Introduction

In the last 20 years, there has been an increasing global attention for ethanol content in wines, influenced mainly by the media and the government programs, due to the marketing-, social-, and health-associated reasons (Saliba et al. 2013). Wine consumption, in light to moderate amounts (1–2 glass of wine per day), has been well demonstrated to be beneficial for the human health (German and Walzem 2000; Yoo et al. 2010). In opposition, high levels of ethanol consumption and irregular drinking have been shown to be casually correlated with more than 60 different medical conditions (Room et al. 2005). The production of well-structured and full-bodied red wines nowadays is more difficult than previously thought, especially in warm climate wine regions (Jones et al. 2005). Usually, winemakers in order to achieve the optimum phenolic maturation and tannin concentration, necessary for the quality of these wines, postpone the harvest time, which results in a high, to excessive, sugar concentration in the over ripe grapes (Mira de Orduňa 2010). As a consequence, the excessive sugar content could be translated to wines with elevated levels of ethanol, increasing the perception of bitterness, hotness, and decreasing the perception of some wine aromas and flavor attributes (Goldner et al. 2009).

Facing the climate change, human health, and the constant growing demand for full bodied red wines, it is important to anticipate further increase. In this way, several technological approaches have been proposed, to reduce ethanol content in wine (Pickering 2000), ranging mainly from pre-fermentation (selection of grapevine clones and vineyard management) to post-fermentation approaches (spinning cone column, reverse osmosis, etc.), which however could increase the production costs and affect negatively wine quality (Pickering 2000). One the other hand, in recent years, intervening on the yeast ecology during must fermentation is gaining more attention and this is carried out mainly by decreasing the sugar-ethanol yield transformation through the selection of wine yeasts (Contreras et al. 2015a, 2015b; Gobbi et al. 2014; Quirós et al. 2014). For non-Saccharomyces yeasts, the quantity of sugar used to produce 1 % (v/v) of ethanol is higher (17.0–40.0 g/L) (Englezos et al. 2015; Magyar and Tóth 2011) due to their ability to utilize the carbon to produce biomass and by-products. As a consequence, through their metabolism ethanol concentration does not increase (Contreras et al. 2014; Contreras et al. 2015a, 2015b; Gobbi et al. 2014; Gonzalez et al. 2013; Morales et al. 2015; Quirós et al. 2014).

Among, the non-Saccharomyces species of oenological interest, Starmerella bacillaris (synonym Candida zemplinina) (Duarte et al. 2012) is considered as one of the most promising species to achieve the objective described above. S. bacillaris is supposed to be one of the best candidates, due to its ability to produce less ethanol from sugar consumed, tolerate high concentrations of ethanol present in the wine, and produce low levels of biogenic amines (Englezos et al. 2015; Magyar and Tóth 2011; Rantsiou et al. 2012; Suzzi et al. 2012; Tristezza et al. 2013). These phenotypic characteristics support the potential use of this wine yeast, in combination with S. cerevisiae either in co-inoculated or sequential inoculated fermentations to reduce the potential ethanol content in wine (Giaramida et al. 2013; Gonzalez et al. 2013; Masneuf-Pomarede et al. 2015). However, strain selection and establishment of inoculation protocols are essential in order to moderate yeast growth and produce wines with the aspects described above.

In this context, the aim of this study was to understand the appropriate time of S. cerevisiae addition after S. bacillaris inoculation in order to achieve a high level of ethanol reduction. A central composite design (CCD) and response surface methodology (RSM) approach were used for this final goal, in order to optimize and find the appropriate inoculation protocol.

Materials and methods

Yeast strains

Two S. bacillaris (FC54 and C.z 03) and one S. cerevisiae (ScBa49) isolate were obtained from the yeast culture collection of the DISAFA (Dipartimento di Scienze Agrarie, Forestali e Alimentari, University of Torino, Italy). S. bacillaris MUT 5705 came from the Mycotheca Universitatis Taurinensis–MUT (Dipartimento di Scienze della Vita e Biologia dei Sistemi (DBIOS)–University of Torino, Italy), while a commercial S. cerevisiae wine yeast Uvaferm BC® (Lallemand SAS, Montreal, Canada) was used as a reference strain (Table 1). S. bacillaris strains were selected for their physiological and enological performance (Englezos et al. 2015) and routinely cultivated on YPD slants (1 % yeast extract, 2 % bacteriological peptone, 2 % glucose, and 2 % agar, all w/v, all from Biogenetics, Milan, Italy) or stored at −80 °C in YPD broth supplemented with 20 % glycerol (Sigma, Milano, Italy).

Table 1 Strains used in this study
Full size table

Wine fermentations

Laboratory scale fermentations

Grape must of Barbera cultivar (Vitis vinifera L.) without the grape skin was obtained from the experimental winery of the University of Torino. Grape must contained 233.2 g/L sugars (116.4 g/L glucose and 116.8 g/L fructose), titratable acidity 8.20 g/L (expressed as tartaric acid), pH 3.20, and absence of ethanol. The initial yeast available nitrogen (YAN) was 197 mg/L composed by the sum of 116 mg/L of AUG (ammonium + urea + L-arginine) and 81 mg/L of PAN (primary amino nitrogen). The must was pasteurized in a water bath at 60 °C for 1 h, and the sterility was checked by plating 100 μL of must on WL Nutrient agar medium (Biogenetics, Milan, Italy) and incubated them at 28 °C for 5 days. Under sterile conditions, 25 mL of the pasteurized must was distributed onto 50-mL tubes with loose screw cap for all the fermentations performed in this work.

Pure fermentations

The oenological performance of the three S. bacillaris and two S. cerevisiae strains was evaluated by microvinification trials in pure culture fermentations. The inoculum of the five yeast strains was prepared by pre-adaptation of the strains in the same must as described above for 48 h at 25 °C. Afterwards, the yeast cells were stained with methylene blue dye and immediately the viable cell population was counted by using a Thoma hemocytometer chamber (BRAND GMBH + CO KG, Wertheim, Germany). Before inoculation, appropriate amounts of inoculum were calculated and subsequently used to inoculate the musts at an initial cell population of 1.0 × 106 cells/mL. All the fermentations were carried out in duplicate under static conditions at 25 °C for 21 days.

Central composite design

Two factorial CCDs were used to understand the appropriate experimental plan to model the delay of S. cerevisiae inoculation and the sampling time for the chemical analyses during the fermentation period, as previously described by Torchio et al. (2011). A matrix was generated with two factors, delay of S. cerevisiae inoculation (hours), and time of chemical analyses (days of fermentation) at five levels (−α, −1, 0, +1, +α), where α was equal to 1.41 factorial units. The corresponding values were calculated in the decoded matrix based on the limit of the design −α and +α. In this study, it was decided that 0 (co-inoculation) and 48 h (sequential inoculation) delay would be the extreme values of the S. cerevisiae addition and 0 and 21 days for the time of chemical analyses (Table S1 in the Supplementary Material).

After running response surface methodology (RSM), a second-order polynomial regression equation was fitted to the subsequent equation:

$$ Y={\mathrm{b}}_0+{\mathrm{b}}_1{\mathrm{X}}_1+{\mathrm{b}}_2{\mathrm{X}}_2+{\mathrm{b}}_{11}{{\mathrm{X}}_1}^2+{\mathrm{b}}_{22}{{\mathrm{X}}_2}^2+{\mathrm{b}}_{12}{\mathrm{X}}_1{\mathrm{X}}_2 $$

where Y is the predicted response of the dependent variable, X1 (inoculation delay) and X2 (sampling time) are the independent variables that correspond to the response of Y, b0 is the value in the central point conditions, b1 and b2 represent the linear regression coefficient associated with each variable, b11 and b22 are the quadratic regression coefficient of each independent value, while b12 is the regression coefficient of the interaction effect between the two variables. The second-order polynomial equations used to generate the surface curves are presented in Supplementary Material Table S2.

Mixed fermentations

Mixed fermentations were carried out by inoculating the three S. bacillaris and two S. cerevisiae strains in combination, according to the X1 of CCD pattern. Five inoculation strategies were carried out: inoculation of the two species simultaneously (co-inoculation) and addition of the S. cerevisiae at 7, 24, 41, and 48 h after S. bacillaris inoculation (sequential inoculation). In each case, the musts were inoculated with 48 h pre-adapted cultures grown in the same must, with the same cell relative density of 1:1 as described above for the pure cultures. Fermentations were carried out in duplicate under static conditions at 25 °C for 21 days.

Pilot-scale fermentations

Fermentations were performed in 2-hL stainless-steel fermenters with Barbera grape, at the experimental winery of the University of Torino. The must composition was as follows: 250.4 g/L of sugars (126.1 g/L glucose and 124.3 g/L fructose), titratable acidity 10.21 g/L (expressed as tartaric acid), pH 3.09, and total sulfur dioxide 20 mg/L. The initial YAN was 145 mg/L composed by the sum of 55 mg/L of AUG and 90 mg/L of PAN. The best performing couple and inoculation strategy according to the laboratory fermentations were selected for these trials: a pure culture fermentation of S. cerevisiae Uvaferm BC® was used as control and a sequential mixed culture which S. cerevisiae Uvaferm BC® was inoculated with 48 h delay after S. bacillaris MUT 5705 inoculation. Both strains were inoculated with an initial cell population of 1.0 × 106 cells/mL as described above for the laboratory scale trials. Fermentations were performed in duplicate at 25 ± 2 °C. Must was pumped up twice a day and racking was carried out when residual sugars were less than 2 g/L. Malolactic fermentation was carried at 20 °C in stainless steel tanks, by inoculating the commercial Oenococcus oeni Lalvin VP41® strain (Lallemand SAS, Montreal, Canada), according the manufacturer’s instructions. At the end of the malolactic fermentation, wines were clarified, supplemented with 50 mg/L of total SO2, and then bottled and subjected to chemical analysis.

Microbiological and molecular analysis

Samples were collected in duplicate at 0, 1, 2, 4, 7, 14, and 21 days from the beginning of fermentation, serially diluted in Ringer’s solution (Oxoid, Milan, Italy) and plated on WL N medium. Plates were incubated at 28 °C for 5 days and the two types of colonies were differentiated visually as described previously (Rantsiou et al. 2012) and subsequently counted. The enumeration of non-Saccharomyces yeasts in the pilot-scale fermentations was carried out using lysine agar medium (Oxoid, Milan, Italy). Concerning pilot-scale trials, five putative colonies of S. bacillaris and S. cerevisiae from each sampling point (30 for each ferment) were isolated and then subjected to molecular characterization by Rep and interdelta-PCR, as suggested by Englezos et al. (2015) and Charpentier et al. (2009), respectively, in order to understand strain dynamics over the fermentation process.

Chemical analysis

Ethanol, glycerol, acetic acid production, as well as the glucose and fructose consumption were determined by HPLC using an Agilent 1260 Infinity HPLC apparatus (Milford, MA, USA) as described by Rolle et al. (2012). The concentration of total YAN was determined by using two enzymatic kits (Megazyme International, Wicklow, Ireland) following the kit manufacturer instructions. In particular, total YAN concentration was calculated by the sum of ammonium, urea, and L-arginine (AUG) and the concentration of the primary amino nitrogen (PAN).

Statistical analyses

The data obtained from the different inoculation strategies were subjected to one-way analysis of variance (ANOVA) by using the statistical software package IBM SPSS Statistics (version 21.0. IBM Corp., Armonk. NY, USA). The ANOVA analysis was coupled by the Duncan test (p < 0.05), in order to evaluate the significant differences between the data obtained.

The RSM was performed with the statistical software STATISTICA™, program version 10.0 (StatSoft Inc. Tulsa, USA), to evaluate the results obtained by the CCD pattern applied. The regression models were performed only with R 2 values greater than 0.8 indicating that the variability could be explained by the second-order model equations.

Results

Laboratory scale fermentations

Growth dynamics

The growth dynamics of the S. bacillaris and S. cerevisiae strains when inoculated in pure cultures in Barbera must are presented in Fig. 1. The five strains grew similarly and reached a cell population of about 108 colony forming units (CFU)/mL on the second day of fermentation. On the 7th day, viable cell population started to decrease and no S. bacillaris population was observed at the last sampling point (<10 CFU/mL on WLN medium plates), while populations of the S. cerevisiae strains (ScBa49 and Uvaferm BC®) ranged from 106 to 107 CFU/mL, respectively.

Fig. 1
figure 1

Growth dynamics of S. bacillaris (a) and S. cerevisiae (b) strains in pure cultures. S. bacillaris strains: FC54 (white circle), MUT 5705 (white diamond), and C.z 03 (white square) and S. cerevisiae strains: ScBa49 (black diamond) and Uvaferm BC® (white diamond). Counts are the mean CFU/mL values ± standard deviations of two independent experiments

Full size image

In Fig. 2, the growth dynamics of the mixed fermentations with S. cerevisiae ScBa49 are illustrated. In the co-inoculated and sequential fermentation (7 h delay), all S. bacillaris and S. cerevisiae couples showed comparable growth dynamics, reaching a population of 107 to 108 CFU/mL in 2 days. Through the rest of the fermentation, S. cerevisiae ScBa49 maintained these counts, while S. bacillaris populations started to decrease. When the other three inoculation strategies (24, 41, and 48 h delay) were carried out, all S. bacillaris strains used in this study competed with S. cerevisiae ScBa49 during the first 7 days of fermentation. After this day, S. bacillaris started to decrease and the population became undetectable (<10 CFU/mL on WLN medium plates) after 14 (24-h delay) and 21 days (41 and 48 h delay), while the viability of the S. cerevisiae cells remained stable at 106 to 107 CFU/mL throughout the whole fermentation process.

Fig. 2
figure 2

Growth dynamics of mixed fermentations performed with the three S. bacillaris strains and the S. cerevisiae strain ScBa49. S. bacillaris/S. cerevisiae couples: FC54 (white circle), ScBa49 (black circle), MUT 5705 (white diamond), and ScBa49 (black diamond), C.z 03 (white square), and ScBa49 (black square). Counts are the mean CFU/mL values ± standard deviations of two independent experiments. The arrow indicates the S. cerevisiae inoculation

Full size image

The population dynamics for the mixed fermentations with S. cerevisiae Uvaferm BC® are shown in Fig. 3. Population dynamics within the first 4 days of fermentation, for the co-inoculated and sequential inoculated trials (7-h delay), were comparable to those observed for S. cerevisiae ScBa49. Afterwards, the population of S. cerevisiae Uvaferm BC® started to decrease, with counts that ranged from 105 to 106 CFU/mL at the end of the monitored period (21 days), in contrast with the S. cerevisiae ScBa49 population that remained stable during all the monitored period. Interestingly, sequential inoculations with the highest delay (24, 41, and 48 h) showed similar population dynamics as for the couples previously tested.

Fig. 3
figure 3

Growth dynamics of mixed fermentations performed with the three S. bacillaris strains and the S. cerevisiae strain Uvaferm BC®. S. bacillaris/S. cerevisiae couples: FC54 (white circle) and Uvaferm BC® (black circle), MUT 5705 (white diamond) and Uvaferm BC® (black diamond), C.z 03 (white square) and Uvaferm BC® (black square). Counts are the mean CFU/mL values ± standard deviations of two independent experiments. The arrow indicates the S. cerevisiae inoculation

Full size image

Chemical composition of the wines

The mean concentration of sugars, glycerol, organic acids, and ethanol in the must and wines obtained from the pure cultures of S. bacillaris and S. cerevisiae, after 21 days of fermentation, is presented in Table 2. S. bacillaris in pure cultures produced partially fermented wines with significant presence of residual sugars (up to 32.6 g/L glucose), while the fructose was totally consumed (<1.0 g/L). Complete fermentation of the sugars was observed only for S. cerevisiae strains in pure cultures. The chemical composition of these wines was characterized from the presence of lower levels of glycerol (7.8–8.3 g/L) and higher levels of ethanol (13.8–14.0 %) (v/v) compared to S. bacillaris pure cultures. Compared to wines produced with S. cerevisiae in pure culture, wines fermented with S. bacillaris presented significantly higher glycerol yields and lower potential ethanol concentrations.

Table 2 Concentration of sugars, glycerol, organic acids and ethanol in the must and wines obtained from pure fermentations of S. bacillaris and S. cerevisiae strains
Full size table

The chemical composition of the wines produced from mixed fermentations carried out with S. cerevisiae strains ScBa49 and Uvaferm BC® are presented in Tables 3 and 4, respectively. In mixed fermentations with ScBa49, the five different inoculation protocols resulted in a different consumption of sugars. As it can be seen, inoculation delay up to 7 h always allowed consumption of all sugars (< 2.9 g/L) from the must at day 21, regardless of the S. bacillaris strain used. On the contrary, inoculation delays of 24, 41, and 48 h always performed poorly, leaving significant higher quantities of sugars (14.1–27.6 g/L), mainly glucose. A different behavior was observed for the Uvaferm BC® commercial strain, since all couples fermented all the sugars from the must (< 3.9 g/L) after 21 days from the beginning of fermentation, independently of the inoculation delay applied.

Table 3 Concentration of sugars, glycerol, organic acids, and ethanol in the wines obtained from mixed fermentations using the S. cerevisiae strain ScBa49
Full size table
Table 4 Mean concentration of sugars, glycerol, organic acids, and ethanol in the wines obtained from mixed fermentations with the S. cerevisiae strain Uvaferm BC®
Full size table

The glycerol production was also influenced by time of S. cerevisiae addition and the sampling time (Table 4). Compared to wines produced by S. cerevisiae Uvaferm BC® in pure culture, wines produced by mixed yeast species contained more glycerol. In particular, with a delay of 48 h, the glycerol content of the wines increased up to 4.2 g/L (Table 4). The modeling of glycerol production with RSM reflects that its increase is correlated with the increase of the inoculation delay of S. cerevisiae (Fig. 4, left panel). The production of this metabolite was linearly increased when S. cerevisiae yeast strain Uvaferm BC® was inoculated in the first 24 h (R 2 = 0.985) after S. bacillaris addition. Conversely, minor differences were found by increasing the sequential inoculation between 24 and 48 h. Acetic acid production was also influenced by the inoculation strategy, however, all the couples tested in this study maintained values at levels lower than 0.50 g/L.

Regarding ethanol production, in sequentially inoculated fermentations, only the couple S. bacillaris MUT 5705 and S. cerevisiae Uvaferm BC® produced wine with 0.7 % (v/v) less ethanol compared to S. cerevisiae Uvaferm BC® in pure culture (Table 4). In this case, the must was initially inoculated with S. bacillaris followed by S. cerevisiae after 24, 41, and 48 h. Additionally, an interesting observation was the increasing quantity of sugars (g/L) consumed by these couples to produce 1 % (v/v) of ethanol, with the increase of inoculation delay. This was particularly evident when the inoculation delay moved to 48 h, highlighting the contribution of S. bacillaris to the ethanol reduction. For all the couples tested, the shape of the surface curves (Fig. 4, right panel) also confirmed this trend, indicating a significant linear decrease of the ethanol content when S. cerevisiae was inoculated with a delay of minimum 4 h. However, this decrease was improved with the inoculation delay, with a maximum value at 48 h (maximum monitored). Taking in consideration these findings and the results from growth dynamics, we hypothesized that the most suitable protocol able to reduce the ethanol at industrial scale could be the sequential inoculation with 48 h delay. This is also in line with previous studies, in which indigenous S. cerevisiae strains started to grow after 2 days from S. bacillaris inoculation (Giaramida et al. 2013). Extended delays were not tested because considered not applicable in real wine-making settings.

Fig. 4
figure 4

Response surface curves fitted to experimental data points corresponding to the glycerol (g/L) (left panel) and ethanol (% v/v) (right panel) production as a function of S. cerevisiae inoculation and time of the chemical analyses. S. bacillaris/S. cerevisiae couples: FC54 and Uvaferm BC® (a), MUT 5705 and Uvaferm BC® (b), and C.z 03 and Uvaferm BC® (c)

Full size image

Pilot-scale fermentations

In order to validate the results obtained at laboratory scale, the best performing couple (MUT 5705 and Uvaferm BC®) and the inoculation strategy of 48 h delay were selected to ferment Barbera must at pilot-scale fermentations at 25 °C. Pure fermentation with Uvaferm BC® was used as control. Both inoculation rates and procedures were as close as possible to laboratory scale fermentations, in order to obtain a better reproducibility. Molecular typing of S. bacillaris and S. cerevisiae isolates by Rep and interdelta-PCR amplification, respectively, revealed that these fermentations were guided by the inoculated strains (data not shown). The cell population of Uvaferm BC® in pure culture reached a concentration of about 108 CFU/mL on the second day, which was maintained to these levels during the whole fermentation period (Fig. S1 in the Supplemental Material, panel a). Indigenous non-Saccharomyces yeasts were detected at concentration of 105–106 CFU/mL during the first 2 days, after which they rapidly decreased to undetectable levels (<10 colonies on lysine medium) on the 4th day. In addition, wild S. bacillaris strains, which were determined to be different from the inoculated MUT 5706 by Rep-PCR profiling, were found in the must at concentrations up to 106 CFU/mL during the first 4 days, after this point a remarkable decrease of cell population was observed. When S. cerevisiae Uvaferm BC® was inoculated with 48 h delay after S. bacillaris MUT 5705 inoculation (Fig. S1 in the Supplemental Material, panel b), Uvaferm BC® cell population was affected by MUT 5705 and it was not able to reach counts of 108 CFU/mL. This allowed MUT 5705 to maintain relative high cell population (about 106–107 CFU/mL) until the 7th day of fermentation. The same pattern was seen for the non-Saccharomyces during the first 4 days, afterwards rapidly decreased to undetectable levels.

The chemical composition of the wines produced from the pilot-scale fermentation is given in Table 5. Sequential fermentations consumed sugars slower than Uvaferm BC® in pure culture (10 vs. 7 days). The wine produced from sequential inoculation contained significantly more glycerol (13.4 vs. 12.0 g/L) than Uvaferm BC®, while the ethanol content was reduced by 0.5 % (v/v). On the contrary, acetic acid production after malolactic fermentation for the sequential inoculation was reduced compared to that observed for Uvaferm BC® (0.34 vs. 0.47 g/L). A significant increase of 0.5 in total acidity was seen for the sequentially inoculated wine, with a parallel decrease of pH.

Table 5 Mean concentration of sugars, glycerol, organic acids, and ethanol in the wines produced from pure (Uvaferm BC®, control) and sequential (MUT 5705 and Uvaferm BC®) culture fermentations at pilot scale (2hL)
Full size table

Discussion

One possible approach to reduce the ethanol content of wines is by fine-tuning yeast ecology during must fermentation. The selection and use of non-Saccharomyces wine yeasts can potentially lead to a reduction of the overall sugar-ethanol yield during alcoholic fermentation. In the study presented here, the possibility of using S. bacillaris in combination with S. cerevisiae was investigated in order to regulate the ethanol production in a must with a high initial sugar concentration (233.2 g/L). In order to find the appropriate time of S. cerevisiae addition after S. bacillaris inoculation, a CCD approach was selected to model the chemical composition of the wine produced with particular attention on the ethanol and glycerol concentration. With RSM, several combinations of S. bacillaris/S. cerevisiae at different inoculation times were tested simultaneously with a small number of experiments able to generate large amounts of information, according to the CCD experiment plan, which permits to uncover interactions between variables (Bezerra et al. 2008).

In pure fermentations, S. bacillaris strains produced wines with residual sugars, composed exclusively by glucose, confirming the preference of this species for fructose (Englezos et al. 2015; Magyar and Tóth 2011). The impact of this non-Saccharomyces yeast on the chemical composition of the wine was evident with a higher production of glycerol, in agreement with previous studies (Englezos et al. 2015; Magyar and Tóth 2011; Magyar et al. 2014), higher amounts of sugars used to produce 1 % of alcohol, and slightly higher production of acetic acid compared to the S. cerevisiae strains, in accordance with previous studies (Sadoudi et al. 2012; Soden et al. 2000).

When mixed fermentation trials were performed, the co-inoculation of the two species did not show a significant reduction of ethanol content and the chemical composition of these wines was very similar to the control wines produced by the S. cerevisiae strains in pure cultures. This behavior is confirming the high competitive ability of S. cerevisiae over non-Saccharomyces yeast cells, probably due to the depletion of nutrients present in the must, cell-to-cell contact-mediated mechanisms, or due to the production of toxic metabolites (Andorrà et al. 2010; Nissen et al. 2003; Pérez-Nevado et al. 2006), and underlines the need to understand better the mechanism of this co-habitation.

In this context, the early growth of S. bacillaris in the sequential inoculations with the highest delays (24, 41, and 48 h) limited the subsequent growth of the two S. cerevisiae strains. One possible explanation for this behavior is that S. bacillaris decreased the nutrient concentration by subtracting large quantities of organic nitrogen from the must (data not shown) (Andorrà et al. 2010; Medina et al. 2012). Indeed, since only the Uvaferm BC® commercial strain totally consumed the sugars in these fermentations, it can be hypothesized that this strain has probably lower demands in nutrients (e.g., nitrogen) compared to ScBa49 wild strain. These results suggest that nutrient concentration and strain selection have a fundamental role on the fermentation rate of the mixed fermentations with S. bacillaris and S. cerevisiae.

Sequential fermentations performed with the strain Uvaferm BC® changed positively the chemical composition of the wines produced, especially in terms of glycerol. Glycerol production was influenced by the time of S. cerevisiae addition and the sampling time (Table 4, Fig. 4 left panel ). The higher concentration of glycerol is in agreement with previous studies (Giaramida et al. 2013; Suzzi et al. 2012).

Interestingly, the inoculation delay changed dramatically the sugar to ethanol conversion rate of alcoholic fermentation. More specifically, in the sequentially inoculated fermentations, yeasts consumed more sugars to produce 1.0 % (v/v) of ethanol, compared to S. cerevisiae Uvaferm BC® in pure culture highlighting the impact of S. bacillaris for ethanol reduction (Bely et al. 2013; Englezos et al. 2015; Giaramida et al. 2013). However, the results revealed that only the couple MUT 5705 and Uvaferm BC® sequentially inoculated with a minimum of 24-h delay was able to consume up to 17.5 g/L of sugars to produce 1.0 % of ethanol, while the official European Economic Community (EEC) ethanol conversion factor is 16.83 g/L (Ribéreau Gayon et al. 2006). The coefficient of determination (R 2) was 0.88 indicating a good correlation between the inoculation delay and fermentation efficiency (g/L of sugar used for 1 % v/v ethanol production).

These results let us to hypothesize that S. bacillaris diverts carbon derived from the glycolytic pathway away from ethanol production to the synthesis of biomass and production of by-products, in order to maintain intracellular NADH/NAD+ redox balance ensuring continuous operation of the metabolic processes. These products include glycerol, monocarboxylic (acetic acid and pyruvic acid), dicarboxylic (succinic acid and α-ketoglutaric acid) and tricarboxylic acids (citric acid and isocitric acid), and aroma volatile compounds (van Dijken and Scheffers 1986). The overproduction of titratable acidity observed in the pilot-scale sequential fermentation could result from the swift of carbon flux towards organic acid production, since S. bacillaris is considered high producer of α-ketoglutaric acid and pyruvic acid (Magyar et al. 2014; Mangani et al. 2011).

In order to confirm laboratory scale fermentations, the best performing couple (MUT 5705 and Uvaferm BC®) and inoculation strategy (48 h delay) were used to ferment Barbera must in pilot-scale fermentations. The presence and dominance of the inoculated yeast strains were confirmed using Rep-PCR and interdelta-PCR amplification for the S. bacillaris and S. cerevisiae, in order to exclude contributions of indigenous strains. Pilot-scale results confirmed the findings observed in laboratory settings, with the only exception of acetic acid production, which was registered to be higher in pure culture fermentation. The effect of MUT 5705 on wine composition was apparent. As in the laboratory scale fermentation, production of glycerol was higher in the sequential trial than in pure culture fermentation, while ethanol production showed a significant reduction. The pH was also lower and the titratable acidity higher for wine produced from sequential than this produced from pure culture fermentation.

In summary, this study presents a fermentation protocol tested under both laboratory and pilot-scale conditions to reduce ethanol levels in wines. This protocol is based on the inoculation of the grape must with S. bacillaris MUT 5705 and S. cerevisiae Uvaferm BC® after 48 h from the beginning of the fermentation. Furthermore, the exploitation of this inoculation protocol could be further investigated using other varieties of grape musts, in order to understand the impact of the co-habitation of these species to wine composition in terms of aroma and flavor. In the future, the decrease of ethanol as described here could help winemakers to decide the appropriate time to harvest their grapes, without the risk of excessive sugar content, which can be converted in high levels of ethanol in wine.