The strains of S. cerevisiae used are listed in Supplementary Table 1. Seven of them belong to the flor group while seven others have been clearly characterized as wine yeasts. Three other strains, SB, FMGS_889, and AC1_191, have a mixed genome related to flor and wine yeasts. The ISVV-2D strain resulted from a cross between the haploid strains CLIB 1770 and CLIB 1769 delivered by the CIRM collection and derived from the flor strain 2D (Coi et al., 2016). S. cerevisiae strains were propagated on YPD 2% (1% peptone, 1% yeast extract, and 2% glucose) at 30°C in both liquid and plate cultures (2% agar). Long-term storage at −80°C was achieved by adding one volume of glycerol to YPD overnight cultures.

Some vials were set aside at 10 and 50% of the maximal CO₂ produced for the extraction of intracellular metabolites. Five milliliters were taken per vial for quenching and metabolites extraction. Measurements of cell density, viability and cell size were taken by flow cytometry for each sample (CytoFlex, Beckman Coulter apparatus). Enzymatic assays of malic acid, acetic acid, glycerol, succinic acid, and residual sugars were performed on all samples. Figure 1 summarizes the experimental steps from fermentations to NMR analysis.

For each sample taken during the alcoholic fermentation and at the end, a volume of 800 μl was manually transferred in Micronics tubes (Novazine, Lyon, France, ref: MP32033L) and stored at −20°C. Concentrations of the following organic metabolites were measured: acetic acid, glycerol, malic acid, and succinic acid using the respective enzymatic kits: K-ACETGK, K-GCROLGK, K-MAL-116A, and K-SUCC (Megazyme, Bray, Ireland) according to the manufacturer’s instructions. Glucose and fructose were assayed at the end of alcoholic fermentation using the enzymatic method described by Stitt et al. (1989). All the enzymatic assays were performed by a robotic platform at the Bordeaux metabolomics facilities.¹

All the statistical and graphical analyses were carried out using R software (R Development Core Team, 2016) and plots were generated using the base or ggplot2 packages. All the raw data are given on the Supplementary Table 4.

In this study, a ¹H-NMR method was developed to quantify the main intracellular metabolites of fermenting yeasts during alcoholic fermentation. We focused our efforts on the characterization of flor and wine yeasts that constitute two distinct S. cerevisiae populations (Coi et al., 2017; Peter et al., 2018) and share similar ecological niches. As recently demonstrated, wines made by these two groups may have statistical differences for several organic compounds including malic, acetic, and succinic acids (Vion et al., 2023). The genetic material used is listed in Supplementary Table 1. Seven flor strains were compared to eight wine strains. In addition, two control strains (FMGS_889 and AC1_191) were included in the experiment. These strains were selected for their opposed malic acid metabolism as previously reported (Vion et al., 2021, 2022). As specified in the methods, yeast cells were collected during alcoholic fermentation and immediately quenched using cold methanol. Samples were taken at two fermentation moments: 10 and 50% of CO₂. According to previous works, the growth of yeast strains during alcoholic fermentation occurs up to 15–20 g/L of CO₂ production (Bely et al., 1990; Marullo et al., 2009; Albertin et al., 2011; Nidelet et al., 2016). The samples taken at 10% of CO₂ were therefore considered to correspond to the growth phase while samples taken at 50% of CO₂ were considered to correspond to the stationary phase.

Just before sampling, the concentration and the cell size of yeasts were estimated by flow cytometry. To ensure the reliability of the quantification, five biological repetitions were carried out per modality. In addition, a small volume of fermenting juice was frozen for enzymatic metabolic analyses. The experimental design of this work is shown in Figure 1.

The average fermentation rate (dCO₂/dt) of the flor and wine groups in function of the CO₂ produced are shown in Figure 2A. The fermentation kinetics of each individual strain and the CO₂ production of each culture are provided in Supplementary Figure 2 and Supplementary Table 4, respectively. An important phenotypic variability was observed within groups for the kinetic parameters computed (lag phase, time to reach the maximal CO₂, and V15_50). Thus, those traits that reflect the fermentation efficiency at the culture level are not statistically different between flor and wine yeast in our conditions (Kruskal test, α = 0.05). A similar conclusion has been done in a recent study using other natural grape juice with a subset of the strains used in this study (Vion et al., 2023).

The quantification of cell growth and viability by flow cytometry at two different time courses of alcoholic fermentation allowed the investigation of possible difference in life-history traits between flor and wine populations. The average viability of fermenting yeast in the growth and the stationary phases are 92 and 81%, respectively. These traits were not statistically different between flor and wine populations. In contrast, the cell concentration reached during the stationary phase is significantly different between flor and wine yeasts, with average values of 6.15 × 10⁷ and 8.6 × 10⁷ (Wilcoxon test, α = 0.05) (Figure 2B). The FSC parameter of the flow cytometer was used as a proxy for determining the cell size of both populations. Interestingly, the evolution of cell size between the flor and wine groups was statistically different. Wine yeasts were smaller than flor yeasts during the growth phase, but they became larger during the stationary phase (Figure 2C), as confirmed by a strong effect of interaction between group and sampling time (two-way ANOVA 21.8% of the variance explained, p-value = 0.009).

Since the cell size can be influenced by the ploidy level, we quantified the DNA content of the 15 strains used in this experiment using the method described by Todd et al. (2018). Except for the ISVV-2D strain, all the strains used in this study were found to be diploids, as shown in Supplementary Figure 3. Strain ISVV-2D was obtained by crossing two haploids derivatives of the strain 2D purchased at the CIRM collection; these strains were haploids (with a correct matting type) and were mixed to obtain a diploid clone named ISVV-2D. Intriguingly, the resulting hybrid showed an unexpected ploidy (4n) with G1 and G2 peak intensity close to the tetraploid control (Hirondelle).

The typical ¹H-NMR spectrum after water suppression is presented in Figure 3. The signals at 0.00 and 1.26 ppm correspond to TSP and DMMA respectively; other signals correspond to intracellular metabolites. The ¹H-NMR spectra were dominated by sugar content and organic acids followed by some amino acids. The chemical shifts and coupling constant used for identification and absolute quantification of 21 organic compounds are listed in Table 1. The abundance of each compound was normalized by the number of cells per milliliter present in the sample. For both sampling times, the average abundance and CV of each compound and phase are indicated in Table 1. The average abundance of each strain at each sampling point is provided in Supplementary Figure 4.

These analytical results demonstrated that the experimental procedure developed can efficiently extract and quantify intracellular metabolites of fermenting S. cerevisiae. The extraction methodology was adapted from previous intracellular metabolite extraction protocols. According to Villas-Bôas et al. (2005), cold methanol acts as a non-selective solvent and quenching agent allowing the efficient extraction of different chemical families’ intracellular metabolites. The addition of galactose in the fermentation medium and its absence in the intracellular extracts confirms the efficiency of our protocol (Supplementary Figure 1). Extracted metabolites were then identified and quantified by ¹H-NMR using the procedures applied in a previous study (Vion et al., 2023).

In order to compare biological samples in the same physiological state, yeast cells were collected according to their progress in the fermentation (10 and 50% of the total CO₂ expected) during the alcoholic fermentation. These sampling points corresponded to the growth and stationary phases since they occurred before and after the peak of CO₂ production rate (Figure 2A). The time monitoring of these metabolites may provide information about the evolution of the physiological status of the cells. For example, trehalose is a typical quiescence marker and a key storage carbohydrate (Gray et al., 2004). The abundance of this compound is 2.2-fold higher during the stationary phase in our experiment (Table 1) (Wilcoxon test, p-value = 1.14e−08 < 0.05). The detection of this accumulation confirmed the overall physiological changes occurring in fermenting yeasts.

Interestingly, a similar approach was applied to characterize the genetic and environmental factors modulating the intracellular content of S. cerevisiae in synthetic media cultures (Airoldi et al., 2015). Compared to this former study, we quantified additional intracellular compounds such as ethanol, 2,3-butanediol, and proline. The detection of additional metabolites would require a higher number of scans and more concentrated biological materials. The analytical parameters are already 256 scans for an analysis time of 35 min per sample. In addition, the evolution of these intracellular compounds was achieved for a larger number of strains (17) compared to Airoldi et al. (only two). By increasing the number of strains, we aim to explore the intraspecific variability of intracellular metabolite abundance that has been poorly investigated so far.

The relative abundance of the 21 intracellular metabolites quantified by ¹H-NMR was represented by a heatmap according to populations and sampling times (Figure 4). Interestingly, the sum of the abundance of each metabolite quantified is statistically different between flor and wine populations during the growth phase. Strains belonging to the wine population displayed a lower pool of intracellular metabolites than strains belonging to the flor group (Wilcoxon test, p-value = 0.05, Figure 4A). In contrast, no significant differences in the sum of abundance were observed between the two populations during the stationary phase. This surprising finding could reflect a real difference of metabolite concentration inside the cell, or it could be due to biases in abundance normalization. As is routinely done, we normalized NMR intensity by the number of viable cells per milliliter. We alternatively normalized data by the total volume of fermenting yeast that is the product of the cell volume and the number of cells. In both cases, the cumulated metabolite load was significantly higher for the flor group, revealing unsuspected variations between S. cerevisiae populations during the growth phase of alcoholic fermentation.

Previous studies reported strong correlations between the specific glycolytic flux, the number of fermenting cells and their average size among yeast strains of different origins. These phenotypic trade-offs were reported at different sugar level concentrations (Spor et al., 2008, 2009) and under different alcoholic fermentation conditions (Albertin et al., 2011). The correlation between these life-history traits defined two styles of growth strategies named “ants” and “grasshoppers.” This trade-off reflects the energetic impossibility of making big and numerous cells at the same time in a context of limited nutrient resources. Wine yeasts have been reported to belong to the “ants” group, since they are small, numerous, and ferment quickly at their populational level (Albertin et al., 2011).

The quantification of the overall metabolite content of fermenting cells provides the opportunity to test the impact of this new metabolic parameter. First, we confirmed the negative relationship between the cell size and the maximum biotic capacity previously reported by Spor et al. (2008; Figure 5A). In addition, a weak yet highly significant positive correlation (rho = 0.49, p-value = 0.0002) was found between the overall metabolite load and the cell size during the growth phase (Figure 5B). Broadly, the biggest cells are those containing the highest intracellular metabolite specific concentration. This trend was only observed during the growth phase when the variation of the cell size between the strains was particularly high (Supplementary Figure 5). Finally, we tested the correlation between the overall metabolite load and the specific glycolytic flux measured between 15 and 50% of CO₂ released (sV15_50) (Figure 5C). There is a strong and very significative positive correlation between the specific glycolytic flux and the overall metabolite concentration observed in both populations. Strains with a high metabolic load have a higher specific CO₂ production rate and could be considered as “grasshoppers.”

In addition to the broad differences found between flor and wine groups, important strain-specific variations were found. Interestingly, strain SB had the highest relative abundance of intracellular metabolites within the wine population and had the biggest size of the wine yeasts in growth phase. This might be explained by its mosaic origin between flor and wine yeasts (Peltier et al., 2021). In contrast, the flor yeast 36.2J (S4V) exhibited a low relative abundance of intracellular metabolites, like most strains of the wine group. Other yeast strains displayed distinctive metabolic profiles, such as the flor yeast ISVV-2D, which had a much higher content of intracellular metabolites in the stationary phase compared to all other strains. Interestingly, this laboratory-made strain was shown to be tetraploid, which could impact its intracellular metabolite levels. Surprisingly, its average cell size is not particularly higher than that of other diploids flor strains, as shown in Supplementary Figure 5. This important strain variability suggested that the cumulated load of metabolites is a complex parameter that remains to be investigated.

A nested analysis of variance was applied to deeply investigate the differences in metabolite content between populations and sampling phases. This analysis estimated the effects of the fermentation phase (Time), genetic factors and their possible interactions. The genetic contribution effect was decomposed in population and strain within population effects as detailed in linear model 2 (see section “2. Materials and methods”). The contribution of each factor on 17/21 quantitative traits is summarized in Figure 6. A Levene test (α = 0.05) was performed to test the variance homogeneity and the four traits were excluded from the analysis.

Some compounds such as 2,3-butanediol, tyrosine and succinic acid are greatly impacted by the sampling moment, which explained almost 50% of the variance for 2,3-butanediol and tyrosine. Intracellular 2,3-butanediol and tyrosine globally increased four times between growth and stationary phases, while succinic acid was three times more concentrated inside the cells at stationary phase. The accumulation of 2,3-butanediol by yeast cells during stationary phase is shown in Figure 7A. This compound is the end metabolite of the degradation of acetoin, a minor by-product of alcoholic fermentation. Besides these few metabolites, most of the compounds showed quite steady intracellular concentration between the two phases. This phenomenon highlighted that yeast cells are able to maintain relative intracellular homeostasis of their metabolites during the alcoholic fermentation process. This observation also implies that the concentrations measured were not contaminated by the extracellular medium. For example, extracellular fructose is supposed to be drastically consumed between the growth and stationary phases due to the ongoing alcoholic fermentation (40 g/L of CO₂ produced). However, the intracellular concentration of fructose is quite similar according to the sampling phases (Figure 7B). Interestingly, most of the metabolic compounds (14/17) assayed are strongly impacted by the genetic component, which represented more than 25% of the total variance. Although the population origin was always significative, the decomposition of the population and strain variances indicated a strong strain variability. In addition, an important effect of genetic per time (GxT) interaction was observed for several compounds. This effect ranged between 8 and 31% of the total variance and highlighted the difficulty of establishing a clear pattern of intracellular metabolic content between the flor and wine populations, as illustrated for fructose, glycerol, and malic acid (Figures 7B–D, respectively).

Population genomic studies have revealed that the species S. cerevisiae is composed of several distinct populations that are characterized by specific genetic and metabolic features. Wine and flor yeasts constitute two distinct groups of strains (Coi et al., 2017). Although they share the same ecological niche (wine environment), they developed alternative metabolic strategies for surviving. The first produces ethanol from grape juice while the second consumes ethanol from wine by forming a velum (Legras et al., 2016). Surprisingly few studies have compared the phenotypic properties of wine and flor yeast during alcoholic fermentation (Esteve-Zarzoso et al., 2001). In addition, a comparative study of diversity of metabolic networks within numerous S. cerevisiae strains encompasses several wine and flor strains. The prediction of metabolic fluxes by applying a flux balance analysis suggested that flor yeast was characterized by a high production and output of acetate, which contrasted with the behavior of other groups such as bread strains (Nidelet et al., 2016). This discrepancy can be explained by a high variability of accessory metabolic fluxes such as the Pentose Phosphate Pathway. More recently, a metabolomic study identified unsuspected differences in the extracellular composition of wines produced by flor and wine strains. Among others, the content of malic acid at the end of alcoholic fermentation clearly discriminated the two groups, with the flor group being a stronger consumer of this organic acid (Vion et al., 2023). Since flor and wine yeast have evolved under different energy metabolisms (fermentation or respiration) we speculated that they might carry intracellular metabolic hallmarks of these very different selective pressures. The previous section highlighted that the cumulated content of the intracellular metabolites is significantly different between these two groups during the first stages of alcoholic fermentation. Taken individually, none of the metabolites are able to clearly discriminate those groups. Among the metabolites identified by our nested ANOVA as having the strongest impacts, including glucose, fructose, malic acid and citric acid, glycerol has a strong population effect accounting for more than 20% of the total variance explained. However, the strain variability effect within groups is always stronger, making it impossible to define whether those metabolic compounds are group specific.

The intracellular metabolic profile of the two control strains (AC1_191 and FMGS_889) were compared in this last section. These strains were selected for their strong discrepancy regarding their malic acid metabolism, which is illustrated in Figure 8A. From the stationary phase sampling point, the extracellular malic acid concentration between the two strains is significantly different (difference of 1.23 g/L) and reached 2.4 g/L by the end of alcoholic fermentation. These differences are much higher than those observed between flor and wine groups that differ by only 0.15 g/L at the end of alcoholic fermentation (Figure 8B). The NMR quantification method developed provided the opportunity to investigate which intracellular compounds could reflect differences in malic acid metabolism observed at the extracellular level. Surprisingly, the various organic acids quantified showed very similar concentrations between the two strains. Thus, regardless of the malic acid consumption and/or production dynamics, the intracellular concentrations of malic acid and other organic acids (succinic, citric, and pyruvic) are quite steady (Figure 9A). In contrast, the intracellular concentrations of valine, alanine, and threonine measured at the stationary phase displayed a statistical difference between the two strains (Figures 9B–D). These amino acids are indirectly derived from pyruvic and alpha ketoglutaric acid and might result from a difference in malic acid assimilation. However, additional experiments will be needed to establish a metabolic link between these compounds and the malic consumption differences observed between AC1_191 and FMGS_889.