Fig 1.
Difference in M- and ME-model scope.
M-models offer a means to comprehensively probe the capabilities of enzymatic conversions possible within an organism. This modeling method is based on the stoichiometry of reactions in the organism’s metabolic network and can be used to predict possible growth supporting nutrient environments (demonstrated in Monk et al.) [25,26]. By mechanistically accounting for enzyme synthesis and activity, ME-models add additional information about the proteome sustaining the growth state. Thus, ME-models offer the ability to study how proteome allocation and cofactor use affects condition-dependent growth. Due to the inconsistency in how the terms cofactor, prosthetic group, and coenzyme are used in the scientific literature, the definitions applied in this study are listed in the table.
Table 1.
Summary of the vitamins synthesized by E. coli K-12 MG1655.
Fig 2.
Comparison of growth-normalized ME- and M-model computed amino acid and cofactor synthesize rates.
The ME-model biomass synthesis demands are a function of the predicted intracellular fluxes provided by the simulation, whereas the M-model values are provided by the biomass objective function. ME-model predictions are shown for aerobic and anaerobic in silico conditions.
Fig 3.
Variation in the synthesis demand (i.e., the amount of each micronutrient that must be synthesized to sustain growth) of enzyme cofactors and amino acids by growth condition.
A) The maximum and minimum biosynthesis demand for each amino acid and cofactor across all growth conditions. B) Stacked bar chart showing the standard deviation in normalized synthesis demand of each nutrient source and aerobicity.
Fig 4.
Differences in the synthesis demand of enzyme cofactors and amino acids by aerobicity.
PCA analysis of all computed growth conditions reveals aerobic (filled points) and anaerobic (outlined points) growth conditions can be resolved by principal component 1. The protein biomass normalized micronutrient demand for L-histidine and NAD are shown next to their component 1 weighting. The two histograms demonstrate the clear separation in aerobicity-dependent demand of these micronutrients. A table of the principal component vector weightings is shown on the right.
Fig 5.
Characterization of aerobic condition-dependent biomass compositions.
A). Outlier analysis was performed to find growth conditions with z-scored biomass constituent synthesis demands with absolute values greater than 3. The colors on the heatmap denote the log2 fold change of the outlier compared to the average biosynthetic demand of all aerobic growth conditions. B) Hierarchical clustering using Ward’s linkage was performed to divide the in silico growth conditions into 6 clusters based on their predicted biomass demands. All 6 clusters contained multiple biomass demand values that were statistically different (p < 1x10-5 by Wilcoxon rank-sum test and log2 fold change >0.15) compared to the non-cluster demands. The significant log2 fold changes are shown comparing growth conditions in each cluster compared to the average for growth conditions not in the cluster. Error bars represent the standard deviation of the log2 fold change values.
Table 2.
Clustering characterization.
Fig 6.
Computed growth rate in auxotrophic models of iJL1678b when the availability of the essential cofactor in the legend is limited.
For each of the 7 metabolites shown in the legend, reactions were imposed into the model creating an auxotrophy for that metabolite (Table C in S1 Text). Top panel: The growth rate is plotted as a function of the availability of the metabolite indicated in the legend. The percent change in the growth rate compared to the wild-type (prototroph) model is shown. Middle panel: Model-predicted metabolic changes in response to tetrahydrofolate limitation. The mass fraction of protein allocated to each metabolic subsystem during tetrahydrofolate limitation is shown (columns) Bottom panel: Heatmap showing fraction of maximum growth rate-normalized reaction fluxes. The 15 reactions with the highest standard deviation are shown and are highlighted in red if the reaction relies on folate activity. If reaction fluxes were perfectly correlated throughout the tetrahydrofolate limitation simulations, then these reactions were grouped together. The number in parentheses shows the number of other reactions represented by the row. The right column depicts a simulation with the highest folate availability and the left column depicts a simulation with the lowest folate availability.