1.1. Temporal evolution of the wine-like feature during alcoholic fermentation

To establish the temporal evolution of the wine-like feature throughout fermentation, fermentations were conducted in a chemically defined juice (CDJ) adapted from Ciani and Ferraro (1996), containing ammonium chloride as the nitrogen source (200 mg/L). Fermentations (1 L) were conducted in modified 2 L Schott bottles fitted with a port for sampling and CO₂ egress. These fermentations were mediated by Saccharomyces cerevisiae QA23 (Lallemand, Montreal, Canada), Zymaflore VL1 (Laffort-Oenologie, Bordeaux, France), a “cachaça” strain (L328) and a yeast isolate from the Douro region, Portugal (ZA). Fermentation vessels were inoculated with a pre-culture in the logarithmic growth phase (OD 640nm ≈ 1) to an OD 640 nm of 0.1 (final cell density of approximately 10⁶ CFU/mL). To evaluate the temporal expression of the wine-like feature, samples were collected daily for sensory analysis and to monitor yeast growth (OD 640 nm).

1.2. Modulating the expression of the wine-like feature at the end of alcoholic fermentation

VIN13 (Anchor Yeast, Cape Town, South Africa), an S. cerevisiae yeast strain, was rehydrated (20 g/hL) for 20 min at 37 °C in warm water and subsequently cooled to within 10 °C of the medium’s temperature before inoculation, according to the supplier’s instructions. Fermentations were conducted using a synthetic grape juice (Henschke and Jiranek, 1993) specifically formulated for wine research (100 g/L glucose and 100 g/L fructose), containing 10 mg/L ergosterol and 0.5 mL/L Tween 80 as the anaerobic factor (SGJ), as previously described (Henschke and Jiranek, 1993) unless indicated otherwise. This synthetic grape juice was formulated to mimic a natural grape must and was used in subsequent fermentations to construct a wine-like aroma.

The nitrogen composition consisted of various amino acid classes, but the concentration of fermentable nitrogen was maintained at 200 mg N/L and each amino acid provided equal amounts of fermentable nitrogen. The nitrogen treatments were classified based on their ability to support yeast growth, namely all amino acids (“all AA”), preferred (“preferred”), branched-chain and aromatic (BCAA), not utilised (“not utilised”), utilised but not preferred (“utilised”) and ammonium chloride (“ammonium”) (Table 1). Static fermentations took place at 20 °C, in duplicate, and were monitored daily (CO₂ evolution).

Based on the data obtained in the previous fermentations, a final set of nitrogen and anaerobic factor treatments were selected, as summarised in Table 2, to be evaluated in combination with each other (1.5 L). Nitrogen treatments include 200 mg N/L of fermentable nitrogen in the form of ammonium chloride, MS200 amino acids (Bely et al., 1990) and the BCAA's (Table 1). The SGJ, ergosterol with oleic acid, phytosterol, and phytosterol with ergosterol treatments were used in combination with the nitrogen treatments as described in Table 2. Static fermentations took place at 20 °C in triplicate and were monitored daily (CO₂ evolution).

2.1. Temporal evolution of the wine-like feature throughout fermentation

The testing sessions were conducted in individual booths under conditions in accordance with ISO 8589 (facilities) and ISO11037 (lighting). Eleven informal sensorial tests were performed (one for each fermentation time point). In each session, 20 untrained panellists, recruited within the department at Escola Superior de Biotecnologia, between the ages of 20 and 46 years old, evaluated the “wine-like aroma” of the four strains. The panellists evaluated the same four samples of 25 mL each in black international tasting glasses (ISO NORM 3591, 1977), coded with random three-digit codes and covered with a watch glass. Samples were randomised across panellists using a balanced complete block design. Panellists were asked whether they perceived a wine-like aroma or not. All data were collected on paper and processed using Microsoft Excel (Microsoft, Redmond, USA).

2.2. Wine-like feature at the end of alcoholic fermentation

The sensory evaluations were conducted (within 3 weeks of the completion of alcoholic fermentation) in compliance with ASTM standards (8589) in an odourless, well-ventilated room, secluded from excess noise, with controlled lighting and temperature (20 °C ±2). All samples and wines were served at room temperature and all assessments were conducted in off-white individual tasting booths. For each experiment, samples (25 mL) were presented simultaneously in black ISO glasses (ISO NORM 3591, 1977) in a randomised order using a balanced complete block design. Each glass was labelled using a 3-digit code and covered with a Petri dish. The commercial wines used were unwooded fruity Chenin blanc wines, without any perceptible faults, in addition to 10 % or 12 % ethanol samples.

Samples were only evaluated via orthonasal olfaction using a free sorting task (Cartier et al., 2006; Chollet et al., 2011; Valentin et al., 2012). Free sorting involves the classification of samples into distinct groups based on their perceived similarities (Valentin et al., 2012), which inherently requires a degree of product comparison; panellists were also asked to describe the reason for the grouping. Finally, each sample was rated with respect to the degree to which they resembled wine on the same unstructured line scale (12 cm or 14 cm) anchored at “wine-like” and “not wine-like”. In the case of the combined evaluation of anaerobic factors and amino acids, panellists first evaluated the synthetic products, and after a short break, they evaluated the second set of replicates of the same synthetic samples in addition to a commercial wine and 12 % ethanol (Figure 1).

For the sensory evaluation of amino acid classes, 30 naive wine consumers were used between the ages of 18 and 65 to affirm that the wine-like character is a precept that does not require defining or training to identify it and the suitability of the wine-like scales used. These naïve wine consumers were staff and postgraduate students recruited from Stellenbosch University who did not have any formal winemaking training. In subsequent evaluations, the panel used was not trained on this matrix but had received regular and extensive training for other white wine descriptive analyses. These 20 participants (24 to 62 years of age) were recruited annually and trained to participate in sensory evaluations. Consequently, we relied on their previous experience with line-scale rating to inform the intuitive rating of the wine-like feature. All data were collected on paper and processed using Microsoft Excel.

3.1. Gas chromatography-flame ionisation detector analyses at the end of alcoholic fermentation

The quantification of several fermentation-derived volatile compounds was conducted using gas chromatography equipped with a flame ionisation detector (GC- FID), as described previously by Louw et al. (2009). Briefly, five millilitres of the sample were used with 4-methyl-2-pentanol (internal standard, 100 µL of 0.5 mg/L solution). The volatile compounds were extracted with diethyl ether (1 ml) and this mixture was then placed in an ultrasonic bath for 5 min. Following this, samples were centrifuged at 4000 g for 3 min. A Hewlett Packard 6890 Plus GC-FID instrument (Agilent, Little Falls, Wilmington, USA) with a split/splitless injector was used for major volatiles quantification. The split flow rate was set at 49.4 mL/min and the split ratio was set to 15:1 at a temperature of 200 °C. The separation of compounds was done using a J and W DB-FFAP capillary GC column (Agilent, Little Falls, Wilmington, USA) with the dimensions of 60 m × 0.32 mm and a 0.5 μL coating film thickness with the flow rate of the hydrogen carrier gas set at 3.3 mL/min. An initial oven temperature of 33 °C was held for 17 min; the temperature was then increased by 12 °C/ min to 240 °C and held for 5 min. Once the FID oven temperature reached the temperature of 240 °C, three microliters of the extracted sample was injected into the gas chromatograph. A post-run step at the end of each sample was carried out at 240 °C for 5 min. The column was cleaned with an injection of hexane after every 10 samples. Authentic reference standards (Ethyl acetate (Sigma-Aldrich, Germany); methanol (Merck, Germany); ethyl 2-methyl propanoate (Sigma-Aldrich); ethyl propionate (Sigma-Aldrich); 2-methyl propyl acetate (Sigma-Aldrich); ethyl butyrate (Fluka, Switzerland), n-propanol (Sigma-Aldrich); ethyl 2-methylbutyrate (Sigma-Aldrich); ethyl isovalerate (Sigma-Aldrich); isobutanol (Fluka); isoamyl acetate (Riedel de Haën, Germany); n-butanol (Fluka); isoamyl alcohol (Sigma-Aldrich); ethyl hexanoate (Sigma-Aldrich); pentanol (Sigma-Aldrich); hexyl acetate (Fluka); acetoin (Sigma-Aldrich); 4-methyl-1-pentanol (Fluka); 3-methyl-1-pentanol (Fluka); ethyl lactate (Fluka); hexanol (Merck); 3-ethoxy-1-propanol (Sigma-Aldrich); ethyl octanoate (Sigma-Aldrich); 1-octen-3-ol (Sigma-Aldrich); acetic acid (Sigma-Aldrich); ethyl-3-hydroxybutanoate (Merck); propionic acid (Sigma-Aldrich); isobutyric acid (Fluka); butyric acid (Fluka); ethyl decanoate (Sigma-Aldrich); isovaleric acid (Fluka); diethyl succinate (Fluka); n-valeric acid (Sigma-Aldrich); ethyl phenylacetate (Sigma-Aldrich); 2-phenylethyl acetate (Sigma-Aldrich); hexanoic acid (Sigma-Aldrich); 2-phenylethanol (Merck); octanoic acid (Sigma-Aldrich); decanoic acid (Sigma-Aldrich)) were used to calibrate for each of the compounds using the internal standard compound 4-methyl-2-pentanol (Fluka, Switzerland) 10 mg/L). A six-point calibration was performed by spiking model wine with the volatile stock solution and performing the liquid–liquid extraction as described above. Manual data collection and peak integration were done using the HP ChemStation software (Rev. B01.03 [204]).

4.1. Statistical analysis of the sensory data relating to the temporal evolution of wine-like aroma

For the sensory data, mean and standard deviations were calculated in Microsoft Excel.

4.2. Statistical analysis of the sensory data at the end of alcoholic fermentation

Participants sorted (grouped) samples based on their similarities and then described the groups made using their own words (Chollet et al., 2011). This sorting data was recorded for each participant in a similarity matrix which was summed for all participants. The descriptors participants provided to describe each group was used to construct a contingency table which summarised how frequently a descriptor was used to describe each of the products. Where appropriate, similar descriptors were combined using lemmatisation and categorisation, and descriptors used by fewer than 15 % of the panel were discarded (Lawrence et al., 2013; Valentin et al., 2012). The summed similarity data matrix was evaluated using multidimensional scaling (MDS) in the case of the sorting data sets using XLSTAT 2017 (XLSTAT, Paris, France). In multidimensional scaling (MDS), samples that are close to each other are similar, whereas those that are further apart are dissimilar (Valentin et al., 2012). Additionally, the descriptors used to describe the products in the sorting task were projected onto the MDS plot using Pearson's correlation coefficients.

The wine-like rating data were evaluated using analysis of variance (ANOVA) (type III) paired with the Fisher LSD post hoc test (P < 0.05) to determine which samples are significantly different with respect to the wine-like rating (Statistica, version 13, Statsoft Inc., Tulsa, USA). A mixed-model ANOVA was used with the judges treated as a random effect and treatments as fixed effects.

4.3. Statistical analysis of the chemical data (GC-FID) at the end of alcoholic fermentation

Principal component analyses (PCA) were performed on the chemical data obtained using XLSTAT 2017, following autoscaling.

2.1. Amino acid classes

Yeast nitrogen metabolism has been shown to directly impact the successful completion of alcoholic fermentation as well as the production of volatile compounds influencing the wine volatilome (Bell and Henschke, 2005). Amino acids assimilated from the medium can be directly incorporated into proteins or catabolised to free the amine group used for the de novo synthesis of other amino acids. The “preferred” amino acid treatment (Table 1) contains amino acids (ARG, ASN, ASP, GLN and GLU) that support yeast growth well when they are the sole nitrogen source (Ljungdahl and Daignan-Fornier, 2012). The amino acids in the “not utilised” treatment (HIS and LYS) do not support growth as well as the others when they are the only nitrogen source (Ljungdahl and Daignan-Fornier, 2012) or require oxygen (PRO) to be catabolised (Duteurtre et al., 1971). The “utilised” but not preferred treatment contains amino acids (ALA, GLY, SER, THR and TRP) that are not fully depleted from grape must during fermentation (Beltran et al., 2004; Beltran et al., 2005; Smit, 2013). Branched-chain and aromatic amino acids (BCAAs) range from average (VAL and PHE) to poor (ILE, LEU and TYR) supporters of growth (Ljungdahl and Daignan-Fornier, 2012). These BCAA amino acids may be catabolised via the Ehrlich pathway (Fairbairn et al., 2017; Hazelwood et al., 2008) into corresponding higher alcohols, esters and volatile fatty acids. At low concentrations, volatile fatty acids contribute positively to wine aroma, but at high concentrations, they may generate a rancid-sweaty character (Francis and Newton, 2005), whereas esters are generally associated with fruity or floral aromas (Bell and Henschke, 2005). In contrast, higher alcohols are greatly affected by the sensory description of the matrix, having little influence when in a neutral background but reducing the intensity of fruity and woody descriptors whilst increasing the intensity of spirit and solvent-like traits (De-La-Fuente-Blanco et al., 2016; Ferreira et al., 2016). Moreover, at concentrations above 284 mg/L higher alcohols are considered to have a negative impact on product preference (De-La-Fuente-Blanco et al., 2017).

We used a similar sensory approach (free sorting and typicality rating) as was used to evaluate the typicality of Chardonnay and Muscadet wines (Ballester et al., 2008) to evaluate a wine-like character. To generate a wide variety of fermentation products with significantly different sensory profiles, entire classes of amino acids were either added in excess to or omitted entirely from the synthetic grape must, as shown in Table 1 (Figure 1). The outcome of the sensory analysis of these fermentation products, together with some real wine samples and a 10 % ethanol solution, is shown in Figure 3A. Panellists were asked to sort the synthetic samples based on their similarities and differences and to describe the reasons for their groupings using their own words (Figure 3A). This sorting data is visualised using a multidimensional scaling (MDS) plot; samples that are close to each other are similar (i.e. frequently grouped), whereas those that are further apart are dissimilar (Valentin et al., 2012). The commercial wines were considered similar to each other and different from the 10 % ethanol sample, with the synthetic products falling between these control samples (commercial wines and 10 % ethanol) in the first dimension (Figure 3A).

Figure 3. The sensory and chemical impact of amino acid class supplementation. The multidimensional scaling (Kruskal’s stress of 0.128) (A) with attributes projected by calculating Pearson's correlation coefficients describing the sorting of treatments: All (all AA), preferred (preferred), branched-chain and aromatic (BCAA), not utilised (not utilised), and the utilised but not preferred amino acids (utilised), in addition to ammonium chloride (ammonium), two commercial wines and 10 % ethanol. Principal component analyses of the GC-FID data (B). The wine-like rating summarises how similar or dissimilar the products are to being wine-like (evaluated on the same 12 cm unstructured line scale); the greater the distance, the more wine-like they are (C).

Save View full size Expand inline Collapse inline

The wine samples were associated with positive descriptors such as fruity and sweet, as well as the descriptor “wine” (Figure 3A). The synthetic products were mostly described as unpleasant, yeasty and chemical. Interestingly, the “not utilised” and “utilised” treatments were more frequently described as fruity and less frequently as chemical than the other synthetic products (Figure 3A).

These differences in sensory perception between synthetic juice samples are not surprising, as the nitrogen treatments were selected to generate distinct chemical profiles, which in this case translated to sensory differences (Figure 3A, Supplementary Table S1). The fermentations treated with BCAA’s contained very high levels of several volatile compounds (2-phenyl ethanol, isoamyl alcohol, isobutanol, isoamyl acetate, isobutyric acid and isovaleric acid), which are directly associated with the catabolism of these amino acids (Supplementary Table S1). Additionally, the yeast is also able to de novo synthesise these volatile compounds; the same volatile compounds are therefore also present in all treatments, albeit at comparatively lower concentrations.

While the synthetic grape juice fermentations clearly and as intended resulted in sensorially and chemically distinct products, they did not differ significantly on a wine-like rating scale. Indeed, all samples, while appearing more wine-like than the 10 % ethanol solution, remained closer to this control than to the real wines, which were readily identified as such (Figure 3D). Nevertheless, the ammonium treatment resulted in the most wine-like synthetic product. This data set suggests that the synthetic grape juice fermentations clearly resulted in distinct products based on the descriptors generated, yet no clear wine-likeness was apparent.

In a similar manner, the decreasing amounts of BCAA were evaluated (Supplementary Figure S1), as was the impact of various combinations of anaerobic factors on the expression of a wine-like aroma, yet the wine-like ratings closely resembled those shown in Figure 3C (Supplementary Figure S1A).

The inclusion of the commercial wines was useful as a means of segmenting the panellists based on whether they could recognise the wine-like concept, but it also served as a reference for comparison. Due to their more chemical nature, the synthetic products may be more negatively judged, as hedonic judgements may take precedence or confound the wine-like ratings (Charters and Pettigrew, 2007; Parr et al., 2010). Therefore, in subsequent sensory evaluations, the synthetic samples were first rated against a memorised abstraction of what a wine-like aroma is, and, following a short break, these same samples were again rated along with a commercial wine (Figure 1).

2.2. Combinations of nitrogen and anaerobic factor treatments

The variation in nitrogen composition led to a series of chemically different fermentation products (Figure 3) but did not significantly impact the wine-likeness of these products. Thus, the unsaturated fatty acid and sterol composition was changed together with the nitrogen composition. Unsaturated fatty acid and sterol composition of grape must have previously been shown to significantly impact fermentative volatile aroma production (Duan et al., 2015; Fairbairn et al., 2019; Mauricio et al., 1997; Rollero et al., 2014; Varela et al., 2012). Panellists were tasked with rating and sorting (also describing) the synthetic products (Figure 4A) and after a short break, these tasks were repeated using a second fermentation replicate with commercial wine and ethanol samples (Figure 4B). In contrast with the previous data set, the wine-like rating data generated using combinations of nitrogen and lipid treatments (Figure 4C and D) showed a progression from less to more wine-like. This evaluation was replicated, and similar trends were observed (Supplementary Figure S2). All treatments in Figure 3 contained the SGJ anaerobic factors and their low wine-like ranking mirrors the trends seen in Figure 4. For each nitrogen treatment, the SGJ sample was rated as comparatively less wine-like (Supplementary Figure S2), particularly the BCAA treatment.

Interestingly, the inclusion of a commercial wine had little impact on the relative rating of the synthetic products (Figure 4B). The commercial wine was rated as being more wine-like (Figure 4B) than the BCAA_SGJ and 12 % ethanol; however, it was not rated as more wine-like than several of the synthetic products.

Figure 4. MDS analyses were performed on the sorting data with aroma attributes projected onto the space using Pearson’s correlation coefficients for the synthetic products (Kruskal’s stress of 0.132) (A), as well as their description relative to 12 % ethanol and commercial wine (Kruskal’s stress of 0.190) (B). The wine-like rating was obtained on the same 14 cm unstructured line scale for the synthetic samples treated with various combinations of anaerobic factors (C) (phytosterol (PHY), ergosterol and oleic acid (EGR+OLE), ergosterol and phytosterol (ERG+PHY), and SGJ anaerobic factor mixture) and nitrogen compositions (BCAA, Ammonium, and MS200 amino acids), as well as rating them relative to ethanol and commercial wine (D).

Save View full size Expand inline Collapse inline

When the synthetic products were evaluated on their own (Figure 4A), the MDS plot shows that each nitrogen treatment largely falls into a separate quadrant. The first dimension illustrates the differences between the BCAA and the ammonia and MS200 treatment grouping, and the second dimension describes the separation of the MS200 treatment from the others. Cluster analysis shows that the BCAA treatment forms a distinct grouping with the ammonia and MS200 treatments forming a second cluster (data not shown). This second cluster generally forms a continuum with the ammonium treatment at one end and the MS200 treatment at the other. The BCAA samples generally correlate with fruity and yellow fruit attributes. The MS200 products were also described as fruity (tropical fruit, citrus, dry fruit), in addition to sweet and floral. The ammonium treatment was associated with an animal aroma. In all instances, the SGJ samples were somewhat removed from the other anaerobic factor treatments. The SGJ modality was also associated with negative sensory attributes (BCAA—chemical/petro-chemical, MS200—forest floor and NH4—animal). It is possible that the SGJ treatment results in a chemical matrix whose sensory impact is enhanced to a greater degree by the high concentration of higher alcohols in the BCAA treatment than in the other nitrogen treatments (De-La-Fuente-Blanco et al., 2016).

With the inclusion of the controls (ethanol and commercial wine), the first dimension describes the separation of the ethanol solution from the synthetic products, and the second dimension shows how the synthetic products and ethanol differ from wine (Figure 4B). As seen in Figure 4A, the BCAA treatments generally clustered together, with the ammonium and MS200 samples forming a second cluster (Figure 4B). The descriptor data shows that the BCAA treatments were more frequently associated with fruity aromas (fruity, white fruit, citrus and dry fruit), except when the BCAA nitrogen was combined with the SGJ lipids. Interestingly, the other SGJ treatments were also associated with petrochemical and forest floor descriptors (Figure 4C, 4D and Supplementary Figure S2), although to a lesser extent. The inclusion of the controls had little impact on the relative distribution of the synthetic samples. Nonetheless, the improved recognition of the wine-like feature is certainly in part because of the changes to the sensory approach used, suggesting that by evaluating the synthetic samples independently first (Figure 1), the panellists were better able to evaluate the less complex synthetic samples for the presence or absence of the wine-like feature. This is confirmed by the poor wine-like rating of various anaerobic factors when evaluated with wine (Supplementary Figure S1).

Overall, the data shown in Figure 4 suggests that although most of the samples were rated as being wine-like, each also had its own specific sensory profile.