,theConseil Interprofessionnel du Vin de Bordeaux various polyols, polyphenols, and protein fraction, and usual (CIVB), and FranceAgriMer program. The work was supported by the analyses provide information on stable isotope and mineral Bordeaux Metabolome Facility and MetaboHUB (ANR-11-INBS-0010 project). concentration. The main advantages of q-NMR reside in the lowvolumerequiredtoperformtheanalysis(0.5mLisneed- Compliance with Ethical Standards ed for the analysis), and the easiness of processing. In order to study the global information brought by the Conflict of Interest Louis Gougeon declares that he has no conflict of analysis, a principal component analysis (PCA) was per- interest. Gregory Da Costa declares that he has no conflict of interest. formed on the two approaches’ results for the three sets of Tristan Richard declares that he has no conflict of interest. François Château Mouton Rothschild’s wines. The principal compo- Guyon declares that he has no conflict of interest. nent analysis is presented on Fig. 5. The data treatment pro- Ethical Approval This article does not contain any studies with human vides a spatial representation which is similar for the two participants or animals performed by any of the authors. approaches. Moreover, this representation confirms the con- clusions presented before, i.e., for A samples, a good corre- Informed Consent Not applicable. spondence between authentic wine (Aa) and suspicious ones Publisher’sNote (As1,As2), and significant differences between authentic and Springer Nature remains neutral with regard to jurisdic- suspicious wines for sets B and C. The interest of PCA repre- tional claims in published maps and institutional affiliations. sentation comes from the close positioning of authentic wines

Ba and Ca. This observation is really interesting as it shows that an estate terroir and wine process provide wines with References similar characteristics as far as the difference of vintage is not too important (in this case 2 years) because aging affects Amargianitaki M, Spyros A (2017) NMR-based metabolomics in wine wine components modifying the analytical characteristics of quality control and authentication. Chem Biol Technol Agric 4:9 wines. As a result, wine A , which comes from a distinct year, Anastasiadi M, Zira A, Magiatis P, Haroutounian SA, Skaltsounis AL, a Mikros E (2009) 1H NMR-based metabonomics for the classifica- is significantly different from wines B and C. The positioning tion of Greek wines according to variety, region, and vintage. of suspicious wines, Bs and Cs, seems to indicate that the same Comparison with HPLC data. J Agric Food Chem 57:11067–11074 wine base has been used to elaborate them. Bharti SK, Roy R (2012) Quantitative 1H NMR spectroscopy. TrAC Trends Anal Chem 35:5–26 Caruso M, Galgano F, Morelli MAC, Viggiani L, Lencioni L, Giussani B, Favati F (2012) Chemical profile of white wines produced from ‘ Bianco’ grape variety in different Italian areas by nuclear Conclusion magnetic resonance (NMR) and conventional physicochemical analyses. J Agric Food Chem 60:7–15 This real case study on wine authentication was performed Cobas C, Seoane F, Dominguez S, Sykora S, Davies AN (2011) A new approach to improving automated analysis of proton NMR spectra using a multitechnique approach (so-called usual wine through global spectral deconvolution (GSD). Spectrosc Eur 23:26–30 analysis) and a q-NMR method. A similarity score, S-score, Consonni R, Cagliani LR, Guantieri V, Simonato B (2011) Identification was developed in order to compare two wines: the authentic of metabolic content of selected Amarone wine. Food Chem 129: and the suspected one. This study showed that, using q-NMR 693–699 results, four values of S-score higher than the absolute value Fotakis C, Kokkotou K, Zoumpoulakis P, Zervou M (2013) NMR me- tabolite fingerprinting in grape derived products : an overview. Food of 2 indicate that the wines are significantly different. This Res Int 54:1184–1194 limit was confirmed by the comparison among 20 commercial Godelmann R, Fang F, Humpfer E, Schütz B, Bansbach M, Schäfer H, wines. When the comparison with authentic wine is not pos- Spraul M (2013) Targeted and nontargeted wine analysis by 1H sible, q-NMR results can be difficultly exploitable for the NMR spectroscopy combined with multivariate statistical analysis. Differentiation of important parameters: grape variety, geographical authentication of a specific wine estate. Nonetheless, this origin, year of vintage. J Agric Food Chem 61:5610–56169 study showed the complementarity of the information provid- Godelmann R, Kost C, Patz CD, Ristow R, Wachter H (2016) ed by the two approaches. Quantitation of compounds in wine using 1H NMR spectroscopy: Author's personal copy

Food Anal. Methods (2019) 12:956–965 965

description of the method and collaborative study. J AOAC Int 99: OIV h Compendium of international methods of analysis: determination 1295–1304 by isotope ratio mass spectrometry 13C/12C of wine ethanol or that Gougeon L, Da Costa G, Le Mao I, Ma W, Teissedre PL, Guyon F, obtained through the fermentation of must, concentrated musts or Richard T (2018) Wine analysis and authenticity using 1H NMR grape . OIV-MA-AS312–06 at http://www.oiv.int. metabolomics data: application to Chinese wines. Food Anal OIV i Compendium of international methods of analysis: method for Methods 11:3425–3434 18O/16O ratio determination of water in wines and must. OIV- Huang Z, Ough CS (1989) Effect of vineyard locations, varieties, and MA-AS2–12 at http://www.oiv.int. rootstocks on the juice amino acid composition of several cultivars. OIV j Compendium of international methods of analysis: method for Am J Enol Vitic 40:135–139 sulfur dioxyde. OIV-MA-AS323-04A at http://www.oiv.int. Larive CK, Barding GA, Dinges MM (2014) NMR spectroscopy for Pauli GF, Gödecke T, Jaki BU, Lankin DC (2012) Quantitative 1H NMR. metabolomics and metabolic profiling. Anal Chem 87:133–146 Development and potential of an analytical method: an update. J Nat López-Rituerto E, Savorani F, Avenoza A, Busto JH, Peregrina JM, Prod 75, 834–851 Enegelsen SB (2012) Investigation of la terroir for wine pro- PereiraGE, GaudillereJ, van LeeuwenC, HilbertG, MaucourtM, duction using 1H NMR metabolomics. J Agric Food Chem 60: DebordeC, MoingA, RolinD (2007) 1H NMR metabolic profiling 3452–3461 of wines from three cultivars, three soil types and two contrasting Mardones C, Hitschfeld A, Contreras A, Lepe K, Gutiérrez L, von Baer D vintages. J. Int Sci Vigne Vin41, 103–109 (2005) Comparison of shikimic acid determination by capillary zone Schlesier K, Fauhl-Hassek C, Forina M, Cotea V, Kocsi E, Schoula R, electrophoresis with direct and indirect detection with liquid chro- van Jaarsveld F, Wittkowski R (2009) Characterization and determi- matography for varietal differentiation of red wines. J Chrom A nation of the geographical origin of wines. Part I: overview. Eur 1085:285–292 Food Res Technol 230, 1–14 Medina B, Salagoity MH, Guyon F, Gaye J, Hubert P, Guillaume F Simmler C, Napolitano JG, McAlpine JB, Chen SN, Pauli GF (2014) (2013) Using new analytical approaches to verify the origin of wine. Universal quantitative NMR analysis of complex natural samples. In “New Analytical Approaches for Verifying the Origin of Food” Curr Opin Biotechnol 25:51–59 ED. P. Brereton, Woodhead publishing limited, 149–188 Smeyers-Verbeke J, Jäger H, Lanteri S, Brereton P, Jamin E, Fauhl- OIV a Compendium of international methods of analysis: alcoholic Hassek C, Forina M, Römisch U (2009) Characterization and deter- strength by volume.-OIV-MA-AS312-01A at http://www.oiv.int. mination of the geographical origin of wines. Part II: descriptive and OIV b Compendium of international methods of analysis: dosage of inductive univariate statistics. Eur Food Res Technol 230, 15–29 sugars in wine by HPLC. OIV-MA-AS311–03 at http://www.oiv.int. Son HS, Kim KM, van den Berg F, Hwang GS, Park WM, Lee CH, Hong OIV c Compendium of international methods of analysis: volatile YS (2008) 1H nuclear magnetic resonance-based metabolomic char- Acidity. OIV-MA-AS313–02 at http://www.oiv.int. acterization of wines by grape varieties and production areas. J Agric OIV d Compendium of international methods of analysis: analysis of Food Chem 56:8007–8016 volatile compounds in wines by gas chromatography. OIV-MA- Son HS, Hwang GS, Ahn HJ, Park WM, Lee CH, Hong YS (2009a) AS315–27 at http://www.oiv.int. Characterization of wines from grape varieties through multivariate OIV e Compendium of international methods of analysis: determination statistical analysis of 1H NMR spectroscopic data. Food Res Int 42: of the principal organic acids of wines and sulphates by capillary 1483–1491 electrophoresis. OIV-MA-AS-313-19 at http://www.oiv.int. Son HS, Hwang GS, Kim KM, Ahn HJ, Park WM, van den Berg F, Hong OIV f Compendium of international methods of analysis: analysis of YS, Lee CH (2009b) Metabolomic studies on geographical grapes mineral elements in wines using ICP - AES (inductively coupled and their wines using 1H NMR analysis coupled with multivariate plasma / atomic emission spectrometry). OIV-MA-AS311–05 at statistics. J Agric Food Chem 57:1481–1490 http://www.oiv.int. Suarez MA, Polo MC, Llugano C (1979) Etude de la composition de vins OIV g Compendium of international methods of analysis : determination mousseux pendant la prise de mousse et au cours du vieillissement of the deuterium distribution in ethanol derived from fermentation of en bouteilles. I. Etude des acides aminés libres, du glycérol, des grape musts, concentrated grape musts, grape sugar (rectified con- sucres et des acides organiques. Connaiss Vigne Vin 3:199–217 centrated grape musts) and wines by application of nuclear magnetic Viggiani L, Castiglione Morelli MA (2008) Characterization of wines by resonance (SNIF-NMR/ RMN-FINS). OIV-MA-AS311–05 at nuclear magnetic resonance: a work study on wines from the http://www.oiv.int. Basilicata region in Italy. J Agric Food Chem 56:8273–8279