Abstract
Chardonnay is a neutral grape variety offering a diversity of wine styles that are popular among consumers. The links between wine production methods and Chardonnay wine volatile composition, as determinants of quality, require further elucidation. Over 80 commercial Australian Chardonnay wines were assessed by expert panelists who were asked to define four distinct levels of quality in a blind tasting. Wine aroma volatiles in each wine were analyzed by solid-phase microextraction-gas chromatography-mass spectrometry, and multivariate statistical techniques were used to examine the relationship between volatile composition and quality as defined by the experts. Of 39 aroma compounds quantified, nine volatiles (including cis- and trans-oak lactones, furfural, and diethyl succinate) correlated significantly and positively with Chardonnay wine quality, while 11 volatiles (including fruity esters and monoterpenoids) correlated negatively. Compounds associated with oak contact and malolactic fermentation were present at highest concentrations in higher-quality wines as perceived by wine experts. Lower scores were assigned to younger but less complex wines, which were richer in fruity esters and other grape-derived compounds. A model was developed using partial least squares regression based on these results, which permitted classification of the Chardonnay wines into high-, medium-, and low-quality brackets depending on their relative concentrations of cis- and trans-oak lactone, ethyl lactate, and 2-methyl-1-propanol (positive) and of 1-propanol and 1-hexanol (negative). There was a significant and positive correlation (r = 0.469, p <0.0001) between retail price and quality score, underlying the usefulness of price as an indicator of quality, although it failed to entirely explain quality (as judged by experts) and should therefore be used in conjunction with other quality cues.
For most goods, value is determined by quality, often measured as nutritional value for food or level of craftsmanship for material objects such as clothes or furniture. For non-commodity hedonic goods such as wine, quality becomes more difficult to define because consumption is not related to nutrition and the steps used in the winemaking process (i.e., craftsmanship) cannot be truly or easily appreciated by the consumer (Schiefer and Fischer 2008). Quality is officially defined by the International Standardization Organization (2008) as “the ability of a set of inherent characteristics of a product, system, or process to fulfil requirements of customers and other interested parties”. However, it is often unclear what these requirements are, particularly for products such as wine, where quality cannot be determined solely by chemical analysis, but instead depends on a range of organoleptic properties (e.g., color, taste, aroma) and the amount of pleasure it affords the consumer (Charters and Pettigrew 2007). Therefore, the quality of a wine cannot be assessed without having tasted the product first, and more often than not, consumers are unable to taste a wine before buying it. Consumers have to rely on a series of quality cues such as brand, price, medals, advertising, packaging, reputation, and the advice and/or judgment of experts, to make a decision (Charters and Pettigrew 2007, Gawel and Godden 2008, Schiefer and Fischer 2008, Lockshin et al. 2009, Sáenz-Navajas et al. 2016).
Sensory judges can be divided into consumers, trained assessors, and experts, according to their level of exposure to the product and sensory training. Although not necessarily formally trained as sensory panelists, wine experts are individuals who, through repeated contact with wine, have honed the ability to focus on individual attributes, identify wine defects, and recognize volatile compounds (Gawel and Godden 2008). In many cases, their experience allows them to recognize wine variety, region, and style, to judge how well a sample complies with these categories, and to produce repeatable and consistent judgments on wine quality (Gawel and Godden 2008).
Expert tasters score the quality of a wine based on the absence of faults and the presence of desirable aromas, among other attributes such as color, texture, taste, balance, and complexity. All aromas perceived in wine depend on the concentrations of multiple volatile compounds which are interpreted and labeled by the brain after olfaction, if and when they occur at concentrations above their odor detection threshold (Rapp and Mandery 1986). A series of studies have confirmed that wine quality depends on physicochemical characteristics such as aroma composition, and these studies have tried to establish correlations with specific volatile components (San Juan et al. 2012, Hopfer et al. 2015). Taking this a step further and determining which compounds correlate to a specific quality level can help develop objective and rapid ways to screen for wine quality. Additionally, such a correlation could enable quality monitoring during production so winemakers could adjust procedures to improve or better target a specific quality level.
Previous research has either concentrated on relating the impact of specific procedures to overall sensory quality or particular marker compounds, or on recognizing which volatiles define the typicity of a variety. For instance, red wines with higher quality levels present higher concentrations of aroma compounds with “pleasant notes” such as ethyl esters, C13-norisoprenoids, and oak-derived components, and lower concentrations of detrimental odorants such as 4-ethylphenol, phenylacetaldehyde, and methional (San Juan et al. 2012). The same remains to be done for white wines such as Chardonnay.
Chardonnay is one of the most widely planted varietals in the world and is grown in most winemaking regions (Gambetta et al. 2014). It is a very flexible variety, with fruit-driven characteristics that lends itself to a number of winemaking techniques, such as barrel fermentation and aging in oak, without these winemaking attributes necessarily becoming the dominant feature. Over 240 different volatile components have been identified in Chardonnay wines, including an assortment of esters, alcohols, acids, lactones, and ketones arising from fermentation or oak storage. Among these compounds, not all impact the overall aroma of the wine; some volatiles have no associated aroma or are present at infrathreshold levels (Welke et al. 2014).
Of particular importance to Chardonnay wines are esters (both straight-chain fatty acid ethyl esters and branched-chain acetate esters), C13-norisoprenoids (β-damascenone, 1,1,6-trimethyl-1,2-dihydronaphthalene (TDN)), and oak volatiles (oak lactones, guaiacol), which, depending on concentration and precise composition, will impact the level of typicity and therefore quality, of the wine (Gambetta et al. 2014). C13-Norisoprenoids are formed during berry ripening from carotenoids in the grapes. Esters are synthesized in wine during vinification through yeast metabolism and oak volatiles come from contact with toasted oak wood during fermentation and/or maturation. Together with duration of oak contact, variables such as grape sunlight exposure, irrigation, yeast strain, vinification technique, and aging affect the concentrations of these important compounds (Gambetta et al. 2014).
Given the importance of aroma to wine quality and the usefulness of expert opinions to determine quality, this study aimed to improve understanding of the link between compositional differences in aroma volatiles measured by solid-phase microextraction-gas chromatography-mass spectrometry (SPME-GC-MS) and quality as determined by an industry expert tasting panel. Samples consisted of commercial Australian Chardonnay wines encompassing a wide range of quality and price. Chromatographic data were aligned and integrated using multivariate curve resolution techniques, and relationships between quality and volatile composition were investigated using chemometrics and network analysis.
Materials and Methods
Samples
Eighty-three commercial Chardonnay wines (three bottles each) from vintages spanning 2010 to 2013 were donated by producers from New South Wales, South Australia, Tasmania, Victoria, and Western Australia. Samples were chosen for having a rating >90 points by James Halliday (Halliday 2013) or for having a sales ranking in the top 10% at one of Australia’s main wine retail chains (G. Hindson, personal communication, 2013). Details of the wine samples, tasting scores, and basic chemical data are reported (Supplemental Table 1). Wineries kindly provided additional proprietary information regarding winemaking and maturation techniques. Wines were stored at 15°C for about two months prior to use.
Reagents, standards, and materials
Reference compounds (purity ≥97%) consisting of ethyl butanoate, ethyl 2-methylbutanoate, ethyl acetate, ethyl hexanoate, ethyl octanoate, ethyl decanoate, ethyl dodecanoate, ethyl furoate, ethyl 2-phenylacetate, diethyl succinate, hexyl acetate, 3-methylbutyl acetate, 2-phenylethyl acetate, 3-methylbutyl octanoate, ethyl lactate, 1-propanol, 2-methyl-1-propanol, 3-methyl-1-butanol, 2-ethyl-1-hexanol, 1-hexanol, 1-octanol, 2-phenylethanol, linalool, α-terpineol, limonene, β-damascenone, hexanoic acid, octanoic acid, decanoic acid, dodecanoic acid, benzaldehyde, nonanal, oak lactone (mixture of isomers), and furfural were purchased from either Sigma-Aldrich or Alfa Aesar. Deuterated internal standards of d4-3-methyl-1-butanol, d3-hexyl acetate, d13-1-hexanol, d5-2-phenylethanol, and d19-decanoic acid were supplied by CDN Isotopes, and d5-ethyl nonanoate was synthesized as described previously (Boss et al. 2015). Absolute ethanol (Merck) and sodium chloride (JT Baker) were analytical grade and water was obtained from a Milli-Q purification system (Millipore).
Wine sensory assessment
Sensory evaluation of samples was conducted during a single day in November 2013 by eight industry professionals (winemakers, professors, and retailers with experience in white wines) who met the criteria defining them as wine experts (Parr et al. 2004). Wine samples (30 mL) were served at room temperature (~20°C) in clear INAO (ISO standard) 215 mL glasses covered with a transparent plastic lid. All wines were coded (three-digit code, Supplemental Table 1) and presented in a randomized order. To cope with fatigue, quality assessment was broken down into three sessions with 1-hr enforced breaks. Within a session, samples were presented in flights of five wines with 10-min breaks between flights. Panelists were provided with water and plain crackers to cleanse their palate, evaluation sheets, and a bucket in which to expectorate the samples. Each panelist was seated at a separate table. Samples were scored using the Australian Wine Show system on a 20-point scale (Iland et al. 2009) and a four-level quality score (A to D) was determined and agreed upon by all members of the panel prior to tasting. Before commencing assessments of samples, the panelists discussed and defined the criteria corresponding to each quality category (Table 1) and tasted and sorted four wines considered to be representative of each bracket (this was an iterative process; wines were provided according to their price and characteristics by the panel leader, where price was used as a proxy quality indicator). This was done to calibrate the panel members with each other and ascertain that they had reached a consensus on the different quality categories. Experts were asked to declass and not score any wine presenting a serious fault (e.g., oxidation or cork taint).
Basic wine composition
The pH and titrable acidity (TA, reported as g/L tartaric acid equivalents) of the wines were measured using a combined pH meter and autotitrator (Crison, CompacTitrator, Crison Instruments), and percent alcohol was determined using an alcoholizer (Alcolyzer Wine ME+DMA 4500M, Anton Paar). All measurements were conducted in duplicate within six months of the sensory evaluation.
Volatile analysis by HS-SPME-GC-MS
Wines were analyzed immediately after sensory analysis from the same bottles as described (Wang et al. 2016a) with modification. Wine (0.5 mL) was aliquoted into a 20 mL SPME vial (Supelco), diluted with water (4.5 mL), and 2 g sodium chloride was added. The vial was spiked with an internal standard solution (10 μL) consisting of a mixture of deuterated standards in absolute ethanol (d4-3-methyl-1-butanol [2380 mg/L], d3-hexyl acetate [25 mg/L], d13-1-hexanol [50 mg/L], d5-2-phenylethanol [500 mg/L], d19-decanoic acid [50 mg/L], and d5-ethyl nonanoate [1.2 mg/L]), sealed tightly with a PTFE-lined cap (Supelco), and the contents were homogenized with a vortex mixer.
Multivariate curve resolution analysis of GC-MS data
Data processing and treatment was performed using MATLAB (version R2012a 7.14.0.739, The Mathworks) after exporting the GC-MS files in netCDF format from Agilent Chemstation (E.02.02.1431). Extracted ion chromatograms of all samples were overlaid, aligned, and integrated using an approach modified from one previously described (Schmidtke et al. 2013). Elution time windows for each analyte of interest, including the internal standards, were chosen by visual inspection of extracted ion chromatograms. Chromatograms were aligned using the icoshift algorithm (Savorani et al. 2010), and peak areas were extracted from the aligned elution profiles for all samples using a trapezoid integration.
Statistical analysis
Data was processed using XLSTAT ver. 2014.05.03 (Addinsoft), Gephi ver. 0.9.1 (Bastian et al. 2009), and The Unscrambler X (CAMO AS, version 10.3). Comparison and correlation of scores, prices, vintage, and analyte concentrations were executed by one-way ANOVA and Pearson’s correlation analysis. ANOVA of compositional variables was accomplished using vintage, quality category, and fermentation vessel as explanatory variables for the differences among samples. Principal component analysis (PCA) was performed on the normalized concentrations of significantly different analytes (using quality category as the explanatory variable) using variables with scores ≥0.7 in the Pearson correlation matrix. Network analysis was carried out on significantly different variables (using score as the explanatory variable) with strong positive (r ≥ 0.6) or negative (r ≤ −0.6) correlations among each other, as described (Wang et al. 2016b). Score (y-variable) was related to all wine compositional data (x-variables) using partial least squares regression (PLS) analysis. The overall set of samples was randomly split into calibration (~2/3 of the samples) and validation (~1/3) sets using the Kennard-Stone algorithm. The prediction ability of the model was tested on the validation set using the root-mean-square error of prediction (RMSEP), the residual predictive deviation (RPD), the correlation coefficient (CC), the slope of the regression curve for the predicted y-variable (m), and the percentage of variance explained by the model (%EV). All variables were normalized before analysis.
Results and Discussion
Sensory analysis by expert tasters
Consensus among expert panelists resulted in descriptions of four different wine quality categories prior to a blind tasting of 83 commercial Chardonnay wines (Table 1). Wines included in category A had a score of 18 points or more, those in category B had 16 to 17.9 points, category C had 14 to 15.9 points, and category D had below 14 points.
Scores for all samples ranged between 14.2 and 18.1 points: only one sample was categorized as A. ANOVA of scores resulted in significant differences (p < 0.0001) among vintages. The highest scores were given to 2010 wines and the lowest to those produced in 2013 (Supplemental Table 1). Given the description determined by the experts for each category (Table 1), this outcome was somewhat expected, as higher scores were given to wines that had been aged and thereby contained more “evolved” aromas. Significant differences (p < 0.01) were observed between scores of wines that underwent barrel fermentation in oak wood and those fermented in stainless steel tanks (Supplemental Table 1), which accords with expert panelists considering that wines fermented in the presence of oak were of higher quality. This agrees with the higher liking reported by panelists for Chardonnay wines fermented and/or aged in the presence of oak over those fermented in stainless steel (Liberatore et al. 2010). The incorporation of oak in the form of barrels, staves, chips, or other alternatives (Gambetta et al. 2014) during alcoholic fermentation not only imparted “smoky” and “woody” characters that were rated favorably by our expert panel, but also decreased the impact of “unripe fruit” and “fresh fruit” aromas and flavors that are typical of young white wines (Pérez-Coello et al. 2000). All younger wines assessed in our study (2013 vintage) corresponded to a fruitier, fresher, usually unoaked style. The panelists in this study classified these wines in category C because they lacked the higher complexity, balance, texture, and aroma profile sought for higher quality categories (Supplemental Table 1).
Volatile composition, typicity, and quality score
Samples were chosen to represent the current available offering of Chardonnay wines in the Australian market (Supplemental Table 1). They originated in the main Australian wine-producing regions from grapes grown in a variety of climates, and encompassed varied winemaking styles (no oak, barrel-fermented, uninoculated fermentation, etc.), were from different vintages, and were sold at a variety of prices. Volatiles analysis was undertaken using SPME-GC-MS and data handling was greatly simplified by employing multivariate curve resolution techniques (see Materials and Methods for details). The relative concentrations of 39 aroma volatiles determined for the 83 Chardonnay wines revealed considerable variability for some compounds (Table 2). Seventeen analytes were significantly different among the different quality categories (p < 0.05): ethyl butanoate; 3-methylbutyl acetate; hexyl acetate; ethyl hexanoate; diethyl succinate; ethyl lactate; 1-propanol; 2-methyl-1-propanol; 2-phenylethanol; linalool; α-terpineol; hexanoic, octanoic, and dodecanoic acids; furfural; and cis- and trans-oak lactones. These aroma volatiles are formed during berry ripening, alcoholic fermentation, and aging (Gambetta et al. 2014), and their variation illustrates the large intravarietal differences among Chardonnay wines resulting from geographic origin, vintage, and viticultural and enological practices. Of these compounds, ethyl butanoate, ethyl hexanoate, 3-methylbutyl acetate, linalool, diethyl succinate, and octanoic acid have correlated positively with the typicity of Chardonnay wines (Gambetta et al. 2014). Conversely, the majority of these compounds, together with several other esters and β-damascenone, correlated negatively with the overall quality score assigned by our experts (Figure 1). Most studies on the typicity of Chardonnay wine have only been conducted on unoaked samples, but oak-derived characteristics were favored by the current study’s tasting panel, which could account for our findings.
PCA analysis was conducted with significantly different variables using quality score as the explanatory variable and production details, vintage, quality score, and price as supplementary variables (Figure 1). Together, PC1 and PC2 explained 76.2% of the variability in the data and showed a segregation of samples according to quality score and fermentation vessel along the F1 axis, where higher scoring and barrel fermented (BF) samples were located to the left of the PCA plot, and lower-scoring samples to the right, were fermented in stainless steel vessels (SSF) or produced with SSF/BF/oak alternatives. Samples in the upper left quadrant had more cis- and trans-oak lactones and, as explained by the supplementary variables, these samples tended to be older (more than one year) and fermented and/or aged in oak barrels. From a sensory perspective, the oak lactones are the most important compounds released by oak into wine and their presence in wine is affected by the age, origin, and volume of the barrel (Pérez-Prieto et al. 2002). These two molecules were recognized as part of a set of 15 key odor-active compounds necessary to reconstitute the aroma of California Chardonnay wines, which are traditionally oaked (Lee and Noble 2006). Lower-scoring samples split into two groups along the F2 axis: located in the lower right quadrant were samples mainly from the 2013 vintage, which contained higher concentrations of hexyl acetate, 3-methylbutyl acetate, and β-damascenone, and in the upper right quadrant was a more heterogeneous group of samples, predominantly produced with SSF/staves or SSF/BF, characterized by higher amounts of hexanoic acid, ethyl hexanoate, and ethyl butanoate. Hexanoic acid, ethyl hexanoate, ethyl butanoate, hexyl acetate, 3-methylbutyl acetate, and β-damascenone either directly impart or enhance “fruity” and “vegetal” (or “green”) aromas (San Juan et al. 2011, Gambetta et al. 2014) that were associated with the lower quality brackets by the experts (Table 1).
As a useful visualization tool (Wang et al. 2016b), network analysis was used to reveal 17 strong positive (r ≥ 0.6) and two strong negative (r ≤ −0.6) correlations or “edges” arising between significantly different aroma volatiles, basic chemical parameters, and quality score (Figure 2, positive correlations only). Positive correlations resulted in three distinct modules: on the far right, a module consisting of hexanoic acid and the fruity esters ethyl butanoate and ethyl hexanoate; in the middle, another module with the fruity acetates 3-methylbutyl, 2-phenylethyl, and hexyl acetate and the varietal compounds linalool, α-terpineol, and β-damascenone; and on the far left, a module with the oak-derived volatiles cis- and trans-oak lactone and furfural, which correlated strongly with quality score and price. As expected, the fatty acid hexanoic acid correlated very strongly with its corresponding ethyl ester (r = 0.88) and to ethyl butanoate (r = 0.72), as a result of their common biosynthetic pathways (Nykänen 1986). The association between the compounds in the middle reflects their higher concentrations in the younger 2013 wines and the common biosynthetic pathways of 3-methylbutyl and hexyl acetates (r = 0.85) and linalool and α-terpineol (r = 0.64; Nykänen 1986). Negative edges were also observed (data not shown) between 2-methyl-1-propanol and ethyl hexanoate and between diethyl succinate and hexyl acetate. Both 2-methyl-1-propanol and diethyl succinate were more abundant in samples older than one year, while concentrations of ethyl hexanoate and hexyl acetate were lower. Fermentation in barrels both suppresses formation of hexyl acetate and promotes that of diethyl succinate (Liberatore et al. 2010).
Effect of wine age
Significantly different variables (p < 0.05) using year of vintage as the explanatory variable (Figure 3) revealed trends consistent with a previous study of red wines from different price points (San Juan et al. 2012), despite the difference in wine type and grape variety studied. In general, varietal compounds such as β-damascenone, linalool, and limonene declined with wine age and were most abundant in the youngest and lowest-scored wines (Table 2 and Figure 3). The average concentration of β-damascenone declined sharply in samples older than one vintage. Under low-pH conditions, β-damascenone is lost due to acid-catalyzed cyclization or nucleophilic attack, particularly in the presence of the nucleophile SO2 (Daniel et al. 2004). Likewise, the concentrations of 3-methylbutyl acetate, hexyl acetate, 2-phenylethyl acetate, ethyl butanoate, and ethyl hexanoate were greatest in the youngest wines. Such changes in the volatile profile of a wine as a function of age depend on the duration and conditions of storage (temperature, oxygen concentration, and exposure to light) (Cejudo-Bastante et al. 2011). In general, “young wine”, “fruity”, and “floral” characters decrease rapidly in white wine during aging, mostly due to loss of acetate esters and ethyl esters of short-chain fatty acids, which undergo acid hydrolysis over time (Guchu et al. 2006, Cejudo-Bastante et al. 2011), and to acid-catalyzed rearrangement of monoterpenoids (e.g., linalool, geraniol) into forms with less-intense aromas such as α-terpineol (Marais 1983). Any increases in temperature or light exposure will accelerate these reactions (Guchu et al. 2006, Cejudo-Bastante et al. 2011).
In contrast, ethyl acetate, diethyl succinate, ethyl lactate, ethyl 2-furoate, ethyl 2-phenylacetate, and ethyl 2-methylbutanoate were more abundant in the older wines (Table 2). Although difficult, a distinction should be made between the effects of bottle and barrel aging, as barrel aging encompasses a series of other phenomena such as extraction of oak volatiles, contact with lees, and processes associated with the “low oxidation” conditions of barrel storage (see section “Winemaking techniques”; Garde-Cerdán and Ancín-Azpilicueta 2006). The presence of ethyl 2-furoate is due largely to oak contact, but also to aging itself (Makhotkina and Kilmartin 2012) and formation of ethyl lactate increases during malolactic fermentation (MLF). Aging time allows esterification of acids such as succinic acid into the corresponding diethyl succinate (Ancín-Azpilicueta et al. 2009) and of branched-chain fatty acids, leading to esters such as ethyl 2-methylbutanoate and ethyl 2-furoate (Makhotkina and Kilmartin 2012). As a general rule, most “young wines” (2013 vintage) were aged in barrels for shorter periods than the older vintages, if at all (Supplemental Table 1), and were on the market in the same year as harvest. Consequently, they also spent less time in bottle, allowing less opportunity for most aging-related changes to take place compared to older wines. cis- and trans-oak lactone concentrations correlated very strongly with score (r = 0.70 and r = 0.64, respectively, p < 0.05) and price (r = 0.70 and r = 0.67, respectively, p < 0.05; Figure 2), and were significantly different (p < 0.05) between the 2013 samples and all others, again mostly because of a decrease or lack of time in contact with oak wood. On the other hand, hexanoic acid, hexyl acetate, and ethyl hexanoate, with their predominantly “green” and “apple” notes (Gambetta et al. 2014), not only correlated negatively to both price (from r = −0.31 to −0.41, p < 0.05) and score (from r = −0.39 to −0.52, p < 0.05), but were also more abundant in the 2013 samples (Table 2). These compounds have been cited as important to unoaked Chardonnay wine typicity (Smyth 2005), so it was of interest that such compounds were associated negatively with price and quality by experts.
Winemaking techniques
Oak volatiles are incorporated into wine either by fermenting and/or aging in barrels or through the presence of oak barrel alternatives such as chips or staves (Gambetta et al. 2014). Roughly 50% of all samples analyzed were completely fermented in oak barrels rather than in stainless steel tanks, and barrel fermentation was the method of choice for the most expensive wines (Supplemental Table 1). Tasting scores revealed a clear association with samples fermented in barrels (Supplemental Table 1), with a predominance of oak-related volatiles associated with the aroma of the highest-rated samples (left hand side of F1-axis, Figure 1). Independently of wine vintage, wine quality scores correlated very strongly and positively with the presence of cis- and trans-oak lactone in the wine, as mentioned above. Oak lactones are among the most important volatile compounds released into Chardonnay wine during contact with oak and contribute a “coconut” and “oaky” aroma when present at concentrations above their detection threshold (Spillman et al. 2004). The concentrations of cis- and trans-oak lactones were five and six times greater, respectively, in 2010 Chardonnay wines than in 2013 samples. However, it should be noted that only one sample was assigned to category A and this sample did not have the highest amounts of these lactones. Further inspection of the data indicated that extremely high quantities of cis- and trans-oak lactone do not contribute further to improving the quality score of a sample. Furfural, which is formed in oak during coopering (Spillman et al. 2004) and extracted during fermentation and aging in barrels, contributed positively to the overall quality score of the wines (r = 0.48, p = 0.05), and was related to the oak lactones (in the same module as score and price) through network analysis (Figure 2).
Unlike stainless steel, oak wood is porous and allows microoxygenation of wine. In addition, oak is not inert: it adsorbs as well as contributes aroma compounds, and alters the production of fermentation volatiles by yeast (González-Marco et al. 2008, Liberatore et al. 2010). For example, barrel fermentation depresses nitrogen consumption and increases production of fermentation volatiles such as 2-phenylethanol and other higher alcohols (González-Marco et al. 2008). Significantly higher average quantities of 2-methyl-1-propanol, 3-methyl-1-butanol, ethyl 2-methyl butanoate, and diethyl succinate were found in barrel-fermented samples than in those fermented in stainless steel (Table 1). Concentrations of most esters, particularly esters of higher alcohols, were appreciably lower in wines fermented and/or aged in barrels (Figure 1A and Table 1), consistent with other published findings (Ancín-Azpilicueta et al. 2009, Liberatore et al. 2010).
A higher proportion of the more expensive wines were produced, exclusively or in part, by fermenting with autochthonous (wild) yeast (Supplemental Table 1). Data provided for these commercial wines revealed a wide range of yeast choices: 49 samples were fermented exclusively with a commercial yeast strain, of which five were priced at or above AU$40, 19 wines were fermented exclusively with autochthonous yeast, of which eight were priced at or above AU$40, and the remaining 15 used a combination of both, of which three were priced at or above AU$40. According to the PCA (Figure 1), samples produced strictly with a commercial yeast strain (and mostly fermented in stainless steel with no barrel aging) scored significantly lower than the rest and could be observed mainly to the right of the F1-axis. However, this association could well be a coincidence of production costs and reflect the more expensive techniques reserved for production of higher-quality, more expensive wines.
Quality models
Quality ratings provided by the panel and the chemical composition of the Chardonnay wines were assessed using PLS (Figure 4). The model explained 67% of the variance in the volatile composition data (x-variable) and 66% of the variance in score (y-variable). RMSEP was 0.53, the CC was 0.84, and the RPD was 1.79. RPD values between 1.5 and 3.0 imply that the model can be used for classifying wines as of low, medium, or high quality (Williams 2001), which was sufficient for the purposes of this study. This low RPD value stems from the small range in quality scores (standard deviation), which ranged from 14.2 to 18.1 despite the large number of samples. Consistent with observations throughout the study, the oak-derived cis- and trans-oak lactones were the two components with the strongest positive effect on the model, followed by ethyl lactate (an MLF metabolite) and 2-methyl-1-propanol (a potential marker of barrel fermentation; Figure 4). The negative contributions of 1-hexanol and 1-propanol were expected based on the details presented above. Although not detracting from our findings, the scope of this work was limited to the effect of aroma compounds on the overall quality of Chardonnay wines. The inclusion of texture and taste attributes would more completely model the full dimensions of Chardonnay wine quality, as they impact important parameters such as mouthfeel, complexity, and balance.
Relationship between expert quality rating and price, wine critic score, and sales
Price is regarded by many consumers as an indicator of quality, which leads to an assumption that more expensive wines are better (Lockshin et al. 2009). Pearson correlation analysis of our results revealed this was partly true: there was significant positive correlation (r = 0.45, p < 0.0001) between retail price and score, with the highest average prices also belonging to the oldest wines. However, although the price of a wine partly explains its quality, other variables such as method of production, aging, vintage, etc., need to be taken into account to completely explain the score given to each wine. Likewise, higher prices do not necessarily equate to better wines: the highest-scored wine in our study set retailed for ~AU$40.00, and prices for wines in the C category ranged between AU$7.00 and AU$85.00 (Supplemental Table 1). Wine prices tend not to reflect quality and depend more on other factors such as winery reputation (25% of price variability) and marketing costs (Combris et al. 1997).
Several specialty magazines in the wine market routinely rank wines and serve as a purchase guide for consumers. Comparison of published scores (Halliday 2013) with those awarded by the expert panel exhibited only a moderate positive correlation (r = 0.44, p = 0.0095), which was expected given the multiple factors that affect judging scores such as setting, order of wine presentation, and number of panelists (Lawless and Heymann 2010). In addition, based on sales volume and ranking obtained from a major Australian liquor retail chain (G. Hindson, personal communication, 2013), consumer purchase behavior is largely explained by retail price (r = 0.73, p = 0.000), as observed elsewhere (Batt and Dean 2000, Cronley et al. 2005, Lockshin et al. 2006, Veale and Quester 2009), followed by wine guide score (r = 0.332, p = 0.0443). These correlations might vary if data from different retail sources was included, such as boutique stores that tend to be frequented by more involved consumers (Batt and Dean 2000, Lockshin et al. 2006).
Conclusion
Expert tasters evaluated 83 commercial Chardonnay wines spanning a range of prices, origins, production methods, and vintages, with the samples being representative of the Australian Chardonnay wines available at the time. Significant differences were found between the chemical compositions of all score brackets. Expert tasters scored more highly those samples fermented in oak barrels, which had higher concentrations of oak-related volatiles, particularly cis- and trans-oak lactones, and lower concentrations of esters and isoprenoids. Younger wines, which also tended to be unoaked, clustered together due to their higher concentrations of β-damascenone and esters such as hexyl acetate, ethyl hexanoate, and ethyl butanoate, consistent with the Chardonnay typicity concept. However, these samples were scored lower by the expert tasters. A model was constructed using PLS to relate expert quality ratings to the chemical composition of the wines. The cis- and trans-oak lactones, 2-methyl-1-propanol, and ethyl lactate had positive effects on quality score, whereas 1-hexanol and 1-propanol were associated with lower quality wines. However, wine quality also depends on mouthfeel and taste properties, which are determined by polysaccharide, tannin, and residual sugar concentrations, among other factors. Future studies should include these nonvolatile constituents to offer a more complete picture. Furthermore, the experts considered that only one wine could be assigned as top quality (category A). The highest-scored group of wines was category B, which were described as more oaked than the A-scored wine. This could explain the high impact of oak lactones on quality scores. As demonstrated with our sample set, wine quality is determined not only by grape composition, but also by a range of decisions made by the winemaker. The use of oak and type of fermentation vessel, yeast employed, the quality of the fruit streamed into each wine quality tier, and the attention given to the operations thereafter, all have an impact on the final quality of the product. Linking compositional differences to Chardonnay wine quality judged by experts has identified a range of targets that may help winemakers tune the quality of their wines by adjusting winemaking protocols.
Acknowledgments
J.M.G. acknowledges financial support provided by the Turner Family Scholarship from the University of Adelaide and the Australian Grape and Wine Authority ( GWR Ph1210). J.W. is supported through the joint scholarship of the UA and China Scholarship Council ( 201206300033) and is also a recipient of the Wine Australia (WA) supplementary scholarship ( GWR Ph1307). The authors are grateful to members of the Australian wine industry for wine donations, provision of information, and ongoing support. We also thank the sensory panelists, along with Albane Gosset-Grainville, Merve Darici, and Clémence Lemba for technical assistance.
Footnotes
Supplemental data is freely available with the online version of this article at www.ajevonline.org.
- Received May 2016.
- Revision received August 2016.
- Accepted August 2016.
- Published online December 1969
- ©2017 by the American Society for Enology and Viticulture