Biological material.
Twenty-six strains of
Saccharomyces cerevisiae coming from different geographical origins and food processing industries (wine making, brewing, baking, and distilling) were initially selected to be representative of the main genetic clusters found in
S. cerevisiae food processing strains (
1). A previous work showed that strains from baking were mostly autotetraploids (
1). To avoid possible effects of ploidy level, we excluded baking strains from the final panel. For each of the nine chosen food processing strains (
Table 1 ), one meiospore was isolated with a micromanipulator (Singer MSM Manual; Singer Instrument, Somerset, United Kingdom). Six out of nine strains were homothallic (
HO/
HO) (294, 328, 382, F10, VL1, and BO213), so that the isolated meiospore gave rise to a fully homozygous diploid strain through mating type switch and further fusion of opposite-mating-type cells. For the three heterothallic strains (
ho/
ho) (963, A24, and 7327), the isolated haploid meiospore was diploidized via transient expression of the HO endonuclease: the strains were transformed with the pHS2 plasmid (kindly given by S. Himanshu) using the lithium acetate transformation protocol described by Gietz and Schiestl (
26). After diploidization, the plasmid was eliminated from the strains through recurrent cultures on yeast extract-peptone-dextrose (YPD) medium. The resulting biological material was constituted by four wine-making strains (E1 to E4) (here called wine strains), two brewing strains (B1 and B2) (here called beer strains), and three distilling strains (D1 to D3) (here called distillery strains) (
Table 1). All strains were currently grown at 30°C in YPD medium containing 1% yeast extract (Difco Laboratories, Detroit, MI), 1% Bacto peptone (Difco), and 2% glucose, supplemented or not with 2% agar.
Synthetic fermentative media.
Three synthetic fermentative media were used; they differed in their levels of sugar and nitrogen, pH, osmotic pressure, and anaerobic growth factors in order to reflect the main changes of fermentation medium between brewing, baking, and wine-making contexts (see
Table 2). Maltose is the main available sugar in wort and dough (
14,
44), and yet most of the strains used in wine making cannot assimilate it (
42). Moreover, previous work shows that the presence of maltose in the medium could affect glycolysis even for strains lacking the maltose permease (
52). Therefore, to compare the basal fermentative behaviors of different food processing strains including those of wine-making origin, we decided to use glucose as the only carbon source in all three media. Although distillery strains were included, no distilling medium (here called distillery medium) was used. Indeed, the distilling process refers to distinct processes, including the production of distilled beverages such as cognac, whisky, tequila, or other spirits and the production of bioethanol fuel from various crops (sugarcane or beet molasses), etc. The compositions of these different distillery media vary greatly regarding sugar content, pH, osmotic pressure, nutrient limitation, or starvation, as illustrated for molasses composition (
2), so that it was not possible to propose a single representative distillery medium.
The enology medium was modified from the work of Marullo et al. (
37) and Bely et al. (
4,
5) and contained glucose (220 g liter
−1), tartaric acid (3 g liter
−1), citric acid (0.3 g liter
−1),
l-malic acid (0.3 g liter
−1), and myoinositol (0.3 g liter
−1). The available nitrogen was 200 mg liter
−1, provided by 315 mg liter
−1 (NH
4)
2SO
4 (corresponding to 66.7 mg liter
−1 nitrogen) and 0.67% (vol/vol) of a mixture of 18 amino acids containing
l-tyrosine (9.4 mg liter
−1),
l-tryptophan (92.3 mg liter
−1),
l-isoleucine (16.9 mg liter
−1),
l-aspartic acid (22.9 mg liter
−1),
l-glutamic acid (62 mg liter
−1),
l-arginine (192.8 mg liter
−1),
l-leucine (24.9 mg liter
−1),
l-threonine (39.1 mg liter
−1),
l-glycine (9.4 mg liter
−1),
l-glutamine (260.2 mg liter
−1),
l-alanine (74.8 mg liter
−1),
l-valine (22.9 mg liter
−1),
l-methionine (16.2 mg liter
−1),
l-phenylalanine (19.5 mg liter
−1),
l-serine (40.4 mg liter
−1),
l-histidine (16.9 mg liter
−1),
l-lysine (8.8 mg liter
−1), and
l-cysteine (6.7 mg liter
−1) in NaHCO
3 (134.8 mg liter
−1), corresponding to 133.3 mg liter
−1 available nitrogen. The mineral salts were provided by KH
2PO
4 (2 g liter
−1), MgSO
4·7H
2O (0.2 g liter
−1), MnSO
4 (4 mg liter
−1), ZnSO
4·7H
2O (4 mg liter
−1), CuSO
4·5H
2O (1 mg liter
−1), KI (1 mg liter
−1), CoCl
2·6H
2O (0.4 mg liter
−1), (NH
4)
6Mo
7O
24·4H
2O (1 mg liter
−1), and H
3BO
3 (1 mg liter
−1); the vitamins were as follows: biotin (0.04 mg liter
−1), thiamine-HCl (1 mg liter
−1), pyridoxine-HCl (1 mg liter
−1), nicotinic acid (1 mg liter
−1), calcium panthothenate (1 mg liter
−1), and para-amino benzoic acid (1 mg liter
−1). Finally, 0.02% (vol/vol) of a mixture of anaerobic growth factors was added, containing 1.5% (wt/vol) ergosterol and 0.5% (wt/vol) sodium oleate in Tween 80-ethanol (1:1, vol/vol). The pH was adjusted to 3.5 using KOH pellets.
The brewery medium was modified from the enology medium. The pH was adjusted to 4.5 using 40 mM citrate buffer, pH 4, replacing tartaric, citric, and malic acids. The glucose content was 80 g liter−1, and nitrogen content was modified as follows: 400 mg liter−1 available nitrogen, one-third coming from 630 mg liter−1 (NH4)2SO4 and two-thirds coming from 13.49% (vol/vol) of the mixture of amino acids described below. Brewery medium was not supplemented with anaerobic growth factors.
The bakery medium was modified from the enology medium. The pH was adjusted to 5.5 using 40 mM citrate buffer, pH 5.2, replacing tartaric, citric, and malic acids. The glucose content was 80 g liter−1, and nitrogen content was modified as follows: 400 mg liter−1 available nitrogen, one-third coming from 630 mg liter−1 (NH4)2SO4 and two-thirds coming from 13.49% (vol/vol) of the mixture of amino acids described below. Sorbitol (150 g liter−1, i.e., 0.82 mol liter−1) was also added in order to increase the osmotic pressure. Bakery medium was not supplemented with anaerobic growth factors.
Fermentation kinetics.
The amount of CO
2 released was determined by automatic measurement of glass reactor weight loss every 20 min (
8,
19). The CO
2 production rate (g liter
−1 h
−1) was calculated using a local polynomial regression fitting (loess function, R program [
51]). Different kinetics parameters were calculated as previously described (
37). The lag-phase time (h) was the time between inoculation and the beginning of CO
2 release (CO
2 production rate higher than 0.05 g liter
−1 h
−1). Similarly, the end of the fermentation was determined to be when the CO
2 production rate dropped below 0.05 g liter
−1 h
−1. This point allowed us to estimate the AF time (
alcoholic
fermentation time; hours), which was the time necessary to ferment the sugars in the medium, excluding the lag phase. All strains were able to achieve the fermentation (i.e., to consume over 98.5% of initial sugar) except strains B1, B2, D1, D2, and D3 in enology medium. However, these five strains displayed sluggish fermentation under winery-type conditions, so that the corresponding AF time was higher than that for strains actually achieving AF. Thus, the AF time parameter was relevant to compare the global fermentation abilities of the strains, whatever the medium.
Vmax (g liter
−1 h
−1) was the maximal CO
2 production rate. Finally, CO
2tot was the total amount of CO
2 released at the end of the fermentation (g liter
−1).
Population dynamics, cell size, and growth recovery.
The population growth and the cell size were monitored regularly using a particle counter (Z2 Coulter counter; Beckman Coulter, Villepinte, France); more than 20 samples per fermentation were taken from the inoculation time until the carrying capacity (maximum population size,
K) was reached. During the exponential phase, samples were taken every 2 or 3 h in order to have a good estimate of the maximum rate of increase of the population (intrinsic growth rate,
r). The experimental points were fitted with a logistic model that allowed estimation of
K and
r:
where
Nt is the population size at time
t,
K is the carrying capacity (cells per ml),
N0 is the initial population size, and
r is the intrinsic growth rate (number of divisions per hour). At the end of the fermentation, the proportion of cells able to form colonies (CFU) after 2 days on YPD agar plates was determined and referred to as the growth recovery (%). This parameter is usually assimilated as an indirect measure of cell viability in microbiology (
62). Finally, the cell size was measured at the end of alcoholic fermentation using a particle counter (Z2 Coulter counter; Beckman Coulter, Villepinte, France) and the mean cell size (diameter [μm]) was calculated and used for further analyses.
Dosage of alcoholic fermentation products.
At the end of the alcoholic fermentation, several dosages were made: ethanol concentration (percent volume) was determined by infrared reflectance (Infra-Analyzer 450; Technicon, Plaisir, France), and residual glucose (g liter−1) and acetic acid production (g liter−1) were measured by colorimetry (A460) in continuous flux (Sanimat, Montauban, France) in the supernatant. For the analyses of variance (ANOVAs), we considered a derived variable, ethanol/glucose ratio (in mol/mol), to compare ethanol yields. External glycerol (g liter−1) and residual nitrogen (g liter−1) were assayed by the enzymatic method (Boehringer kits 10 148 270 035 and 11 112 732 035; R-Biopharm, Darmstadt, Germany), and residual nitrogen was used to determine nitrogen consumption (%). Biomass dry weights were measured from 200 ml of final fermentation medium using a desiccator (SMO 01; Scaltec Instruments GmbH, Göttingen, Germany) and were expressed in g liter−1.
For internal glycogen and trehalose dosages, 10 ml of medium was sampled at the end of AF and washed twice with 4 ml of 5% (wt/vol) NaCl. After centrifugation (5 min, 2,750 ×
g), yeast pellets were chilled at −20°C. Trehalose was extracted by twice washing the pellets for 1 h at 4°C in 8 ml of 0.5 M trichloroacetic acid, with an additional wash with 4 ml of cold H
2O. The pellets were kept at 4°C for further glycogen dosage (see below). After centrifugation (5 min, 2,750 ×
g), the washing supernatants were merged and adjusted to 25 ml with H
2O. Trehalose was determined with anthrone as described by Roustan and Sablayrolles (
51,
56). Briefly, 2.5 ml of cold anthrone solution (190 mg anthrone [Sigma-Aldrich, Lyon, France] in 83% [vol/vol] sulfuric acid) was added to 250 μl of the supernatant solution, boiled for 12 min, and immediately cooled on ice to stop the reaction. Optical density at 625 nm (OD
625) was measured using a spectrophotometer (Lambda EZ201; Perkin-Elmer, Courtaboeuf, France). Gradual dilutions of trehalose dehydrate (Sigma) were used as standards. For glycogen dosage, the residual pellets were suspended in 0.5 ml of 8 M HCl, and glycogen was solubilized for 30 min at 60°C using 2 ml of dimethyl sulfoxide (DMSO; Sigma, Lyon, France). The suspension was cooled to room temperature, neutralized using 0.5 ml of 8 M NaOH, and buffered using 17 ml of 0.11 M citrate buffer (pH 4.5). Five hundred microliters of sample was used for an overnight hydrolysis at 37°C with 25 μl of amyloglucosidase (Roche). The released glucose was determined using a glucose oxidase kit (Sigma). Trehalose and glycogen concentrations were expressed in glucose equivalents per cell (g cell
−1). As glycogen extraction is not considered to be exhaustive, minor variations may be meaningless.
Statistical analysis.
The variation of each trait was investigated using the lme4 package (R program), through the following mixed model of ANOVA:
Z = μ + medium
i + strain
j + block
k + position
l + medium × strain
ij + ε
ijkl, where
Z is the variable, medium is the medium effect (
i = 1, 2, 3), strain is the strain effect (
j = 1, …, 9), block is the random block effect (effect of each week of experimental repetition,
k = 1, …, 11), position is the random position effect (glass reactor position,
l = 1, …, 15), medium × strain is the interaction effect between medium and strain factors, and ε is the residual error. For further statistical analyses (principal component analysis [PCA], linear discriminant analysis [LDA], Spearman's correlations, etc.), the data were corrected for block and position effects (random effects). Since classical significance tests are controversial for fixed effects in mixed models, we used an alternative method based on Markov chain Monte Carlo (MCMC) sampling by means of R's languageR package version 1.0 (
51; R. H. Baayen, 2009, “languageR: data sets and functions with
Analyzing Linguistic Data: a Practical Introduction to Statistics” [
http://cran.r-project.org/web/packages/languageR/index.html]).
P values were then adjusted for multiple testing using Benjamini-Hochberg methods by means of R's multtest package, version 2.1.1 (
7,
27,
51). For each variable, the homogeneity of the variance was assessed using a Levene test by means of R's
car package, version 1.2-15 (
51; J. Fox, 2009, “
car: Companion to Applied Regression” [
http://cran.r-project.org/web/packages/car/index.html]), as well as the normality of residual distribution using a Shapiro test (
51). For AF time,
r, acetic acid, trehalose, and nitrogen consumption variables, a log transformation was necessary to obtain normally distributed residues; for glycerol and ethanol/glucose variables, an inverse transformation was applied, while a square root transformation was used for glycogen and growth recovery.
To get a general overview of the data, a PCA was performed from the following variables:
Vmax, lag-phase time, AF time,
K,
r,
Jmax, CO
2tot, ethanol, acetic acid, glycerol, glycogen, trehalose, biomass, cell size, growth recovery, and nitrogen consumption. PCA was run using the ade4 package from the R program on the ANOVA predicted mean for each trait and for each medium-strain combination. The data were corrected for the random effects and standardized, i.e., mean centered and scaled (
12,
51).
For the food origin study, LDA was performed using R's
mda package, version 0.3-4 (F. Leisch, K. Hornik, and B. D. Ripley, 2009, “mda: mixture and flexible discriminant analysis” [
http://cran.r-project.org/web/packages/mda/index.html]) on the whole set of data corrected for random effects and using the same variables as those for the PCA.
Pairwise correlations between variables (mean of three replicates; data corrected for random effects) were studied using Spearman's rank correlations (ρ). Since 120 multiple correlations were computed,
P values were corrected for multiple testing using Benjamini-Hochberg methods (
7) by means of R's multtest package, version 2.1.1 (
27,
51).
For
Vmax study, multiple linear regressions were performed within each medium to estimate the contribution of
K and
Jmax to
Vmax using the lm function in the R program (
51). The entire data set and summarizing graphs for 15 kinetics and metabolic and life history traits are available as supplemental material (Table S2 and Fig. S1).