Torulaspora delbrueckii May Help Manage Total and Volatile Acidity of Santorini-Assyrtiko Wine in View of Global Warming

Non-Saccharomyces (NS) yeasts are gaining popularity in modern winemaking for improving wine quality. Climate change is one of the biggest challenges winegrowing now faces in warm regions. Here, Lachancea thermotolerans LtS1 and Torulaspora delbrueckii TdS6 combined with Saccharomyces cerevisiae ScS13 isolated from Assyrtiko grapes from Santorini island were evaluated in grape must fermentation with the aim to mitigate major consequences of temperature rise. Different inoculation protocols were evaluated, including simultaneous and sequential mixed-strain inoculations, displaying significant variation in the chemical and kinetic characteristics. Both LtS1 and TdS6 could raise the titratable acidity (TA). TdS6 also reduced the volatile acidity (VA) and was thus chosen for further evaluation in microvinifications and pilot-scale fermentations. Consistent with lab-scale trials, sequential inoculation exhibited the longest persistence of TdS6 resulting in minimum VA levels. Diethyl succinate, ethyl propanoate, and ethyl isobutyrate were significantly increased in sequential inoculations, although a decline in the net total ester content was observed. On the other hand, significantly higher levels of TA, succinic acid, and 2-methylpropanoic were associated with sequential inoculation. The overall performance of TdS6 coupled with a high compatibility with S. cerevisiae suggests its use in the fermentation of Santorini-Assyrtiko or other high sugar musts for the production of structured dry or sweet wines.


Introduction
The yeasts residing on grape berries are of high importance in winemaking. The majority of these vine-associated populations are brought into the winery along with the harvested grapes, while winery yeasts may find their way back to the vineyard [1]. This dynamic yeast consortium of the grape-wine biological system contributes to wine quality, and has been recently recognized as a potential contributor to the terroir expression of wines. Bokulich et al. [2] by applying metagenomics targeting the whole fungal community were among the first to demonstrate that geographical influence might drive regional grapeassociated microbial biodiversity. Recently, Chalvantzi et al. [3] focusing exclusively on the wine yeast populations, showed that different viticultural appellations may indeed possess distinct species diversity characterized by particular enological features. These results endorse the plausible effect of indigenous yeasts on the regional typicity of wines, the socalled microbial terroir effect. However, the addition of commercial Saccharomyces cerevisiae

Pilot-Scale Fermentations
Pilot scale fermentations were performed in duplicate (triplicate in IS) in 220-L tanks filled with 170 L of Assyrtiko-Santorini must (sugars 221 g/L; pH 3.13; titratable acidity 6.35 g/L, as tartaric acid; 50 ppm total SO 2 ; initial yeast assimilable nitrogen 100.8 mg/L) at 18 • C. Forty g/hL of organic nitrogen was added into the grape must prior to yeast inoculation and another 40 g/hL of nitrogen in both organic and inorganic form were added at 50 g/L sugar depletion. Yeasts were grown on yeast extract peptone dextrose (YPD) agar at 26 • C and then resuspended in 1/4 strength Ringer's solution. Yeasts were added at 6 Log CFU/mL in fermenting musts of the IS, SMT and SQT inoculation protocols, as above. Duplicate spontaneous fermentations (SP) were also performed for comparison reasons. Fermentation courses were monitored by density measurements.

Microbiological Analysis
Total yeasts, S. cerevisiae, and non-Saccharomyces kinetics were daily followed by plating on Wallerstein laboratory nutrient agar (WL), ethanol sulfite agar (ESA) and yeast extract-peptone-dextrose agar (YPD), and lysine medium agar (LA), respectively. Colonies were enumerated after incubation of plates at 28 • C (37 • C for YPD plates) for 2-5 days. T. delbrueckii and S. cerevisiae colonies recovered from the initial, middle and final stages of non-sterile fermentations were genotyped as described above.

Chemical Analysis
Reducing sugars, total acidity, volatile acidity, pH, total SO 2 and free SO 2 were estimated as described in the Compendium of International Methods of Analysis of Musts and Wines [21]. The formol method [22] was applied for the estimation of yeast assimilable nitrogen (YAN). Organic acids (citric acid, tartaric acid, malic acid, succinic acid, lactic acid, and acetic acid), sugars (glucose, fructose), glycerol and ethanol were determined by HPLC as previously described [23]. The major volatile compounds (acetaldehyde, ethyl acetate, methanol, 1-propanol, 2-methyl-1-propanol [Isobutanol], 3-and 2-methyl-1butanol) were determined by direct injection of wines into a gas chromatograph according to Nisiotou et al. [23]. The minor volatiles were determined by a headspace SPME/GC-MS method, as described previously [24] with minor modifications [23]. Peaks were quantified relative to internal standards using the peak area of an extracted ion.

Statistical Analysis
Analysis of Variance (ANOVA) and Tukey's HSD tests were used to determine significant differences between the chemical profiles of wines produced by different inoculation protocols. To investigate the correlations between samples and variables, principal component analysis (PCA) was performed to chemical parameters using the software PRIMER Version 7 (www.primer-e.com, accessed on 30 October 2022). The permutational multivariate analysis of variance (PERMANOVA) was applied to compare between different inoculation protocols. Pairwise distances between inoculation protocols were calculated using the Jaccard metric, and F-statistics were computed using a random sample of 4999 permutations. All tests were conducted with the PAST software (version 3.11) [25].

Fermentation Kinetics and Yeast Population Dynamics in Mixed Lab-Scale Fermentations
The performance of T. delbrueckii TdS6, L. thermotolerans LtS1 and S. cerevisiae ScS13 strains was examined in small-scale fermentations of sterile Assyrtiko grape must under different inoculation schemes, i.e., simultaneous inoculations of ScS13 with TdS6 (SMT) or LtS1 (SML) and sequential inoculations (SQ) of TdS6 (SQT) or LtS1 (SQL) followed by ScS13. Single inoculations with ScS13 (IS) or the commercial S. cerevisiae Lalvin QA23 (CS) were also performed as control fermentations. Fermentation kinetics of the various inoculation protocols are shown in Figure 1. The rate of fermentation was lower in SQ than in SM or single-strain inoculations, resulting in a longer fermentation process in SQ (12.4 d in SQL and 13.6 d in SQT) compared to the other inoculation schemes (ca. 9 d). The kinetics of S. cerevisiae ScS13 were not influenced by the simultaneous addition of either non-Saccharomyces (NS) strain ( Figure 1). ScS13 reached a plateau at day 2 of fermentation with an average maximum population value (Log CFU/mL) of 7.91 ± 0.09 in IS, 7.95 ± 0.12 in SMT, and 7.89 ± 0.09 in SML. As opposed to this, the kinetics of ScS13 were altered in sequential inoculations. The maximum population of ScS13 in SQ was ca. 0.5 Log CFU/mL lower than in SM or single inoculations, corresponding to maximum average Log CFU/mL values of 7.30 ± 0.09 in SQT and 7.35 ± 0.08 in SQL.
As opposed to ScS13, the NS populations reached higher counts in SQ than in the SM fermentation schemes. More specifically, the maximum population of Tds6 was 0.53 Log CFU/mL higher in SQT than in SMT (i.e., 7.72 ± 0.07 vs. 7.18 ± 0.04, respectively), while the LtS1 population was 0.94 Log CFU/mL higher in SQL than in SML (i.e., 7.98 ± 0.01 vs. 7.04 ± 0.10, respectively). Both NS strains persisted much longer in SQ than in SM fermentations before starting to decline (i.e., 10.8 d in SQT vs. 2.3 d in SMT and 7.0 d in SQL vs. 2.0 d in SML), with Tds6 persisting longer than LtS1 by 3.8 d. Similar death rates were observed for the two NS strains under either SM or SQ inoculation protocol. The populations of both TdS6 and LtS1 declined upon the release of ca. 60 g CO 2 (corresponding to 8.0% vol ethanol) and 88 g CO 2 (11.7% vol ethanol), respectively. As opposed to ScS13, the NS populations reached higher counts in SQ than in the SM fermentation schemes. More specifically, the maximum population of Tds6 was 0.53 Log CFU/mL higher in SQT than in SMT (i.e., 7.72 ± 0.07 vs. 7.18 ± 0.04, respectively), while the LtS1 population was 0.94 Log CFU/mL higher in SQL than in SML (i.e., 7.98 ± 0.01 vs. 7.04 ± 0.10, respectively). Both NS strains persisted much longer in SQ than in SM fermentations before starting to decline (i.e., 10.8 d in SQT vs. 2.3 d in SMT and 7.0 d in SQL vs. 2.0 d in SML), with Tds6 persisting longer than LtS1 by 3.8 d. Similar death rates were observed for the two NS strains under either SM or SQ inoculation protocol. The populations of both TdS6 and LtS1 declined upon the release of ca. 60 g CO2 (corresponding to 8.0% vol ethanol) and 88 g CO2 (11.7% vol ethanol), respectively.

Chemical Profile of Laboratory-Scale Ferments
The chemical characteristics and major volatiles of the ferments produced at laboratory-scale fermentations of sterile must are shown in Tables 1 and 2, respectively. Significant differences were detected in the chemical characteristics of various ferments. The total acidity was significantly higher in SQL compared to other ferments. Concomitantly, SQL exhibited the lowest pH value accompanied by a considerable production of lactic acid. Significant differences were also detected in the levels of volatile acidity, with CS producing the highest levels. Interestingly, the prolonged activity of NS yeasts in SQ

Chemical Profile of Laboratory-Scale Ferments
The chemical characteristics and major volatiles of the ferments produced at laboratoryscale fermentations of sterile must are shown in Tables 1 and 2, respectively. Significant differences were detected in the chemical characteristics of various ferments. The total acidity was significantly higher in SQL compared to other ferments. Concomitantly, SQL exhibited the lowest pH value accompanied by a considerable production of lactic acid. Significant differences were also detected in the levels of volatile acidity, with CS producing the highest levels. Interestingly, the prolonged activity of NS yeasts in SQ inoculations caused a significant drop in volatile acidity, with SQT exhibiting the lowest values. The ferments also showed high variation in major volatiles. Acetaldehyde, ethyl acetate, propanol and isobutanol levels were elevated in SQ fermentations. LtS1 was associated with the highest production of ethyl acetate and propanol. On the other hand, 3-methyl-1-butanol was lower in SQ-than in SM-inoculated ferments. Total acidity (as tartaric acid g/L) 9.9 ± 0.1 b 9.9 ± 0.1 b 10.0 ± 0.0 b 11.6 ± 0.1 a 9.9 ± 0.0 b 9.4 ± 0.1 c pH 3.23 ± 0.01 b 3.25 ± 0.01 a 3.20 ± 0.01 c 3.16 ± 0.01 d 3.26 ± 0.01 a 3.20 ± 0.01 c Volatile acidity (as acetic acid g/L) 0.43 ± 0.01 a 0.34 ± 0.01 b 0.39 ± 0.02 c 0.18 ± 0.01 d 0.30 ± 0.01 e 0.12 ± 0.01 f Free SO2 (mg/L) 11.5 ± 2.5 a 9.7 ± 1.9 ab 11.4 ± 1.4 a 9.8 ± 0.7 ab 9.0 ± 0.0 ab 6.8 ± 2.0 b Total SO2 (mg/L) 24 Overall, the chemical profiles of the ferments were affected by the inoculation protocol (PERMANOVA F = 13.2, p < 0.01). Higher F values were observed between SQ and other ferments, exhibiting their high differentiation compared to other treatments. A Principal Component Analysis (PCA) (Figure 2) was applied to depict the similarities between the chemical profiles of the different ferments. The PCA graph represented 54.5% of total variation on PC1and 15.8% on PC2.
CS, IS, SML and SMT ferments were in close proximity, showing high levels of volatile acidity, acetic acid and SO 2 . SQT and SQL formed two well separated groups distantly located on the negative side of the PC1 axis. SQL was differentiated by the highly negative values of lactic acid, total acidity, glycerol and propanol on both PC1 and PC2 axis. SQT showed high positive values on PC2 for succinic acid, acetaldehyde and isobutanol.
Overall, the chemical profiles of the ferments were affected by the inoculation protocol (PERMANOVA F = 13.2, p < 0.01). Higher F values were observed between SQ and other ferments, exhibiting their high differentiation compared to other treatments. A Principal Component Analysis (PCA) (Figure 2) was applied to depict the similarities between the chemical profiles of the different ferments. The PCA graph represented 54.5% of total variation on PC1and 15.8% on PC2. CS, IS, SML and SMT ferments were in close proximity, showing high levels of volatile acidity, acetic acid and SO2. SQT and SQL formed two well separated groups distantly located on the negative side of the PC1 axis. SQL was differentiated by the highly negative values of lactic acid, total acidity, glycerol and propanol on both PC1 and PC2 axis. SQT showed high positive values on PC2 for succinic acid, acetaldehyde and isobutanol.

Fermentation Kinetics and Yeast Population Dynamics in Mixed Microvinifications
Based on the above results (i.e., the longer persistence of TdS6 than LtS1 in sequential fermentations and its potential to reduce volatile acidity and acetic acid levels) the strain TdS6 was selected to be used in mini-vinifications of Assyrtiko must along with ScS13 ( Figure 3). Similar fermentation kinetics were recorded for the CS (QA23), IS (ScS13), and SM (TdS6 and ScS13) inoculation schemes, where the course was completed by day 9. The fermentation rate in the SQ protocol (TdS6 followed by ScS13) was much lower than the other schemes showing a 6-day delay in the completion of the course. The fermentation rate in spontaneous fermentations (SP) was comparable to SQ until day 11, after which it slowed down leaving residual sugars at the end of the course (Table 2). SP lasted up to day 20, mainly due to its prolonged lag phase. ScS13 showed identical kinetic behavior to the commercial S. cerevisiae starter (QA23), with maximum population rising up to 8.01 Log CFU/mL for either strain. The kinetics of ScS13 were unaltered in the SMT scheme, but its maximum population count was lessened by 0.5 Log CFU/mL in SQT ( Figure  3C,D). The indigenous S. cerevisiae reached higher population levels in SP than in SQ (7.80 ± 0.13 Log CFU/mL) ( Figure 3E). ScS13 was the only S. cerevisiae strain that was recovered from IS, SM and SQ fermentations, while it was not encountered in SP.

Fermentation Kinetics and Yeast Population Dynamics in Mixed Microvinifications
Based on the above results (i.e., the longer persistence of TdS6 than LtS1 in sequential fermentations and its potential to reduce volatile acidity and acetic acid levels) the strain TdS6 was selected to be used in mini-vinifications of Assyrtiko must along with ScS13 ( Figure 3). Similar fermentation kinetics were recorded for the CS (QA23), IS (ScS13), and SM (TdS6 and ScS13) inoculation schemes, where the course was completed by day 9. The fermentation rate in the SQ protocol (TdS6 followed by ScS13) was much lower than the other schemes showing a 6-day delay in the completion of the course. The fermentation rate in spontaneous fermentations (SP) was comparable to SQ until day 11, after which it slowed down leaving residual sugars at the end of the course ( Table 2). SP lasted up to day 20, mainly due to its prolonged lag phase. ScS13 showed identical kinetic behavior to the commercial S. cerevisiae starter (QA23), with maximum population rising up to 8.01 Log CFU/mL for either strain. The kinetics of ScS13 were unaltered in the SMT scheme, but its maximum population count was lessened by 0.5 Log CFU/mL in SQT ( Figure 3C,D). The indigenous S. cerevisiae reached higher population levels in SP than in SQ (7.80 ± 0.13 Log CFU/mL) ( Figure 3E). ScS13 was the only S. cerevisiae strain that was recovered from IS, SM and SQ fermentations, while it was not encountered in SP.
The initial population size of the indigenous NS yeasts in grape must was 2.01 ± 0.02 Log CFU/mL, reaching 6.52 ± 0.57 Log CFU/mL in SP by day 7, while quickly diminishing in CS and IS ( Figure 3). In contrast, TdS6 dominated in SM and SQ at maximum population counts of 7.31 ± 0.04 and 7.93 ± 0.12 Log CFU/mL, respectively. Much longer persistence of TdS6 was recorded in SQ (11 d) than in SM (3 d) before its population began to drop. At this point, TdS6 was recovered at 100% frequency from both SM and SQ inoculation schemes. The initial population size of the indigenous NS yeasts in grape must was 2.01 ± 0.02 Log CFU/mL, reaching 6.52 ± 0.57 Log CFU/mL in SP by day 7, while quickly diminishing in CS and IS ( Figure 3). In contrast, TdS6 dominated in SM and SQ at maximum population counts of 7.31 ± 0.04 and 7.93 ± 0.12 Log CFU/mL, respectively. Much longer persistence of TdS6 was recorded in SQ (11 d) than in SM (3 d) before its population began to drop. At this point, TdS6 was recovered at 100% frequency from both SM and SQ inoculation schemes.

Chemical Profile of Wines Produced in Microvinifications
The chemical profiles of wines produced in microvinifications are shown in Table 3. The chemical characteristics of the various wines showed significant variations (Table 3). Interestingly, SQT and SP exhibited much higher TA values than the other ferments. The addition of Tds6 significantly reduced the levels of volatile acidity (VA), especially in the SQT inoculation scheme, as compared to IS. In contrast, significantly higher levels of citric acid and succinic acid were recorded in SQT compared to other inoculation schemes.

Chemical Profile of Wines Produced in Microvinifications
The chemical profiles of wines produced in microvinifications are shown in Table 3. The chemical characteristics of the various wines showed significant variations (Table 3). Interestingly, SQT and SP exhibited much higher TA values than the other ferments. The addition of Tds6 significantly reduced the levels of volatile acidity (VA), especially in the SQT inoculation scheme, as compared to IS. In contrast, significantly higher levels of citric acid and succinic acid were recorded in SQT compared to other inoculation schemes.
Further differences were observed in the volatile profiles of ferments ( Table 4). The inoculation of S. cerevisiae at high counts at the onset of fermentation was associated with higher levels of esters as compared to either the SQT or SP ferment. In particular, SQT was associated with the lowest levels of esters recorded, diminishing the concentration of ethyl acetate, isoamyl acetate and ethyl octanoate. SQT ferment also exhibited the lowest amount of acetic acid, hexanoic acid and octanoic acid. In contrast, a striking increase was recorded in 2-methyl propanoic acid. Increased levels of linalool, b-citronellol and phenylethyl alcohol were also associated with the SQT inoculation scheme. The highest amounts of terpenes were observed in IS and the lowest respective concentrations in SP.  71 ± 6 a 57 ± 9 ab 53 ± 10 ab 29 ± 2 c 49 ± 8 bc 1-Heptanol 11 ± 0 b 10 ± 0 b 10 ± 0 b 11 ± 3 b 28 ± 6 a 1-Octanol 1437 ± 76 a 1257 ± 32 ab 1207 ± 85 ab 827 ± 81 c 1040 ± 135 bc 1-Octen-3-ol 0 ± 1 a 1 ± 1 a 1 ± 0 a 1 ± 1 a 1 ± 0 a 1-Decanol 2 ± 0 ab 2 ± 0 ab 2 ± 0 b 2 ± 0 a 2 ± 0 ab 1- PERMANOVA test indicated significant differences between the inoculation schemes (F = 25.75, p < 0.001). SQT ferments exhibited the highest F values in pairwise comparisons with other ferments, revealing high differentiation of SQT ferments from all the other inoculation schemes. The differences between the various inoculation schemes were better visualized on a PCA graph (Figure 4; 50.9% of total variation on PC1 and 21.6% on PC2). CS, IS and SMT ferments were grouped closely to each other, showing positive values on PC1 for numerous characteristics, including damascenone, ethyl acetate, isoamyl acetate and ethyl octanoate. In contrast, SQT and SP were placed on the negative side of the PC1 axis. SQL was well separated from all the other ferments due to highly positive values of 2-methylbutanoic acid, succinic acid and linalool on PC2. SP ferments showed highly negative scores for 2-methyl-1-propanol, methyl esters and residual sugars on both the PC1 and PC2 axes.

Fermentation Kinetics and Yeast Population Dynamics in Mixed Pilot-Scale Fermentations
Fermentation kinetics and yeast population dynamics in pilot-scale fermentations with the strains ScS13 and TdS6 are depicted in Figure 5. Similar fermentation dynamics were observed in IS and SMT inoculation schemes, which exhibited the highest fermentation rates resulting in the fastest completion of the course 2 days earlier than in the other treatments. SP showed a lower fermentation rate compared to either IS or SMT but higher than SQT. The kinetic behavior of yeasts varied among the different inoculation schemes. In both IS and SM fermentations, ScS13 followed comparable kinetics, reaching maximum population densities of 8.06 ± 0.08 and 8.05 ± 0.03 Log CFU/mL, respectively ( Figure 5A,C). The maximum population count reached by ScS13 was lower in SQT (7.79 ± 0.24 Log CFU/mL) and SP (7.79 ± 0.19 Log CFU/mL) compared to those in IS and SMT fermentations ( Figure 5B,D). ScS13 prevailed in both IS and SM fermentations at percentages of 100% and 86-100%, respectively. The respective percentages were lower in SQT fermentations (23-35%) in which two indigenous yeast populations also evolved. The ScS13 strain was not detected in SP fermentations.

Fermentation Kinetics and Yeast Population Dynamics in Mixed Pilot-Scale Fermentations
Fermentation kinetics and yeast population dynamics in pilot-scale fermentations with the strains ScS13 and TdS6 are depicted in Figure 5. Similar fermentation dynamics were observed in IS and SMT inoculation schemes, which exhibited the highest fermentation rates resulting in the fastest completion of the course 2 days earlier than in the other treatments. SP showed a lower fermentation rate compared to either IS or SMT but higher than SQT. The kinetic behavior of yeasts varied among the different inoculation schemes. In both IS and SM fermentations, ScS13 followed comparable kinetics, reaching maximum population densities of 8.06 ± 0.08 and 8.05 ± 0.03 Log CFU/mL, respectively ( Figure 5A,C). The maximum population count reached by ScS13 was lower in SQT (7.79 ± 0.24 Log CFU/mL) and SP (7.79 ± 0.19 Log CFU/mL) compared to those in IS and SMT fermentations ( Figure 5B,D). ScS13 prevailed in both IS and SM fermentations at percentages of 100% and 86-100%, respectively. The respective percentages were lower in SQT fermentations (23-35%) in which two indigenous yeast populations also evolved. The ScS13 strain was not detected in SP fermentations.
Significant differences were observed in the non-Saccharomyces (NS) yeast kinetics among the different inoculation protocols applied. In the IS scheme, the indigenous NS yeast populations, which were at 4.96 ± 0.16 Log CFU/mL, were quickly suppressed by the early addition of ScS13 reaching a maximum population of 6.07 ± 0.13 Log CFU/mL by day 1, and thereafter started to decline sharply. In contrast, in SP the NS yeast population reached higher counts (7.23 ± 0.07 Log CFU/mL) and had longer persistence. The addition of TdS6 increased the initial counts in LA medium by more than 1 Log CFU/mL in both SMT and SQT as compared to IS and SP ferments ( Figure 5). In SM ferments, the NS counts peaked at day 1 at higher average values (7.19 ± 0.09 Log CFU/mL) than in either IS (6.00 ± 0.05 Log CFU/mL). In SP, NS counts peaked at 7.06 ± 0.30 Log CFU/mL Significantly higher NS counts (7.84 ± 0.04 Log CFU/mL) and also persistence were recorded in SQT compared to the other protocols ( Figure 5). At day 6 of the alcoholic fermentation, the strain TdS6 was isolated at frequencies of 100% and 60% from LA agar in SQT and SMT ferments, respectively. Significant differences were observed in the non-Saccharomyces (NS) yeast kinetics among the different inoculation protocols applied. In the IS scheme, the indigenous NS yeast populations, which were at 4.96 ± 0.16 Log CFU/mL, were quickly suppressed by the early addition of ScS13 reaching a maximum population of 6.07 ± 0.13 Log CFU/mL by day 1, and thereafter started to decline sharply. In contrast, in SP the NS yeast population reached higher counts (7.23 ± 0.07 Log CFU/mL) and had longer persistence. The addition of TdS6 increased the initial counts in LA medium by more than 1 Log CFU/mL in both SMT and SQT as compared to IS and SP ferments ( Figure 5). In SM ferments, the NS counts peaked at day 1 at higher average values (7.19 ± 0.09 Log CFU/mL) than in either IS (6.00 ± 0.05 Log CFU/mL). In SP, NS counts peaked at 7.06 ± 0.30 Log CFU/mL Significantly higher NS counts (7.84 ± 0.04 Log CFU/mL) and also persistence were recorded in SQT compared to the other protocols ( Figure 5). At day 6 of the alcoholic fermentation, the strain TdS6 was isolated at frequencies of 100% and 60% from LA agar in SQT and SMT ferments, respectively.

Chemical Profile of Wines Produced in Pilot-Scale Fermentations
The chemical characteristics of wines produced at pilot scale are presented in Table  5. The most striking differences were observed in the levels of VA and acetic acid that were lower in SQT compared to the other ferments. On the other hand, the succinic acid concentration was significantly higher in SQT than in other wines. Concerning the major volatile constituents of wines, the use of TdS6 caused a significant increase in the levels of isobutanol, 2-methyl-1-butanol and 3-methyl-1-butanol (Table 6). Significant differences were also recorded in the minor volatile profiles of wines ( Table 7). The addition of TdS6 reduced the levels of esters, similarly to the laboratory-scale fermentations. The decline in ester concentrations was most prominent in SQT, showing high decreases in the levels of ethyl acetate, isoamyl acetate, hexyl acetate and ethyl octanoate. Overall, despite differences that were not as pronounced, the chemical profiles of wines made using various

Chemical Profile of Wines Produced in Pilot-Scale Fermentations
The chemical characteristics of wines produced at pilot scale are presented in Table 5. The most striking differences were observed in the levels of VA and acetic acid that were lower in SQT compared to the other ferments. On the other hand, the succinic acid concentration was significantly higher in SQT than in other wines. Concerning the major volatile constituents of wines, the use of TdS6 caused a significant increase in the levels of isobutanol, 2-methyl-1-butanol and 3-methyl-1-butanol (Table 6). Significant differences were also recorded in the minor volatile profiles of wines ( Table 7). The addition of TdS6 reduced the levels of esters, similarly to the laboratory-scale fermentations. The decline in ester concentrations was most prominent in SQT, showing high decreases in the levels of ethyl acetate, isoamyl acetate, hexyl acetate and ethyl octanoate. Overall, despite differences that were not as pronounced, the chemical profiles of wines made using various inoculation protocols in pilot-scale fermentations followed the same trends as in laboratory-scale fermentations.
Significant variations between the chemical profiles of the different wines were observed by PERMANOVA (F = 3.27, p = 0.001). A PCA graph was constructed to better depict chemical variation of the wines produced at pilot-scale fermentation (Figure 6; 26.8% of total variation on PC1 and 20.2% on PC2). IS and SMT ferments were grouped closely to each other on the negative side of PC2. IS was characterized by highly negative values of 3-methyl pentanol and 4-methyl pentanol while SMT of citronellol and ethyl hexanol. The SQT ferments were located on the positive side of PC1 with high positive values for several metabolites, including nerolidol and citric acid. SP formed a separate cluster scoring positively on PC2 for several volatiles, such as ethyl acetate and acetic acid.

Discussion
'Assyrtiko' is one of the greatest white Greek grape varieties originated from the island of Santorini in the Aegean Sea. It is most often used to produce high quality dry white wines of complex texture, body and structure rather than aromatic style wines. Assyrtiko-Santorini is highly correlated with the volcanic terroir of Santorini, its original place of origin, where lean, highly acidic and less aromatic but mineral and very concentrated wines are produced. When cultivated in other regions, Assyrtiko wines tend to be less dense in structure or minerality. Due to the detrimental effects of global warming on diminishing wine acidity in warm regions, the unique and distinctive terroir of Assyrtiko-Santorini is currently under threat. To increase the acidity while boosting the terroir expression of this emblematic volcanic wine, two indigenous NS yeast species with promising technological characteristics along with an indigenous S. cerevisiae strain were investigated in various fermentation volumes and inoculation schemes.
S. cerevisiae is known to be highly competitive against NS yeasts and often outcompetes their population when added at the onset of the fermentation course. Accordingly, in this study, both NS yeasts tested quickly declined when co-inoculated with S. cerevisiae. The drop of NS yeasts at early addition of S. cerevisiae has been attributed to the increased sugar fermentation ability and nitrogen absorption of S. cerevisiae compared to NS wine yeasts [26,27]. In sequential fermentations, however, both NS yeasts exhibited an extended persistence with T. delbrueckii persisting the longest. The prolonged viability of both S. cerevisiae and T. delbrueckii when sequentially inoculated in grape must has been observed previously [28]. According to Álvarez-Fernández et al. [29] , since T. delbrueckii utilizes nitrogen through different metabolic pathways than S. cerevisiae, it may be easier for both species to coexist in mixed fermentations for a longer period of time. Therefore, T. delbrueckii strains may exhibit good compatibility with S. cerevisiae in mixed fermentations, a fact that may facilitate their use as yeast starter cultures in commercial applications. However, the fermentative power of each particular T. delbrueckii strain needs to be carefully considered since it appears to be highly strain-dependent, with evidence supporting the existence of strains showing either good [15,16,30] or poor fermentation performance [31][32][33][34].
In the present study it was shown that the sequential inoculation scheme with either L. thermotolerans LtS1 or T. delbrueckii TdS6 had greater differentiating effect as compared to simultaneous co-inoculation on the wine chemical profile in small-scale laboratory fermentations. The use of L. thermotolerans caused a significant rise in the titratable acidity (TA) of the ferment driven by the high production of lactic acid. L. thermotolerans is well  Table 5. Chemical parameters of wines produced in pilot scale fermentations (mean ± SD, n = 2 or n = 3 in IS and SMT). Values not sharing a letter are significantly different (p < 0.05).

Chemical Parameter
Inoculation Protocol 1

IS SMT SQT SP
Total acidity (as tartaric acid g/L) 6.6 ± 0.1 ab 6.7 ± 0.3 ab 7.0 ± 0.2 a 6.4 ± 0.2 b pH 3.14 ± 0.03 a 3.11 ± 0.01 ab 3.12 ± 0.00 ab 3.00 ± 0.08 b Volatile acidity (as acetic acid g/L) 0. 36  3.2 ± 0.3 a 3.3 ± 0.2 a 3.7 ± 0.4 a 3.2 ± 0.5 a Malic acid (g/L) 1.7 ± 0.0 ab 1.7 ± 0.1 a 1.6 ± 0.1 b 1.7 ± 0.0 ab Succinic acid (g/L) 0.7 ± 0.1 bc 0.7 ± 0.1 ab 0.9 ± 0.1 a 0.6 ± 0.0 c Lactic acid (g/L) <0.6 <0.6 <0.6 <0.6 Glycerol (g/L) 6.2 ± 0.2 b 6.1 ± 0.5 ab 6.6 ± 0.1 ab 6.8 ± 0.0 a Acetic acid (mg/L) 273 ± 47 b 208 ± 4 b 188 ± 8 b 509 ± 23 a Ethanol (g/L) 105.6 ± 0.8 a 104.2 ± 8.0 a 100.1 ± 4.5 a 108.9 ± 1.6 a Table 6. Major volatiles (mg/L) of wines produced in in pilot scale fermentations (mean ± SD, n = 2 or n = 3 in IS and SMT). Values not sharing a letter are significantly different (p < 0.05). In the present study it was shown that the sequential inoculation scheme with either L. thermotolerans LtS1 or T. delbrueckii TdS6 had greater differentiating effect as compared to simultaneous co-inoculation on the wine chemical profile in small-scale laboratory fermentations. The use of L. thermotolerans caused a significant rise in the titratable acidity (TA) of the ferment driven by the high production of lactic acid. L. thermotolerans is well known for its capacity to produce elevated levels of lactic acid and for this reason it has been considered as a powerful means to achieve biological acidification in wines [11,35]. Although to a lesser extent, T. delbrueckii TdS6 also showed the ability to increase TA through the production of succinic acid. Furthermore, the reduced levels of volatile acidity (VA) observed in sequential inoculations with T. delbrueckii is of high importance in fermentation of Assyrtiko grape must, which is often characterized by high sugar content and elevated VA levels. The use of a low-VA producing strain, such as T. delbrueckii TdS6 here, would significantly ameliorate the VA levels of 'Visanto', a prestigious naturally sweet wine of Santorini, which is mostly made from Assyrtiko must. The use of several NS strains, mainly of Starmerella bacillaris, in wine fermentation has been recently considered as a strategy to counteract the elevated acetic acid production mostly at the beginning of fermentation by S. cerevisiae [23,36,37]. Among other yeasts, T. delbrueckii has been shown previously to generate reduced levels of acetic acid, even during fermentation at very high sugar concentrations [38]. In the present study, T. delbrueckii showed much higher potential to reduce VA than L. thermotolerans LtS1. The strain TdS6 was further shown to largely decrease both VA and acetic acid levels in pilot-scale production of wine.

Chemical
On the other hand, T. delbrueckii significantly increased the levels of succinic acid in wine. This elevation of succinic acid levels has been proposed to be coupled with acetic acid decline, as described above. Accordingly, it has been suggested that a decrease in acetic acid redirects carbon flux to the overproduction of other metabolites, such as succinate [39]. Although succinic acid is the second most abundant non-volatile by-product of alcoholic fermentation [40], its relevance in wine is sometimes neglected due to its low acidity. Yet, succinic acid is considered to be the principal driver of titratable acidity rises in the finished product. Added to this, significant increases in the TA were observed in wines with high succinic acid production in this study. Taken together, succinate-producing T. delbrueckii strains may be considered as a significant candidate NS yeast starter for biological acidification of must. This biological activity is of particular interest in the wine industry to address low acidity in wines, an emerging threat in the grape/wine sector due to global warming [9].
Succinic acid has also a very important sensorial role. Besides increasing sourness, it brings salty and bitter tastes in wine [41]. High concentrations of succinic acid were associated with increased perceived minerality in white wines [42]. Therefore, the use of T. delbrueckii could highlight the mineral notes and enhance the perceived salinity that characterize Assyrtiko wines. Furthermore, succinic acid is also a precursor for aromaactive esters which bring mild and fruity aromas in wines, such as diethyl succinate in aged wines or sherry wines [43].
Ester formation is a key qualitative characteristic of yeast activity, adding pleasant notes to wine aroma [44]. In the present study, diethyl succinate (fruity), ethyl propanoate (pineapple) and ethyl isobutyrate (sweet, fruity), which are considered to be activity markers of T. delbrueckii [45], were significantly increased in mixed sequential inoculations. Some other ethyl esters were also increased, such as ethyl heptanoate (fruity, apple, grape) [46], ethyl lactate (fruity, buttery and creamy aromas, with sense of roundness in the mouth) [47,48], and ethyl 2-methylbutanoate. Despite the increase in the aforementioned esters, a decline in the net total ester content was observed in T. delbrueckii-inoculated sequential fermentations. More precisely, the use of TdS6 lowered the concentration of important esters in wine, such as ethyl acetate, isoamyl acetate or ethyl octanoate. A lower ester content in mixed fermentations with T. delbrueckii as compared to single-inoculation with S. cerevisiae has been also reported previously [37,49]. In some other cases, however, the use of T. delbrueckii in mixed fermentations with S. cerevisiae was shown to increase the total ester concentration [32,45,50]. A plausible explanation for those divergent outcomes is the high intra-species diversity of T. delbrueckii [51]. In fact, accumulating data reveal a significant diversity in the enological phenotypes among T. delbrueckii isolates [15].
In conclusion, the present results show that the use of indigenous S. cerevisiae in mixed culture with L. thermotolerans or T. delbrueckii strains may introduce significant diversification of the wines compared to the use of a commercial starter. T. delbrueckii TdS6 in particular showed several important technological characteristics that may be of certain usefulness in the production of Assyrtiko-Santorini wine. Among others, the capacity of TdS6 to diminish VA and to increase TA coupled with good compatibility with S. cerevisiae dictates its use in the fermentation of high sugar musts for the production of structured dry or sweet wines rather than aromatic style wines.