1 Introduction

Wine has been consumed by humans for thousands of years and for the majority of that time it was produced by crushing grapes and allowing them to ferment using the organisms present on the grapes and in the surrounding environment. There was relatively little a winemaker could do to control the quality of the final product. However, after Louis Pasteur discovered that yeasts were responsible for the conversion of sugars into ethanol more than 150 years ago, the wine industry slowly began to move away from its more unpredictable production methods (Hutkins 2006). The use of spontaneous fermentation for example slowly gave way to intentional inoculation of meticulously selected and maintained Saccharomyces cerevisiae starter cultures to ensure a more consistent and predictable product vintage to vintage. We now understand that wine is the result of a complex biological process that takes place between grapes, microorganisms (yeasts, bacteria and fungi), vinification and the wine cellar environment (Fleet 2003). Of all the yeasts found to be associated with the winemaking process, S. cerevisiae is indeed by far the most capable and reliable ethanol producer. When it is inoculated at high cell density, it can drastically reduce the chances of stuck fermentation or the production of off-flavors that can come from the unwanted growth of other organisms (Fleet 1993). This simply owes to the fact that it can rapidly outcompete other yeast and bacterial species as well as quickly produce an environment inhospitable to most other organisms primarily through the production of ethanol. However, while this may reduce sources of microbial spoilage, some winemakers feel that this has resulted in a lack of organoleptic complexity. It has been shown in recent years that certain indigenous non-Saccharomyces yeasts can contribute to distinct regional and desirable characteristics of wine when inoculated at high concentrations (Jolly et al. 2006).

This has prompted an interest in beginning to understand the specific influences of non-Saccharomyces yeasts in winemaking (Andorrà et al. 2010, 2012; Benito et al. 2015; Ciani and Comitini 2010a; Comitini et al. 2011; Dashko et al. 2015; Jolly et al. 2014; Sadoudi et al. 2012; Sun et al. 2014; Zott et al. 2011). Even though the impact of non-Saccharomyces yeasts is usually limited because of the fast fermentative metabolism of S. cerevisiae, research has shown that this impact may be enhanced when non-Saccharomyces yeasts are inoculated at high cell density. However, because most non-Saccharomyces yeasts cannot ferment to dryness, S. cerevisiae must also be inoculated along with the non-Saccharomyces yeast when they are used intentionally. Two modes of inoculation are usually envisaged: staged (sometimes called sequential) and co-inoculations. In co-inoculation, all yeasts are added to the must at the same time while in staged inoculation, the non-Saccharomyces yeasts are added first, allowed to ferment for a given amount of time, and the Saccharomyces yeasts are added to finish the fermentation. Staged inoculations are of particular interest since they can ostensibly allow for even greater control over the species fermentation progress and thus the aroma and flavor profile of a fermentation. Both strategies have been shown to mimic the results of natural fermentations in having more complex aromas (Ciani and Maccarelli 1998; Romano et al. 2003a). The principal outcomes of fermentations conducted with the aid of non-Saccharomyces yeasts have been documented in literature already mentioned here. Nevertheless, the description of the impact of these yeasts is usually restricted to a few specific attributes such as enzyme, acetic acid, glycerol, ethyl acetate, and higher alcohol production (Andorrà et al. 2012; Charoenchai et al. 1997; Clemente-Jimenez et al. 2004; Gobbi et al. 2013; Pina et al. 2004; Rojas et al. 2001; Romano et al. 2003b; van Breda et al. 2013; Villena et al. 2007; Wang et al. 2015).

There are between 9 and 15 different yeast genera that are typically reported to be associated with the winemaking process (Johnson and Echavarri-Erasun 2011). Many of these were originally studied in the context of spoilage but this work slowly began to shed light on some potentially beneficial aspects of these yeasts. For example, early work showed that the Candida, Debaryomyces, Hanseniaspora, Hansenula, Kloeckera, Metschnikowia, Pichia, Saccharomyces and Torulaspora genera isolated from wines could produce extracellular enzymes such as pectinases, amylases, lipases, proteases and glucosidases (Charoenchai et al. 1997). β-glucosidases are of particular interest for their ability to liberate otherwise bound terpenes and thus have a direct impact on wine aroma. This work was expanded on and complemented by investigations of the specific behaviors of certain species in grape must (Ciani and Maccarelli 1998; Esteve-Zarzoso et al. 1998). Studies began to characterize the macronutrient consumption as well as macromolecule production in single and mixed fermentations. This in turn gave way to more targeted studies of the potential impact of specific yeast (Andorrà et al. 2012; Anfang et al. 2009; Azzolini et al. 2012; Ciani et al. 2006; Clemente-Jimenez et al. 2004, 2005; Dias et al. 2003; Gobbi et al. 2013; Pina et al. 2004; Romano et al. 2003a, b; Wang et al. 2015; Zott et al. 2008). Based on this research, commercial non-Saccharomyces starter cultures have recently been developed for use in wine production and are comprised of the following yeast species: Torulaspora delbrueckii, Lachancea thermotolerans, Pichia kluyveri and Metschnikowia pulcherrima. Nevertheless, compared to S. cerevisiae, little research has been conducted that can indicate specifically what metabolic profiles to expect from these yeasts under various fermentation conditions. Indeed, though the mounting evidence supports the use of these yeasts to help improve wine aroma, the majority of the previously mentioned studies are somewhat limited in scope. They focus either on enzyme production or target ester and alcohol production and only Gobbi et al. (2013) complemented their targeted chemical analysis of L. thermotolerans and S. cerevisiae co-fermentation with sensory work. Therefore there is still a knowledge gap on the impact of these yeasts during wine fermentations.

In this study we specifically compared untargeted volatile compound profiles and sensory analysis data of Sauvignon blanc wine fermented sequentially with six different non-Saccharomyces yeasts. Of the six non-Saccharomyces species used, four were commercial starter strains, Torulaspora delbrueckii, Lachancea thermotolerans, Pichia kluyveri, and Metschnikowia pulcherrima, while the other two, Candida zemplinina, and Kazachstania aerobia, are laboratory strains. The goal of this study was to expand on previous work where only the profile of the non-Saccharomyces yeasts in single fermentation were characterized (Beckner Whitener et al. 2015). This study completed the wine fermentations through the addition of S. cerevisiae in order to gain a better understanding of the aroma compounds present in the final wine following the use of the selected non-Saccharomyces yeasts in sequential inoculation. The potential metabolic implications, as well as how these compounds might contribute to the perceived sensory attributes of the finished wine product were assessed.

2 Materials and methods

2.1 Grapes, yeasts, and chemicals

Sauvignon blanc grapes (vintage 2014) were obtained from the vineyards at Welgevallen Experimental Farm, Stellenbosch University, Stellenbosch, South Africa. The clone was SB 316 and rootstock was R110, the vineyard was planted in 1991. The trellis system used was a seven wire hedge trellis with moveable foliage wires and grapevines were spaced at 2.7 × 1.5 m with a east–west row direction. Grapevines were unilateral cordon-trained and spur pruning was applied. The grapevines were not irrigated and the vineyard was established on a duplex Hutton/Glenrosa soil according to the 1992 South African Binomial Soil Classification system. S. cerevisiae (Enoferm M2®, Lallemand Inc., Quebec, Canada), T. delbrueckii (Biodiva®, Lallemand Inc., Quebec, Canada), M. pulcherrima (Flavia®, Lallemand Inc., Quebec, Canada), P. kluyveri (Viniflora® FROOTZEN™, Chr. Hansen, Horsholm, Denmark), L. thermotolerans (Viniflora® CONCERTO™, Chr. Hansen, Horsholm, Denmark), C. zemplinina (Institute of Wine Biotechnology (IWBT)-Y1082) and K. aerobia (IWBT-Y845) were used. Twenty-milliliter glass screw cap vials, sodium chloride (ACS grade), sodium azide, internal standard 2-octanol, a divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) coating 50/30 μm, 2-cm length SPME fiber was purchased from Supelco by Sigma-Aldrich S.r.l., Milan, Italy.

2.2 Winemaking procedure

Fermentations were carried out using Sauvignon blanc grape must. The must was evaluated for initial sugar (21.7 Brix), titratable acidity (5.8 g/L) and yeast assimilable nitrogen (YAN) (170 mg/L) content, as well as pH (3.39). YAN was adjusted by adding 40 mg/L of diammonium phosphate (DAP) to the must. The yeasts were grown in yeast extract peptone dextrose (YPD) medium (Biolab-Merck, Wadeville, South Africa). They were shaken to ensure aerobic conditions at 30 °C in successively larger batches using a 1 % transfer rate starting from 10 mL and ending at 1 L at which point necessary cell concentrations for wine inoculation were obtained via centrifugation. The 11 L stainless steel fermentation vessels containing 10 L of must were inoculated with a volume of yeast determined from the pre-culture by plate count and optical density to obtain a level of 106 cfu/mL. The inoculation levels were confirmed and yeast growth monitored via plate count on WL Nutrient agar (Fluka, Sigma-Aldrich) which allows for visual differentiation of the yeast strains. Fermentations were carried out in triplicate at 15 °C. The non-Saccharomyces yeasts were allowed to ferment until approximately 2 % ethanol concentration was reached. At this point, S. cerevisiae was added at 106 cfu/mL concentration to finish the fermentations after being grown up in the same manner as the non-Saccharomyces yeasts. Samples were taken daily to track fermentation progress via plate count and by Fourier-transform mid-infrared spectroscopy (FOSS WineScan FT120, Foss Analytical, Denmark) in accordance with the protocol outlined in Nieuwoudt et al. (2006). The apparatus measured levels of glucose, fructose, titratable acidity, volatile acidity, pH, acetic acid and malic acid. The final wines were bottled after clarification via cold rest for 1 week at −4 °C in 750 mL glass bottles with screw caps. Wines were then transported to the laboratory of the Department of Food Quality and Nutrition, Research and Innovation Center, Fondazione Edmund Mach (FEM) for chemical analysis. Sensory analysis was performed at the Department of Viticulture and Oenology, Stellenbosch University.

2.3 Sensory evaluation

General Descriptive Analysis was used as the method to evaluate the experimental wines. A panel of 10 judges was selected; all had moderate to good experience in wine evaluation, in particular Sauvignon blanc. The panel was composed of eight females ranging in age from 25 to 55; and two males (aged about 25). A session was completely dedicated to the taste component of the wines and the panel was trained on sweetness, acidity, bitterness and astringency intensities. For this purpose, a commercial Sauvignon blanc wine was spiked with increasing levels of sugar, tartaric acid, quinine and alum respectively. All were over the counter items purchased at a local grocery store. To score the intensity of the attributes of the experimental wines a 100-mm unstructured scale was used, demarked with ‘None’ and ‘Intense’ at the extreme left and right sides, respectively. Panel performance was evaluated using Panel Check (Tomic et al. 2009). The descriptive study was performed in two sessions. Panelists were asked to taste in isolated booths and each treatment was presented to them covered in ISO black glasses and marked with three-digit codes. A complete Block Design was used to randomise the distribution of the wines presented to the panellists (Lawless and Heymann 2010). Each judge evaluated each treatment in triplicate.

2.4 SPME extraction and GCxGC-TOF-MS analysis

Vials were prepared as follows: 5 mL of wine and 50 μL of 0.5 mg/L 2-octanol were added to 20 mL screw cap vials containing 1.5 g NaCl. A Gerstel MPS autosampler (GERSTEL GmbH & Co. KG) equipped with the standard sample agitator and SPME fiber conditioning station was used to extract the volatiles from the sample vial headspace. GCxGC-TOF-MS analysis of the extracts was performed using a LECO Pegasus-4D system consisting of an Agilent 6890N (Agilent Technologies) coupled to a LECO Pegasus 4D detector. The system employed a consumables free modulation system. The samples were incubated for 5 min at 35 °C under 500 rpm rotation at 10 s intervals. Extraction took place for either 10 s, 5 min, or 30 min prior to desorption in the inlet for 180 s at 250 °C. Quality control (QC) vials containing an equal mix of all wines were spaced at the beginning and every third sample thereafter within each time batch. Each extraction time consisted of only one batch as all samples and spaced QCs fit into a single cooling tray. Helium carrier gas was used with a flow set at 1.2 mL/min and a splitless time of 180 s. The oven was equipped with a 30 m x 0.25 mm × 0.25 μm VF-WAX MS primary column (Agilent Technologies) and a 1.5 m × 0.15 mm × 0.15 μm RXI 17Sil MS secondary column (Restek Corporation, Bellefonte, PA, USA). The GC oven parameters were as follows: initial temperature was 40 °C held for 2 min, followed by an increase to 250 °C at a rate of 6 °C/min, the oven was then held at 250 °C for 5 min before returning to the initial temperature (40 °C). The total cycle time, was 42 min. The modulation period was set to 7 s with a hot pulse time of 1.4 s. The modulator was offset by 15 °C. The MS protocol consisted of electron ionization at 70 eV with ion source temperature at 230 °C, a detector voltage of 1543 V with a voltage offset of 200 V, mass range of m/z 35-350, and acquisition rate of 200 spectra per second. There was an acquisition delay of 120 s.

2.5 Data processing and alignment

ChromaTOF software version 4.32 was used to perform baseline correction, deconvolution and peak picking of the raw data. The baseline offset was set to 1, just above the noise level. The first dimension peak width was set to 43 s while the second dimension peak width was set to 0.1 s. A factor of 500 was set as the match required to combine peaks in the second dimension. A signal to noise (S/N) of 10 was used for the 10 s and 5 min extraction times data with a minimum S/N of 6 for sub peak retention. A S/N of 100 was used for the 30 min extraction time data with a minimum S/N of 60 for sub peak retention. Traditional, not adaptive, integration was used. Forward library searching was used with the following parameters: Hits to return were set to 10, minimum molecular weight was set to 40, maximum molecular weight was set to 350, the mass threshold was set to 50 and the minimum similarity match was set to 700. The NIST and Wiley libraries were used to achieve level II identification as defined by Sumner et al. 2007. For alignment the following parameters were used: a mass threshold of 10, a minimum similarity match of 600, the maximum number of modulation periods matching peaks could be apart was set to 1, a maximum retention time difference was set to 7 s, for peaks not found by initial peak finding the signal to noise ratio was set to 5 for the 10 s and 5 min extractions and to 50 for the 30 min extractions, for analytes to be kept they had to be found in all biological replicates within a class. Each yeast species was given its own class.

2.6 Statistical analysis

Each extraction time, 10 s, 5 and 30 min, was treated as a separate data set in the following way. First, to avoid underestimation of the variance of the data, zero intensity values (undetected features) were replaced feature-wise by a random number between the lowest detected intensity and zero. Following this, for each feature, a fixed effects linear model was fitted with yeast strain as the fixed effect. This model was used for pairwise comparisons between all wines without correction for multiple testing. Subsequently, the collection of p-values for all comparisons were corrected for multiple testing by controlling the false discovery rate (FDR) and q-values calculated (Strimmer 2008a, b).

To select the compounds of interest (COI) a filter with three requirements was applied to the data. Compounds were selected if the q-values for any of the comparisons between any of the wines were below 0.05 and at least one comparison had a fold change greater than 2.5. In addition, the QC samples were used to calculate the relative coefficient of variance (%CV) for each feature across the whole analysis. Only features with %CV lower than 50 % in the QC samples were selected. A venn diagram was generated to illustrate this filtering process for each extraction time Fig. 1. The features that fell into the center of these diagrams were considered significant COIs for each extraction time. The peak area values for each of these compounds were used to generate heat maps and PCA plots to better illustrate the data (Figs. 2, 3). Unit variance scaling was used for PCA and heat map generation as well as the values seen in Table 1 and Table 1S. Values outside the range of 3 standard deviations were reassigned to 3 in the case of the heat maps. The Pearson correlation coefficient and Ward’s minimum variance method were used for hierarchical clustering (Murtagh and Legendre 2014). The PCA bi-plots from the sensory data were generated from the analysis performed using panel check (Fig. 3d).

Fig. 1
figure 1

Venn diagrams depicting the three extraction times a 10 s, b 5 min and c 30 min. The total number of features identified in each extraction time was 988 for (a), 1510 for (b), and 1048 for (c). The center values in each diagram represent the significant compounds of interest with q-values below 0.05, a fold change of 2.5 or higher and %CV in quality control samples lower than 50 %

Full size image
Fig. 2
figure 2

Heat map of the Venn diagram center metabolites for the a 10 s, b 5 min, and c 30 min extraction time data. Ward’s minimum variance was used for hierarchical clustering. SC S. cerevisiae fermentations, TD T. delbrueckii fermentations, CZ C. zemplinina fermentations, KA K. aerobia fermentations, LT L. thermotolerans fermentations, PK P. kluyveri fermentations, MP M. pulcherrima fermentations and QC quality control samples

Full size image
Fig. 3
figure 3

PCA after unit variance scaling of the Venn diagram center features for the a 10 s, b 5 min, and c 30 min extraction time data. d PCA bi-plots of the sensory data. SC S. cerevisiae fermentations, TD T. delbrueckii fermentations, CZ C. zemplinina fermentations, KA K. aerobia fermentations, LT L. thermotolerans fermentations, PK P. kluyveri fermentations, MP M. pulcherrima fermentations and QC quality control samples

Full size image
Table 1 Average peak areas after unit variance scaling for each metabolite found to be in highest relative concetration amoung the treatments across all extraction times
Full size table

3 Results and discussion

3.1 Fermentation progress and primary metabolite production

All fermentations progressed at slightly different rates. S. cerevisiae was the quickest fermenter, reaching 2 % ethanol in 3 days (Fig. 4). The first of the non-Saccharomyces fermentations to reach 2 % ethanol was L. thermotolerans, 4 days after inoculation. Next was T. delbrueckii 1 day later followed by C. zemplinina on day six. K. aerobia and M. pulcherrima each took 7 days and the P. kluyveri was the slowest at 8 days. The order of fermentation speed is comparable to results in our previous study for L. thermotolerans, T. delbrueckii, and M. pulcherrima (Beckner Whitener et al. 2015). Once the fermentations reached 2 % ethanol, S. cerevisiae was added to complete the fermentations. The musts inoculated with only S. cerevisiae fermented to dryness in 12 days while the rest of the fermentations took between 19 and 24 days with L. thermotolerans again finishing first among the non-Saccharomyces fermentations. Almost all of the non-Saccharomyces fermentations showed a similar pattern of sugar consumption in which glucose was consumed faster than fructose. The C. zemplinina fermentation stood out in that it was the only fermentation in which fructose was consumed more rapidly than glucose even after S. cerevisiae addition. This was not surprising since C. zemplinina is known to be fructophilic and able to survive to the end of fermentation due to its high ethanol tolerance (Rantsiou et al. 2012). It has also been reported that sequential inoculation of C. zemplinina produced a wine lower in acetic acid compared to a S. cerevisiae pure culture but this was not the case in our study (Englezos et al. 2015). Of all the fermentations conducted, the C. zemplinina fermentations produced the most acetic acid (1.37 g/L while the T. delbrueckii produced the least (0.07 g/L) (Fig. 5). Despite the relatively large amount of acetic acid in the C. zemplinina fermentations the sensory panel did not note an acetic acid fault in the wine. It is worth mentioning however, that the two fermentations that showed the highest amounts of acetic acid did score the closest to the ‘sour’ descriptor, those being C. zemplinina and K. aerobia (Fig. 3d). L. thermotolerans fermentations were characterized by the least amount of overall titratable acidity as well as the least amount of malic acid at the end of the fermentation. In fact, all of the co-fermentations had lower overall levels of malic acid than the S. cerevisiae control (Fig. 5). This confirms previous findings that S. cerevisiae is characterized as a poor metabolizer of l-malate (Salmon 1987). The other yeasts in this study have not been investigated for their l-malate metabolism or their ability to metabolize other TCA cycle intermediates as a sole carbon source (Saayman and Viljoen-Bloom 2006). Given the results in this study however it is likely that all of the non-Saccharomyces yeasts used here are able to transport and metabolize l-malate as has been shown for Candida sphaerica, Candida utilis, Hansenula anomala, Pichia stipitis and Kluyveromyces marxianus (Saayman and Viljoen-Bloom 2006).

Fig. 4
figure 4

Each graph indicates the progress of the fermentations by each species. SC S. cerevisiae, CZ C. zemplinina, KA K. aerobia, LT L. thermotolerans, MP M. pulcherrima, PK P. kluyveri, TD T. delbrueckii. Each graph shows glucose consumption (square shape), fructose consumption (diamond shape), and ethanol production (triangle shape). All of these lines are an average of the three biological replicates and the standard error of the mean is shown by error bars. The solid vertical line indicates where the ethanol concentration reached 2 % and in the case of the non-Saccharomyces fermentations S. cerevisiae was added. The dashed vertical line indicates where the non-Saccharomyces yeast was no longer detectable by plate count

Full size image
Fig. 5
figure 5

Bar graph indicating the final average acidity and pH levels of each fermentation. TA titratable acidity while VA volatile acidity. SC S. cerevisiae, CZ C. zemplinina, KA K. aerobia, LT L. Thermotolerans, MP M. pulcherrima, PK P. kluyveri, TD T. delbrueckii

Full size image

Even though there were obvious differences in growth patterns and macro metabolite production, all wines did eventually reach approximately the same ethanol concentration of 14 % v/v. The presence of the non-Saccharomyces yeasts was monitored during the fermentations and it should be noted that the non-Saccharomyces yeast populations began to decline as soon as the S. cerevisiae was added but that they remained detectable via plate count for between 6 and 10 days (Fig. 4). This indicates that the non-Saccharomyces yeasts remained viable and detectable for over half of the total fermentation time. For C. zemplinina, L. thermotolerans, and T. delbrueckii this is in agreement with literature (Azzolini et al. 2012; Kapsopoulou et al. 2006; Maio et al. 2012). For M. pulcherrima, P. kluyveri, and K. aerobia this has not been previously reported in a sequential wine fermentation. In all likelihood, the yeasts remained metabolically active and thus able to contribute to the organoleptic profile for even longer than this since the non-Saccharomyces colonies became difficult to count due to the overcrowding of the S. cerevisiae.

3.2 Aroma compound presence in Sauvignon blanc due to specific species: chemistry and sensory analysis

GCxGC-TOF-MS is becoming more common in the field of metabolomics as it has proven to be a powerful tool that can increase separation, detection and identification of a wide variety of metabolic analytes compared to 1D GC (Zhang et al. 2012). When Solid-Phase-Microextraction (SPME) is used as an extraction method it is possible to study, with great chemical selectivity and sensitivity, the volatile profile of samples. The different compounds that make up the headspace of wine samples will be present in a broad concentration range, with varying vapor pressures and have different adsorption affinities to the SPME fiber. Therefore, this study employed three separate extraction times to increase compound coverage while limiting chromatographic and detector saturation. The 10 s extraction time proved useful for obtaining peak shapes conducive to consistent integration for the most highly concentrated analytes such as esters and alcohols. The 30 min extraction time was used to characterize the smaller but no less important peaks that represent aroma compounds such as terpenes, volatile phenols, thiols and some of the less concentrated esters and alcohols. The 5 min extraction severed as a good average between compounds found in saturation at 30 min but not detected by the 10 s extraction time. Figure 1S illustrates this finding by showing two compounds. Compound 1 is in saturation at 30 min (Fig. 1Sc) but measurable in both the 5 min (Fig. 1Sb) and the 10 s (Fig. 1Sa) extraction times. The peak shape is however best for measurement in the 10 s chromatogram. Peak 2 in this figure shows the opposite trend. Some compounds were only measurable at 30 min and too small in 5 min and 10 s to be reliable. However, there were some compounds that were reliably measurable at two or all three extraction times and in this case the data was combined and represented as such in Table 1 and Table 1S. Each extraction time consisted of only one batch and the intra-batch reproducibility was assessed by comparing the peak area of the internal standard in each sample. Figure 2S shows the normalized mean peak area of the internal standard of each sample from each extraction time batch. Together, these three data sets, along with the sensory analysis, provided a highly detailed volatile compound and aroma profile of Sauvignon blanc wines generated in this study.

It is well known that certain sulfur compounds such as 4-mercapto-4-methylpentan-2-one, 4-mercapto-4-methylpentan-2-ol, 3-mercaptohexanol, and 3-mercaptohexyl acetate give Sauvignon blanc its characteristic tropical and green aromas (Tominaga et al. 1998). Though these compounds were not reliably detectable with the analytical method used in this study, likely due to their relatively low concentration, the compounds have a very low sensory detection threshold and thus are easily distinguished by the human olfactory sense at much lower concentrations, the parts per trillion range, than the SPME fiber is capable of detecting (Dubourdieu et al. 2006). It is for this reason that tropical aromas such as guava and passion fruit were a critical part of the sensory panel evaluation as can be seen in Table 2S. The sensory analysis not only mirrors the untargeted volatile profile but complements and expands it.

Both the sensory and the analytical methods were able to show a distinct separation of the wines co-fermented sequentially with the different yeasts based on their detectable aroma features; this can be seen clearly in the principal component analyses (PCAs) (Fig. 3). The sensory analysis focused on 16 typical Sauvignon blanc aromas. Only 12 of these (guava, passion fruit, grapefruit, banana, apple, pineapple, cooked vegetable, solvent, sherry, fermented apple, dried peach and stewed fruit) proved to be consistently evaluated and significantly different across all samples according to ANOVA analysis (p < 0.01). Thus it is not surprising that, with so few parameters, the sensory PCA is able to account for more than 90 % of the total variance. The first principal component axis is largely defined by a difference in the fruity aroma profiles (Fig. 3d). Esters are primarily responsible for the bulk of fruity aromas and flavors in wine and this result could indicate a significant difference in ester production between the yeast species. In fact, the analytical method showed significantly different ester profiles for each fermentation. The basic flavors of sweet, bitter and sour were also found to be significantly different across all samples and were distributed more along the second component. The SPME method on the other hand was able to detect thousands of volatile aroma compounds which after our feature selection was applied to cut the number of features down from over 1000 total identified features to 336 compounds found to be statistically significantly different across the fermentations. The breakdown according to extraction time is as follows: 78 compounds for the 10 s extractions, 196 for the 5 min extractions and 239 for the 30 min extractions. Some compounds were reliably extracted by more than one extraction time and their unit variance scaled values were combined and are shown in Table S1. It is clear from the PCA plots that like the sensory analysis the yeasts showed distinct profiles with strong grouping of the biological replicates. This same result is also confirmed by the hierarchical clustering (Fig. 2a–c) in which it is clearly shown that the yeast replicates grouped with themselves and each grouping had a distinct chemical signature. It should be noted that with so many chemical compounds it was only natural that the yeasts grouped together so well and showed such distinct profiles in the chemical data versus the sensory data. There are two possible explanations for this: either by focusing on only 16 compounds, the tasters “missed” significant odors in the wine, or the compounds produced (or not) by the different yeasts contribute to the overall wine complexity but were not distinguishable as individual aromas because they remained below detection threshold. In all likelihood the explanation lies somewhere between the two. To put this into perspective the most prominent details of these profiles of each fermentation are discussed below on a yeast by yeast basis.

3.2.1 Saccharomyces cerevisiae

The panel associated the S. cerevisiae fermentations most closely with guava, passion fruit, grapefruit, banana, pineapple and apple. It was least associated with fermented apple, dried peach and stewed fruit (Fig. 3d). Chemically, the S. cerevisiae fermentations were distinguished mostly by a group of 65 compounds found to be in the highest relative concentration across all fermentations. These compounds are seen in red in the heat maps (Fig. 2) meaning they consistently showed the highest relative concentration among those samples. Of these, the majority were alcohols and esters associated with green, fruity, and tropical notes (Table 1). This correlates well with the previously mentioned panel findings, specifically the banana, pineapple and apple aromas. A large portion of the most significant compounds have currently no documented aromas or flavors. Some compounds were wholly unidentified features and all of them represent an area of possible future study. Out of all the fermentations, the S. cerevisiae showed the highest number of distinguishing esters, alcohols, and other compounds and this is in agreement with the literature (Dubourdieu et al. 2006; Lambrechts and Pretorius 2000; Majdak and Herjavec 2002; Zalacain et al. 2007). Metabolically speaking, there was nothing out of the ordinary for these fermentations and they served well as a control.

3.2.2 Candida zemplinina

The sensory panel found that the C. zemplinina fermentations had the most distinct aroma profile next to the S. cerevisiae fermentations. They were characterized by the guava, fermented apple, sherry, dried peach/apricot, and stewed fruit descriptors. This is not surprising given its profile of compounds found to be significantly higher, which can be seen in Table 1, and are represented in red in the heat maps in Fig. 3. There were 49 features with statistically significantly larger relative peak areas that separated the C. zemplinina fermentations from the rest. Of these, 12 were esters, one of which, 2-methyl-propanoic acid ethyl ester, has a very high odor strength and is characterized as sweet, ethereal and fruity with pungent, alcoholic, fusel and rummy descriptors as well. This is likely one of the main contributors to the ‘fermented apple’ aroma described by the panel. All other yeast fermentations showed almost none of this compound comparatively. Also worth noting is the statistically significant presence of relatively large acetic acid and hydroxyl acetic acid peaks in the SPME–GCxGC-TOF-MS analysis. As previously mentioned, the Fourier-transform mid-infrared spectroscopy analysis revealed a relatively high level of acetic acid (Fig. 5), and the sensory panel noted this fermentation to be more sour than others. This shows all three analysis methods to be both cohesive and complimentary to one another. It is however, in direct contrast to previously published work which indicates that C. zemplinina had the capacity to reduce the amount of acetic acid in a wine fermentation especially when used in conjunction with S. cerevisiae (Englezos et al. 2015; Rantsiou et al. 2012; Sadoudi et al. 2012). These differences however, could be due to biological variability between different strains used. Indeed, as noted by Englezos et al. (2015), within this species the strain diversity is significant.

The C. zemplinina fermentations were also characterized by the largest number of terpenes and sesquiterpenes. Of the 49 significant compounds, 11 were either a terpene or sesquiterpene including geraniol, nerol, α-pinene, α-farnesene, ocimene, and linalool (Table 1). In general, these compounds are responsible for floral, pine and citrus aromas. In wine, rather than being produced directly by the yeast through a metabolic pathway, terpenes are released when glucosidases such as β-glucosidase free bound glycosylated precursors (Carrau et al. 2005). Two previous studies, Englezos et al. 2015 and Sadoudi et al. (2012), looked specifically at terpene content in single and mixed culture fermentations of C. zemplinina and S. cerevisiae. Englezos et al. (2015) tested 63 different strains and found that only 5 % of the isolates showed β-glucosidase activity. Sadoudi et al. (2012) found that, in monoculture, C. zemplinina produced more norisoprenoids and terpenols but this trend did not hold in mixed fermentation with S. cerevisiae. Our results however, indicate that the strain of C. zemplinina used in this study may produce relatively high amounts of β-glucosidase even in the presence of S. cerevisiae resulting in a wine richer in terpenes. Further screening should be carried out on this strain to confirm and quantitate enzyme production.

In summary, the C. zemplinina and S. cerevisiae fermentations were both the most sensorially and chemically distinct with C. zemplinina displaying the highest number of terpenes and sesquiterpenes as well as some more uncommon esters and presenting more dried fruit rather than fresh fruit aromas. Unfortunately, of all the fermentations it also produced the largest amount of acetic acid.

3.2.3 Kazachstania aerobia

The strain of K. aerobia used for these fermentations was isolated from Cabernet Sauvignon grape must at the IWBT and here for the first time we outline the chemical and organoleptic properties that this yeast is capable of producing in a finished wine product. Chemically speaking, the K. aerobia only showed 30 compounds to be statistically significantly different from the other fermentations. Though less than C. zemplinina’s 49, they still provide an interesting picture of what this yeast can bring to a wine fermentation.

The sensory panel agreed that the K. aerobia fermentations were driven more by solvent and bitter characteristics and slightly by the dried or stewed fruit aromas than the fresh ones (Fig. 3d). The chemical analysis revealed that the bulk of the compounds, 12 out of the 30, found to be positively different from the other fermentations were ethyl and acetate esters including ethyl acetate. This is most likely the cause of the solvent aroma. This correlates well with the fermentation data which revealed that K. aerobia fermentations had the second highest volatile acidity level of which ethyl acetate is a contributor (Fig. 5). 2-phenethyl acetate and 6-methyl-2-heptanol acetate were two other acetate esters found to be in higher relative concentration. It is interesting to note that the higher alcohols corresponding to the acetate esters in these fermentations were not shown to be significantly higher. The next largest group of compounds found to be significantly positively different was terpenes. α-farnesene, γ-terpinene, nerol, m-cymene, and terpinolene all showed only trace peaks in the S. cerevisiae fermentations but much more substantial peaks in the K. aerobia fermentations. Though not responsible for the majority of a Sauvignon blanc flavor profile, terpenes are beneficial in their ability to provide complexity via subtle earthy, woody, citrus and floral undertones. They enhance and complement the more known fruity and floral notes provided by the esters. Besides a few alcohols, acids, aldehydes, and alkenes the rest of the K. aerobia’s chemical profile was made up of six compounds which could not be identified based on their mass spectra.

Since K. aerobia’s genome has yet to be fully sequenced, it is difficult to point to a specific cause for the abundant presence of these compounds relatively to the other fermentations. However, Kazachstania’s nearest genetic relative is the Saccharomyces genus. It stands to reason that they share many of the same genes and thus regulatory pathways (Kurtzman 2003).

To conclude, the K. aerobia fermentations showed relatively high ethyl acetate, ester and terpene production and a few compounds that could not be identified. No major off-flavors were noted either chemically or sensorially.

3.2.4 Torulaspora delbrueckii

T. delbrueckii has been used in winemaking for years and is one of a few non-Saccharomyces species commercially available for use in wine and beer production. While it may be the best studied species of the genus, like all wine-related non-Saccharomyces species, it remains poorly understood. Of the studies that have been conducted, it has been reported that wine fermented with T. delbrueckii in co-culture with S. cerevisiae were typically characterized by low volatile acidity, higher terpenols, 2-phenylethanol and C6 compound production (Ciani and Maccarelli 1998; Renault et al. 2009; Sadoudi et al. 2012; van Breda et al. 2013). Further metabolic and sensory evaluation of this yeast has yet to be done however.

Our study showed that sensorially T. delbrueckii fermentations were similar to the L. thermotolerans, P. kluyveri and M. pulcherrima all of which were most significantly characterized by the bitter attribute and equidistant from the fresh and dried fruit aromas (Fig. 3d). Fermentation data confirms previous reports in that T. delbrueckii produced the least amount to acetic acid and volatile acidity (Fig. 2) (Sadoudi et al. 2012). Chemically, its unique profile was most closely related to the L. thermotolerans across all extraction times (Fig. 2). Where it differed from L. thermotolerans and in fact all of the other fermentations was that it showed relatively higher concentrations of the sulfur containing compounds 3-(methylthio)-1-propanol, 3-[(2-hydroxyethyl)thio]-1-propanol, thietane, 3-(methylthio)propanoic acid ethyl ester, and 1,3-oxathiane. Moreira et al. (2002) showed that increased amounts of methionine in grape must lead to increase in 3-(methylthio)-1-propanol and 3-(methylthio) propanoic acid ethyl ester among other unidentified sulfur compounds. They also showed that wines made from must generally low in amino acids had the highest total amount of sulfur compounds. As such, there are two likely causes of the increased sulfur compounds seen in our T. delbrueckii fermentations. Either T. delbrueckii itself assimilates and catabolizes methionine more readily than S. cerevisiae or T. delbrueckii creates an amino acid poor environment and facilitates the formation of these compounds by S. cerevisiae. As already stated, sulfur containing compounds have generally very low sensory thresholds and these are no exception. 3-(methylthio)-1-propanol has been described as having a raw potato, sulfurous, onion, soup, vegetable odor and 3-(methylthio)propanoic acid ethyl ester has been described as sulfurous, metallic, pineapple, fruity, and ripe pulpy tomato. They both have very high odor strengths and in too high a concentration would undoubtedly contribute to a wine fault. Given that the sensory panel did not identify a sulfurous fault in the T. delbrueckii fermentations, it is likely that though they were identified in the chemical analysis as significantly different these compounds were not in high enough concentration to be detected by the human palate. This however, does indicate the need for further study of amino acid catabolism by non-Saccharomyces yeasts with a specific focus on how differences may affect the metabolism and volatile compound production of S. cerevisiae. Besides these findings, it should be noted that like the K. aerobia fermentations there were two analytes found to be significantly higher in the T. delbrueckii fermentations that could not be identified.

To summarize, while T. delbrueckii may reduce acetic acid in the final fermentation, it does little else to positively enhance the overall aroma profile. The wine showed higher levels of off-odor causing thiol compounds compared to the other fermentations which, while not noted by the sensory panel, could be detrimental to a final product if concentrations become too high.

3.2.5 Lachancea thermotolerans

Various studies have investigated its potential use in winemaking with regards to acetaldehyde, lactic acid, glycerol, 2-phenylethanol, and polysaccharide production as well as β-glucosidase activity. It is well established that this strain is capable of producing lactic acid and increasing the pH of wine while reducing its volatile acidity. It has also been shown to increase glycerol and 2-phenylethanol concentrations while being a low acetaldehyde producer (Ciani and Comitini 2010b; Ciani et al. 2006; Comitini et al. 2011; Cordero-Bueso et al. 2012; Kapsopoulou et al. 2006). Gobbi et al. (2013) is the most extensive study of this species in wine to date. They report that even in sequential inoculation, L. thermotolerans was the dominant species during fermentation and that these fermentations showed reduced 2-methyl-1-propanol and 3-methyl-1-butanol, higher 2-phenylethanol, reduced acetate esters but higher ethyl acetate. The ethyl acetate was below the sensory threshold, however. Some of this is in direct contrast to our findings where our results indicate that the L. thermotolerans population was slowly over taken by S. cerevisiae after its addition. Another difference was that over half of the esters found to be higher in the L. thermotolerans fermentations in our case were acetate esters. The L. thermotolerans fermentations were also characterized in our case by the lowest amount of both titratable acidity and malic acid out of all the fermentations (Fig. 5). Sensorially, these fermentations were mostly characterized along PC2 in the PCA, specifically the pineapple and bitter descriptor and as previously mentioned grouped closely with T. delbrueckii, P. kluyveri, and M. pulcherrima (Fig. 3d). Chemically, the L. thermotolerans and T. delbrueckii showed the most similar profiles according to the PCAs and hierarchical clusters. Of the 34 compounds shown to be significantly higher in the L. thermotolerans fermentations, 12 of them had no suitable matches in the NIST library. Many of these were small peaks that were only found in the 30 min extractions. The L. thermotolerans fermentations contained the largest number of unknown analytes. Only eight esters were shown to be higher, four of those were acetate esters one of which was citronellol acetate. Farnesol, geraniol, α-ionene, and cosmene were found to be highest in the L. thermotolerans fermentations. This is supported by previous research which has shown that certain strains of L. thermotolerans can have high β-glucosidase activity (Cordero-Bueso et al. 2012).

In short, the L. thermotolerans fermentations showed a relatively high number of acetate esters and certain terpenes as well as the lowest amount of both titratable acidity and malic acid out of all the fermentations. There were no notable off-flavors in high relative concentration but there were 12 unidentified compounds, the highest number out of all the fermentations.

3.2.6 Pichia kluyveri

Despite the fact that this species is commercially available, comparatively even less research than on the other non-Saccharomyces yeasts has been published on its specific contributions to the winemaking process. Anfang et al. (2009) co-fermented Sauvignon blanc with a specific P. kluyveri isolate from New Zealand and showed that the resulting wines had elevated levels of 3-mercaptohexyl acetate (3MHA), indicating that the specific isolate was capable of releasing more favorable volatiles thiols from the Sauvignon blanc must. By contrast, the isolate used in this study did not show a sensorially significant increase in the tropical fruity aromas characterized by 3MHA. In fact, the P. kluyveri fermentations fell close to the center of PC1 being equally defined by both fresh and dried fruit aromas (Fig. 3d). Chemically, previous research had shown that Pichia membraenifaciens was a good acetate ester producer (Viana et al. 2008). However, this trait does not seem to carry over to P. kluyveri when compared to the other yeast in this study. This is unsurprising given the high amount of biodiversity observed in the Pichia genus (Domizio et al. 2011). Our study shows for the first time an in depth chemical profile of P. kluyveri. In both the PCAs and heat maps the P. kluyveri grouped most closely with the M. pulcherrima (Figs. 2, 3a–c). There were only 23 compounds found to be significantly higher in the P. kluyveri over all of the other fermentations. Eight of these were esters with significantly fruity aromas, three of which were 3-methylbutyl esters of three different organic acids (Table 1). 3-methyl-butanoic acid (isovaleric acid) was also relatively high. This compound is associated with an off-putting sour, sweaty, and cheesy aroma and in too high a concentration is considered a wine fault. It is a product of l-leucine catabolism and can undergo esterification to create 3-methyl-butanoic acid ethyl ester which has a much more pleasant, fruity aroma. This compound was one of the esters present in relatively high concentration in the P. kluyveri fermentations. Another potentially fault inducing compound found to be higher was phenethylamine. Metabolically, there are two enzymes responsible for the conversion of the amino acid phenylalanine to phenethylamine: Aromatic-l-amino-acid decarboxylase and phenylalanine decarboxylase, either of which could have been up-regulated in either the P. kluyveri or the S. cerevisiae. Ultimately, neither of these potential fault compounds was in high enough concentration to have a sensory impact as the sensory panel did not note an off aroma in the wine. However, given these issues, combined with the lack of notable positive sensory attributes, this particular strain of P. kluyveri is conceivably not as good a candidate for Sauvignon blanc production as others covered by this study.

3.2.7 Metschnikowia pulcherrima

The M. pulcherrima fermentations were, sensorially, closest to the P. kluyveri fermentations and similarly not strongly associated with either the fresh or dried fruit aromas but fell closer to the sweet, bitter and solvent traits. Unlike the P. kluyveri fermentations however, chemically, there were no discernible off-aromas. A common isolate in vineyards and from grape must, M. pulcherrima has long been associated with grapes and wine and early research into the potential of this species showed that certain isolates displayed a high β-glucosidase activity (Fernández et al. 2000). Our study indicates that while some terpenes were higher in the M. pulcherrima fermentations when compared strictly to the control other yeasts showed higher amounts (Table 1S). Clemente-Jimenez et al. (2004) reported that M. pulcherrima produced high amounts of 2-phenyl ethanol and our findings support that as well. Of the thirty compounds found to be relatively higher in the M. pulcherrima fermentations, over half were esters most of which being either methyl butyl, methyl propyl, or phenethyl esters. Most of these however, have no recorded aroma. Similarly, there were six compounds that could not be identified, making the M. pulcherrima fermentations difficult to characterize both from a sensory and a metabolic standpoint.

Sadoudi et al. (2012) is, to date, the most comprehensive study of M. pulcherrima in co-culture with S. cerevisiae. They observed that fructose was consumed more slowly over the course of co-culture fermentation. This was not the case in our study but S. cerevisiae was added much later in our fermentations than the reported 48 h post M. pulcherrima inoculation of Sadoudi et al. (2012). They also reported that the co-cultures showed lower acetic production compared to the S. cerevisiae mono-culture. In our case, the opposite was true though in the M. pulcherrima fermentation the acetic acid level, though higher, remained below the sensory threshold (Fig. 5). These differences could be due to any number of variables such as yeast–yeast interactions, or changes in regulation of acetic acid metabolism in one or both species as a result of different fermentation stresses, to name a few. Like many of the other yeasts in this study, M. pulcherrima strain differences might be a possible reason for the discrepancies observed between studies.

4 Concluding remarks

In conclusion, of all the yeasts used in this study, S. cerevisiae and C. zemplinina had the most distinct and remarkable fermentation profiles. However each of the six non-Saccharomyces yeast co-fermentations displayed a unique sensory and metabolic profile. We were able to show that the sensory and chemistry methods complemented each other well and gave a much more detailed profile of these yeasts than any previously published work. Overall, our results would suggest that while the non-Saccharomyces yeasts produced wines that were unique, S. cerevisiae in single culture produced a product with the strongest positive sensory components thanks to high ester production. While it is true that our results are not fully in line with previously published results, this study was strongly dependent on the wine matrix composition, especially amino acids, terpene and thiol precursors, and thus is not 100 % reflective of the non-Saccharomyces capabilities. Given how little is currently known about these yeasts in wine and their contribution to wine aroma this study served to greatly increase the body of knowledge and understanding of these yeasts and their metabolism in the wine matrix used.