Figure 1.
Schematic description of integrating a small example metabolic network with gene expression data.
Construction of the altered metabolic state used gene expression data to constrain and alter the reference fluxes obtained from a metabolic network compatible with the reference condition. The example network contained six metabolites (A–F), two uptake reactions, six enzymatic reactions, and one biomass reaction. In the reference condition, the biomass function contained equal amounts of metabolites E and F, set to 1.0 millimoles per gram dry weight of the organism (mmol/gDW), and the uptake rates for the metabolites A and B were each assigned an upper limit of 2.0 mmol/(h·gDW). In Step I, we obtained the minimum and maximum fluxes under the optimal biomass production rate via flux variability analysis and calculated the average normalized flux for the reference metabolic network. In Step II, the gene expression ratios were mapped to their corresponding reactions. In Step III, we initially set constraints for reactions that were associated with altered gene expression values. These constraints were based on the normalized reference network with the biomass production rate set to one and resulted in increased normalized fluxes through reactions related to up-regulated genes (reactions A→B and D→F) and decreased fluxes related to down-regulated genes (reaction C→D). Because biological activities other than gene transcription can influence reaction fluxes, we introduced a set of non-negative slack variables (L1, L2, and L3) to account for possible violations of the constraints. In Step IV, we further performed a number of optimizations subject to the constraints from the previous step and obtained a new minimum and maximum normalized flux for each reaction. We first minimized the overall violation of the developed constraints in the form of the sum of the slack variables (highest priority). We then minimized the modifications in the biomass objective function and those in the upper limits of metabolite uptakes (medium priority), and, last, we minimized and maximized each reaction flux (lowest priority). Finally, in Step V, we constructed the new metabolic state by calculating the new average normalized flux for each reaction as the mean of its new minimum and maximum fluxes. This metabolic state was representative of the new condition and in this case was associated with altered uptake rates, pathway preferences, and an altered biomass composition.
Figure 2.
Phenotypic characteristics of wild type Mycobacterium tuberculosis H37Rv and ΔdosR under normoxia and hypoxia.
(A) The predicted normalized oxygen uptake rates of Mycobacterium tuberculosis H37Rv and the ΔdosR deletion mutant under normoxia and hypoxia. The oxygen uptake rates were normalized by each strain's biomass production rate. Supplemental Table S1 gives the biomass production rates of each strain under different conditions, which were calculated as described in the Materials and Methods Section. We based the metabolic network models of the hypoxic state on differential gene expression data associated with the change from normoxic air to hypoxic nitrogen gas with 0.2% oxygen (1.5 mm Hg) after 2 hours [25]. The wild type metabolic response involved reducing its oxygen requirement to cope with the low-oxygen stress, while the ΔdosR deletion mutant was not capable of adjusting. (B) The predicted ATP production levels for the same systems as in panel A showed a slight reduction for the wild type and a much larger decrease for the ΔdosR deletion mutant in response to hypoxia. Note that the ATP production rates were not normalized so to facilitate a direct comparison with the experimental data in Refs. [16] and [49]. (C) The modeled growth characteristic of the wild type and ΔdosR deletion mutant were compared with the corresponding experimental data [49]. Following the experimental data presentation, the x-axis plots two different time intervals, 0–20 and 20–200 days, using two different time scales. The initial aerobic growth phase for the first 5 days was followed by a slight decrease in cell concentration upon switching to hypoxic conditions on day 5. Our metabolic model interpretation was also compatible with a slight decrease in cell concentration for wild type and a substantial decrease for the deletion mutant. Because the gene expression data was compatible with the immediate hypoxic response, the validity range of the metabolic model cannot be expected to capture genotypic and phenotypic adaptations beyond an initial adaptation. Here, the calculated growth reductions for the wild type and ΔdosR deletion mutant mimicked the experimental data up to days 60 and 12, respectively.
Figure 3.
Variations of oxygen uptake of wild type Mycobacterium tuberculosis H37Rv are compatible with experimental fluctuations.
The solid vertical bars indicate the wild type normoxic- and hypoxic-normalized oxygen uptake rates based on our integrated metabolic network model and correspond to the same values given in Figure 2A. We calculated the distribution of uptake values by sampling each expression value from a normalized distribution based on its mean and standard deviation and calculating the resultant uptake rates. This distribution, labeled “Fluctuations” in the graph, captured the experimental gene expression variability and was centered on the value derived from using the mean expression values. In contrast, the distribution labeled “Random selection” was derived from randomizing all gene expression data and was far away from either the normoxic or hypoxic conditions. This confirmed that the gene expression data carried sufficient information to guide the metabolic network model to describe the organism's adaptations under hypoxia. The units are mmol per gram dry weight of Mycobacterium tuberculosis.
Figure 4.
Classification of biomass metabolites predicted to change under hypoxia.
Our metabolic network model predicted different adjustments in the biomass composition of wild type Mycobacterium tuberculosis and the ΔdosR deletion mutant under hypoxic stress. We classified these metabolites into the following categories: amino acids (AA), cofactors (COF), mycolates and related derivatives (MYC), nucleotides (NUC), phosphatidyl-myo-inositol mannosides (PIM), precursors of peptidoglycan (PTD), and other (OTHER). The pie charts indicate the numbers of metabolites that changed in each category. In total, the wild type was associated with 51 changes and the ΔdosR deletion mutant with 42 changes. The top-right portion in each chart represents the metabolites that were predicted to increase under hypoxia, while the bottom-left portion represents those predicted to decrease. Supplemental Table S2 provides detailed information for all predicted biomass composition changes.
Figure 5.
Predictions of hypoxia-induced changes in fluxes through central carbon metabolism.
The left panel shows the flux ratios, i.e., the ratios of reaction fluxes under hypoxia to those under normoxia, of wild type Mycobacterium tuberculosis H37Rv and the right panel shows those of the ΔdosR deletion mutant. If the normoxic flux of a reaction was close to zero, we did not calculate the flux ratio for this reaction due to the numerical uncertainty associated with creating the corresponding ratio. The results indicated that the wild type strain activated glucose processing pathways and the predominant reaction flow was on the reductive side of the tricarboxylic acid (TCA) cycle. Conversely, the ΔdosR deletion mutant was not able to cope under hypoxic conditions as evident by an overall reduced activity in the TCA cycle. NAD, nicotinamide adenine dinucleotide; NADP, nicotinamide adenine dinucleotide phosphate; FAD, flavin adenine dinucleotide. NADH, NADPH, and FADH2 are the reduced forms of NAD, NADP, and FAD, respectively. NAD(P), NAD or NADP; NAD(P)H, NADH or NADPH.
Figure 6.
Genes predicted to be essential for Mycobacterium tuberculosis H37Rv to adapt to hypoxia.
Shown are the genes predicted to be nonessential under normoxia but essential under hypoxia for the wild type strain. Given the metabolic state shown in Figure 5, the genes predicted to be essential for hypoxic adaptation were mostly located in the glucose/glycerol processing pathways and on the reductive side of the tricarboxylic acid (TCA) cycle. dctA, Na+/H+-dicarboxylate symporter; eno, enolase; fba, fructose-bisphosphate aldolase; frdA, frdB, frdC, frdD, fumarate reductase; gap, glyceraldehyde-3-phosphate dehydrogenase; gpm, phosphoglycerate mutase; pfKA, pfkB, phosphofructokinase; pgk, phosphoglycerate kinase; ppgK, polyphosphate glucokinase; sdhA, sdhC, sdhD, succinate dehydrogenase; tpi, triosephosphate isomerase.