Fig 1.
Metabolic map of the iCH360 model.
The map was created with the metabolic visualisation tool Escher [19] and shows the metabolic subsystems included in the model. Shaded areas denote metabolic subsystems already present in the ECC model [15]. Reaction and metabolite names were omitted from the plot for clarity. Overlaid onto the map is a flux distribution for aerobic growth on glucose, computed via parsimonious FBA.
Table 1.
The main metabolic subsystems covered by iCH360.
Table 2.
Biosynthesis pathways outside of the iCH360 model that were considered to construct a biomass reaction equivalent to the biomass reaction in the iML1515 model.
The right column shows the model identifiers of the main precursors present in iCH360. Note that only the main precursors are shown here, but the equivalent biomass reaction computed also accounts for any net production or consumption of metabolites in the reduced model.
Fig 2.
The iCH360 model shows similar, but more realistic metabolic capabilities than iML1515.
Considering glucose as a feedstock and studying ethanol, lactate, and succinate production, a production envelope analysis yields similar results in the two models (note that the dashed line representing the production envelope of iML1515 is sometimes hidden behind the coloured lines). Growth rate and production fluxes were computed by limiting the glucose uptake rate to a maximum of 10 mmol/gDW/h, so that the production yield (in moles of product per mole of carbon source) can be obtained by dividing the production flux (y-axis) by 10. In the scenario of acetate production (top right panel) iCH360 avoids an unrealistically high production flux [16] as predicted by iML1515. An extended set of production envelope comparisons between the two models is available online in the code repository supporting this manuscript.
Fig 3.
Layers of annotation and biological knowledge supporting the stoichiometric model in iCH360.
A: Annotations for the model reactions point to the BioCyc, MetaNetX, and KEGG databases. Bars show the number of annotations, highlighting the share of annotations that were added to or updated from the parent model iML1515. B: Some of the biological knowledge parsed from EcoCyc (and manually curated) included in the model-supporting functional annotation graph. The graph captures catalytic relationships between reactions and enzymes, protein subunit compositions, protein-gene mappings, and small-molecule regulation interactions, among others. Shown here as an example are the branches of the graph corresponding to the Glutamate Dehydrogenase (GLUDy) and Glutamate Synthase (GLUSy) reactions. C: Examples of catalytic relationships functionally annotated as either primary or secondary in the graph. Note that all catalytic relationships were classified as primary by default, unless sufficient evidence was found to annotate them as secondary. D: Functional annotation of catalytic edges as primary or secondary can be used to improve phenotypic predictions. Left: Classification of catalytic edge disruptions in the network resulting from simulated knockout of genes associated with essential reactions in the model across 9 growth conditions (see text for a description of each disruption class). Right: Comparison of predicted disruption outcomes against a large dataset of mutant fitness data [27] shows that the different types of disruption tend to lead to significantly different fitness changes. Whiskers in the box plot denote the range of data located 1.5 times above and below the interquartile range. Black dots represent data points lying outside this range.
Fig 4.
Enzyme allocation predictions obtained with the model variant EC-iCH360 after adjusting the turnover parameters.
A: Predicted vs measured enzyme abundances for aerobic growth on eight different carbon sources. Each data point represents an enzyme-condition pair. A total of 325 data points corresponding to zero predictions (enzymes associated with zero-flux in the enzyme-constrained FBA solution for a given condition) were omitted from the plot. B: Geometric mean across conditions of predicted vs measured enzyme abundances. For each enzyme, the geometric mean was computed across the conditions with non-zero predicted abundance. A total of 27 data points, corresponding to enzymes with zero predictions across all conditions, were omitted from the plot.
Fig 5.
Growth rates and biomass yields achieved by different elementary flux modes of iCH360red for growth on glucose.
The inset on the top left (corresponding to the area of the plot enclosed by the red rectangle) highlights the front of Pareto-optimal EFMs (squares), with the maximum-growth and maximum yield modes lying at the extremes of the front. The growth rate of each mode was estimated by assuming that the metabolic enzymes in the model occupy, by mass, a constant fraction of the cell’s dry weight (see Methods).
Fig 6.
Saturation FBA enables the exploration of the optimal switching across elementary modes as a function of the growth environment.
A: satFBA predictions for the growth rate as a function of external glucose concentration, showing a typical Monod curve. Note that satFBA computes the cell’s growth rate by assuming a fixed total enzyme mass budget while varying the saturation of the substrate transporter as a function of external substrate concentration. Importantly, although the curve is continuous and smooth, it comprises many smaller sections, each dominated by a different elementary mode. B: satFBA predictions for the acetate excretion flux, showing progressively higher use of fermentative metabolism in the optimal solution as external glucose availability increases. C: The biomass yield of the optimal satFBA solution (in gDW/mol glucose) progressively decreases in a step-like manner as external glucose availability increases. Each jump represents a switch in the optimal elementary flux mode.
Fig 7.
Thermodynamic analysis of the model via the curated thermodynamic parameter set.
A Probabilistic max-min driving force (MDF) analysis of flux distributions obtained by parsimonious flux balance analysis for a total of 12 growth conditions. All flux distributions tested have a positive MDF, implying they are thermodynamically feasible under physiologically reasonable metabolite concentration ranges. The computed MDF values cluster in three groups, corresponding to glycolytic aerobic, gluconeogenic aerobic, and anaerobic growth conditions. B: Fluxes relative to the glucose uptake flux (EX_glc__D_e) and flux-force efficacies computed by probabilistic metabolic optimisation (PMO). The labelled data points represent examples of reactions (excluding transport and spontaneous reactions) with low predicted flux-force efficacy (here, below 20%), but carrying high relative flux in the optimal solution (here, more than 5% of the glucose uptake flux). xyl__D: D-xylose; glc__D: D-glucose; rib__D: D-ribose; glyc: glycerol; akg: alpha-keto-glutarate; ac: acetate; pyr: pyruvate; lac__D: D-lactate; succ: succinate. fum: fumarate.
Table 3.
A summary of knowledge captured by the iCH360 model, as well as example simulations and analyses shown in this article.
Fig 8.
The flux force efficacy (η) as a function of the (scaled) negative Gibbs free energy of reaction, .
The efficacy of the flux force corresponds to the ratio between the net flux (forward minus backward flux) and the total flux (forward plus backward flux) of a reaction, which approaches 1 for reactions operating far from chemical equilibrium ().