Metabolic profiling as a tool for revealing Saccharomyces interactions during wine fermentation


  • Kate S. Howell,

    1. Food Science and Technology, School of Chemical Engineering and Industrial Chemistry, University of New South Wales, Sydney, NSW, Australia
    2. The Australian Wine Research Institute, Glen Osmond, Adelaide, SA, Australia
    Search for more papers by this author
  • Daniel Cozzolino,

    1. The Australian Wine Research Institute, Glen Osmond, Adelaide, SA, Australia
    2. The Cooperative Research Centre for Viticulture, Glen Osmond, Adelaide, SA, Australia
    Search for more papers by this author
  • Eveline J. Bartowsky,

    1. The Australian Wine Research Institute, Glen Osmond, Adelaide, SA, Australia
    Search for more papers by this author
  • Graham H. Fleet,

    1. Food Science and Technology, School of Chemical Engineering and Industrial Chemistry, University of New South Wales, Sydney, NSW, Australia
    Search for more papers by this author
  • Paul A. Henschke

    1. The Australian Wine Research Institute, Glen Osmond, Adelaide, SA, Australia
    Search for more papers by this author

  • Editor: Lex Scheffers

Paul A. Henschke, The Australian Wine Research Institute, PO Box 197, Glen Osmond, Adelaide SA 5064, Australia. Tel.: +61 8 8303 6600; fax: +61 8 8303 6601; e-mail:


The multi-yeast strain composition of wine fermentations has been well established. However, the effect of multiple strains of Saccharomyces spp. on wine flavour is unknown. Here, we demonstrate that multiple strains of Saccharomyces grown together in grape juice can affect the profile of aroma compounds that accumulate during fermentation. A metabolic footprint of each yeast in monoculture, mixed cultures or blended wines was derived by gas chromatography – mass spectrometry measurement of volatiles accumulated during fermentation. The resultant ion spectrograms were transformed and compared by principal-component analysis. The principal-component analysis showed that the profiles of compounds present in wines made by mixed-culture fermentation were different from those where yeasts were grown in monoculture fermentation, and these differences could not be produced by blending wines. Blending of monoculture wines to mimic the population composition of mixed-culture wines showed that yeast metabolic interactions could account for these differences. Additionally, the yeast strain contribution of volatiles to a mixed fermentation cannot be predicted by the population of that yeast. This study provides a novel way to measure the population status of wine fermentations by metabolic footprinting.


The fermentation of grape juice into wine is carried by yeasts. The population dynamics and diversity of yeast species associated with the fermentation are quite complex and variable (Fleet, 2003). Generally, various species of Hanseniaspora, Candida, Torulaspora, Metschnikowia, Kluyveromyces and Saccharomyces, which originate from the grape berry and winery environment, grow during the first stages of fermentation (Fleet et al., 1984; Fleet & Heard, 1993). As the ethanol concentration increases, the non-Saccharomyces yeasts die off, leaving Saccharomyces cerevisiae and Saccharomyces bayanus to dominate and complete the fermentation (Fleet & Heard, 1993; Loureiro & Querol, 1999). The predominance of S. cerevisiae in fermentations has led to its recognition as the principal wine yeast, and various strains of S. cerevisiae have been commercialized as starter cultures for wine production (Fleet, 2003). Grape juice is not sterilized or pasteurized, and the added starter culture must therefore compete with indigenous yeasts. Consequently, most wines are the products of fermentation with mixtures of yeast species (Fleet & Heard, 1993; Fleet, 2003). It is now realized that the ecological complexity and variability of these fermentations extend beyond the species level. Within the one fermentation, several strains of each species may be involved. Over 100 genetically distinct strains of S. cerevisiae have been reported in some fermentations (Pramateftaki et al., 2000; Torija et al., 2001), but in one study, only one strain was found to persist (Frezier & Dubourdieu, 1992).

The impact of the yeasts upon wine flavour is largely determined by the array of volatile substances (e.g. higher alcohols, acids, esters, carbonyls, thiols) produced by the metabolism of grape juice components (Berry & Watson, 1987; Dumont & Dulau, 1997; Lambrechts & Pretorius, 2000). The profile of these flavour-active volatiles varies with the yeast species and strains contributing to the fermentation (reviewed by Suomalainen & Lehtonen, 1979; Soles et al., 1982; Moreno et al., 1991; Longo et al., 1992; Lema et al., 1996; Henick-Kling et al., 1998; Antonelli et al., 1999; Heard, 1999; Marais, 2001; Garcia et al., 2002).

With the current understanding of the yeast ecology of wine fermentation, winemakers are seeking to enhance the flavour diversity and appeal of wines by controlled fermentation with multiple species or strains of yeasts (Lambrechts & Pretorius, 2000; Fleet, 2003). Several studies have already described fermentation with mixtures of non-Saccharomyces yeasts and S. cerevisiae (Moreno et al., 1991; Soden et al., 2000; Garcia et al., 2002), and one study has examined fermentation with a mixture of different strains of S. cerevisiae (Cheraiti et al., 2005). It is apparent from the chemical data reported in these studies that the profile of volatile aroma substances produced in mixed culture ferments can differ from the profile produced by the simple addition of substances expected of the constituent, single cultures. It could be concluded that, in mixed-culture ferments, one species or strain may impact on the metabolic behaviour of others (Cheraiti et al., 2005). The concept of such metabolic interactions is new in the field of wine science and requires more precise description and understanding to enable practical development and application.

Yeasts produce many metabolites (over 60) that are known to have an impact on wine flavour (Berry & Watson, 1987; Dumont & Dulau, 1997; reviewed by Lambrechts & Pretorius, 2000). Generally, these are analysed by gas chromatography–mass spectrometry (GC–MS) (Ebeler, 2001). The large number of peaks and fragments that are found in the analysis of one sample of wine presents a logistical challenge when comparisons between several wine samples are required, for example between mixed-culture and single-culture ferments of grape juice. Metabolomic footprinting has been defined as gaining ‘enough information to unravel (otherwise hidden) metabolic alterations, without aiming to get quantitative data for all biochemical pathways’ (Jolliffe, 1986). Because analytical footprinting techniques (for example GC–MS) generate a large amount of information, chemometric methods are generally employed to reduce the size of the data. Both supervised and unsupervised multivariate analysis, such as cluster analysis, principal-component analysis (PCA) or discriminant partial least squares (DPLS) (Martens & Martens, 2000), have been used to generate biochemical footprints from samples ranging from olive oil and cocoa butter to tomatoes and various plants. Chemometric methods have been successfully used to obtain biochemical fingerprints of fruits and beverages (Jellum et al., 1991; Munck et al., 1998; Arvantoyannis et al., 1999; Brereton, 2000; Sunesson et al., 2001; Cordella et al., 2002; Eide et al., 2002; Fiehn, 2002; Fiehn & Weckwerth, 2003; Johnson et al., 2004), as well as metabolic footprints of yeasts (Raamsdonk et al., 2001; Eglinton et al., 2002; Allen et al., 2003).

In this study, we use chemometric analysis of GC–MS data to elucidate differences in the volatile components of wines fermented with different strains of S. cerevisiae and S. bayanus. To investigate potential metabolic interactions between the yeasts, wines made using each yeast strain grown in monoculture were blended and compared with wines made with a mixed inoculum of the strains.

Materials and methods

Yeast strains and culture preparation

Saccharomyces cerevisiae strains AWRI 796, AWRI 838, AWRI 835 and AWRI 1434 were obtained from The Australian Wine Research Institute (Adelaide, Australia). Strain AWRI 796 is a commercial yeast from AB Mauri Yeast (Australia). Strain AWRI 838 was isolated from the commercial preparation Lalvin EC1118 (Lallemand, Adelaide, Australia). Saccharomyces cerevisiae strain AWRI 1434 was isolated from the commercial preparation Zymaflore VL3 from Laffort Oenologie (Australia). Saccharomyces cerevisiae strains ICV D47 and QA23 were provided by Lallemand, and are available as commercial preparations. All yeast strains have the killer phenotype. Saccharomyces bayanus strain AWRI 1176 was isolated from a spontaneously fermented cold stored grape juice (Eglinton et al., 2000). Yeasts were maintained by culture on plates of YPD agar (28°C) and stored at 4°C. YPD consisted of yeast extract (10 g L−1), peptone (20 g L−1) and glucose (20 g L−1) with the inclusion of agar (20 g L−1) for solid media.

Yeast colonies were taken from YPD plates, inoculated into liquid YPD and incubated with shaking at 25°C overnight. The overnight culture was sub-cultured in grape juice diluted with double-distilled water (1 : 1) and incubated in cotton wool-plugged conical flasks at 25°C, with agitation. When the culture reached 1–2 × 108 cells mL−1 (as determined by haemocytometer counts), yeast cells were inoculated into fermentors to give an initial density of approximately 1 × 106 cells mL−1. Mixed cultures were inoculated with about 3.3 × 105 cells mL−1 for each strain of yeast.

Wine fermentations

Mixed culture fermentations (coded M1, M2, M3 and M4) were conducted using four different mixtures of strains of S. cerevisiae or S. bayanus, as indicated in Table 1. Two combinations of mixed fermentations were examined in this study. The first combination examined the effect of varying a single strain in a mixture of S. cerevisiae strains. These fermentations were inoculated with two constant S. cerevisiae strains (AWRI 796 and ICV D47) and a third, variable strain which was QA23 for M1, AWRI 838 for M2 and AWRI 835 for M3 (Table 1). The second combination examined the effect of varying the Saccharomyces species in mixed cultures. Saccharomyces bayanus strain AWRI 1176 was used in conjunction with two S. cerevisiae strains, AWRI 1434 and AWRI 838, to make M4 (Table 1). Monoculture ferments of each of the seven yeast strains were performed in either duplicate or quadruplicate. Fermentations were conducted in 3-L batches of grape juice contained in Bellco fermentors at 15°C. Fermentors were fitted with a magnetically driven stirrer, water air-lock and sampling and filling ports, and were sterilized prior to use. The grape juice was a gift from The Hardy Wine Company (Australia) and had an initial sugar concentration of 230 g L−1 and a pH of 3.2. The Chardonnay juice was sterilized by cross-flow membrane filtration (0.2 μm, S & F Fabrications, Victor Harbour, South Australia), stored at 4°C and transferred aseptically to the fermentors using a peristaltic pump. Fermentations were conducted in duplicate (M1, M2 and M3) or quadruplicate (M4), including the respective monoculture fermentations.

Table 1.    Strain composition of yeast cultures used in fermentation trials. Each number represents a strain of Saccharomyces cerevisiae, except AWRI 1176, which is a strain of Saccharomyces bayanus
Fermentation trialCombination of yeast strains
M1AWRI 796. ICV D47, QA23
M2AWRI 796, ICV D47, AWRI 838
M3AWRI 796, ICV D47, AWRI 835
M4AWRI 1434, AWRI 1176, AWRI 838

Monitoring of fermentation

The progress of fermentation was followed by analysis for sugar utilization and yeast populations. Glucose and fructose were assayed using an enzymatic kit (Roche, Mannheim, Germany). Wine samples (1 mL) were taken aseptically throughout fermentation, using nitrogen flushing to prevent ingress of air. A portion of the sample was immediately used for determination of yeast population and the remainder was stored at −20°C until analysis. Yeast viability was determined by surface plating on WL Nutrient Agar (Amyl Media, Sydney, Australia). Samples were appropriately diluted in 0.1% peptone (Amyl) and then 0.1 mL was inoculated onto the plates. Plates were incubated at 27°C for 3 days and colonies were examined for species and strain diversity. Species differentiation was determined on the basis of colony morphology. Colonies of S. bayanus stained green, and S. cerevisiae colonies were green with a white margin. The validity of this species differentiation technique was confirmed by plating mixed cultures on WL Nutrient agar plates, and identifying colonies by morphology and PCR amplification of the MET2 gene (Masneuf et al., 1996).

For S. cerevisiae strain differentiation, PCR amplification of the SC9182X locus was performed (Howell et al., 2004). The amplified PCR products were separated on precast 12% polyacrylamide gels (Gradipore, New York, NY), in 0.5 × Tris borate EDTA buffer (100 V for 1 h 30 min), stained with ethidium bromide, visualized with ultraviolet light and photographed (Sambrook & Russell, 2001). Strain-specific banding profiles were then tallied. For each treatment replicate, 45 colonies were analysed per sample point (beginning, middle and end of fermentation). The proportion of each yeast strain was calculated as a percentage of the total colonies counted. The midpoint of fermentation was considered to be when 50% of the initial sugar had been metabolized.

Postfermentation wine handling and adjustment

When the sugar concentration in each fermentor was less than 2 g L−1, the temperature of the culture was decreased to 4°C and the yeast lees allowed to settle for 5 days. Excluding air by nitrogen gas cover and solid carbon dioxide additions, the wine was racked into 5-L carboys, and sulphur dioxide (as potassium metabisulphite, AnalaR, Gibbstown, NJ) was added incrementally to achieve a concentration of 10–15 mg L−1 free sulphur dioxide (Iland et al., 2000). Next, the wine was filtered under nitrogen pressure through a glass prefilter (Millipore, Bedford, MA), and a 0.2-μm membrane filter (Millipore) in a 2-L capacity dead-end filter apparatus (Sartorious, Göttingen, Germany). The wine was placed into glass bottles (750 mL), crown-sealed, and stored at 15–20°C for up to 10 weeks until chemical analysis.

Wine blending

To investigate the effects of yeast co-fermentation, blended wines were made. Blended wines corresponding to mixed fermentations M1, M2 and M3 were named B1, B2 and B3. These wines consisted of equal proportions of monoculture wines corresponding to the yeast which fermented them. For example, blended wine B1 contained equal proportions of monoculture wines prepared with strains AWRI 796, ICV D47 and QA23 (see Table 1). The volatile profile of the blended wines will be different from the corresponding mixed culture wine (for example B1 vs. M1) if the constituent yeasts interact metabolically.

A second blending experiment was carried out based on the population of each yeast strain found at the endpoint of fermentation. Here, proportionately blended wines (B1P, B2P and B3P) were constructed using a ratio obtained by reference to Fig. 2. For example, blended wine B1P was prepared using 18% AWRI 796, 75% ICV D47 and 7% QA23 (Fig. 2). If a yeast dominates the fermentation numerically, there may be a flavour impact on that wine which may not be reflected in the proportionately blended wines. The proportionately blended wines allow comparisons with the mixed-culture wines, taking into consideration potential flavour impact by the numerically dominating yeast.

Figure 2.

  Proportions of each yeast strain at the beginning (a), middle (b) and end (c) of fermentation. Yeast strain mixtures are the same as in Table 1. Error bars indicate the standard error between the duplicate (M1, M2 and M3) and quadruplicate (M4) fermentations. ▪, AWRI 796; inline image, ICV D47; inline image, QA23; inline image, AWRI 838; inline image, AWRI 835; inline image, AWRI 1434; inline image, AWRI 1176.

Gas chromatography–mass spectrometry

An HP 6890 gas chromatographer (GC) coupled to an HP 5973 mass spectrometer (MS) (Agilent, Forest Hill, Victoria, Australia) was used to determine the concentrations of volatile components in wines. Immediately upon opening the bottle, 5 mL of wine was extracted at 20–25°C into an equal volume of pentane : ether after addition of a mixture of standards (ethyl nonanoate, octanol and nonanoic acid). The GC was fitted with a 30 m × 0.25 mm fused silica capillary column DB-1701, film thickness 0.25 μm (J&W Scientific, Folsom, CA). The oven temperature was 50°C and was held at this temperature for 1 min before being increased by 10°C per min to 250°C and then kept at that temperature for a further 20 min. The injector was held at 220°C and the transfer line at 280°C throughout the analysis. The sample volume injected was 2 μL. The carrier gas was helium, with a flow rate of 1.2 mL min−1. The inlet was in pulse splitless mode. Positive ion impact spectra at 70 eV were recorded in the range m/z 50–350 for scan runs.

Data analysis and interpretation

Multivariate models of GC–MS data were constructed to describe compositional changes that occurred in the samples due to the different yeast mixtures. All the GC–MS data files (nonprocessed, CSV format) were exported to the unscrambler software (version 7.5, CAMO ASA, Oslo, Norway) for chemometric analysis. Before performing PCA, GC–MS data were pre-processed in order to avoid baseline influence, retention time drifts, variations in peak shapes and differences in recovery between the analysed samples (Jonsson et al., 2004).

In this study, smoothing (moving average on each of seven data points) and normalization (mean normalization) provided by the unscrambler software were used as pre-processing methods. The moving average reduced the noise and made it easier to observe the start and end peaks (Jonsson et al., 2004). Mean normalization consisted of dividing each row of a data matrix by its average, thus neutralizing the influence of any hidden factor. It is equivalent to replacing the original variables by a profile centred on one. Only the relative values of the variables were used to describe the sample, and the information carried by their absolute level was dropped. This is indicated in the specific case where all variables are measured in the same unit, and their values assumed to be proportional to a factor, which cannot be directly taken into account in the analysis. Such transformation is required to express the results in the same units for all samples (Jonsson et al., 2004).

PCA was used for reducing the dimensionality of data, detecting the number of components and visualizing the outliers (Martens & Naes, 1989; Jonsson et al., 2004). It is a mathematical procedure for resolving sets of data into orthogonal components whose linear combinations approximate the original data to a desired degree of accuracy. PCA was used to derive the first 20 principal components from the GC–MS data and arrange samples into groups. DPLS is a supervised pattern recognition technique, used to locate differences between members of different logical groups by searching for structural information that can discriminate between the groups. DPLS models were developed using a dummy variable. With this technique, each sample in the calibration set is assigned a dummy variable as a reference value, which is an arbitrary number. Samples of mixed fermentations (M) were assigned a numeric value of 1 and blended fermentations (B) a value of 2. The DPLS model was then developed by regressing the GC–MS data against the assigned dummy value.

In empirical modelling, it is essential to determine the correct complexity of the model. With numerous and correlated GC–MS variables, there is a substantial risk of over-fitting, where a well-fitting model has little or no predictive power (Martens & Martens, 2000). Hence, a strict test of the predictive significance of each PCA or PLS component is necessary. Cross-validation was performed by arranging the data into four groups and then developing a number of parallel models from the reduced data when one of the groups had been deleted.

The loadings correspond to the total ion chromatograph (as a numeral) measured in one scan by the mass spectrometer. They do not correspond to a single compound, but to the total ions produced by all the compounds detected in that scan. Therefore, sample grouping was based on the score plots and the loadings after each PCA model. The procedure does not aim to identify GC–MS peaks, but to provide groupings of samples.


Mixed yeast culture fermentation dynamics

The progress of fermentations with mixed cultures is shown in Fig. 1. The end of fermentation was taken as when less than 2 g L−1 of glucose plus fructose remained in the wine. The length of fermentation varied from 9 to 11 days for M1, M2 and M3, and 17 days for M4. Monoculture fermentations of the corresponding yeast strains used in mixtures M1, M2 and M3 required 8–15 days to complete, whilst the strains constituting M4 needed 15–18 days to complete (data not shown). However, the monoculture fermentation with strain AWRI 1176 did not give final sugars below 20 g L−1, despite attempts to restart the fermentation using air sparging and nitrogen addition (Bisson, 1999; data not shown).

Figure 1.

  Utilization of sugars during mixed-culture fermentations with wine yeasts. Each point is the average of duplicate determinations of samples from duplicate or quadruplicate fermentations. Error bars indicate the standard deviation, where it was possible to be resolved on the graph.

Change in relative proportions of yeast strains during fermentation

Total yeast populations reached 5–9 × 107 cells mL−1 within 5 days for M1, M2, M3 and associated monoculture fermentations, and 5–6 × 107 cells mL−1 within 4 days for fermentation M4 and associated monocultures (data not shown). The viability of the total yeast population did not decline over the course of fermentation (data not shown). The data presented in Fig. 2 show shifts in the proportions of each yeast strain in the four mixed culture fermentations. The proportion of each yeast in the mixed culture at inoculation ranged from 25% to 50% for each strain, although an optimal inoculation density of 33% for each strain was desired. For mixtures M1, M2 and M3, the proportion of the three yeast strains at inoculation did not change until after the midpoint of fermentation, whereas M4 showed a larger proportion of AWRI 1434 when 50% of the sugar was utilized. For all mixed cultures, dominance by one strain was evident at the end of fermentation. M1, M2 and M3 were dominated by AWRI 796, whereas M4 was dominated by AWRI 1434. Interestingly, the two strains (AWRI 796 and ICV D47) that were used across the three ferments (M1, M2 and M3) had similar relative proportions at the end of fermentation, irrespective of the third strain used in these mixtures (QA23, AWRI 838 or AWRI 835). For M1, M2 and M3, strain AWRI 796 dominated from inoculation through to the completion of fermentation. M4 was dominated by strain AWRI 1434, which was the yeast present in the lowest concentration at the beginning of fermentation. Although S. bayanus AWRI 1176 represented 45% of the yeast population at the beginning, it had decreased to 18% by midfermentation and was not detected at the end of fermentation in M4.

The volatile composition of mixed-culture wines differs from that in blended wines

Volatile wine components were measured by GC–MS. Although there are more sensitive methods to measure individual compounds, such as stable-isotope dilution analysis with GC–MS, (Kotseridis et al., 2000; Steinhaus et al., 2003), the approach taken here can examine more compounds irrespective of whether they can be identified (Allen et al., 2003; Jonsson et al., 2004). The volatiles were extracted, and injected into a GC–MS, which collected data from the total ion chromatograph during the scan interval.

The collated ions, and therefore the volatiles measured by GC–MS, were able to differentiate mixed-culture ferments (M) and blended (B) wines (Fig. 3a). The PCA plot shows that 97% of the variation is explained by the first three principal components, and that replicate fermentation measurements group together. The spread grouping of the replicates could be due to inherent biological variation among three fermentation replicates (see error bars, Fig. 2).

Figure 3.

  Principal-component analysis (PCA) score plots and corresponding loadings of volatiles measured by gas chromatography–mass spectrometry (GC–MS). Wines were made by inoculating Saccharomyces cerevisiae strains (a) or Saccharomyces cerevisiae with Saccharomyces bayanus strains (b). The value of the principal component is given as a percentage in each of the dimensions. The data points correspond to a mixed-culture (M) wine or a blended wine of monocultures (B or BP). The loading plots show the magnitude of the eigenvector used to construct the PCA, by the scan number of the GC–MS data. The data points at each scan number are the sum of the total ion chromatograph total at that point.

The PCA scores for both B2 and M2 wines plotted in different quadrants. The basis for the observed separation is the diverse PCA loadings (Fig. 3a). Further, the blended wine made with the proportions of monoculture wines corresponding to the yeast ratios at the endpoint of fermentation (B2P) was separated from both B2 and M2. A similar pattern was observed for B3, M3 and B3P, where blended wines were clearly different from the corresponding mixed culture (Fig. 3a). The PCA plots for M1 and the corresponding blended wines are not shown due to the poor data resolution from GC–MS.

The preliminary hypothesis that blending wines proportionately would imprint the wine with an aroma profile related to the dominating yeast in the mixed cultures was investigated. However, as proportionately blended wines were different from the mixed-culture wines and the equal-proportion blended wines, this hypothesis did not appear to apply (Fig. 3a).

When a different species of yeast (S. bayanus) was included in the fermentations (M4), wines from mixed-culture fermentation and the corresponding blended wines (B4) were separated into opposite quadrants of the PCA plot (Fig. 3b). By investigating the loadings for the PCA, Fig. 3b shows that more ions produced by the GC–MS are responsible for separating the S. cerevisiae/S. bayanus wines than wines produced only with strains of S. cerevisiae.

In addition, the results show that the volatile yeast products in a mixed-culture wine differ from those produced in a monoculture wine (Fig. 4a), as well as in the blended monoculture wines.

Figure 4.

  Principal-component analysis (PCA) score plots and corresponding loadings of volatiles measured by gas chromatography–mass spectrometry (GC–MS). Wines made using a monoculture of Saccharomyces cerevisiae were compared to mixed-culture wines (M2 and M3) (a) or the monocultures alone (b). The value of the principal component is given as a percentage in each of the dimensions. The data points correspond to a monoculture (strain name) or mixed-culture inoculum wine (for composition see Table 1). The loading plots show magnitude of the eigenvector used to construct the PCA, by the scan number of the GC–MS data. The data points at each scan number are the sum of the ion chromatograph total at that point.

Wines made with AWRI 835 have volatile profiles different from those in other wines

The degree of difference between wines from mixed-culture and blended monoculture wines prompted further analysis of the volatile profiles of the monoculture fermentations. A PCA plot compared the differences between the monoculture and the mixed wines for M2 and M3 only (Fig. 4a). The score plot confirms that monoculture wines are different from mixed-culture wines. The loadings show that the separation could be related to several variables (ions) not observed when mixed-culture or blended wines were analysed (cf. Figs 3b and 4b). Monoculture fermentation data were re-analysed to distinguish the grouping of monocultures seen in Fig. 4a. As shown in Fig. 4b, wines made with strain AWRI 835 are different from the other monocultures. The wines made with QA23, although scattered, group together, and wines made with strains AWRI 838, AWRI 796 and ICV D47 are close to one another (Fig. 4b). The latter combination of yeasts (AWRI 838, AWRI 796 and ICV D47) was used for mixed-culture fermentation M2 (Table 1). The M2 inoculum consists of yeasts which, when grown in monoculture, provide a similar GC–MS volatile profile. Despite the similarity in volatile production between the yeast strains, the data presented in Fig. 4a show that M2 separates from the group of monoculture wines. Interactions between these yeasts provide a fermentation outcome dissimilar to that predicted by the monocultures.

Confirmation of the PCA modelling analysis by DPLS

Figure 5 shows the PLS score plots for the B4 and M4 cultures. The explained variation in the X matrix (GC–MS data) is around 87%, and the explained variation in the Y matrix (dummy values) is 92%. To validate the DPLS models, samples not included in the calibration model were used to test the predictive ability. Blended samples were all correctly (100%) classified, whereas only 66% of the samples belonging to the mixes were correctly classified. The patterns in the loadings for the DPLS models were similar to those observed for the PCA models. These observations show that wines made by mixed cultures were more complex than the corresponding blended wines. Additionally, the results show that differences in the GC–MS measurements (replicates) did not explain the variation observed in the monocultures, meaning that real differences separated the samples.

Figure 5.

  Discriminant partial least squares (DPLS) score plots for the mixed culture (M) and blended culture (b) for fermentations containing Saccharomyces cerevisiae and Saccharomyces bayanus (M4 and B4). For strain composition see Table 1. Samples not included in the calibration model were used to test the predictive ability as validation of the model. Loadings as described previously.


It is now widely accepted that many wines, whether produced with or without inoculated yeasts, are the outcome of a mixed fermentation that involves contributions from many species and strains (Henschke, 1997; De Vos, 2001; Fleet, 2003). However, research investigating winemaking practices is generally based on laboratory experiments where a sterile or near-sterile juice or must is used (see for example Heard & Fleet, 1988; Ciani & Maccarelli, 1998; Eglinton et al., 2000; Soden et al., 2000). The final flavour of the wine is determined in part by the composite of volatile aroma compounds produced by the mixed-culture reaction (Lambrechts & Pretorius, 2000; Fleet, 2003). Building on this knowledge, winemakers may now seek greater control over yeast contribution to wine flavours and its predictability by conducting fermentation with defined mixtures of yeast species and strains. Such a mixed-culture product has recently been proposed (Grossmann et al., 1996). For this aspiration to become a practical reality, more information is needed to understand how particular species and strains of wine yeasts grow in mixed-culture, and how such culture impacts on their production of aroma volatiles.

In a previous publication (Howell et al., 2004), we have reported a convenient molecular method of yeast strain differentiation that can be used to monitor the population dynamics of strains of wine yeasts during fermentation. Here, we have combined this method with several chemometric methods to demonstrate that mixed cultures of Saccharomyces wine yeasts give a combination of volatile aroma substances distinctly different from wines made by blending together monoculture wines made with the same component yeast strains. These results indicate metabolic interaction between component strain and species. The profile of volatile aroma compounds produced by yeasts could be considered as a partial or volatile-fraction ‘metabolome’. Processing of these data by the methods of Jonsson et al. (2004) has facilitated the comparative analysis and interpretation of these large sets of data. This method could be further developed as a quality control tool to monitor rapidly the aroma footprint given by yeasts during fermentation.

In the first series of trials, M1, M2 and M3, the potential interaction of yeast strains was investigated. As monocultures, yeasts AWRI 796, ICV D47, QA23, AWRI 838 and AWRI 835 gave profiles of aroma components that were clearly differentiated from those in mixed culture (Fig. 4a). This was evident despite the clear dominance of yeast strain AWRI 796 in all three mixed-culture ferments.

This dominance might be due to its having the highest population at the beginning of fermentation. Although the intention was to inoculate the fermentations with equal proportions of each strain, this was difficult to achieve in practice. Inoculum populations were determined by microscopic counts, but subsequent monitoring of the fermentations was done by viable plate counts. It seems that, compared with other strains, inoculum cultures of AWRI 796 always had a higher proportion of viable to non-viable cells. The reason for this was not investigated. Apart from the inoculum influence, the dominance of AWRI 796 throughout fermentation could be due to other factors such as faster growth favoured by the cultural conditions used in this study.

Wines made solely with AWRI 835, or with a mixed-culture that contained AWRI 835 (M3), were distinctly different from wines made with the other strains studied (Fig. 4a). All mixed-culture inocula (M1, M2 and M3) had the two strains AWRI 796 and ICV D47 in common, whereas the third strain was varied (Table 1). Although the proportion of AWRI 835 in the M3 total population was small, the aroma profile of the M3 mixed-culture wine was different from that of the other mixed-culture wines that also included AWRI 796 and ICV D47. Relative to the monoculture wines made with ICV D47 or AWRI 796, the AWRI 835 monoculture wine had a very different composition (Fig. 4b). As yeasts AWRI 796 and ICV D47 produced wines with similar profiles, we suggest that the wines made with AWRI 835 contain a different profile of aroma compounds, or at least some compounds are present in significantly different concentrations. Indeed, wines made with AWRI 835 have been shown to display distinctive aroma and sensory profiles (Rankine & Lloyd, 1963; Monk, 1982; Jane et al., 1996), suggesting that the distinctive metabolic activity of composite strains of a wine fermentation can affect the metabolic outcome. Therefore, the volatile contribution of a yeast strain to mixed-strain fermentation cannot be predicted by the population of that yeast alone. The accepted view is that the most numerous strain of wine yeast dominates the fermentation outcome (Fleet & Heard, 1993; Lambrechts & Pretorius, 2000; Fleet, 2003). Sensory studies to examine the contribution of less populous strains in mixed culture will show whether aroma modifications are detectable.

Saccharomyces bayanus is another species of the genus Saccharomyces that can ferment grape juice to completion (Eglinton et al., 2000). This yeast was included in the mixed culture M4 to examine yeast interactions at a species level. However, populations of S. bayanus were not detected at the end of fermentation (Fig. 1). The mechanism of S. bayanus elimination was not determined, but killer interaction can be dismissed as this strain is killer neutral (data not shown). The metabolic activity of S. bayanus AWRI 1176, nonetheless, affected the aroma profile of the wine when grown in mixed culture with S. cerevisiae AWRI 1434 and AWRI 838, as shown in Fig. 3b. The wines made by mixed and blended cultures of this yeast combination were clearly distinguishable, and represent another example of a significant contribution to the metabolic profile by a numerically inferior yeast.

We hypothesize that yeasts, which metabolically interact with one another during fermentation will give a mixed-culture wine with a composition different from that made by blending the monoculture wines in equal proportions. The mixed-culture wines presented here were different and distinguishable from the corresponding blended wines (Figs 3a and b). Further, as some yeasts dominated the fermentation population, a proportionate blend of wine was prepared in which wines were blended in the ratios of the yeast population present at the end of sugar utilization (Fig. 3a). These results add further weight to our hypothesis that yeasts can modify the products of fermentation when grown in mixed culture.

When a yeast strain produced a compound, it could be taken up and used by other yeasts present. In this way, yeast interaction and sharing of metabolites could occur. A recent study by Cheraiti et al. (2005) has demonstrated that redox interactions can occur between yeasts in co-culture and that acetaldehyde produced by one yeast is metabolized by the other. This observation provides an explanation as to why modulation of wine flavour in mixed culture cannot be replicated by blending wines together, as the modification arises from complex, largely unknown, interactions between wine yeasts.

This study has examined the hypothesis that different yeasts growing together in wine fermentations interact to change the volatile outcome, and that this interaction is measurable by investigating the metabolome. This research has important outcomes for mixed-culture research in winemaking, and indeed in other fermentations as well. However, further work is required to elucidate the molecular mechanisms by which these interactions take place, and could involve the use of gene arrays or cell-wide protein assays to identify the key modified compounds with an impact on the aroma.


The authors would like to thank Professors Peter Høj and Sakkie Pretorius for support and advice during this study. Ms Heather Smyth is thanked for advice on statistical analysis and interesting discussions in the preparation of this manuscript. Ms Dimi Capone and Dr Alan Pollnitz provided assistance for the GC–MS instrumentation and analysis. This project is supported by Australia's grapegrowers and winemakers through their investment body, the Grape and Wine Research Development Corporation, with matching funds from the Australian Government. K.S.H. is a recipient of an Australian Postgraduate Award stipend and AWRI scholarship.