Two key metabolic activities of yeast relevant to wine fermentations are nutrient utilization efficiency and wine aroma development. For nutrient utilization efficiency (NUE), variability in yeast cell metabolism results from modulation of cellular processes that include changes in membrane composition along with a range of other metabolic pathways that are not fully understood. This variability often affects the completeness of a fermentation (characterized as “dry”, ”sluggish” or ”stuck”). Moreover, variability in yeast species or strains used in wine production results in different concentrations of aroma compounds, which can lead to distinct sensory characteristics. Controlling factors affecting nutrient utilization efficiency and wine aroma profile and mouthfeel characteristics related to yeast requires a detailed understanding of cellular metabolism. To develop such understanding, studies often use large-scale data approaches (e.g. genomics and metabolomics), along with multivariate statistics, to identify key metabolic fluxes or metabolites whose presence favors a specific fermentation outcome.
Although these studies are useful in exploring variation between yeasts, they are often not comprehensive enough, especially considering that they are labor intensive and costly. An alternative method is to use genome-scale metabolic models combined with dynamic FBA (flux balance analysis) to predict the flux distribution of all the metabolites within the cell over the course of an entire fermentation. As a part of this grant, our goal is to show that this computational approach can be used to predict experimental wine fermentation data, to understand differences between commercial strains, and to suggest genetic modification strategies towards increasing strain performance and control aroma characteristics. To date, we have been able to simulate anaerobic, nitrogen-limited yeast fermentations with the latest genome-scale yeast model. Behavior predicted for changing initial nitrogen concentration matches qualitatively with experiment. We simulated fermentation of three commercial yeast strains with highly varied NUE. Utilizing multivariate statistics, we have used the simulation results to identify the metabolic pathways that differ the most between these strains. On first analysis, the results are in agreement with existing experimental data. It is also clear that having an accurate biomass composition will be critical to a good quantitative fit of the data. Therefore, we are currently pursuing measurement of these key parameters as a function of fermentation time and strain.