The alcohol content of wine has steadily increased in recent years [1
], in part due to the warmer temperatures experienced during the growing season, as a consequence of climate change. This has financial implications for winemakers, not only because wines with higher alcohol levels attract higher import duties and taxes, but because wine producers and consumers alike, are increasingly mindful of the health and social issues associated with excessive alcohol consumption [2
]. The wine industry is therefore adopting various strategies for correcting the alcohol content of wine, including the partial dealcoholization of wine using reverse osmosis-evaporative perstraction (RO-EP) [3
]. This process employs membrane technologies to fractionate wine and remove ethanol from the resulting permeate, before the (partially) dealcoholized permeate is recombined with the retentate, to give reduced alcohol wine (RAW). Partial dealcoholization of wine has also be achieved using pervaporation and spinning cone column distillation technologies [4
]. It is still unclear, however, how to determine optimal ethanol levels for RAW from a sensory perspective.
Some wine producers employ a process referred to as ‘alcohol sweetspotting’ to optimize wine alcohol levels [6
]. This involves partial dealcoholization of wine (by RO-EP) and subsequent blending of the base wine and RAW to generate a series of wines (typically 8–10) comprising alcohol levels that differ by small, incremental amounts (e.g., 0.2% alcohol by volume (abv)). The blends are then evaluated (typically in ascending or descending order of ethanol concentration) by three or four winemakers, who identify the wine (or wines) they considered to exhibit superior organoleptic properties (i.e., optimal aroma, flavor, taste, mouthfeel and balance). To date, however, the existence of an alcohol sweetspot phenomenon has not been scientifically validated.
The impacts of ethanol on wine sensory properties tend to support the concept of an alcohol sweetspot. Depending on its concentration, ethanol can affect the perception of taste and mouthfeel properties by enhancing or suppressing sweetness, sourness, bitterness, saltiness, astringency and hotness [7
]. Moreover, ethanol can enhance wine aroma by masking undesirable attributes and/or harmonising imbalances, but the perception of desirable aromas and flavors can also be diminished by ethanol [11
]. At lower levels. i.e., 5–12% abv, ethanol facilitates the release of volatile compounds under dynamic conditions and maintains their headspace concentrations to enhance wine aroma [13
]. Conversely, higher ethanol levels, i.e., 10–18% abv, are negatively correlated with volatile headspace concentrations, reflecting the increased solubility of volatiles in ethanol, relative to water [14
Several previous studies have attempted to validate the alcohol sweetspot phenomenon. A 2013 study investigated alcohol sweetspots in three white wines and a red wine, using an expert panel [15
]. The ethanol content of the wines evaluated via ranking and triangle tests ranged by 3% abv, spaced at 0.5% intervals, e.g., 12.1, 12.6, 13.1, 13.6, 14.1, 14.6 and 15.1% abv for a bracket of Riesling wines. Panelists ranked wines according to their individual preferences, with the most preferred wine being ranked 1 and the least preferred wine ranked 7. Each bracket was evaluated in triplicate, using three different orders of presentation: increasing ethanol content; decreasing ethanol content; and a randomized presentation order. Although there were no clear preferences for wines based on alcohol content, wines with ethanol levels above 15% abv were consistently ranked lower than other samples. It is not clear to what extent the spacing between wine ethanol levels (i.e., 0.5%) may have influenced the identification of any alcohol sweetspots.
In a subsequent study, consumers evaluated subsets of Chardonnay wine comprising samples for which the ethanol content had been adjusted to span a concentration range of approximately 1% abv (at 0.2% increments), using a wine that was partially dealcoholized by spinning cone column distillation [16
]. Consumers were randomly allocated to wine subsets and asked to rate their liking of samples using a 9-point hedonic category scale, but only one consumer group yielded results (i.e., a positive quadratic curve) that suggested the existence of an alcohol sweetspot (being 13.8 to 14.0% abv) for their subset of wines. Consistent liking scores were not obtained for other wine subsets and therefore did not enable identification of a sweetspot for that wine.
The alcohol sweetspotting process is likely affected by a combination of factors, including the order in which samples are presented, the incremental difference in alcohol levels between samples, and the composition of the sensory panel (i.e., consumers vs. winemakers/experts). Standardized sensory methodology typically requires randomized presentation of samples, whereas industry based sweetspotting trials usually present samples in sequential order (i.e., increasing or decreasing ethanol concentration [5
]). Randomization of samples may confound the perception of subtle differences between samples of similar alcohol content, whereas sequential presentation of samples may introduce bias (i.e., away from the samples of highest and/or lowest alcohol content).
This study aimed to investigate to what extent different methods of presenting samples (i.e., ordered vs. randomized, and linear vs. circular) might influence the outcome of alcohol sweetspotting trials. These approaches were intended to evaluate the ability of panelists to choose the same sample in replicate brackets, as well as to investigate whether the order in which samples were assessed influenced the sensory perceptions of wines. Different approaches to statistical analysis of data from sweetspotting trials were also evaluated, i.e., chi-square tests for ‘goodness of fit’ vs. ‘one proportion’, in an attempt to further provide scientific evidence of an alcohol sweetspot.
2. Materials and Methods
2.1. Wine Samples
Three 2015 Barossa Valley red wines, a Shiraz Cabernet Sauvignon blend (60:40) and two Shiraz wines, hereafter wines A, B and C, respectively, were sourced from a commercial winery who had deemed the wines to be in need of alcohol adjustment. The wines were partially dealcoholized using an industrial scale RO-EP unit (VA Filtration, Nuriootpa, Australia), in accordance with manufacturer operating instructions [3
]; their alcohol concentrations before and after RO-EP treatment were: 16.0 and 14.4% abv for wine and RAW A; 16.0 and 14.2% abv for wine and RAW B; and 16.3 and 14.0% abv for wine and RAW C.
Wines A and B were subsequently blended with different proportions of their corresponding RAW to generate a series of samples with ethanol concentrations that differed by 0.2% abv, for use in alcohol sweetspotting trials, according to the practices typically employed by the winery. Trial 1 comprised nine blends of wine A and RAW A, with alcohol percentage levels: 16.0 (wine A), 15.8, 15.6, 15.4, 15.2, 15.0, 14.8, 14.6, and 14.4 (RAW A); while Trial 2 comprised nine blends of wine B and RAW B, with alcohol percentage levels: 16.0 (wine B), 15.8, 15.6, 15.4, 15.2, 15.0, 14.8, 14.6, and 14.2 (RAW B). Wine C was blended with its corresponding RAW to achieve samples which differed in ethanol concentration by 0.2, 0.5 and 1.0% abv from both wine C (i.e., 16.1, 15.8 and 15.3% abv) and RAW C (i.e., 14.2, 14.5 and 15.0% abv), for use in difference tests.
Wines, RAWs and blends thereof were subsequently bottled (under screw cap) in 750 mL glass bottles and cellared (in darkness at 15 °C) prior to chemical and sensory analyses (performed approximately 1 month after RO-EP treatment, blending and bottling).
2.2. Sensory Analysis of Wines
For each of the sensory analyses performed, i.e., alcohol sweetspotting trials and difference tests: samples (30 mL) were served at ambient temperature (i.e., 22–24 °C), in covered, three-digit coded XL5 wine glasses (International Organization for Standardization, ISO 3591:1977), under LED white lighting. Alcohol sweetspotting trials were held in an open-plan sensory laboratory (with each panelist on a separate bench) at a commercial winery in the Barossa Valley wine region. Difference tests were held in a sensory laboratory at the University of Adelaide’s Waite Campus. Sensory trials were approved by the Human Research Ethics Committee of the University of Adelaide (H-2015-094).
2.2.1. Alcohol Sweetspotting Trials
Expert panels, each comprising 14 winemakers, were assembled for alcohol sweetspotting trials. Prior to evaluation, each panelist completed a short survey comprising demographic questions. For each trial, a higher proportion of panelists were male (64–79%) and aged between 31 and 50 years (72–85%), but the majority had more than 10 years industry experience, all had wine judging experience, and 79% had previously undertaken alcohol sweetspotting trials (Table S1
Panelists were presented with brackets of nine samples at a time (comprising wine, RAW and blends thereof), using four different orders of presentation: (i) samples presented in a linear format, in random order, hereafter ‘row, randomized’; (ii) samples presented in a linear format, with alcohol content decreasing from left to right, hereafter referred to as ‘row, ordered’; (iii) samples presented in a circular format (i.e., evenly spaced around a circular tray), in random order, hereafter ‘circular, randomized’; and (iv) samples presented in a circular format with alcohol content increasing in a clockwise direction, hereafter referred to as ‘circular, ordered’ (Figure 1
Panelists were instructed to evaluate samples presented in rows from left to right, whereas for samples presented in a circle, panelists were instructed to evaluate samples in a clockwise direction, starting with any sample. For each trial, panelists evaluated eight brackets of samples in total, i.e., four presentation orders in duplicate, with brackets also being presented in a randomized order. Short breaks (2 min) were enforced between brackets, with longer breaks (30–60 min) enforced after the first four brackets. In Trial 1, panelists were asked to identify one or more samples from each bracket that they considered exhibited superior overall sensory properties; but in Trial 2, the panel was asked to identify only one sample. Trials 1 and 2 were held on different days.
2.2.2. Difference Tests
A series of triangle tests [17
] were conducted with a panel of 18 assessors comprising wine science staff and students from the University of Adelaide (nine males and nine females, aged between 18 and 55 years) to determine the change in ethanol concentration that resulted in perceptible differences in sensory properties for wine C and RAW C (using blends thereof, prepared as outlined above). Samples were presented in a balanced, randomized presentation order comprising all possible configurations (i.e., XXY, XYX, YXX, YYX, YXY and XYY, where X denotes wine C or RAW C, and Y denotes blends that differed in ethanol concentration by 0.2, 0.5 and 1.0% abv), an equal number of times. Panelists were asked to smell and taste the samples presented in each bracket, and to identify the sample that was different. Sensory data were collected and processed using RedJade online software (RedJade, Redwood Shores, CA, USA).
2.3. Chemical Analysis of Wines
The alcohol content, density, pH, titratable acidity (TA, as g/L of tartaric acid equivalents, to an endpoint of pH 8.2) and volatile acidity (VA, as g/L of acetic acid equivalents) were determined (in duplicate) by the Australian Wine Research Institute’s (AWRI) Commercial Services laboratory (Adelaide, Australia), using a Foss WineScan analyzer (Mulgrave, Australia). Glucose, fructose, glycerol and organic acids were determined (in duplicate) by high-performance liquid chromatography (HPLC), as described previously [18
]. Briefly, an Agilent 1100 series HPLC (Agilent Technologies, Forest Hill, Australia), fitted with diode array and refractive index detectors was used, with separation achieved using an Aminex HPX-87H cation exchange column (Bio-Rad Laboratories, Gladesville, NSW, Australia) and 2.5 mM sulfuric acid as the mobile phase. Calibration curves relating concentrations to optical density or refractive index were fitted using ChemStation software (Agilent Technologies). Wine color measurements, including CIELab, were performed (in duplicate) using a Cintra 4040 spectrophotometer (GBC Scientific Equipment, Melbourne, Australia). Samples were filtered through 0.45 µm filters (Acrodisc, Sigma-Aldrich, Castle Hill, Australia) after which absorbance was recorded at 420, 520 and 800 nm. Wine color density and hue were calculated as: color density (au) = (A520
) + (A420
) and hue = (A420
]. CIELab measurements determined L*, a* and b*, being coordinate values corresponding to the degree of lightness, and the intensity of red (when a* > 0), green (when a* < 0), yellow (when b* > 0) and blue (when b* < 0) hues [19
The concentrations of several fermentation volatiles (acids, alcohols and esters) were determined (without replication, i.e., n
= 1) by Metabolomics Australia (AWRI) using an Agilent 7890A gas chromatograph, equipped with a Gerstel MPS2 multipurpose autosampler and coupled to an Agilent 5875C mass selective detector, and previously reported stable isotope dilution analysis (SIDA) methods [20
]. Headspace solid phase micro-extraction (HS-SPME) sampling of diluted wine (1 in 10 dilution in water) was performed in a 20 mL vial containing 2 g of sodium chloride, with the SPME fiber being exposed to the headspace for 10 min prior to desorption. Separation was achieved using a Phenomenex ZB-Wax column (60 m × 0.25 mm i.d. × 0.25 µm film thickness) and helium as the carrier gas (2.0 mL/min in constant flow mode). Preparation of isotopically labelled internal standards, method validation and instrument operating conditions are described extensively in the aforementioned publication [20
2.4. Statistical Analysis
Basic compositional data were analyzed by one-way analysis of variance (ANOVA) using GenStat (15th Edition, VSN International Limited, Herts, England, UK). Mean comparisons were performed by a least significant difference (LSD) multiple comparison test at α
<0.05. Volatile data were analyzed via an ANOVA F
-test using the lmerTest package in R statistical software (www.R-project.org/
]. Mixed effect linear models were fitted individually for each volatile, with the response variable being the concentration at each treatment level. A fixed effect predictor was included for treatment, together with a random intercept for wine, to account for the repeated measures on each wine. The fitting was performed using the lme4 package in R [22
]. Two statistical analyses were employed using XLSTAT (version 2015.4.1, Addinsoft, NY, USA): (i) a chi-square goodness of fit test, to determine statistical significance of the observed distribution of sample preferences vs. random choice; and (ii) a one sample proportion test, to determine statistical significance between the preferred sample, i.e., the alcohol sweetspot, vs. random choice.
Partial dealcoholization of wines A and B decreased alcohol levels by 1.6 and 1.8%, respectively, with no significant changes to key compositional parameters such as pH, TA, VA or color, but some loss of ethyl esters which could affect wine aroma. Chi-square goodness of fit and one proportion tests indicated preference data were statistically significant for some sweetspot determinations, with the one proportion test discriminating preferences for individual samples relative to random choice, thereby improving the likelihood of statistically significant preferences being identified. However, the outcomes of sweetspotting trials could not be replicated (either by panel or by individual panelists), suggesting identification of samples comprising superior sensory properties was challenging for this set of samples, irrespective of presentation order. It is unclear to what extent this can be attributed to the absence of perceptible differences in sensory properties due to moderate differences in (incremental) ethanol concentrations and/or the order of sample presentation, vs. the non-existence of alcohol sweetspot phenomena. Regardless, the global wine industry employs various strategies to adjust the alcohol content of wine in response to environmental, financial and market challenges, and it is likely that winemakers will continue to use alcohol sweetspotting practices to inform decisions regarding dealcoholization. The use of statistical analyses such as one proportion tests are therefore recommended, to validate any significance of outcomes from alcohol sweetspotting. Future research involving paired comparisons of samples could also be undertaken as an alternate approach to discrimination of samples that exhibit optimal overall sensory properties (i.e., to facilitate identification of alcohol sweetspots, if the alcohol sweetspot phenomena does exist).