Inferring carbon sources from gene expression profiles using metabolic flux models

Abstract

Background: Bacteria have evolved the ability to efficiently and resourcefully adapt to changing environments. A key means by which they optimize their use of available nutrients is through adjustments in gene expression with consequent changes in enzyme activity. We report a new method for drawing environmental inferences from gene expression data. Our method prioritizes a list of candidate carbon sources for their compatibility with a gene expression profile using the framework of flux balance analysis to model the organism's metabolic network.

Principal findings: For each of six gene expression profiles for Escherichia coli grown under differing nutrient conditions, we applied our method to prioritize a set of eighteen different candidate carbon sources. Our method ranked the correct carbon source as one of the top three candidates for five of the six expression sets when used with a genome-scale model. The correct candidate ranked fifth in the remaining case. Additional analyses show that these rankings are robust with respect to biological and measurement variation, and depend on specific gene expression, rather than general expression level. The gene expression profiles are highly adaptive: simulated production of biomass averaged 94.84% of maximum when the in silico carbon source matched the in vitro source of the expression profile, and 65.97% when it did not.

Conclusions: Inferences about a microorganism's nutrient environment can be made by integrating gene expression data into a metabolic framework. This work demonstrates that reaction flux limits for a model can be computed which are realistic in the sense that they affect in silico growth in a manner analogous to that in which a microorganism's alteration of gene expression is adaptive to its nutrient environment.

Figure 1. The principles of our method illustrated using a simple metabolic network.

Flux limits in this figure are represented by a thin black outline. Reaction fluxes are represented as shaded regions with flux magnitude proportional to thickness. Flux direction is not indicated. Panel A. Creation of the baseline flux limits (corresponding to in Figure 2, Panel A). Each reaction is given a flux limit corresponding to the maximum optimal flux solution over the two in silico nutrient uptake conditions. The shading is orange for in silico glucose growth, blue for in silico acetate and grey where the two solutions overlap. Panel B. Creation of the glucose expression-derived flux limits (corresponding to in Figure 2, Panel B). Each flux limit shown in Panel A has been scaled by the level of gene expression for in vivo growth on glucose relative to the maximum gene expression for that reaction over both nutrient conditions. The arrows indicate two reactions for which gene expression was significantly lower on glucose than on acetate, resulting in significantly reduced flux limits. Panel C. Effect of the glucose expression-derived flux limits of Panel B on in silico glucose growth. The glucose optimal flux from Panel A (orange region) lies within the limits; biomass production is not changed. Panel D. Effect of the glucose expression-derived flux limits of Panel B on in silico acetate growth. The acetate optimal flux from Panel A (blue region) exceeds the flux limits for several reactions. (This is analogous to the optimal flux vector lying outside the flux cone in Figure 2, Panel B.) Hence the flux limits will lead to smaller optimal fluxes for these reactions and reduced biomass production. Relative biomass production is therefore smaller for in silico acetate than for in silico glucose, and we conclude that glucose is the more likely carbon source for the expression data.

Figure 2. The principles of our method illustrated with flux cones.

Only three of the many reaction fluxes are shown. For simplicity only two in silico candidate nutrients are represented. The figure does not correspond to actual experimental data. Panel A. Creation of the baseline flux limits, represented as a rectangular parallelpiped. Reaction fluxes must lie within the flux cone (grey area). Flux vectors producing maximal biomass for candidate nutrient i are indicated by colored asterisks and labeled These solutions of the baseline FBA model constrained by in silico nutrient uptake lie on the surface of the flux cone. For each dimension j the baseline upper flux limit is denoted Panel B. Creation of the expression-derived flux limits by scaling the baseline flux limits. The upper flux limit for dimension j derived using expression data for the unknown in vitro nutrient l is denoted and the solution vectors are denoted The baseline flux limits are indicated with dashed lines, the scaled limits are indicated by solid lines. In this hypothetical example the expression of the gene for the enzymatic reaction producing flux v₂ is 40% of the maximal expression level for that gene under the other nutrient condition. The maximal flux for this reaction is set to 40% of its original level. This smaller flux cone represents the metabolic capabilities of the organism under the corresponding growth condition. The solution vector producing optimal biomass for nutrient 1 has not changed with the new flux limits, but the solution vector for nutrient 2 has been reduced in magnitude, with a consequent reduction in biomass production. Relative biomass production will be larger for nutrient 1 than for nutrient 2. We would therefore conclude that the in vitro nutrient l that gave rise to the expression profile is probably nutrient 1, rather than nutrient 2.

Figure 3. Priority ordering of candidate nutrients for each expression set.

A panel for each of the six challenge expression sets presents the candidate nutrients ordered from top to bottom in order of decreasing relative biomass production. The length of the horizontal bars indicates the relative biomass production for the corresponding candidate nutrient for that challenge expression set. The bar for the in silico nutrient that corresponds to the in vitro carbon source for which expression was measured (the matching nutrient) is colored green. The other bars (for the non-matching nutrients) are blue. The order of candidate nutrients with equal relative biomass production is not meaningful. In particular this holds for the candidate nutrients with zero relative biomass production. Candidate nutrients with 100% relative biomass production are highlighted in bold.

Figure 4. Near optimal growth is supported by most matching expression profiles.

Each grouping of bars represents the relative biomass production of an in silico candidate nutrient with respect to the six challenge expression sets, which are in decreasing order from top to bottom. The vertical dotted line represents 95% relative biomass production. As in Figure 3, the length of the horizontal bars indicates the relative biomass production for the corresponding candidate nutrient for the indicated challenge expression set. The matching nutrients (green bars) reach the 95% threshold with higher frequency than the non-matching nutrients (blue bars), indicating the matching expression profiles are well adapted to that nutrient, while many of the non-matching profiles are not. The order of the candidate nutrients is that used in Table 1, and is not based on relative biomass production. Only in silico nutrients with a corresponding in vitro nutrient expression set are included in the figure. Candidate nutrients with 100% relative biomass production are highlighted in bold.