Development and analysis of an in vivo-compatible metabolic network of Mycobacterium tuberculosis

Abstract

Background: During infection, Mycobacterium tuberculosis confronts a generally hostile and nutrient-poor in vivo host environment. Existing models and analyses of M. tuberculosis metabolic networks are able to reproduce experimentally measured cellular growth rates and identify genes required for growth in a range of different in vitro media. However, these models, under in vitro conditions, do not provide an adequate description of the metabolic processes required by the pathogen to infect and persist in a host.

Results: To better account for the metabolic activity of M. tuberculosis in the host environment, we developed a set of procedures to systematically modify an existing in vitro metabolic network by enhancing the agreement between calculated and in vivo-measured gene essentiality data. After our modifications, the new in vivo network contained 663 genes, 838 metabolites, and 1,049 reactions and had a significantly increased sensitivity (0.81) in predicted gene essentiality than the in vitro network (0.31). We verified the modifications generated from the purely computational analysis through a review of the literature and found, for example, that, as the analysis suggested, lipids are used as the main source for carbon metabolism and oxygen must be available for the pathogen under in vivo conditions. Moreover, we used the developed in vivo network to predict the effects of double-gene deletions on M. tuberculosis growth in the host environment, explore metabolic adaptations to life in an acidic environment, highlight the importance of different enzymes in the tricarboxylic acid-cycle under different limiting nutrient conditions, investigate the effects of inhibiting multiple reactions, and look at the importance of both aerobic and anaerobic cellular respiration during infection.

Conclusions: The network modifications we implemented suggest a distinctive set of metabolic conditions and requirements faced by M. tuberculosis during host infection compared with in vitro growth. Likewise, the double-gene deletion calculations highlight the importance of specific metabolic pathways used by the pathogen in the host environment. The newly constructed network provides a quantitative model to study the metabolism and associated drug targets of M. tuberculosis under in vivo conditions.

Figure 1

Main steps for the development of the iNJ661v network. In Step I, we compared the gene essentiality of iNJ661m with experimental in vivo data and identified the set of false positive (FP) and false negative (FN) predictions. In Step II, for each incorrect prediction, we attempted to obtain a set of possible modifications. In Step III, we combined all the suggested modifications for each different incorrect prediction and screened the network realizations to obtain adequate and consistent metabolic modifications. In Step IV, we analyzed the availability and blockage of nutrient uptakes. In Step V, we reviewed the relevant literature to verify the biochemical and biological veracity of the introduced modifications. TN, true negative; TP, true positive.

Figure 2

Procedure to correct false positive (FP) gene essentiality predictions. For each predicted FP gene, we attempted to correct the prediction by (I) removing metabolites from the biomass objective function and (II) introducing new nutrient uptakes and/or changing irreversible reactions to reversible (using the optimization model developed by Kumar and Maranas [39]). When a modification was successful, as determined by the criteria shown in Figure 4, we recorded and collected it in a set of possible modifications. TN, true negative; RXN, reaction; Y, yes.

Figure 3

Procedure to correct false negative (FN) gene essentiality predictions. For each reaction associated with a FN gene, we first examined whether the reaction required the presence of both the FN gene and one or more TN genes. If this was not the case, we attempted to correct the FN prediction by blocking the functions of isozyme(s). Next, we examined whether the reaction was in a dead-end pathway, i.e., a pathway containing metabolites that cannot be produced, metabolites that cannot be consumed, or both. If this was the case, we added (I) metabolite uptakes or (II) metabolites to the biomass objective function. The last attempt was to correct the FN prediction by suppressing one or more reactions (using the optimization model developed by Kumar and Maranas [39]). When a modification was successful, as determined by the criteria shown in Figure 4, we recorded and collected it in a set of possible modifications. RXN, reaction; TN, true negative; TP, true positive; Y, yes; N, no.

Figure 4

Criteria to judge whether a modification for an incorrect prediction is adequate. A modification was deemed to be adequate if, after applying the modification to iNJ661 m, the following criteria were met: 1) the calculated wild-type growth rate was greater than the minimal rate (taken to be 0.027 h^-1, according the growth rate of M. tuberculosis in mouse macrophages [32]), 2) the false positive (FP) or false negative (FN) prediction was corrected, and 3) no true positive (TP) or true negative (TN) prediction became FN or FP, respectively. Y, yes; N, no.

Figure 5

Procedures to analyze combined modifications for different incorrect predictions. Each network realization included one modification for every incorrect prediction. A plausible network realization satisfied the following criteria: 1) the network contained no contradictory modification, and, after the combined modifications to iNJ661m were applied, 2) the wild-type growth rate (calculated from flux balance analysis) was greater than the minimal rate (taken to be 0.027 h^-1 according the growth rate of M. tuberculosis in mouse macrophages [32]), 3) incorrect predictions were corrected, and 4) existing correct predictions before the applied modification were still correct. Finally, from all plausible network realizations, we selected the networks that were associated with minimum adjustment. TN, true negative; TP, true positive; FN, false negative; TN, true negative; Y, yes; N, no.

Figure 6

Receiver operating characteristic (ROC) curves for gene essentiality predictions of Mycobacterium tuberculosis. Sensitivity [TP/(TP + FN)] and 1 minus specificity [1 - TN/(TN + FP)] (where TP: true positive, FN: false negative, TN: true negative, and FP: false positive) were calculated as a function of the growth ratio thresholds used to determine gene essentiality in three different network models: iNJ661 (dotted curve), iNJ661m (dashed curve), and iNJ661v (solid curve).

Figure 7

Number of essential gene pairs predicted using iNJ661m and iNJ661v. Flux balance analysis of iNJ661m under in vivo conditions predicted 78 essential gene pairs, whereas iNJ661v predicted 166 essential gene pairs. There were 35 gene pairs predicted to be essential by both network descriptions; 131 gene pairs were only predicted to be essential using iNJ661v, whereas 43 gene pairs were only predicted to be essential using iNJ661 m. Most of the jointly predicted gene pairs were involved in amino acid and nucleotide metabolism.

Figure 8

Metabolic responses of the iNJ661v network to fatty-acid-limited growth. Metabolite flow was characterized for enzymes in the tricarboxylic acid cycle and the glyoxylate shunt pathway. The numbers in the graph indicate ratios of normalized flux-range midpoints. These were calculated based on flux variability analysis for slow and fast growth conditions, where the fluxes were normalized by dividing by the corresponding total growth rates. This normalization removes artifacts introduced by the lower absolute reaction fluxes associated with induced slow growth [16]. CS, citrate synthase; ACONT, aconitase; ICDHy, isocitrate dehydrogenase; OXGDC, 2-oxoglutarate decarboxylase; SSAL, succinate-semialdehyde dehydrogenase; FRD, fumarate reductase; SUCD, succinate dehydrogenase; FUM, fumarase; MDH, malate dehydrogenase; MALS, malate synthase.

Figure 9

Predicted effects of a double-reaction inhibition on the in vivo growth of Mycobacterium tuberculosis. The two inhibited reactions were isocitrate lyase (ICL) and glycerol-3-phosphate dehydrogenase (G3PD). The growth rates (in units of h^-1) were calculated based on flux balance analysis with different upper limits of the fluxes through the two reactions. The upper limits are in unit of mmol/h/gDW, i.e., mmol per hour per gram dry weight of M. tuberculosis. Panel A shows the results of the combinational inhibition using the iNJ661v network; panel B shows the in vitro iNJ661m results.