Introduction

One of the main challenges that face the modern wine science is how to optimise grape and wine production in order to have a minimum environmental footprint, lower production costs, as well as how to improve or maintain wine quality (Dambergs et al. 2015; Gishen et al. 2005; Rossouw and Bauer 2009a; b; Fotakis et al. 2013). The traditional wine science, combining viticulture and oenology, has accumulated and generated knowledge and know-how that has resulted in a significant optimisation of the vine-growing and wine-making process (Rossouw and Bauer 2009a, b). However, much of these processes remain based on empirical and even anecdotal evidence (e.g. trial-and-error approaches) accumulated along years by grape-growers and winemakers (Rossouw and Bauer 2009a, b). Only a small fraction of all the interactions and cause-effect relationships between individual inputs and outputs during grape and wine making production is scientifically well understood where the complexity of these processes has prevented a deeper understanding of such interactions and causal relationships (Rossouw and Bauer 2009a, b; Fotakis et al. 2013). The aim of this paper is to provide with a general overview on the applications of omics approaches in grape and wine research, focusing on metabolomics fingerprinting. Examples from the scientific literature published in the past 10 years that have reported the use of metabolomic approaches to analyse grapes and wines will be discussed. Although applications related with geographical origin and authenticity are also based in the metabolomics approach, they were not considered in the current overview.

From Reductionism to Complexity

It has been generally accepted by several researchers that a single analytical technique will not provide sufficient information about the wine metabolome; therefore, a holistic approach is needed for a more comprehensive analysis (Burlingame 2004; Fardet 2014; Cozzolino 2015; Munck et al. 1998; Munck 2007; Kelly et al. 2011; Cozzolino et al. 2007; Lindon and Nicholoson 2008).

For decades, research in food, including grape and wine, has been focused on the reductionist approach (Fardet 2014). In recent years, different studies stated that this approach has prompt with an ‘unreal world view’ where compounds are analysed independently of the food matrix (e.g. cereals, meat, fruit juices) (Burlingame 2004; Fardet 2014; Cozzolino 2015; Munck et al. 1998; Munck 2007; Kelly et al. 2011). Complex systems such as grapes and wine require complex answers, and research in grape and wine need to move towards a more holistic and integrative approach (systems approach). This type of analysis ‘omics approach’ determines a high degree of complexity that has not been extensively used before in wine science (Burlingame 2004; Fardet 2014; Cozzolino 2015; Munck et al. 1998; Munck 2007; Kelly et al. 2011).

Omics studies are normally concerned with multifactorial problems, and it is in this context that makes good sense to explore and to measure the same sample on complementary, synergistic analytical platforms that comprise multifactorial sensors and separation methods (Munck 2007; Kelly et al. 2011; Capozzi and Bordoni 2013; Garcia-Canas et al. 2012; Castro-Puyana et al. 2013; del Castillo et al. 2013; Cevallos-Cevallos et al. 2009; Wishart 2008). However, the challenge of exploring, extracting and describing the data in this way increases exponentially and the risk of becoming flooded with non-informative data increases concomitantly (Munck 2007; Kelly et al. 2011; Capozzi and Bordoni 2013; Garcia-Canas et al. 2012; Castro-Puyana et al. 2013; Cevallos-Cevallos et al. 2009; Wishart 2008). Although the acquisition of data from different analytical platforms provides with new opportunities in grape and wine research, issues such as the validity of the information from the data generated, the comparison of the data between different analytical platforms, and the need for a rigorous control of the integrity of the data in the context of the models generated are still some of the main constraints facing omic approaches in many food applications (including wine), as discussed by several authors (Munck 2007; Kelly et al. 2011; Capozzi and Bordoni 2013; Garcia-Canas et al. 2012; Castro-Puyana et al. 2013; Cevallos-Cevallos et al. 2009; Wishart 2008; Roullier-Gall et al. 2014; Silvestri et al. 2014). In addition, the limitations generated by the current state of the art in the available bioinformatic tools, the limited information in databases (e.g. on the identity of many metabolites), our poor knowledge on many molecular processes that take place or the difficulty to combine the huge data generated by the so called ‘omics’ technologies such as transcriptomics, proteomics and metabolomics are still critical issues to be resolved (Munck 2007; Kelly et al. 2011; Capozzi and Bordoni 2013; Garcia-Canas et al. 2012; Castro-Puyana et al. 2013; Cevallos-Cevallos et al. 2009; Wishart 2008; Roullier-Gall et al. 2014; Silvestri et al. 2014). Omics studies generally attempt to cover a broader range of more holistic research questions, such as how grapes grown under limiting water conditions will change its chemical composition and how this in turn will affect wine production and sensory properties and ultimately the consumer’s perception about wine. The questions behind most omics studies require the utilisation of untargeted analytical methods as well as the extraction of as much information as possible from these platforms (Munck 2007; Kelly et al. 2011; Capozzi and Bordoni 2013; Garcia-Canas et al. 2012; Castro-Puyana et al. 2013; Cevallos-Cevallos et al. 2009; Wishart 2008; Roullier-Gall et al. 2014; Silvestri et al. 2014).

Metabolomics is generally defined as the study of as many metabolites as possible in a system (Cevallos-Cevallos et al. 2009; Wishart 2008; Roullier-Gall et al. 2014; Silvestri et al. 2014). More specifically, it has been defined as a field in omics research concerned with the high-throughput identification and quantification of small molecule metabolites in the metabolome (Cevallos-Cevallos et al. 2009; Wishart 2008), where the metabolome is defined as the collection of all small molecule metabolites or chemicals that can be found in a cell, organ, or organism (Cevallos-Cevallos et al. 2009; Wishart 2008; Roullier-Gall et al. 2014; Silvestri et al. 2014). Figure 1 illustrates the steps in wine metabolomics, from wine to the end use application.

Fig. 1
figure 1

Metabolomics in wine from data rich to the application

Full size image

The Hardware and Software Used in Metabolomics

The sheer size of the generated data in omics often becomes too big to be effectively evaluated by conventional univariate approaches and requires multivariate data analysis (MVA) and pattern recognition (PR) methods that utilise all measured variables simultaneously. These methods are able to identify underlying latent factors that carry information (e.g. biochemistry, chemistry, process) about the sample (Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Allen et al. 2003; Dunn and Ellis 2005; Nielsen and Oliver 2005; Oliver et al. 1998, 2002; Sumner et al. 2003; Sweetlove et al. 2004; Cozzolino 2011; Unger 2009; Wolfender 2009; Roullier-Gall et al. 2015; Hong 2011; Gromski et al. 2015). The most important and effective way to show the best of the success in how different teams tackle complexity is by the use of the ‘holistic’ or ‘systems’ approach. Without the multidisciplinary team approach and the use of the multivariate tools, the development of these types of applications will not be possible. The term ‘omic’ is derived from the Latin suffix ‘ome’ meaning mass or many (Ellis et al. 2015). Metabolomics is still considered as an emerging field in research which enables the chemical and biochemical profiling of samples from living organisms in order to obtain insight into biological processes (Sweetlove et al. 2004; Cozzolino 2011; Unger 2009; Wolfender 2009; Roullier-Gall et al. 2015; Hong 2011; Gromski et al. 2015; Khakimov et al. 2014).

In the context of grape and wine research, metabolomics can be defined as a discipline that studies the main grape and wine domains through the application and integration of advanced omics technologies (Dambergs et al. 2015; Gishen et al. 2005; Rossouw and Bauer 2009a, b; Fotakis et al. 2013; Roullier-Gall et al. 2014; Silvestri et al. 2014). Different analytical methods have successfully been used to characterise wine as well as to find an association of wine metabolite with environmental and fermentative factors in vineyard and making wine. Development of metabolomics has depended on advances in a diverse range of instrumental techniques such as liquid chromatography (LC), electrospray ionisation mass spectrometry (ESI-MS), capillary electrophoresis (CE), gas chromatography (GC), NMR spectroscopy, high-performance liquid chromatography (HPLC), mass spectrometry (MS) and vibrational spectroscopy (e.g. NIR, MIR, Raman) among other techniques (Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Allen et al. 2003; Dunn and Ellis 2005; Nielsen and Oliver 2005; Oliver et al. 1998, 2002; Sumner et al. 2003; Sweetlove et al. 2004; Cozzolino 2011; Unger 2009; Wolfender 2009; Roullier-Gall et al. 2015; Hong 2011; Gromski et al. 2015). Recently, the increase in resolution on mass measurements using direct injection Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) and ultra-high-performance liquid chromatography coupled to mass spectrometry (UPLC/MS) allowed an improvement on the separation ability of isomeric and isobaric substances, increasing the scope of detectable unknown metabolites in wines (Gromski et al. 2015; Khakimov et al. 2014). Table 1 summarises some advantages of the most common analytical methods reported by different authors on the use of metabolomics in grape and wine research.

Table 1 Characteristics of analytical techniques using in grape and wine metabolomics
Full size table

These methods are considered to be useful for fingerprinting as well as for comparing natural and synthetic samples, and to identify single active compounds, and each of them provides unique capabilities to separate different chemical classes of metabolites from several types of samples (Cevallos-Cevallos et al. 2009; Wishart 2008; Gromski et al. 2015; Khakimov et al. 2014).

These technologies and methods in the biological and chemical sciences, combined with MVA, open new opportunities to assess the entire vine-growing and wine-making process from a more holistic perspective (Dambergs et al. 2015; Gishen et al. 2005; Rossouw and Bauer 2009a, b; Fotakis et al. 2013). Developments in mathematics, statistics, software and computers have provided algorithms and techniques such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) that enabled the analysis and interpretation of the complex data generated during metabolomics analysis (Cevallos-Cevallos et al. 2009; Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Gromski et al. 2015).

These mathematical techniques employed in omic applications typically have two purposes, one used as exploratory analysis and/or classification of samples in different categories (e.g. PCA) where the second is commonly used to develop a model (calibration) (Cevallos-Cevallos et al. 2009; Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Gromski et al. 2015). To develop mathematical models that are suitable for identifying and isolating these groups of classes using MVA, pattern recognition techniques are usually incorporated as part of the analysis (metabolomics tool = instrumental method + MVA) (Cevallos-Cevallos et al. 2009; Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Gromski et al. 2015).

The main task in discriminant or classifications methods allows to assign unclassified objects into predefined categories. In most of omic applications, prior to the interpretation or analysis of the data collected, the data often undergo different pre-processing steps (e.g. scattering, baseline corrections, peak alignment) allowing the signal from different samples to be compared and further used in routine applications (Cevallos-Cevallos et al. 2009; Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Gromski et al. 2015). The choice of the most appropriate method (e.g. algorithm, pre-processing) for each of these steps is highly dependent on the sample, instrumentation technique and purpose of analysis, and is beyond the purpose of this review. A recent article by Khakimov and collaborators provide with a summary of the current MVA methods used or applied in omics (Khakimov et al. 2015). This review also provides with a guide of practical and pragmatic tools to validate and to deal advantageously with data from more than one analytical platform, as well as emphasises the need for complementary correlation studies within and between blocks of data to ensure proper data handling, interpretation and dissemination (Cevallos-Cevallos et al. 2009; Wishart 2008; Khakimov et al. 2014, 2015; Skov et al. 2014; Gromski et al. 2015).

Examples on the Current Status of Metabolomics Applications in Grape and Wine Research

As metabolomics is becoming an emerging field in grape and wine research, different analytical techniques (e.g. NMR, GC-MS, HPLC etc.) were used to produce information-rich, highly reliable and reproducible data in a non-targeted or global way where MVA is often applied to analyse and interpret the data generated. Examples of such applications in viticulture and wine making published by several researchers are provided below. Note that the use of metabolomics to classify wines from different geographical origins or varieties is not addressed in this overview.

Metabolomics Applied in Viticultural Studies and Grape Composition

It has been reported that the analysis of grape (Vitis vinifera) berries at the transcriptomic, proteomic and metabolomic levels can provide great insights into molecular events underlying berry development as well as postharvest drying (withering) (Zamboni et al. 2010). However, the large and very different sets of data produced by this type of investigations are difficult to integrate using traditional methodologies. Zamboni et al. (2010) have identified putative stage-specific biomarkers for berry development and withering as well as integrating systems-level studies of these processes using omics approaches. An integrated transcriptomic, proteomic and metabolomic approach using two different strategies, namely hypothesis free and hypothesis driven, was used (Zamboni et al. 2010). In this study, a multistep hypothesis-free approach was applied to the data sourced from four grape developmental stages and three withering intervals, with integration achieved using a hierarchical clustering strategy based on the multivariate bidirectional orthogonal projections to latent structures technique (Zamboni et al. 2010). This approach allowed the identification of stage-specific functional networks of linked transcripts, proteins and metabolites, providing important insights into the key molecular processes that determine the quality characteristics of the wine (Zamboni et al. 2010). The proposed methodology also allowed the identification of transcripts and proteins that were modulated during withering as well as specific classes of metabolites that accumulated at the same time and used these to select several sets of variables (Zamboni et al. 2010). Overall, this paper highlighted the importance of this holistic approach in order to improve the current knowledge of grapevine berry development and withering (Zamboni et al. 2010).

The use of NMR spectroscopy to analyse the unaltered metabolic profile of Sardinian Vermentino grape berries was reported by Mulas and co-workers (Mulas et al. 2011). In this study, seven selections of Vermentino grapes were harvested from the same vineyard where berries were stored and extracted following an unbiased extraction protocol. The extracts were analysed to investigate variability in metabolite concentration as a function of the clone, the position of berries in the bunch or growing area within the vineyard (Mulas et al. 2011). The combination of NMR spectroscopy with PCA analysis, correlation analysis and ANOVA allowed to investigate the sources of variation in the data associated with the different positions of the berries within the bunch. Results from this study indicated that the position of the grape within the bunch is the main variable influencing the metabolic profile of the berries, while growing area and clone do not have an effect (Mulas et al. 2011). The variability in the composition of amino acids such as arginine, proline as well as organic acids (malic and citric) can be used to characterise the rapid rearrangements of the metabolic profile in the berry in response to environmental stimuli (Mulas et al. 2011).

Metabolic patterns of different grapevine cultivars, grown in the same greenhouse, were characterised using NMR spectroscopy as metabolomic approach (Son et al. 2014). Pattern recognition based on PCA analysis revealed clear dependence of the grape metabolome on the grape cultivar (Son et al. 2014). Results from this study indicated a high concentration of proline in the grapes from Cabernet Sauvignon, Merlot and Chardonnay, whereas the proline levels were depleted in the crossbred grapes of cultivars of Steuben, Campbell Early and Seibel (Son et al. 2014). Intrinsic levels of alanine, glutamine and trans-feruloyl derivative were highest in the Campbell Early cultivar, which grows easily in a wild vineyard, suggesting that their levels play important roles in the improvement of resistance or adaptation of the plant to environmental stress, such as freezing stress during the winter season in Korea (Son et al. 2014). This study also highlighted the importance of metabolomics as a powerful approach for better understanding the differences of intrinsic metabolic variables of grape berries among various grape cultivars and their associations with the plant physiological mechanisms (Son et al. 2014).

Grape polyphenols are well-known antioxidant compounds, and they were used as markers in vine chemotaxonomy as well as markers to monitor colour stabilisation of red wines from the Rabioso Piave variety (Panighel et al. 2015). In this study, ultra-high-performance liquid chromatography/high-resolution mass spectrometry (UHPLC/QTOFMS) was used to study the grape metabolomics (Panighel et al. 2015). The use of MS/MS fragmentation also allowed the identification of new polyphenol compounds in the set of grape samples analysed (Panighel et al. 2015). Three putative p-coumaroyl flavonoids were identified in the grape extracts, namely dihydrokaempferide-3-O-p-coumaroylhexoside-like flavanone, isorhamnetin-3-O-p-coumaroylglucoside and a chrysoeriol-p-coumaroyl hexoside-like flavone (Panighel et al. 2015). The elucidation of the structural characteristics of the functional groups was also achieved using this methodology combined with MS fragmentation (Panighel et al. 2015). However, stereochemistry and the definitive position of substituents in the molecule can only be confirmed by isolation and characterisation or synthesis of each compound (Panighel et al. 2015).

Suspect screening analysis is an intermediate approach in metabolomics between the so-called targeted and untargeted analyses (Rosso et al. 2015). In order to develop new methods of grape metabolomics, a new database of putative grape and wine metabolites (GrapeMetabolomics) has been developed containing more than a thousand compounds (Rosso et al. 2015). By performing high-resolution mass spectrometric analyses of grape extracts in positive and negative ionisation mode, and using GrapeMetabolomics, more than 400 putative compounds have been identified (Rosso et al. 2015). Most of these compounds were grape metabolites involved in the determination of different organoleptic characteristics of wine (Rosso et al. 2015). This method also allowed the identification of 30 anthocyanins in hybrids grapes (Rosso et al. 2015).

Anthocyanin degradation has been proposed as one of the primary causes for reduced colour and quality in red wine grapes grown in a warm climate (Chassy et al. 2015). An untargeted metabolomics approach was aided by filtering the MS data using different algorithms to extract all M and M+6 isotopic peak pairs, allowing the analysis to focus solely on the metabolites of phenylalanine (Chassy et al. 2015). A paired-comparison t test that was performed using biological replicates revealed that 13 metabolites were statistically different between thermal treatments (25 and 45 °C) (Chassy et al. 2015). In addition, it was found that resveratrol was significantly reduced following heat treatment, where only five metabolites related with the degradation of anthocyanins products increased following the 45 °C treatment (Chassy et al. 2015).

It is well known that grapevine berries undergo complex biochemical changes during fruit maturation, many of which are dependent upon the variety and its environment (Degu et al. 2014). In order to elucidate the varietal dependent developmental regulation of primary and specialised metabolism, berry skins of Cabernet Sauvignon and Shiraz were analysed using GC-MS and LC-MS in order to obtain the metabolite profile of the samples from pre-veraison to harvest (Degu et al. 2014). The analysis of the metabolite data revealed similar developmental patterns of change in primary metabolites between the two cultivars analysed (Degu et al. 2014). Nevertheless, towards maturity, the extent of change in the major organic acid and sugars such as sucrose, trehalose, malate as well as precursors of aromatic and phenolic compounds such as quinate and shikimate was greater in Shiraz compared to Cabernet Sauvignon (Degu et al. 2014). Using PCA analysis, the authors found that the differences between the two cultivars towards maturation when using the specialised metabolite profiles were apparent, suggesting a cultivar-dependent regulation of the specialised metabolism (Degu et al. 2014). Correlation analysis confirmed the tightly coordinated metabolic changes during development and suggested a source-sink relation between the central and specialised metabolism, which is stronger in Shiraz than Cabernet Sauvignon (Degu et al. 2014). RNAseq analysis also revealed that the two cultivars exhibited a distinct pattern of changes in genes related to abscisic acid (ABA) biosynthesis enzymes. Compared with Cabernet Sauvignon, Shiraz showed a higher number of significant correlations between metabolites, which together with the relatively higher expression of flavonoid genes supports the evidence of increased accumulation of coumaroyl anthocyanins in that cultivar. Enhanced stress-related metabolism, trehalose, stilbene and ABA in Shiraz berry skin are consistent with its relatively higher susceptibility to environmental cues (Degu et al. 2014).

Metabolomics Applied in Grape Juice and Must

The utility and efficiency of using metabolic profiling using unsupervised and untargeted methodologies were explored in order to extract information from data generated using HS-SPME/GC-MS (Silva Ferreira et al. 2014). In this study, the use the of the proposed methodology enabled the ‘real time’ monitoring of the metabolic changes inherent to the biochemical dynamics of a perturbed complex biological system as well as the extraction of molecular candidates that are later validated on its biochemical context (Silva Ferreira et al. 2014). To evaluate the efficiency of the proposed methodology, five different microfermentations were carried out using synthetic media combined with different levels of nitrogen (Silva Ferreira et al. 2014). The fermentations were monitored online using HS-SPME/GC-MS (small volume), allowing the collection of the metabolic profiles and the identification of several molecular candidates (Silva Ferreira et al. 2014). Non-target analysis was applied using MS data in two ways using one dimension (1D), where the total ion chromatogram per sample was used, or two dimensions (2D), where the integrity time versus m/z per sample was used (Silva Ferreira et al. 2014). The published results indicated that the 2D protocol was able to capture the most relevant information more efficiently than the 1D (Silva Ferreira et al. 2014). Overall, this study allowed the identification of different metabolic pathways and precursors responsible for the production of volatile compounds associated with wine quality (Silva Ferreira et al. 2014). In addition, the authors highlighted that the results obtained using small-scale fermentation did not differ from those obtained using large-scale fermentation (Silva Ferreira et al. 2014).

Metabolomics in Wine Chemistry

Several authors highlighted that the molecules present in the wine matrix include a large number of primary (e.g. sugars, amino acids, organic acids, lipids) and secondary metabolites (phenolics, alkaloids, sterols, lignans, terpenes, fatty acids) that confer and modulate the quality and sensory properties of the wine (Rossouw and Bauer 2009a, b; Fotakis et al. 2013; Arbulu et al. 2015). While hundreds of metabolites (volatile and non-volatile) can be detectable in the wine matrix, only a fraction of them have been identified (Rossouw and Bauer 2009a, b; Fotakis et al. 2013; Arbulu et al. 2015). All of these compounds have a strong influence on the quality and sensory properties (aroma and taste) of the wine and are therefore not only important for the characterisation and differentiation of wines but also used in targeting issues related with authenticity, traceability and fraud as reviewed by Alañón et al. (2015).

Overall, the whole chemical composition of a wine reflects the history of the wine-producing process, including the grape variety, the yeast strain, the containers used for alcoholic and malolactic fermentations, storage, ageing and different oenological practices (Rossouw and Bauer 2009a, b; Fotakis et al. 2013; Arbulu et al. 2015; Alañón et al. 2015).

A search strategy using the Phenol-Explorer database in order to obtain the widest range of metabolites related to wine consumption was evaluated (Boto-Ordoñez et al. 2014). In this study, a total of 97 metabolites have been described in intervention studies with wine and related products (n = 37) and after consumption of pure compounds known to be wine constituents (n = 90) (Boto-Ordoñez et al. 2014). These 97 metabolites, derived from host and microbial metabolism of several classes of polyphenols, were found in plasma and urine samples, and some of them have demonstrated to have high or low biological activity compared to the parent compounds analysed in in vitro studies (Boto-Ordoñez et al. 2014). The metabolites have been used to develop a global pathway map of polyphenol metabolism in wine. The described metabolic pathway map could aid in the identification of possible biomarkers of wine and improve the current understanding of the health effects of wine consumption (Boto-Ordoñez et al. 2014).

A rapid, sensitive and selective analysis method using ultra-high-performance liquid chromatography coupled to triple-quadrupole MS (UHPLC-QqQ-MS) has been developed for the quantification of polyphenols in rose wines (Lambert et al. 2015). The detection of different polyphenol compounds was based on specific MS transitions in Multiple Reaction Monitoring (MRM) mode, where the present method allowed the selective quantification of up to 152 phenolic and two additional non-phenolic wine compounds in a short period of time (30 min) without sample purification or pre-concentration, even at low concentrations (Lambert et al. 2015). This method was repeatable and was used to analyse a set of 12 rose wines where this methodology was considered to be suitable for high-throughput and large-scale metabolomics studies (Lambert et al. 2015).

A comprehensive untargeted metabolomic fingerprinting method for the non-volatile profile of the Graciano (Vitis vinifera) wine variety, using LC-electrospray ionisation time of flight MS (LC-ESI-QTOF), was proposed by Arbulu et al. (2015). In this study, pre-treatment of samples, chromatographic columns, mobile phases, elution gradients and ionisation sources were evaluated for the extraction of the maximum number of metabolites in the wine samples analysed (Arbulu et al. 2015). Putative compounds were extracted from the raw data using the extraction algorithm, molecular feature extractor (MFE) (Arbulu et al. 2015). For the metabolite identification, the WinMet database was designed based on electronic databases and literature research and includes only the putative metabolites reported to be present in oenological matrices (Arbulu et al. 2015). The results obtained using WinMet were compared with those in the METLIN database in order to evaluate how much these databases overlap in performing identifications (Arbulu et al. 2015). The reproducibility of the analysis was assessed using manual processing following replicate injections of wine spiked with external standards where more than 400 different metabolites in samples of Graciano red wine were identified using this approach (Arbulu et al. 2015). Wine metabolites reported included compounds such as sugars (4 %), amino acids (23 %), biogenic amines (4 %), fatty acids (2 %) and organic acids (32 %), and secondary metabolites such as phenols (27 %) and esters (8 %). Significant differences between varieties Tempranillo and Graciano were related to the presence of 15 specific compounds (Arbulu et al. 2015).

NMR spectroscopy was used to investigate the metabolic differences in wines produced from different grape varieties and different regions of Australia, France, Korea and USA (Son et al. 2008). A significant separation among wines from Campbell Early, Cabernet Sauvignon and Shiraz grapes was observed using PCA and PLS-DA techniques (Son et al. 2008). The metabolites contributing to the separation were assigned to 2,3-butanediol, lactate, acetate, proline, succinate, malate, glycerol, tartarate, glucose and phenolic compounds after analysing the PCA and PLS-DA loading plots (Son et al. 2008). Wines produced from Cabernet Sauvignon grapes harvested in the continental areas of Australia, France and California were also separated using this methodology (Son et al. 2008). The interpretation of the PLS-DA loading plots revealed that the level of proline in Californian Cabernet Sauvignon wines was higher than wines from Australian and French Cabernet Sauvignon, Australian Shiraz and Korean Campbell Early wines, showing that the chemical composition of the grape berries varies with the variety and growing area (Son et al. 2008). This study highlighted the applicability of NMR-based metabolomics with MVA in determining wine quality and product origin (Son et al. 2008).

NMR spectroscopy has been also used for metabolomic analysis of Riesling and Mueller-Thurgau white wines from the German Palatinate region (Ali et al. 2011). Two-dimensional NMR spectroscopy has been applied for the identification of metabolites, including phenolic compounds. It is shown that sensory analysis correlates with NMR-based metabolic profiles of wine (Ali et al. 2011). The NMR data in combination with PCA, PLS and bidirectional orthogonal projections to latent structures (OPLS) analysis were employed in an attempt to identify the metabolites responsible for the taste of wine, using a non-targeted approach (Ali et al. 2011). The high-quality wines were characterised by elevated levels of compounds like proline, 2,3-butanediol, malate, quercetin and catechin (Ali et al. 2011). Characterisation of wine based on type and vintage was also achieved using OPLS analysis (Ali et al. 2011). Riesling wines were characterised by higher levels of catechin, caftarate, valine, proline, malate and citrate whereas compounds like quercetin, resveratrol, gallate, leucine, threonine, succinate and lactate were found discriminating for ‘Mueller-Thurgau’ (Ali et al. 2011). The wines from 2006 vintage were dominated by leucine, phenylalanine, citrate, malate and phenolics, while valine, proline, alanine and succinate were predominantly present in the 2007 vintage (Ali et al. 2011). It was concluded that the NMR spectroscopy-based metabolomics offers an easy and comprehensive analysis of wine and, in combination with MVA, can be used to investigate the source of the wines and to predict certain sensory aspects of wine (Ali et al. 2011).

The contribution of volatile aroma compounds to the overall composition and sensory perception of wine is well recognised (Schmidtke et al. 2013). The classical targeted measurement of volatile compounds in wine using GC-MS is laborious, and only a limited number of compounds can be quantified at any time (Schmidtke et al. 2013). Application of an automated multivariate curve resolution (MCR) technique to non-targeted GC-MS analysis of wine makes possible to detect several hundred compounds within a single analytical run (Schmidtke et al. 2013). In this study, a metabolomic approach to wine analysis, using MCR combined with GC-MS profiles coupled with full descriptive sensory analysis, was used to determine the objective composition of various styles of Semillon wines sourced from the Hunter valley (Australia) (Schmidtke et al. 2013). Over 250 GC-MS peaks were extracted from the wine profiles where sensory scores were analysed using PARAFAC prior to development of predictive models using PLS regression (Schmidtke et al. 2013). Good predictive models of the sensorial attributes honey, toast, orange marmalade and sweetness were determined from the extracted peak tables (Schmidtke et al. 2013). Compound identification for these rated attributes indicated the importance of a range of ethyl esters, aliphatic alcohols and acids, ketones, aldehydes, furanic derivatives and norisoprenoids in the development of the different styles of Semillon from the Hunter valley (Schmidtke et al. 2013). The development of automated metabolomic data analysis of GC-MS profiles of wines will assist in the development of wine styles for specific consumer segments and enhance understanding of production processes on the ultimate sensory profiles of the product (Schmidtke et al. 2013).

Wine micro-oxygenation is a well-known winemaking technique. The effects of micro-oxygenation in wine samples were studied using untargeted LC-MS fingerprint (Arapitsas et al. 2012). Different variables were tested such as addition of oxygen (four levels) and iron (two levels) to a Sangiovese wine, before and after malolactic fermentation (Arapitsas et al. 2012). Data analysis using supervised and unsupervised MVA highlighted some known candidate biomarkers, together with a number of metabolites which had never previously been considered as possible biomarkers for wine micro-oxygenation (Arapitsas et al. 2012). Various pigments and tannins were identified among the known candidate biomarkers. Additional new information was obtained suggesting a correlation between oxygen doses and metal contents and changes in the concentration of primary metabolites such as arginine, proline, tryptophan and raffinose, and secondary metabolites such as succinic acid and xanthine (Arapitsas et al. 2012). Based on these findings, new hypotheses regarding the formation and reactivity of wine pigment during micro-oxygenation have been proposed (Arapitsas et al. 2012). This experiment also highlighted the feasibility of using unbiased, untargeted metabolomic fingerprinting to improve the overall understanding of wine chemistry during micro-oxygenation (Arapitsas et al. 2012).

Network reconstruction (NR) has proven to be useful in the detection and visualisation of relationships among the compounds present in a Port wine ageing data set. The use of this methodology provided with a considerable amount of information that allowed the researchers to better understand the kinetic context of the molecules represented by peaks in each chromatogram (Castro et al. 2014; Monforte et al. 2015). The use of NR together with the determination of kinetic parameters allowed to extract more information about the mechanisms involved in Port wine ageing. The volatile compounds present in samples of Port wines spanning 128 years in age were measured with the use of GC-MS. After chromatogram alignment, a peak matrix was created, and all peak vectors were compared to one another to determine their Pearson correlations over time (Castro et al. 2014; Monforte et al. 2015). A correlation network was created and filtered on the basis of the resulting correlation values. Some nodes in the network were further studied in experiments on Port wines stored under different conditions of oxygen and temperature in order to determine their kinetic parameters. The resulting network can be divided into three main branches (Castro et al. 2014; Monforte et al. 2015). The first branch is related to compounds that do not directly correlate to age, the second branch contains compounds affected by temperature and the third branch contains compounds associated with oxygen (Castro et al. 2014; Monforte et al. 2015). Compounds clustered in the same branch of the network have similar expression patterns over time as well as the same kinetic order and thus are likely to be dependent on the same technological parameters (Castro et al. 2014; Monforte et al. 2015). Network construction and visualisation provides more information with which to understand the probable kinetic contexts of the molecules represented by peaks in each chromatogram. This approach was able to demonstrate that metabolomics might be a powerful tool for the study of mechanisms and kinetics in complex systems and should aid in the understanding and monitoring of wine quality (Castro et al. 2014; Monforte et al. 2015).

As defined in the previous section, metabolomics aims at gathering the maximum amount of metabolic information for a total interpretation of biological systems (Arapitsas et al. 2014). A process analytical technology pipeline, combining GC data pre-processing with MVA, was applied to a Port wine ‘forced ageing’ process under different oxygen saturation regimes at 60 °C (Arapitsas et al. 2014). It was found that extreme ‘forced ageing’ conditions promote the occurrence of undesirable chemical reactions by production of dioxane and dioxolane isomers, furfural and 5-hydroxymethylfurfural, which affect the quality of the final product through the degradation of the wine aromatic profile, colour and taste. High kinetical correlations between these key metabolites with benzaldehyde, sotolon and many other metabolites that contribute to the final aromatic profile of the Port wine were also reported (Arapitsas et al. 2014). The use of the kinetical correlations in time-dependent processes as wine ageing can further contribute to biological or chemical systems monitoring, discovery of new biomarkers and metabolic network investigations. Storage conditions and duration have a considerable influence on wine quality (Arapitsas et al. 2014). Optimum temperature and humidity conditions may improve wine quality through ageing, while incorrect or excessively long storage leads to negative results. In order to evaluate the global effects of storage on red wine composition, 20 Sangiovese wines were stored in two different conditions (cellar or house) for a period of 2 years and analysed every 6 months (Arapitsas et al. 2014). Untargeted LC-MS analysis showed various putative markers for the type and length of conservation (e.g. pigments, flavanols, pantothenic acid, among others), while targeted LC-MS confirmed and expanded these results within specific metabolic groups. The results of MVA showed that wines stored in the cellar changed little even after 2 years of storage, while wines stored in typical domestic conditions (house) developed approximately four times faster, reaching a composition similar to wines stored in the cellar for 2 years after 6 months (Arapitsas et al. 2014). The formation of several monosulfonated flavanols during domestic ageing provided the first evidence of a reaction between wine tannins—both catechins and proanthocyanidins—and the exogenous antioxidant bisulphite in wine. Moreover, ageing in domestic conditions appeared to induce an accelerated decrease in wine pigments, while specifically promoting the formation (Arapitsas et al. 2014).

The pathway for the biogenesis of varietal thiols such as 3-mercaptohexanol (3MH), 3-mercaptohexyl acetate (3MHA) and 4-mercapto-4-methylpentan-2-one (4MMP) in Sauvignon blanc (SB) wines is still an open question (Pinu et al. 2014). Varietal thiol development requires yeast activity, but poor correlation has been found between thiols and their putative respective precursors. Metabolomics was used to unravel metabolites in the grape juice that affect the production of varietal thiols in wines (Farhana et al. 2014). Comprehensive metabolite profiling of 63 commercially harvested SB juices was performed by combining GC-MS and NMR spectroscopy (Pinu et al. 2014). These juices were fermented under controlled laboratory conditions using a commercial yeast strain (EC1118) at 15 °C where correlation of thiol concentration in the wines with initial metabolite profiles identified 24 metabolites that showed positive correlation (R > 0.3) with both 3MH and 3MHA, while only glutamine had a positive correlation with 4MMP (Pinu et al. 2014). The results from the analysis of the juice samples confirmed metabolomics hypotheses and revealed grape juice metabolites that significantly impact on the development of three major varietal thiols and other aroma compounds of SB wines (Pinu et al. 2014).

NMR spectroscopy was used to examine the molecular profile of a white wine Fiano di Avellino obtained through fermentation by either a commercial or a selected autochthonous Saccharomyces cerevisiae yeast starter (Mazzei et al. 2013). The latter was isolated from the same grape variety used in the wine-making process in order to strengthen the relationship between wine molecular quality and its geographical origin (Mazzei et al. 2013). The NMR spectral data combined with MVA showed that the two different yeasts led to significant diversity in the wine metabolomes. These results indicated that metabolomic fingerprints from NMR spectroscopy combined with MVA enables wine differentiation as a function of yeast species and other wine-making factors, thereby contributing to objectively relate wine quality to the terroir (Mazzei et al. 2013).

Sensory analysis of wine is expensive and time consuming, and alternatives to characterise aspects of wine flavour and aroma are highly desirable (Rochfort et al. 2010). This study demonstrated that certain mouth-feel parameters identified from sensory analysis can be strongly correlated to NMR spectroscopy-based metabolomics analysis of wine (Rochfort et al. 2010). Wines were made from Cabernet Sauvignon and Shiraz grapes (V. vinifera L.) subjected to different levels of sun exposure in a commercial vineyard in the Sunraysia region of Victoria, Australia (Rochfort et al. 2010). Descriptive analysis revealed that the wines from the shaded treatment were significantly different from other treatments for a number of flavour and aroma characters, particularly those related to mouth-feel. Metabolomic analysis allowed classification of the wines based on grape variety and shade treatment in a manner similar to that by sensory analysis (Rochfort et al. 2010). The NMR analysis described in this study was considered to be rapid and inexpensive, allowing the simultaneous assessment of multiple metabolites that contribute to wine quality. Metabolomic analysis of wine may therefore offer a more affordable technique to investigate certain sensory aspects of wine (Rochfort et al. 2010).

Metabolomics in Yeast Metabolism Related with Wine Aroma and Flavour

System-wide ‘omics’ approaches have been widely applied to study a limited number of laboratory strains of S. cerevisiae (Rossouw and Bauer 2009a, b). Recently, industrial S. cerevisiae strains have become the target of such analyses, mainly to improve our understanding of biotechnologically relevant phenotypes that cannot be adequately studied in laboratory strains (Rossouw and Bauer 2009a, b). Most of these studies have investigated single strains in a single medium. However, this experimental layout cannot differentiate between generally relevant molecular responses and strain- or media-specific features (Rossouw and Bauer 2009a, b). In this study, the authors analyse the transcriptomes of two phenotypically diverging wine yeast strains in two different fermentation media at three stages of wine fermentation (Rossouw and Bauer 2009a, b). The data generated showed that the intersection of transcriptome datasets from fermentations using either synthetic MS300 (simulated wine must) or real grape must (Colombard) can help to delineate relevant from ‘noisy’ changes in gene expression in response to experimental factors such as fermentation stage and strain identity (Rossouw and Bauer 2009a, b). The obtained differences in the expression profiles of strains in the different environments also provided with relevant insights into the transcriptional responses toward specific compositional features of the media (Rossouw and Bauer 2009a, b). The data also suggested that MS300 is a representative environment for conducting research on wine fermentation and industrially relevant properties of wine yeast strains (Rossouw and Bauer 2009a, b).

Dekkera bruxellensis is a yeast known for its ability to produce ethyl phenols from hydroxycinnamic acid in wine, affecting the quality of its flavour (Conterno et al. 2013). In wine, D. bruxellensis is not responsible for producing ethanol; however, it is able to survive and sometimes also to grow in the presence of large amounts of ethanol (Conterno et al. 2013). Because of its endurance, D. bruxellensis poses a serious threat to the wine industry and can cause substantial financial losses. In order to analyse yeast activity in the presence of different amounts of ethanol, the metabolic profile of a D. bruxellensis strain isolated from wine was outlined in defined chemical conditions in model wines (Conterno et al. 2013). The metabolic profile of model wines with 10, 11, and 12 % ethanol after D. bruxellensis growth was studied. Several ethyl esters and phenyl ethanol, together with 4-ethyl guaiacol, were produced in significantly higher amounts in response to the increase in ethanol stress (Conterno et al. 2013). It was shown how the cell metabolism of specific compounds increased in response to higher ethanol content, although yeast cell growth was limited (Conterno et al. 2013).

Temperature is one of the most important parameters affecting the length and rate of alcoholic fermentation and final wine quality (Lopez-Malo et al. 2013). Wine produced at low temperature is often considered to have improved sensory qualities. However, there are certain drawbacks to low temperature fermentations such as reduced growth rate, long lag phase and sluggish or stuck fermentations (Lopez-Malo et al. 2013). To investigate the effects of temperature on commercial wine yeast, a range of temperatures between 12 and 28 °C in a synthetic must were evaluated (Lopez-Malo et al. 2013). Some species of the Saccharomyces genus have shown better adaptation at low temperature than S. cerevisiae. This is the case of the cryotolerant yeasts Saccharomyces bayanus var. uvarum and Saccharomyces kudriavzevii (Lopez-Malo et al. 2013). In an attempt to detect inter-specific metabolic differences, the metabolome of these species growing at 12 °C, which we compared with the metabolome of S. cerevisiae (not well adapted at low temperature) at the same temperature (Lopez-Malo et al. 2013). These results showed that the main differences between the metabolic profiling of S. cerevisiae growing at 12 and 28 °C were observed in lipid metabolism and redox homeostasis (Lopez-Malo et al. 2013). Moreover, the global metabolic comparison among the three species revealed that the main differences between the two cryotolerant species and S. cerevisiae were in carbohydrate metabolism, mainly fructose metabolism. However, these two species have developed different strategies for cold resistance. S. bayanus var. uvarum presented elevated shikimate pathway activity, while S. kudriavzevii displayed increased NAD(+) synthesis (Lopez-Malo et al. 2013).

Summary and Future Prospects

During the past 10 years, advances in instrumentation (hardware) and multivariate data manipulation techniques made possible the development of different metabolomic applications in grape and wine. Both target and untarget approaches have been reported in the literature in order to better characterise different metabolites in either grapes or wines.

Most of these studies have highlighted the importance of metabolomics in wine science, as well as emphasised on the need of a multidisciplinary team approach where the participation of scientists from different disciplines such as biology, biochemistry, chemistry and chemometrics (mathematics and statistics) being equally important to deliver successful and reliable data, in order to improve our knowledge about wine. The combination of different techniques provides both the research and industry with powerful and complementary tools that differ from the conventional routine methods currently in use.

However, some roadblocks still exist on the use and development of the metabolomic approach in grape and wine studies. Examples of this can be related with the lack of integration between chemistry and maths (chemometrics is often ignored), and no formal (academic) education in the holistic approach to analyse complex systems (the reductionist approach is always favoured). In addition, a proper metabolomics study needs to define very well the experimental conditions (design of experiments), optimisation and validation of the models. Other issues related with sampling, sample and data pre-processing still need to be optimised and are highly dependent on the application.

Without doubt one of the biggest challenges that face the use of metabolomics approaches in grape and wine is the ability to interpret the information contained in the instrumental methods and the mathematical models generated through multivariate analysis in order to develop an application.