<sup>1</sup>H-NMR metabolomics for wine screening and analysis

Inès Le Mao; Grégory Da Costa; Tristan Richard

doi:10.20870/oeno-one.2023.57.1.7134

Review articles

Inès Le Mao, Grégory Da Costa, Tristan Richard

Vol. 57 No. 1 (2023): OENO One

Received : 10 August 2022; Accepted : 16 November 2022; Published : 3 January 2023

DOI: https://doi.org/10.20870/oeno-one.2023.57.1.7134

Abstract

The number of metabolomic studies has grown steadily over the last twenty years. Among the fields of application, food sciences are broadly represented. Proton NMR (¹H-NMR) is a commonly used technique for metabolomics and is particularly suitable for wine analysis, because the major wine constituents are highly dependent on biotic and abiotic conditions. ¹H-NMR-based metabolomics were used first to guarantee the authenticity of wines, and more recently to determine the impact of viticultural or oenological practices using both targeted and untargeted protocols. This state-of-the-art review covers the different analytical methodologies developed to ensure wine traceability from sample preparation to ¹H-NMR spectrum analysis. The potential applications of 1H-NMR spectroscopy in oenology, from wine authenticity control to the monitoring of winemaking, are described. The challenges and perspectives of the deployment of NMR for oenological monitoring are also discussed.

Introduction

Metabolomics is the scientific study of the small molecules (metabolites) of a biological system based on a complete chemical analysis (omics technologies) in order to detect as many substances as possible (Cevallos-Cevallos et al., 2009; Rochfort, 2005). It is therefore based on the chemical analysis of a biological matrix coupled with multivariate data analysis.

NMR is an analytical technique that has been used in food sciences for several decades (Hatzakis, 2019). Site-specific natural isotopic fractionation by NMR (SNIFNMR) is widely used to detect wine chaptalization (Viskić et al., 2021). NMR-based metabolomics based on ¹H-NMR spectroscopy really began in the 80s and is now used to characterise human body fluids (Wishart, 2019). NMR has become one of the most widely used techniques in metabolomics to analyse complex mixtures, such as body fluids and natural extracts. In the last twenty years, the application of NMR-based metabolomics in food sciences in general and for vine products in particular has stimulated keen interest in this technique, as shown in Figure 1. Its use in wine is of central importance (Amargianitaki and Spyros, 2017), with approximately 20 % of NMR-based metabolomic studies on food focusing on vine or wine.

Compared to other food products, the chemical analysis of wine is a major challenge, since its chemical composition is complex and can evolve over time. The main advantage of NMR spectroscopy is its ability to identify and quantify in a single experiment a wide range of chemical compounds, such as amino acids, organic acids, alcohols, sugars and phenolic compounds (Gougeon et al., 2018). NMR signals are directly proportional to the number of resonating nuclei and compound concentrations, and the relevance of ¹H-NMR analysis as a methodology for quantitating wine components was recently demonstrated in an international collaborative trial (Godelmann et al., 2016).

Figure 1. NMR-based metabolomics in food sciences and infringing rate in food.

Save View full size Expand inline Collapse inline

A: Number of publications including the keywords [Food and NMR and Metabolomics] (in blue) or [Food and NMR and [Vine or Wine]] (in red), indexed in Scopus. B: Infringing market and infringement rate by product class in Europe (Wajsman et al., 2016).

The chemical information obtained by ¹H-NMR spectroscopy, a.k.a. the wine metabolome, is affected by several winemaking factors such as agronomic practices and pedoclimatic conditions (Mazzei et al., 2010), grape variety (Son et al., 2008), fermentation process (López-Rituerto et al., 2009; López-Rituerto et al., 2022) and geographical origin (Gougeon et al., 2019a; Papotti et al., 2013). The NMR spectrum of a wine sample can be considered as a molecular fingerprint and can be used for traceability and authentication purposes (Solovyev et al., 2021; Valls Fonayet et al., 2021). The purpose of this review is to establish state of the art on the potential use of ¹H-NMR-based metabolomics in oenology, from the issue of establishing authenticity to ascertaining wine quality.

Experimental methodology for wine analysis

Initially, NMR was mainly used for the structural elucidation of compounds in organic and inorganic chemistry. It is used in metabolomics largely thanks to its ability to quantitate compounds. Several approaches, from sample preparation to data processing, have been described in the literature. To discuss these approaches, their analytical steps are divided into three parts: sample preparation, acquisition of spectra, and post-acquisition and data processing.

1. Wine sample preparation

The preparation of wine samples is a decisive step for the quantitative NMR (qNMR) analysis of wine. Several protocols have been developed to prepare samples including extraction steps or the use of internal standards and buffers. These methods are summarised in Table 1.

Table 1. Summary of wine sample preparation methods.

Expand inline Collapse inline

Method	Solvent/Buffer	pH adjusted	Standards	References
Liquid-liquid extraction	methanol-d₄	no	no	(Ali et al., 2011)
XAD column extraction	methanol-d₄	no	no	(Anastasiadi et al., 2009)
Solid-phase extraction	methanol-d₄	no	TSP	(Ocaña-Rios et al., 2021)
Evaporation with nitrogen	D₂O	no	TSP	(Amaral and Caro, 2005)
Evaporation with nitrogen	D₂O/phosphate	yes (pH 6.0)	TSP	(Chang et al., 2014)
Evaporation with centrifugal evaporator	D₂O	no	TSP	(De Pascali et al., 2014)
	D₂O/phosphate	no	TSP	(Aru et al., 2018)
	D₂O	yes (pH 3.0)	TSP	(Alves Filho et al., 2019)
Evaporation with centrifugal evaporator and lyophilisation	D₂O	no	TSP	(da Silva Neto et al., 2009)
Lyophilisation	D₂O	no	TSP	(Pereira et al., 2007)
	D₂O	no	DSS	(Hu et al., 2015; Viggiani and Morelli, 2008)
	D₂O	yes (pH 2.0)	TSP	(Papotti et al., 2013)
	D₂O/oxalate	no	DSS	(Zhu et al., 2018)
Direct analysis	D₂O	no	formic acid	(Mazzei et al., 2010)
	D₂O	no	TSP, formic acid	(Mazzei et al., 2013)
	D₂O	yes (pH 3.2)	TSP	(Ragone et al., 2015)
	D₂O	yes (pH 3.0)	TSP	(López-Rituerto et al., 2012; López-Rituerto et al., 2022)
	D₂O/oxalate	no	TSP	(Baiano et al., 2012)
	D₂O/acetate	no	TSP	(Martin-Pastor et al., 2016)
	D₂O/PBS	no	no	(Kioroglou et al., 2020)
	D₂O/phosphate	yes (pH 3.1)	TSP	(Godelmann et al., 2016)
	D₂O/phosphate	yes (pH 3.0)	TSP	(Fan et al., 2018)
	D₂O/phosphate	no	TSP	(Crook et al., 2021; Magdas et al., 2019)
	D₂O/phosphate	no	TSP, calcium formate	(Gougeon et al., 2018)
	D₂O/phosphate	yes (pH 3.1)	TSP, calcium formate	(Gougeon et al., 2019b)
	D₂O/phosphate	Yes (pH 4.0)	DSS	(Mascellani et al., 2021)

Wine is a beverage roughly comprising 83 % of water, 12 % of ethanol and 5 % of other compounds. To overcome the problem of water and ethanol contents, several authors use wine samples that are pre-concentrated by drying, lyophilisation or evaporation with nitrogen or argon. Although this improves the detection of compounds present in low concentrations in wine by increasing the signal-to-noise ratio, the drying process causes a loss of volatile and semi-volatile compounds, hence modifying the chemical composition of the samples (Amaral and Caro, 2005; Aru et al., 2018). Moreover, Amaral and Caro demonstrated that freeze-drying is time-consuming and poses reproducibility issues. As a result, most studies use wine directly after filtration or centrifugation (to remove solid residues) and with the addition of at least 10 % deuterated solvent. Nevertheless, the quantification of minor constituents and specific families, such as polyphenols, poses a challenge. Recently, Ocaña-Rios et al. combined solid-phase extraction (SPE) and NMR metabolomics to investigate phenolic acids and flavonoids (Ocaña-Rios et al., 2021).

The pH shift from one sample to another induces a variation in the chemical shifts of certain compounds (Son et al., 2008). White wines are generally more acidic than red wines, so they are more acidic with pH ranging from 2.8 to 4.2. Many authors adjust pH to be able to compare spectra, even if this issue may be circumvented by post-acquisition data processing (detailed in the post-treatment paragraph). Most authors adjust the pH of wines using a buffer solution. Oxalate and phosphate buffer, which are of variable pH and ionic strength, are the two buffers mainly used for the analysis of wines by ¹H-NMR. It is important to adapt the buffer concentration, because the ionic strength can make the tuning of the probe difficult (Bharti and Roy, 2012). To simplify and automatise this step, the use of a titration pH robot is becoming more widespread. The titration pH robot helps to add small amounts of concentrated acid and/or base to finely adjust the pH (Godelmann et al., 2016; Gougeon et al., 2019a). A reference compound is commonly added to calibrate spectra for NMR analyses. This step facilitates both the identification of metabolites and the comparison of the spectra. The references commonly used are 3-(trimethylsilyl)propionic-2,2,3,3-d₄ acid sodium salt (TSP) and 2,2-dimethyl-2-si-lapentane-5-sulfonate-d₄ sodium salt (DSS). In addition to the calibration reference, an internal standard, such as formic acid (Mazzei et al., 2013) or calcium formate (Gougeon et al., 2019a), can be added for quantification purposes.

2. 1H-NMR spectrum acquisition

¹H-NMR experiments are performed with different magnetic fields, from benchtop NMR (62 MHz) to 800 MHz. However, wine analyses are mainly carried out at 400 and 600 MHz. The sequence parameters are crucial to obtain an effective quantitative analysis by NMR (Bharti and Roy, 2012). The main acquisition parameters used (sequence, temperature, etc.) are listed in Table 2 and are classified by spectrometric frequency.

Owing to the composition of wine, a ¹H-NMR spectrum is dominated by the water signal (δ 4.8 ppm) and those of ethanol, a quadruplet at δ 3.6 ppm (CH₂) and a triplet at δ 1.2 ppm (CH₃). The concentration of the compounds of interest is low compared to these two major compounds, and it is thus impossible to detect and accurately measure them with a simple proton sequence. Therefore, there are two possibilities for studying the composition of wine by NMR: either physically removing water and ethanol by evaporation and freeze-drying and then using simple proton sequences, or removing the solvents by using specific pulse sequences. The former has been widely used and NMR analysis consists of a single-pulse ¹H-NMR sequence that includes a 90 ° radio frequency pulse, followed by a signal acquisition time and a relaxation delay (d₁). In this case, the NMR analysis is simple and fast since only one sequence is needed, but the evaporation step can lead to the loss of other volatile compounds, as mentioned previously. While this technique is therefore limited for wine metabolomics, it can be useful for specific compounds of interest, such as polyphenols (Ocaña-Rios et al., 2021).

Table 2. Main acquisition parameters used in studies of metabolomic analysis of wines by 1H-NMR.

Expand inline Collapse inline

Field (Hz)	T (K)	Sequence	Sequence parameters	PC^a	BC^b	References
62	-	presaturation	ns 16, td 32k, d₁ 30 s	-	-	(Matviychuk et al., 2021)
200	298	presaturation	ns 160, sw 25 ppm, td 16k, d₁ 3 s	-	-	(Amaral and Caro, 2005)
300	293	proton	ns 32, sw 15 ppm, td 32k, d₁ 3 s	-	-	(da Silva Neto et al., 2009)
400	298	wet1d	ns 128, sw 12 ppm, td 32k, d₁ 1.5 s	manual	automatic	(Caruso et al., 2012)
	298	zgpr	ns 128, sw 10 ppm, td 64k, d₁ 60 s	manual	automatic	(López-Rituerto et al., 2009)
	298	zgpr	ns 128, sw 20 ppm, td 64k, d₁ 10 s, rg fixed	-	-	(López-Rituerto et al., 2012)
	300	zgpr	ns 8, sw 20 ppm, td 64k, d₁ 4 s, rg 1	-	-	(Godelmann et al., 2013)
	300	zgpr	ns 4, sw 20 ppm, td 64k, d₁ 4 s, rg 1	-	-	(Ragone et al., 2015)
	300	zgpr	ns 16, sw 20 ppm, td 64k, d₁ 1 s	-	-	(Geana et al., 2016)
	298	zgesgp/mt1ir	ns 96, sw 16 ppm, td 32k, d₁ 3.5 s	-	automatic	(Mazzei et al., 2013)
	300	noesygpps1d	ns 32, sw 20 ppm, td 64k, d₁ 4 s, rg 16	automatic	automatic	(Ehlers et al., 2022; Godelmann et al., 2013)
	300	noesygpps1d	ns 16, sw 18 ppm, d₁ 6 s, rg 16	local	-	(Godelmann et al., 2016)
	300	noesygpps1d	ns 16, sw 15 ppm, td 64k, d₁ 4 s, rg 16	automatic	automatic	(Fan et al., 2018)
500	298	proton	ns 128, d₁ 1.5 s	manual	manual	(Ali et al., 2011)
	300	noesypr1d	ns 256, sw 20 ppm, td 16k, d₁ 2 s	manual	manual	(Martin-Pastor et al., 2016)
	298	noesypr1d	ns 128, sw 16 ppm, td 64k, d₁ 1 s, rg 18	-	-	(Mascellani et al., 2021)
600	298	zgpr	ns 256, sw 20 ppm, td 64k, d₁ 4 s, rg fixed	-	-	(Aru et al., 2018)
	293	zgpr	ns 8, sw 20 ppm, td 64k, d₁ 5 s, rg 5	manual	automatic	(Gougeon et al., 2019a)
	298	zgpr	ns 32, sw 20 ppm, td 64k, d₁ 20 s, rg fixed	manual	automatic	(Alves Filho et al., 2019)
	298	zgpr	ns 64, sw 20 ppm, td 64k, d₁ 10 s, rg fixed	manual	automatic	(López-Rituerto et al., 2022)
	298	noesypr1d	ns 16, sw 16 ppm, td 32k, d₁ 1.5 s	automatic	automatic	(Son et al., 2009a)
	298	noesypr1d	ns 256, sw 10 ppm, td 32k, d₁ 2 s	-	-	(Zhu et al., 2018)
	300	noesypr1d	ns 256, sw 20 ppm, td 64k, d₁ 5 s	manual	automatic	(Kioroglou et al., 2020)
	293	noesygpps1d	ns 32, sw 20 ppm, td 64k, d₁ 5 s, rg 64	manual	automatic	(Gougeon et al., 2019a)
	298	noesygpps1d	ns 256, sw 12 ppm, td 32k, d₁ 2 s	-	-	(Hu et al., 2020)
700	300	proton	ns 32, sw 20 ppm, td 64k, d₁ 2 s, rg 8	-	-	(Ocaña-Rios et al., 2021)
	-	noesypr1d	ns 128, sw 20 ppm, td 64k, d₁ 4 s	automatic /manual	-	(Crook et al., 2021)
	300	noesygpps1d	ns 128, sw 30 ppm, td 64k, d₁ 10 s, rg 16, lb 0.3 Hz	automatic	automatic	(Baiano et al., 2012)
	300	noesygpps1d	ns 32, sw 20 ppm, td 64k, d₁ 4 s, rg 16	-	-	(Ocaña-Rios et al., 2021)
800	300	zgpr	ns 32, sw 10 ppm, td 14k, d₁ 2 s, rg 64	manual	manual	(Rochfort et al., 2010)

^a PC: phase correction; ^b BC: baseline correction.

ns = number of scans, sw = window size, td = number of acquired points, d_{1 =} relaxation time, rg = receiver gain.

One of the major advantages of NMR is that it is possible to modify pulse sequences to suppress unwanted signals. Indeed, multiple solvent removal techniques are used to increase the signal-to-noise ratio of the compounds of interest. Regarding the removal of solvents, several methods can be used: solvent presaturation, Water suppression Enhanced through T₁ effects (WET) and Nuclear Overhauser Effect SpectroscopY (1DNOESY) sequences being the most popular (Giraudeau et al., 2015; Kew et al., 2017). The simplest method of solvent removal is to pre-saturate the solvent before the 90 ° pulse with continuous irradiation during the relaxation time (zgpr sequence). This method is not recommended for quantitative analysis, because it is not sufficiently selective in frequency, and it can induce signal saturation. Among the other available pulse sequences, multi-solvent suppression with 1D-NOESY is very often used to analyse wines (Monakhova et al., 2014a; Ragone et al., 2015). This sequence makes it possible to significantly increase the receiver gain (rg), and therefore to increase the sensitivity of the analysis (Bharti and Roy, 2012). Selective suppression of the water and ethanol signals is achieved from specific pulse sequences not only during the relaxation time but also during the mixing time. This sequence is easy to optimise and calibrate and is suitable for high-throughput NMR-based metabolomics.

Acquisition parameters depend on the method of preparation (sample concentration, direct analysis or drying, internal or external reference, etc.) and the chosen pulse sequence. However, Table 2 shows several general characteristics of these parameters for wine analysis. The spectral width (sw) analysed is, in general, between 10 and 20 ppm. The number of iterations of the experiment (number of scans, ns) ranges from 16 to 128. For accurate quantitative analysis, the number of scans needs to be adjusted to reach a signal over noise ratio (SNR) higher or equal to 250:1 (Bharti and Roy, 2012). In NMR metabolomics, the smallest peaks (SNR < 15:1) provide the largest coefficients of variation and should be examined carefully (Wang et al., 2013). For quantitative analysis, the relaxation time should be at least 5 × T₁ (Bharti and Roy, 2012), where T₁ is the constant time characteristic of the relaxation process for spin to return to its thermal equilibrium value after pulsing. This process is called longitudinal relaxation. The relaxation time (d₁) is generally fixed between 4 s and 6 s for direct wine analyses. The number of points acquired in the time domain (td) is usually between 32k and 64k for wine analyses. Bharti and Roy consider that a lower number of points does not provide sufficient resolution for the quantisation of signals, especially those that are partially superimposed, and that 32k data points are sufficient for quantitative analyses. Finally, temperature is also a critical factor that affects the reproducibility of the results (variations in chemical shifts). It is important to maintain a constant temperature for all analyses, especially for those using external quantitation standards, since a change in temperature can affect the relaxation properties, which can lead to quantitation errors. Analysis temperatures are generally set between 290 and 300 K.

3. 1H-NMR spectrum processing

The acquired free precession signal (free induction decay or FID) then undergoes a Fourier transform. To increase the resolution, a zerofilling is generally applied. This consists in increasing the number of points, which are generally doubled. Finally, an exponential window function is often applied. This function allows the attenuation of background noise and thus improves the SNR, albeit with a loss of resolution. This negative effect can be offset by using a single parameter called line broadening (lb), which is usually adjusted between 0.3 and 1.0 Hz (Bharti and Roy, 2012).

After the Fourier transform, a phase correction of the spectrum is performed to obtain symmetrical signals over the entire spectrum. This is an important step because mis-phased signals can lead to integration errors and thus induce incorrect quantisation. Manual phase correction is preferable to automatic phase correction in metabolomic studies, because small signals are distorted during automatic phase correction (Bharti and Roy, 2012). Baseline correction is often required to correctly account for the area of the integrated peak. This correction can be carried out manually or automatically by various mathematical functions with the help of processing software. Finally, the chemical shift of the spectrum is calibrated to allow the processing and comparison of the different ¹H-NMR spectra acquired. In general, calibration consists in adjusting the reference signal of the internal calibration standard (DSS, TSP) to δ 0.00 ppm.

1H-NMR spectrum analysis

Two different methods can be applied to ¹H-NMR wine spectra (Vignoli et al., 2019). The first is targeted analysis, which is based on the identification and quantification of a panel of wine constituents. These compounds must be identified and quantified without ambiguities on the ¹H-NMR spectrum. The advantage of identifying the compounds present in wine is to be able to monitor their evolution according to different parameters and to determine the metabolites involved in the differentiations. The second method is untargeted analysis (fingerprinting), which provides a global view of all the observable wine constituents previously identified or not. This approach allows spectral pattern comparison to discriminate specific vine and wine behaviours (e.g., variety, vintage, geographic origin, winemaking process, etc.). Finally, data obtained from ¹H-NMR spectra are usually combined with multivariate analysis using supervised and unsupervised methods. The application of chemometrics on the ¹H-NMR metabolomics data allows wine classification.

1. Targeted analyses

1.1 Identification of wine constituents

Wine is a complex matrix composed of many metabolites. For metabolomics purposes, most of the studies reviewed used wine directly. This method is a more reliable quantitative approach, because extraction, freeze-drying and evaporation can lead to the loss of all or some compounds (Amaral and Caro, 2005; Aru et al., 2018). In addition, it limits the number of wine manipulations, thereby increasing the reproducibility of the analyses. Nevertheless, with this approach, only the major constituents are observable on the ¹H-NMR spectrum. Figure 2 shows a representative ¹H-NMR spectrum of wine obtained by direct analysis. The signals of water (δ 4.8 ppm) and ethanol at (δ 3.6 and 1.2 ppm) were suppressed in a 1D-NOESY experiment.

Dozens of compounds from different families are superimposed onto the spectra obtained: alcohols and polyols, amino acids, organic acids, phenolics, sugars, esters, aldehydes and ketones. An initial approach is to compare signal assignments of the wine constituents with literature data and libraries provided by NMR data banks, such as the Biological Magnetic Resonance Bank (BMRB, https://bmrb.io/), the Yeast Metabolome Database (YMDB, http://www.ymdb.ca/) and the Birmingham Metabolite Library (BML, http://www.bml-nmr.org/).

Figure 2. Typical wine 1H-NMR spectrum using multi-solvent suppression based on 1D-NOESY experiment.

Save View full size Expand inline Collapse inline

Spectrum is divided into three regions: 0.5-3.5 ppm, 3.5-6.5 ppm and 6.5-9.5 ppm. Wine mixed with phosphate buffer and deuterated buffer (7:2:1 v/v). pH automatically adjusted to 3.1. ¹H-NMR spectrum recorded at 293 K using the following parameters: ns 32, td 64k, sw 16 ppm, d₁ 5 s.

Table 3. Chemical shifts and coupling constants used for compound identification (Le Mao et al., 2021).

Expand inline Collapse inline

Family	Peak	Compound	δ¹_H (multiplicity, J in Hz, assignment)
Family	Peak	Compound	Used for quantification	Other signals
Alcohols	1	ethanol	1.18(t, 7.2, CH₃)	3.65 (q, 7.1, CH₂)
	2	isopentanol	1.65 (m, CH)	0.88 (d, 6.7, 2CH₃), 1.44 (q, 6.9, CH₂), 3.61 (t, 6.7, CH₂)
	3	myo-inositol	3.27 (t, 9.7, CH)	3.52 (dd, 10.0, 2.8, 2CH), 3.61 (t, 2.8, 2CH), 4.05 (t, 2.8, CH)
	4	methanol	3.35 (s, CH₃)	0.87 (d, 6.7, 2CH₃), 1,73 (m, CH)
	5	isobutanol	3.37 (d, 6.7, CH₂)
	6	phenethyl alcohol	7.33 (m, 5CH)	2.85 (t, 6.6, CH₂), 3.74 (t, CH₂)
Amino acids	7	leucine	0.96 (d, 6.2, 2CH₃)	1.71 (m, CHCH₂), 3.74 (m, CH)
	8	isoleucine	0.99 (d, 7.0, CH₃)	0.93 (t, 7.4, CH₃), 1.24 (m, CH₂), 1.45 ( m, CH₂), 1.97 (m, CH), 3.66 (d, 3.9, CH)
	9	valine	1.04 (d, 7.3, CH₃)	0.99 (d, 7.3, CH₃), 2.28 (m, CH), 3.66 (d, 4.3, CH)
	10	threonine	1.33 (d, 6.7, CH₃)	2.58 (d, 4.9, CH), 4.24 (m, CH)
	11	alanine	1.50 (d, 7.2, CH₃)	3.76 (q, 7.2, CH)
	12	arginine	1.89 (m, CH₂)	1.70 (m, CH₂), 3.23 (t, 6.9, CH₂), 3.75 (t, 6.5, CH)
	13	proline	1.99(m, CH₂)	2.06 (m, CH), 2.33 (m, CH); 3.32 (dt, 14.0, 7.0, CH), 3.42 (dt, 11.6, 7.0, CH); 4.11 (dd, 8.6, 6.4, CH)
	14	tyrosine	6.88 (d, 8.4, 2CH)	3.02 (dd, CH₂), 3.17 (dd, CH₂), 3.92 (dd, CH), 7.17 (d, 8.6, 2CH)
Organic acids	15	lactic acid	1.40 (d, 7.0, CH₃)	4.31 (q, 7.0, CH)
	16	acetic acid	2.08 (s, CH₃)
	17	pyruvic acid	2.35 (s, CH₃)
	18	succinic acid	2.65 (s, 2CH₂)
	19	malic acid	2.89 (dd, 16.3, 4.5, CH)	2.78 (dd, 16.3, 7.0, CH), 4.53 (dd, 7.0, 4.5, CH)
	20	citric acid	2.94 (d, 15.6, CH₂)	2.79 (d, 15.6, CH₂)
	21	tartaric acid	4.60 (s, 2CH)
	22	fumaric acid	6.78 (s, 2CH)
	23	syringic acid	7.36 (s, 2CH)	3.84 (s, 2CH₃)
Phenolics	24	catechin	6.01 (d, 2.0, CH)	2.53 (dd, 15.4, 5.7, CH₂), 2.85 (m, CH₂), 4.15 (m, CH), 4.41 (d, 7.0, CH), 6.08 (d, 2.3, CH), 6.84 (d, 8.6, CH), 6.92 (m, 2CH)
	25	epicatechin	6.10 (d, 2.0, CH)	2.76 (m, CH₂), 2.90 (m, CH₂), 4.32 (m, CH), 4.95 (m, CH), 6.93 (m, CH₂), 7.03 (d, 2.0, CH)
	26	gallic acid	7.16 (s, 2CH)
	27	tyrosol	6.85 (m, 8.4, 2CH)	2.77 (t, CH₂), 3.77 (t, CH₂), 7.17 (m, 8.4, 2CH)
Polyols	28	2,3-butanediol	1.13 (d, 6.2, 2CH₃)	3.61 (m, 2CH)
	29	glycerol	3.55 (dd, 11.8, 6.5, CH₂)	3.64 (dd, 11.7, 4.3, CH₂), 3.77 (m, CH)
	30	mannitol	3.84 (dd, 11.9, 2.8, CH₂)	3.65 (dd, 11.7, 6.2, CH₂), 3.73 (m, CH), 3.77 (d, 9.0, CH)
Sugars	31	fructose	4.02 (dd, 12.8, 1.0 CH₂)	3.56 (m, CH₂), 3.70 (m, 2CH₂), 3.77 (m, CHCHCH₂), 3.87 (dd, 9.9, 3.4, CH), 3.97 (m, CH), 4.09 (m, 2CH)
	32	arabinose	4.50 ( d, 7.7, CH)	3.51 (dd,CH), 3.68 (m, CHCH₂), 3.83 (dd, CH), 3.90 (m, CHCH₂), 3.95(m, CH), 4.02 (m, CHCH₂), 5.25 (d, CH)
	33	xylose	5.18 (d, 3.7, CH)	3.21 (dd, 9.3, 7.9, CH), 3.31 (t, 11.4, CH₂), 3.42 (t, 9.25, CH), 3.51 (dd, 9.3, 3.7, CH), 3.63 (m, CHCHCH₂), 3.91 (dd, 11.5, 5.5, CH₂), 4.57 (d, 7.9, CH)
	34	glucose	5.23 (d, 3.6, CH)	3.23 (dd, 9.2, 8.0, CH), 3.39 (m, CH), 3.45 (dd, 9.8, 3.7, CH) 3.72 (m, CHCH₂), 3.82 (m, CHCH₂), 3.88 (dd, 12.2, 2.1, CH₂), 4.63 (d, 7.9, CH)
Others	35	acetoin	1.37 (d, 7.0, CH₃)	2.21 (s, CH₃), 4.42 (q, 7.2, CH)
	36	ethyl lactate	4.21 (q, 7.06, CH)	1.28 (t, CH₃), 1.42 (d, 7.0, CH₃), 4.39 (q, 7.0, CH)
	37	ethyl acetate	2.07 (s, CH₃)	1.26 (t, 7.2, CH₃), 4.12 (q, 7.1, CH₂)
	38	ethanal	2.23 (d, 3.0, CH₃)	9.67 (q, 3.0, CH)
	39	γ-aminobutyric acid	2.50 (t, 7.3, CH₂)	1.96 (m, CH₂), 3.05 (m, CH₂)
	40	choline	3.19 (s, 3CH₃)	3.51 (dd, CH₂), 4.05 (m, CH₂)
	41	trigonelline	9.14 (s, CH)	4.42 (s, CH₃), 8.07 (m, CH), 8.82 (m, 2CH)
	42	galacturonic acid	5.32 (d, 3.8, CH)	3.49 (dd, 8.0, 10.0, CH), 3.69 (dd, 9.9, 3.5, CH), 3.80 (dd, 10.3, 3.8, CH), 3,92 (dd, 10.3, 3.4, CH), 4.24 (dd, 3.6, 1.2, CH), 4.26 (d, 1.2, CH), 4.31 (dd, 3.3, 1.4, CH)
	43	shikimic acid	6.82 (dt, CH)	2.21 (dd, 18.2, 7.0, CH₂), 2.75 (dd, 18.0, 5.3, CH₂), 3.74 (dd, 8.6, 4.3, CH), 4.01 (m, CH), 4.42 (t, 4.1, CH)

To confirm the signal identification, spiking experiments directly in wine or in wine-like matrices using standard molecules can be performed, especially for compounds present in low concentrations (Cassino et al., 2017). Finally, besides the addition of pure standards, a combination of 2D NMR spectra can be used to confirm metabolite identification, including J-resolved spectroscopy (JRES), COSY COrrelated SpectroscopY (COSY) and TOCSY Total Correlation SpectroscopY (TOCSY) (Vignoli et al., 2019). Table 3 shows the main wine constituents identified in wine and their chemical shifts, signal multiplicities and coupling constants (Le Mao et al., 2021).

1.2. Quantification of wine constituents

The integration of the signal area is one of the crucial steps in qNMR analysis. The wine ¹H-NMR spectrum consists of many signals, each corresponding to the chemical shifts of the different non-exchangeable protons present in the analysed extract. The area under each signal is directly proportional to the concentration and proton numbers of the corresponding compound (Bharti and Roy, 2012). The relative or absolute concentration of a compound can be obtained by comparing the area of the peak corresponding to that of the reference signal. Nevertheless, the integration procedure needs to be performed carefully due to overlapping signals and under- or over-estimation effects requiring correction factors (Godelmann et al., 2016).

Various procedures can be performed to provide the reference signal used for quantification, including internal standards, external standards, calibration curve methods and even electronic methods (Bharti and Roy, 2012). An internal standard needs to be stable in the wine matrix. Ideally, its signal should be isolated from the metabolites of interest. Gougeon et al. used calcium formate for the internal standard in wines (Gougeon et al., 2018). Some studies have used succinic acid for the external reference (López-Rituerto et al., 2022). A third approach is to use an electronic reference to avoid the problems posed by internal or external standards. Several specific procedures have been developed and applied to wine, such as electronic reference to access in vivo concentration (ERETIC), quantification by artificial signal (QUANTAS), and pulse length-based concentration determination (PULCON) (Bharti and Roy, 2012). These methods are very sensitive to variations in the physicochemical properties of the samples to be analysed (salt concentrations, analyte concentrations, etc.).

Overlapping signals cannot be integrated accurately using global integration methods. A signal deconvolution can be performed to determine the contribution of an individual peak to the total area (Cobas et al., 2011). NMR software has implemented algorithms to solve the problems of overlap by deconvolution of the signals (Gougeon et al., 2019a). Although useful in some circumstances, this method cannot solve all overlaps seen across the spectrum. Moreover, it does not predict the number of components hidden under a series of superimposed signals (Monakhova et al., 2014b).

Finally, to perform absolute quantification, specific constant response factors may be introduced for each compound depending on the measuring conditions (Godelmann et al., 2016). These factors depend on various parameters, including NMR sequences for solvent suppression and duplet roof effect.

2. Untargeted analyses (fingerprinting)

The untargeted methods are based on global analysis of the ¹H-NMR spectral data (Alonso et al., 2015; Riedl et al., 2015). This approach seeks to take advantage of all the data contained in the NMR spectrum to build patterns for classifying wines. The aim is to discriminate different wine fingerprints based on their geographical origins, varieties or vintages (Magdas et al., 2019). Untargeted analyses require many wine samples to be relevant, so they are combined with chemometrics to discriminate the specific signatures of each class of wines.

Typically, untargeted methods consist in transforming ¹H-NMR wine spectra into matrices of data by what is known as binning or bucketing (Ehlers et al., 2022). Each bin (bucket) represents a small area of the spectrum (between 0.01 and 0.05 ppm). This approach reduces the number of variables and smooths small shift fluctuations between spectra.

One of the main limitations of this procedure is the integration of fixed buckets, applied independently of potential deviations (variation in chemical shifts, local deformation of the baseline, etc.). For example, pH or salt concentrations of wines may distort ¹H-NMR spectra. Several methods have been developed to correct these fluctuations and allow advanced bucketing (Monakhova et al., 2013). For example, NMRProcFlow allows a semi-automatic procedure to be carried out based on adaptive intelligent bucketing (Jacob et al., 2017).

Chemometrics

Irrespective of whether the analysis is targeted, all metabolomics techniques produce a very large volume of data. NMR data are generally subjected to chemometrics based on multivariate data analysis, especially for food authentication studies (Borràs et al., 2015; Granato et al., 2018). Two types of multivariate analysis are commonly used: unsupervised and/or supervised approaches. Although non-linear approaches are used in foodomics, the multivariate statistical analysis models generally used are principal component analysis (PCA) and partial least squares regression models (Bona et al., 2018).

Unsupervised methods, such as principal component analysis (PCA) and hierarchical cluster analysis (HCA), are generally used to highlight patterns in the global data set. They make it possible to classify wines without allocation of samples to a membership group. PCA is the most widely used unsupervised method. It reduces the number of variables by linear combination of the initial variables, providing a smaller set of variables (principal components). PCA used first to identify trends, clusters and outliers (Le Mao et al., 2021; Mascellani et al., 2021).

In contrast to unsupervised methods, the clusters are known in the supervised approaches. Partial least squares discriminant analysis (PLS-DA) and orthogonal projection to latent structures (OPLS-DA) are the most popular supervised methods. Like PCA, PLS-DA allows the dataset to be reduced and simplified, but it differs in its supervised nature. It uses learning sets with a priori-known information (grape variety, geographical origin, etc.) to build a classification model (Amargianitaki and Spyros, 2017; Gougeon et al., 2019a). PLS-DA is characterised by its high discriminatory power, but it can lead to the artificial separation of groups with no real difference between them (Hatzakis, 2019).

A model validation procedure is of crucial importance in order to avoid overly optimistic classification results, which is currently one of the pitfalls of authenticity studies (Kjeldahl and Bro, 2010). The model validation procedure generally involves evaluating the acquired data, the variables selected to build the model, and its predictive capacity and relevance. Depending on the models and statistical packages used (R Project, Matlab, SIMCA, etc.), various validation procedures are available. The best way to estimate a model is the external validation procedure using a training set to build the model and an independent test set to estimate its relevance. In foodomics - an internal validation procedure - cross-validation is often applied when the number of samples is limited (Gougeon et al., 2019a; Magdas et al., 2019; Triba et al., 2015). The most commonly used method is k-fold cross-validation and its derivative leave-one-out cross-validation (LOOCV) for very small datasets (Spyros and Dais, 2013). LOOCV consists of excluding only one object at a time, the others being used to construct the model which is then applied to the discarded sample. In the k-fold method, it is a part of the dataset that is excluded and used as a test set for the rest of the data.

Application of NMR-based metabolomics to wine authenticity

The first highlighted application of NMR-based metabolomics is to guarantee the authenticity of wine. A major new issue in recent decades, both customers and producers require wine to be authentic and traceable. As wine is a product with high added value, it is the target of numerous counterfeits; therefore, customers demand better traceability of the products they consume. So far, three wine characteristics have been addressed by NMR-based metabolomics: geographical origin, grape variety and vintage.

1. Geographical origin

As mentioned by Amargianitaki and Spyros (2017), NMR-based metabolomics has been widely used to classify wines according to their geographical origin. Initial studies first showed that wines from different countries or regions were dissociated by NMR analysis (Brescia et al., 2002; Du et al., 2007). However, as this separation could be due to the grape varieties used, studies showed that the same grape variety vinified in different world regions gave a different metabolome (Caruso et al., 2012; Gougeon et al., 2019; Magdas et al., 2019; Son et al., 2009a; Son et al., 2008). These studies clearly demonstrated the impact of the soil to discriminate the geographical origin of wine. Finally, on a smaller scale, studies have shown that NMR-based metabolomics can also discriminate between wines from regional trademarks in the same area (Gougeon et al., 2019; López-Rituerto et al., 2012; Mazzei et al., 2010; Pereira et al., 2007). For example, Pereira et al. showed for the first time the impact of soil composition on the metabolic profile of wines by highlighting its impact on amino acids, phenolic compounds, glycerol and some organic acids (Pereira et al., 2007). These results were more recently confirmed by (Gougeon et al., 2019). Various studies have been carried out to establish the profile of wines from different regions of the world (Table 4).

Table 4. Geographical origin discrimination by 1H-NMR-based metabolomics.

Expand inline Collapse inline

Geographical origin		Cultivars	Methods	Reference
Countries	Region	Cultivars	Methods	Reference
Australia, New Zealand	-	Pinot noir	¹H-NMR and ICP-MS targeted analysis	(Duley et al., 2021)
China	Shanxi	Cabernet-Sauvignon, Shiraz	¹H-NMR targeted analysis	(Zhu et al., 2018)
	-	Cabernet-Sauvignon, Beihong	¹H-NMR targeted analysis	(Gougeon et al., 2018)
	Shacheng	Cabernet-Sauvignon, Merlot, Ruby cabernet, Syrah, Zinfendel	¹H-NMR targeted analysis	(Hu et al., 2015)
	Yeongdong, Yeongcheon and Chochiwon	Muscat bailey	¹H-NMR targeted analysis	(Son et al., 2009a)
Czech Rep	-	Riesling, Chardonnay, Pinot gris, Sauvignon blanc, Welschriesling, Pinot noir, Grüner veltliner, Gewürtztraminer, Pinot blanc, Blaufränkisch, Pálava, Cabernet-Sauvignon, Hibernal, Zweigeltrebe, Grüner silvaner, Saint laurent, Neuburger, Merlot, Muskat moravsky, Andre, Müller-Thurgau, Cabernet moravia, Blauer portugieser, Dornfelder, Kerner, Aurelius, Alibernet	¹H-NMR targeted and untargeted analysis	(Mascellani et al., 2021)
France	Bordeaux, Beaujolais, Burgundy, Côtes du Rhône, Languedoc-Roussillon, Loire Valley	-	¹H-NMR targeted analysis	(Gougeon, da Costa, Guyon, et al., 2019a)
France, USA, Australia, South Korea	-	Cabernet-Sauvignon, Shiraz, Campbell early	¹H-NMR targeted analysis	(Son et al., 2008)
Germany	Rheinpfalz, Rheinhessen, Mosel, Saar, Ruwer, Baden and Württemberg	Riesling, Pinot noir, Müller-Thurgau, Pinot blanc, Pinot gris, Pinot meunier, Dornfelder, Gewürtztraminer, Silvaner, Lamberger	¹H-NMR targeted and untargeted analysis	(Godelmann et al., 2013)
Greece	-	Red: Mandilaria, Agiorgitiko White: Moschofilero, Asyrtiko	¹H-NMR targeted analysis	(Anastasiadi et al., 2009)
Italy	Verona	Amarone	¹H-NMR untargeted and targeted analysis	(Consonni et al., 2011)
Italy	Basilicata and Campagnia	-	¹H-NMR targeted analysis	(Viggiani and Morelli, 2008)
Hungary	Villány, Eger	Cabernet-Sauvignon, Blaufränkisch, Merlot, Pinot noir	¹H-NMR untargeted analysis	(Nyitrainé Sárdy et al., 2022)
Romania France	Romanian: Transylvania, Oltenia, Moldova	Romanian: Sauvignon blanc, Riesling, Chardonnay, Pinot gris French: Sauvignon blanc, Chardonnay	¹H-NMR untargeted analysis	(Magdas et al., 2019)
Romania	Murfatlar	Cabernet-Sauvignon, Merlot, Feteasca neagra, Pinot noir, Mamaia	¹H-NMR untargeted analysis, HPLC and isotopes targeted analysis	(Geana et al., 2016)
Spain	Galicia	Albariño, Godello, Treixadura, Palomino	¹H-NMR and SPME-GC untargeted analysis	(Martin-Pastor et al., 2016)
Spain	La Rioja	-	¹H-NMR targeted analysis	(López-Rituerto et al., 2012)

The main compounds related to the geographical discrimination of wines are listed in Table 5. There is a consensus that proline, one of the major amino acid in wines, is also the one whose content varies the most depending on the region, followed by alanine, leucine, threonine and histidine (Duley et al., 2021; Gougeon et al., 2018). Grape acids, such as malic acid or citric acid, as well as acids resulting from fermentation, such as lactic or succinic acid, are also impacted (Mazzei et al., 2010; Son et al., 2009a). In their previous study, Son et al. (2008) also found that - and -glucose were discriminative in Cabernet-Sauvignon wines from France, California and Australia. Other metabolites resulting from fermentation are regularly mentioned as being discriminative according to their geographical origin, such as 2,3-butanediol, phenethyl alcohol and glycerol (Duley et al., 2021; Gougeon et al., 2018; Viggiani and Morelli, 2008). Of the phenolic compounds, gallic acid seems to be linked to geographical differences in wines (Gougeon et al., 2018; Nyitrainé Sárdy et al., 2022; Son et al., 2008). Interestingly, an untargeted NMR-based metabolomics approach showed that the region of the phenolic compounds (between 5.1 and 9.8 ppm) is the most discriminating area regarding geographical origin (Magdas et al., 2019).

2. Grape variety

The influence of grape variety on wine chemical composition has been widely studied. Numerous studies have shown that ¹H-NMR-based metabolomics combined with multivariate statistical analysis allows the successful classification of wines according to grape variety. Several comparisons conducted on wines produced in various countries have been performed on red grape varieties (Anastasiadi et al., 2009; Fan et al., 2018; Geana et al., 2016; Gougeon et al., 2019) and white grape varieties (Ali et al., 2011; Anastasiadi et al., 2009; Fan et al., 2018; Godelmann et al., 2013). Recently, a study performed on almost one thousand Czech wines showed the contribution of ¹H-NMR-based metabolomics to the classification of wines according to grape variety (Mascellani et al., 2021). The statistical analyses showed the classification percentage of wines from thirteen different grape varieties. Of the grape varieties widely used in the world, the authors demonstrated that Pinot noir, Riesling, Cabernet-Sauvignon and Chardonnay wines are well discriminated (good classification rate ranging between 76 and 96 %), while Sauvignon blanc and Pinot gris wines were less well classified (45 and 48 % respectively). These results confirm the specificity of the metabolome of wines made from certain grape varieties. The ¹H-NMR-based comparison of wines is able to discriminate between closely related varieties (Hu et al., 2015). As some commercialised wines may be the result of a blend of several grape varieties (blended wine), one study also showed that ¹H-NMR-based metabolomics can discriminate wines made with various proportions of different grape varieties (Imparato et al., 2011).

Table 5. Main discriminating metabolites of geographical origin, grape variety and vintage.

Expand inline Collapse inline

Factors	Organic acids	Alcohols	Sugars	Amino acids	Phenolics
Geographical origin	malic acid, citric acid, lactic acid, succinic acid	2,3-butanediol, phenethyl alcohol, glycerol	glucose	proline, alanine, leucine, threonine, histidine	gallic acid
Grape variety	tartaric acid, citric acid, malic acid, lactic acid, succinic acid, acetic acid, shikimic acid	ethyl acetate, 2,3-butanediol, glycerol, methanol, acetone, isopentanol	glucose, fructose	proline, arginine, alanine, valine, leucine, isoleucine, choline, threonine, γ-aminobutyric acid	gallic acid, catechin, syringic acid, caffeic acid
Vintage	lactic acid, tartaric acid, fumaric acid, malic acid, citric acid, succinic acid	2,3-butanediol, ethyl acetate, glycerol	glucose, xylose	proline, alanine, leucine, valine, choline, γ-aminobutyric acid	catechin, gallic acid, syringic acid, epicatechin, caffeic acid

As shown in Table 5, most studies agree that amino acids are the main compounds involved for grape variety classification, in particular proline, arginine, alanine and valine (Anastasiadi et al., 2009; Zhu et al., 2018). There is also a consensus that other compounds, such as organic acids (malate, tartrate, citrate, succinate, acetate and lactate) and alcohols (2.3-butanediol, glycerol), are markers of grape varieties (Geana et al., 2016; Godelmann et al., 2013; Hu et al., 2015). This is because the initial levels of primary metabolites from the grape berry vary greatly depending on the grape variety used (Cosme et al., 2016; Liu et al., 2006). These compounds undergo numerous chemical processes leading to different wine metabolomic profiles. Finally, shikimic acid, which is extracted from skin during winemaking, has also been identified as a grape variety marker (Godelmann et al., 2013; Magdas et al., 2019; Nyitrainé Sárdy et al., 2022).

3. Vintage

Finally, the vintage is a crucial criterion for guaranteeing wine authenticity. Indeed, since climatic and environmental conditions vary from one year to another, it strongly affects the chemical composition of grape berries. Vintage therefore plays an important role in evaluating the metabolic profile of wines. Several publications have shown that ¹H-NMR-based metabolomics reveals differences in the metabolic profiles of wines depending on the vintage (Anastasiadi et al., 2009; Consonni et al., 2011; Lee et al., 2009a; López-Rituerto et al., 2012), even if the percentage of good classification can depend on the vintage (Gougeon et al., 2019). A recent study showed the impact of four vintages (2009-2012) on the chemical composition of Cabernet-Sauvignon wine; it demonstrated that it is not so much the compositions of metabolites that change, but rather the contents (Zhang et al., 2021).

As indicated in Table 5, all authors agree that most organic acid, sugars, and amino acid contents vary with vintage (Cassino et al., 2017; Gougeon et al., 2019). Some studies have also shown that the contents of phenolic compounds can change from vintage to another (Anastasiadi et al., 2009; López-Rituerto et al., 2012). While initial levels depend on climatic and environmental conditions, a significant decrease in the bottle can occur, mainly due to condensation reactions involving anthocyanins and flavonols (Consonni et al., 2011).

Although the vintage has a clearly demonstrable effect on the chemical composition of wine, aging greatly complicates the issue. Wine continues to evolve in the bottle, leading to variations in the results obtained over time. Cassino et al. conducted a study on two white wines and 10 red wines for two and four years respectively (Cassino et al., 2019). They found that some compounds analysed by ¹H-NMR-based metabolomics were impacted. Overall, they showed that, in red wines, 2,3-butanediol, acetic acid, ethyl lactate, ethyl acetate and gallic acid increase during aging, while acetoin, lactic acid, galacturonic acid, histidine, leucine, glucose, xylose, catechin and epicatechin decrease. They also observed an increase in ethyl acetate and ethyl lactate in white wines, but a decrease in malic acid, lactic acid and succinic acid. They attributed these findings to oxidation, reduction, hydrolysis and precipitation phenomena in wine (Cassino et al., 2019). Similar results were obtained by (Gougeon et al., 2019), who compared young wines (2013-2016) to older wines (2004-2007) from the Bordeaux area. They found that vintage and aging effects were closely linked. Thus, for the purpose of wine authentication, they proposed using a z-score system based on the evolution of compound levels in bottles over time (Gougeon et al., 2019).

Application of NMR-based metabolomics to control winemaking

Metabolic NMR may also be used to control winemaking by studying the impact of various viticultural and oenological practices. This application has received less attention than controlling for wine authenticity, but it demonstrates the usefulness of ¹H-NMR-based metabolomics as a comprehensive tool for studying the impact of different practices commonly used in winemaking.

1. Viticultural practices

The impact of several viticultural practices on the chemical composition of wine has been increasingly studied by using ¹H-NMR-based metabolomics over the past decade. To our knowledge, one of the first studies demonstrating the value of ¹H-NMR-based metabolomics was performed on Cabernet-Sauvignon wine produced by different cultivation techniques (Todasca et al., 2011). Variations in the wine metabolome depending on the viticultural practices were observed. The effects of tilling the soil, fertilisation and the training system were also studied and the wines analysed showed a different chemical composition depending on the process (Ciampa et al., 2019; De Pascali et al., 2014). ¹H-NMR-based metabolomics has proven to be an efficient tool for studying the influence of organic and biodynamic cultivation on the wine metabolome (Laghi et al., 2014; Picone et al., 2016). Several studies have recently used it to analyse wines produced with grapes at different stages of maturity (Alves Filho et al., 2022; Chang et al., 2014; Le Mao et al., 2021). Since the primary metabolism is directly impacted by maturity, the use of ¹H-NMR-based metabolomics is relevant. Each one of these studies was conducted on different grape varieties and showed an effect on amino acids, organic acids, sugars and phenolic compounds. In a context of climate change, this avenue of research is of particular interest to be able to continue to produce quality wines in the future.

2. Winemaking practices

¹H-NMR-based metabolomics has also been used to study the influence of oenological processes on wine metabolism to better understand and control the impact of practices commonly used in oenology. One of the first uses was to study the impact of different fermentation processes and evaluate the fermentation characteristics of different yeast strains (Hanganu et al., 2011; Mazzei et al., 2013; Son et al., 2009b) and bacteria (Lee et al., 2009b). The influence of Botrytis cinerea attack on grape berries used in Champagne wines has also been demonstrated (Hong et al., 2011). NMR-based metabolomics analyses have also evaluated the interest and influence of using innovative winemaking technologies such as cryomaceration, reductive winemaking, and ultrasound (Baiano et al., 2012; De Pascali et al., 2014). Recently, studies have shown that ¹H-NMR-based metabolomics is an effective tool to monitor the evolution of the majority compounds in wine during fermentation and barrel aging (López-Rituerto et al., 2022), or to evaluate the impact of the use of different glues or enzymes on the chemical composition of wines (Le Mao et al., 2021). Maceration time was also studied by ¹H-NMR-based metabolomics and although it showed a tendency to impact certain compounds, but ANOVA results were found to be non-significant (Alves Filho et al., 2022). An interesting non-targeted approach on Mexican Merlot wines was used by focusing on the phenolic compounds region (5.58-8.00 ppm) to study different aging processes (Herbert Pucheta, 2019). Finally, as previously indicated, ¹H-NMR-based metabolomics is an efficient tool for monitoring the wine aging effect (Cassino et al., 2019; Gougeon et al., 2019).

Challenges and perspectives

¹H-NMR-based metabolomics can be used both to guarantee the authenticity of wines and to control wine-growing parameters, whether the approach is targeted or non-targeted and quantitative or not. NMR has important advantages for the analysis of complex mixtures compared to other usual metabolomics tools: easy sample preparation, short analysis times, good reproducibility and adequate specificity (Wishart, 2019). Nowadays, ¹H-NMR-based metabolomics enables the rapid and efficient quantification of several wine constituents from different chemical classes: organic acids, amino acids, carbohydrates, alcohols and phenolics. In combination with chemometrics including multivariate data analysis, ¹H-NMR spectroscopy allows the discrimination of fundamental wine parameters including geographical origins, grape varieties, and vintages.

However, our review found that many different protocols for preparing wine samples have thus far been used. Most of the wines studied were either freeze-dried, evaporated or directly used. Therefore, depending on the studies, the wine samples may or may not have undergone the addition of a buffer. In addition, they may have used different buffers, with the adjustment or not of the pH of the sample, and with different target pH depending on the study. This protocol variability leads inexorably to different spectra, so caution is required when comparing spectra and results. Almost two decades ago, Amaro and Caro showed that the evaporation of wine produced better results than freeze-drying. Both methods provide good spectral resolution, but they are time-consuming and can lead to reproducibility problems (Amaral and Caro, 2005). The WineScreener^TM system (Bruker Corporation), which is based on ¹H-NMR metabolomics, provides a commercial facility for wine traceability using a proprietary database (Spraul et al., 2015); however, it has a lack of transparency in terms of data validation and chemometric workflow. These factors represent a major limitation to its use by wine control laboratories. There is a pressing need for model validation approaches to verify the robustness of methods, including the long-term stability of the instruments used and the evolution of wine, in order to guarantee their usability and transparency.

Wine authenticity studies can become extraordinarily complex when sophisticated counterfeits are being analysed. Innovative analytical techniques are always required and it is only through the careful combination of various technologies that subtle differences between wines can be revealed (Valls Fonayet et al., 2021). ¹HNMR-based metabolomics now need to be combined with other techniques. Recent studies have attempted the coupling of ¹H-NMR, LC-MS and GC-MS data for rums (Belmonte-Sánchez et al., 2020) and wines (Kioroglou et al., 2020), and 1D 1H-NMR and ICP-MS data for wines (Duley et al., 2021). These studies demonstrated how such data could lead to better prediction efficiency, thus underling the contribution of such approaches. The pooling of data obtained with different analytical techniques increases the reliability of authentication; however, it is not straightforward and presents major methodological challenges, such as the combination of these data with those obtained by multiblock chemometrics (Borràs et al., 2015).

¹H-NMR-based metabolomics has proven its usefulness for monitoring viticultural and oenological practices, and NMR metabolomics is sufficiently rapid and sensitive to monitor the main components of wine. The major wine metabolites can be tracked throughout the winemaking process from the grape berry to the bottled wine. However, there are challenges that still need to be addressed. First, the relationship between wine quality and ¹H-NMR-based metabolomics remains unclear. Rochfort et al. (2010) investigated the relationship between NMR analysis and some sensory aspects of wine quality. They demonstrated that ¹H-NMR-based metabolomics could serve to predict some specific traits of interest correlated with the organoleptic quality of wine. However, wine quality involves many other compounds that are not currently analysed by NMR. Among the non-volatile compounds, phenolic compounds are important markers of organoleptic properties. Interestingly, Ocaña-Rios et al. have developed a solid-phase extraction method for ¹H-NMR-based metabolomics (Ocaña-Rios et al., 2021). By eliminating the major polar wine constituents, the method allows several compounds, including hydroxybenzoates and flavonols, to be assessed. Such approaches could be useful to assess the constituents that directly affect wine quality, such as phenolic compounds. Second, there is a need for simpler, more economically viable NMR systems for use in oenology laboratories. In this respect, benchtop NMR spectrometers are promising (Giberson et al., 2021). They are less expensive and more compact, thus offering new perspectives to a wide range of new users in oenological laboratories. Recently, Matviychuk et al. (2021) quantified more than fifteen wine constituents including alcohols, organic acids and amino acids by applying ¹H-NMR-based metabolomics to wine using a 60 Mhz benchtop spectrometer. These spectrometers could be deployed very shortly where conventional NMR remains inaccessible in terms of cost, technical complexity or physical constraints.

Acknowledgements

This work was supported by Association Nationale de la Recherche et de la Technologie (Inès Le Mao was the recipient of a CIFRE PhD fellowship from ANRT and Baron Philippe de Rothschild S.A.). We would also like to thank Château Mouton Rothschild and Fondation de Bordeaux for their financial support (donors: Baron Philippe de Rothschild SA, Chateau Cheval Blanc, Château Lafite Rothschild, Le Domaine Clarence Dillon, Château Petrus). The work was supported by the WAPNMR project (ANR-21-CE21-0014 project) and MetaboHUB (ANR-11-INBS-0010 project).

References

Authors

Inès Le Mao

Affiliation : Univ. Bordeaux, Bordeaux Sciences Agro, Bordeaux INP, INRAE, OENO, UMR 1366, ISVV, F-33140 Villenave d’Ornon

Country : France

Grégory Da Costa

Affiliation : Univ. Bordeaux, Bordeaux Sciences Agro, Bordeaux INP, INRAE, OENO, UMR 1366, ISVV, F-33140 Villenave d’Ornon

Country : France

Tristan Richard

tristan.richard@u-bordeaux.fr

https://orcid.org/0000-0002-5308-8697

Affiliation : Univ. Bordeaux, Bordeaux Sciences Agro, Bordeaux INP, INRAE, OENO, UMR 1366, ISVV, F-33140 Villenave d’Ornon

Country : France

Attachments

No supporting information for this article

Article statistics

Views: 3932

Downloads

XML: 32

Citations

PlumX