Research Article

Systems-Based Approaches to Probing Metabolic Variation within the Mycobacterium tuberculosis Complex

  • Emma K. Lofthouse,

    Affiliations: Animal Health and Veterinary Laboratories Agency (Weybridge), Department for Bovine Tuberculosis, New Haw, Surrey, United Kingdom, Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Paul R. Wheeler,

    Affiliation: Animal Health and Veterinary Laboratories Agency (Weybridge), Department for Bovine Tuberculosis, New Haw, Surrey, United Kingdom

  • Dany J. V. Beste,

    Affiliation: Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Bhagwati L. Khatri,

    Affiliation: Animal Health and Veterinary Laboratories Agency (Weybridge), Department for Bovine Tuberculosis, New Haw, Surrey, United Kingdom

  • Huihai Wu,

    Affiliation: Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Tom A. Mendum,

    Affiliation: Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Andrzej M. Kierzek,

    Affiliation: Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Johnjoe McFadden mail

    Affiliation: Department of Microbial and Cellular Sciences, Faculty of Health and Medical Sciences, University of Surrey, Stag Hill, Guildford, Surrey, United Kingdom

  • Published: September 17, 2013
  • DOI: 10.1371/journal.pone.0075913


The Mycobacterium tuberculosis complex includes bovine and human strains of the tuberculosis bacillus, including Mycobacterium tuberculosis, Mycobacterium bovis and the Mycobacterium bovis BCG vaccine strain. M. bovis has evolved from a M. tuberculosis-like ancestor and is the ancestor of the BCG vaccine. The pathogens demonstrate distinct differences in virulence, host range and metabolism, but the role of metabolic differences in pathogenicity is poorly understood. Systems biology approaches have been used to investigate the metabolism of M. tuberculosis, but not to probe differences between tuberculosis strains. In this study genome scale metabolic networks of M. bovis and M. bovis BCG were constructed and interrogated, along with a M. tuberculosis network, to predict substrate utilisation, gene essentiality and growth rates. The models correctly predicted 87-88% of high-throughput phenotype data, 75-76% of gene essentiality data and in silico-predicted growth rates matched measured rates. However, analysis of the metabolic networks identified discrepancies between in silico predictions and in vitro data, highlighting areas of incomplete metabolic knowledge. Additional experimental studies carried out to probe these inconsistencies revealed novel insights into the metabolism of these strains. For instance, that the reduction in metabolic capability observed in bovine tuberculosis strains, as compared to M. tuberculosis, is not reflected by current genetic or enzymatic knowledge. Hence, the in silico networks not only successfully simulate many aspects of the growth and physiology of these mycobacteria, but also provide an invaluable tool for future metabolic studies.


The pathogenic microorganisms constituting the Mycobacterium tuberculosis complex are associated with important human and animal diseases. M. bovis is the causative agent of bovine tuberculosis, a chronic and occasionally fatal infectious disease primarily infecting cattle and other livestock; but is capable of infecting a wide range of mammals and other vertebrates, including humans [1,2]. Bovine tuberculosis causes immense economic loss in many countries, either from loss of livestock, disease testing, or compensation. Worldwide, agricultural losses are estimated to be around $3 billion a year [3]. M. bovis is very closely related to M. tuberculosis, a virulent tubercle bacillus estimated to infect a third of the world’s population and cause the deaths of 1.4 million people each year [4]. In an attempt to prevent tuberculosis infections more than 3 billion individuals [5] have been immunised with M. bovis BCG, a live attenuated derivative of M. bovis.

M. tuberculosis, M. bovis and M. bovis BCG are characterised by 99.9% similarity at the nucleotide level [3,5,6]. However, the genetic deletions, rearrangements and duplications that M. bovis and M. bovis BCG have undergone relative to M. tuberculosis results in widely differing host tropisms, phenotypes and pathogenicity [2,3,5-9]. Whilst large deletions, such as the regions of difference identified in M. bovis and M. bovis BCG, have been shown to encode virulence factors and result in attenuation of infection [9-12], the genetic basis of these profound variations are mostly undefined. Defining the metabolic differences between the three species is of particular importance as metabolic adaptation to the host environment has been highlighted as a key component of the pathogenic strategy of M. tuberculosis [13-17] and is also likely to be important for the virulence of M. bovis.

Previous targeted studies have identified the genetic basis of some observed metabolic differences, for example, the inability of M. bovis to generate energy from glycolytic intermediates [3,18,19]. This defect is thought to be due to the inactivation of pyruvate kinase which causes a disconnection in central metabolism between glycolysis and the Tricarboxylic acid cycle (TCA cycle) [18,19]. However, although there have been focused investigations into the metabolic differences between the human and bovine tubercle bacillus, systems level comparisons of metabolism have not yet been undertaken.

A systems biology approach provides very effective methods for studying metabolism. Genome scale metabolic reaction networks incorporate all known biochemical reactions within a cell and represent a global system where reaction pathways are defined within the context of whole-cell metabolism. These networks are able to predict phenotypic behaviour, aid in hypothesis generation, identify missing reactions and provide information on the robustness of the metabolic networks, which can be used to identify vulnerable pathways that may be targeted with novel drugs. The creation of M. tuberculosis genome scale reaction networks [20-23] has provided a mechanism to study its metabolism in a systemic manner and a basis for the modelling and metabolic comparison of M. bovis and M. bovis BCG strains.

In this study we present the first genome scale metabolic networks for M. bovis and M. bovis BCG, along with a phenotypic analysis of M. tuberculosis, M. bovis and M. bovis BCG. The networks are freely available in the Supporting Information (Tables S1-S3; Models S1-S3) and online for use with our interactive software ( [24]. To our knowledge this is the first time the metabolism of a pathogen and its vaccine strain have been compared on a systems level and any identified differences have the potential to aid investigations into causes of M. bovis BCG attenuation and the suitability of novel vaccines. The models qualitatively predict phenotype data with between 87-88% agreement, gene knockout results with 75-76% accuracy, and quantitative assessment of measured carbon uptake rates as a function of growth rate show that in silico growth rates are comparable to in vitro results. Therefore, these reaction networks successfully simulate many aspects of mycobacterial metabolism and can be used to examine the metabolic differences between these strains.

Materials and Methods

Bacterial strains and growth conditions

M. tuberculosis H37Rv, M. bovis AF2122/97 and M. bovis BCG Pasteur were used for this study. Frozen stocks were maintained in 10% (vol/vol) glycerol at −80°C. Middlebrook 7H9 broth containing 5% (vol/vol) albumin-dextrose-catalase enrichment medium supplement (ADC) (Becton Dickenson), 2 g/L pyruvate and 0.05% (vol/vol) Tween 80 was used to grow cultures from frozen stocks at 37°C, either statically or rolling. Brain heart infusion agar was used to assess culture purity (Becton Dickinson).

For the Roisin’s agar experiments cultures were grown until late exponential phase (OD600 = 1.0) in 7H9 containing ADC, pyruvate and Tween 80 (as described), washed twice with Ringer’s solution with 0.2% (vol/vol) tyloxopol and plated in triplicate onto Roisin’s agar containing sole carbon and nitrogen sources. When testing sole carbon source assimilation, cells were grown on Roisin’s minimal media [25] with 10 g/L agarose containing sole carbon sources at 5 g/L (2-oxoglutarate, L-alanine, D-arabinose, L-arginine, L-asparagine, L-aspartic acid, citrate, D-fructose, fumarate, D-galactose, D-glucose, L-glutamate, L-glutamine, glycerol, glycine, L-histidine, L-isoleucine, D-lactose, L-leucine, L-lysine, malate, D-maltose, D-mannose, L-methionine, L-phenylalanine, L-proline, propanoate, D-raffinose, D-rhamnose, D-ribose, L-serine, D-serine, succinate, D-sucrose, L-threonine, D-trehalose, L-valine, D-xylose). When testing sole nitrogen source assimilation Roisin’s minimal media was used with 5 g/L pyruvate, 10 g/L agarose and 5.9 g/L sole nitrogen source (L-alanine, ammonia, L-arginine, L-asparagine, L-aspartic acid, L-cysteine, L-glutamate, L-glutamine, glycine, L-histidine, L-isoleucine, L-leucine, L-lysine, L-methionine, L-phenylalanine, L-proline, L-serine, D-serine, L-threonine, L-tryptophan, L-tyrosine, urea, L-valine). Plates were incubated for up to 12 weeks. To test for any mutations which may have given false results any positives after 6 weeks were independently tested a further two times.

For growth curves inoculum cultures of M. bovis were grown statically until late exponential phase (OD600 = 1.0) in Roisin’s minimal media with 5 g/L pyruvate and 0.2% (vol/vol) tyloxapol. Using a 1% inoculum 100 ml rolling (2 rpm) cultures were set up in Roisin’s minimal media with 5 g/L D-glucose, 0.2% (vol/vol) Tween 80 or 5 g/L D-glucose and 0.2% (vol/vol) Tween 80 as carbon sources. A Biomate, ThermoSpectronic spectrophotometer was used to take daily OD readings until they started to fall, presumably due to bacterial death.

Construction of GSMN-MB and BCG

The first M. tuberculosis genome scale metabolic network [20] (GSMN-TB) (Table S1; Model S1) created for the H37Rv strain formed the basis for the reconstruction of M. bovis AF2122/97 (GSMN-MB) (Table S2; Model S2) and M. bovis BCG Pasteur (GSMN-BCG) (Table S3; Model S3). Genolist [26] was used to assign respective M. bovis and M. bovis BCG gene numbers to the genes present in GSMN-TB and alter annotations which since the GSMN-TB was published have been assigned with an alternative enzymatic function. Using data compiled by Genolist [26] any mutations between M. tuberculosis, M. bovis and M. bovis BCG leading to changes in protein sequence were identified. Published scientific literature was used to investigate these genes and to identify metabolic differences not caused by DNA mutations within enzyme sequences. This information was used to alter the networks to reflect M. bovis or M. bovis BCG metabolism (Table S4).

Biolog phenotype data

Biolog Phenotype MicroArray [27] experiments which examined the ability of M. tuberculosis H37Rv and M. bovis AF2122/97 to respire in the presence of 190 carbon and 95 nitrogen sources (PM1, 2a, 3b [27]), were obtained from Khatri, B et al. 2013 [8]. Equivalent experiments for M. bovis BCG were carried out using the described method [8]. Three to four biological replicates were performed for each plate and a cell free control plate was also included to test for abiotic dye reduction. The raw data was analysed using OmniLog PM kinetic plot software [27], with final colour intensity values and manual analysis of kinetic curves used to assess respiration. Wells were considered positive if the final colour intensity value minus the negative control and minus the standard deviation of the negative controls was a positive value [28]. Results were verified by analysing the kinetic curves. Substrates considered to be respired had at least 50% of duplicate wells showing a positive result. The confidence levels applied to the data were determined by the number of concordant results for each biological replicate [28]. High, medium and low confidence results were categorised as substrates where 100%, >75% and >50% of replicates produced the same result.

Modelling of phenotype experiments

The composition of Biolog base medium [27] is not published, however, the mycobacteria were not able to respire in this medium without the added carbon or nitrogen source. To model the carbon source experiment we therefore simulated the media as a modified form of Roisin’s minimal media [25] containing unlimited quantities of ammonia, phosphate, iron, sulfate, carbon dioxide and a Biolog carbon source influx of 1 mmol/g DWt/h. Similarly the nitrogen source experiment was simulated using a modified form of Roisin’s media, where ammonia was replaced with 1 mmol/g DWt/h of the Biolog nitrogen source and pyruvate was used as a carbon source (influx at 1 mmol/g DWt/h). Flux balance analysis (FBA) was performed under aerobic conditions using biomass as the objective function.

Radioactive glucose uptake in M. bovis AF2122/97 and M. tuberculosis H37Rv

Cultures of M. tuberculosis and M. bovis were grown until exponential phase (OD600 0.6-0.8) in Roisin’s minimal media with 5 g/L pyruvate and 0.2% (vol/vol) tyloxapol, washed twice and then resuspended in Roisin’s media (no carbon, tyloxapol). 1 ml of culture was added to a universal tube containing 3700 becquerels of [6-14C] D-glucose and 1 ml of 1 M sodium hydroxide and incubated for 4 h at 37°C. Samples were carried out in triplicate with negative controls (identically prepared heat killed suspensions) tested in parallel. Samples were filtered through Whatman GFC glass microfibre filters, washed 3 times with 0.025% (vol/vol) tyloxapol and placed in scintillation tubes with 250 µl of 70% ethanol and 5 ml of BDH Scintran Fluoran flow scintillation fluid. The sodium hydroxide was directly added to the scintillation fluid. Radioactivity was counted using a Packman tri-carb liquid scintillation counter and dpm calculated from a quench curve. Dpm for negative controls were subtracted from dpm for the live samples before the mean average uptake was calculated. For the analysis of [6-14C] D-glucose radioactivity incorporation in M. bovis when both [6-14C] D-glucose and non-radiolabelled Tween 80 were available as carbon sources, the experiment carried out as described above, except that the incubation medium was supplemented with 0.2% (vol/vol) Tween 80.

Modelling of growth rate experiments

To model in vitro growth rate experiments in GSMN-BCG and TB the media was simulated as a modified form of Roisin’s minimal media (see modelling of phenotype experiments) with carbon sources constrained to experimentally derived values [29] and the biomass composition for slow or fast growth used as the objective function.

Gene essentiality predictions and comparison with TRASH and deep sequencing data

The maximal theoretical growth rate of each in silico gene knock out was calculated by removing single genes from the network and performing FBA linear programming as described in Beste et al., 2007 [20]. The computational predictions were compared to the gene essentiality findings of Transposon site hybridization (TraSH) mutagenesis [30] and deep sequencing [31] experiments.


The genome-scale metabolic networks of M. bovis and M. bovis BCG

The genome scale metabolic network of M. tuberculosis, GSMN-TB [20], (Table S1; Model S1) was used as a starting point for the reconstruction of the genome scale networks of M. bovis (Table S2; Model S2) and M. bovis BCG (Table S3; Model S3). Reactions were adapted to reflect M. bovis and M. bovis BCG reactions based on genome annotations, protein sequences and published biochemical data [3,5,6,18,26,32-49] (Table S4). For orthologous genes GSMN-TB gene assignments were changed to M. bovis and M. bovis BCG gene numbers, with each DNA sequence analysed for changes relative to M. tuberculosis. The analysis identified 228 and 204 genes with non-synonymous sequence differences in M. bovis and M. bovis BCG and these genes catalysed approximately 30% of reactions within each network. Only a small proportion (~19%) of sequence differences were predicted to lead to metabolic differences requiring network modifications, the vast majority of which were the removal of genes and reactions (Table 1). For GSMN-MB 42 genes and 14 reactions were deleted from the original GSMN-TB network, whilst slightly fewer genes (40) and more reactions (16) were deleted from the GSMN-BCG. These alterations appear to support the theory that M. bovis has evolved from a progenitor of the M. tuberculosis complex and that M. tuberculosis is more closely related to this common ancestor [50].

Glycerol kinase (glpK)1-1
Glycerol-3-phosphate dehydrogenase (glpD)223
GDP-D-rhamnose biosynthesis (gca, gmdA)211
GDP-4-dehydro-6-deoxy-D-mannose epimerase100
UTP-hexose-1-phosphate uridylyltransferase (galT)211
β-glucosidase (bglS)1--
Pyruvate kinase (pykA)1-1
Isocitrate lyase (icl)122
(S)-2-hydroxy-acid oxidase1--
Nitrate reductase (nar)5--
Fumarate reductase (frd)877
Glycine dehydrogenase (gcvB)<=>=>=>
L-serine ammonia-lyase (sdaA)11-
Alanine dehydrogenase (ald)1--
Precorrin-6Y C5,15-methyltransferase1--
Molybdopterin biosynthesis protein (moaE)3--
enoyl-CoA hydratase/isomerases (echA)212121
Methoxy mycolic acid synthase (mmaA3)11-
Synthesis of methoxy mycolic acidsBiomassBiomass
Polyketide synthase (pks15/1)-11
Sulfolipid-1 synthesis1--
Mas-like gene (msl3, msl4, msl5)433
Nitrate transporter (narK2)1--
Phosphate transport via ABC system8--
Glycerol-3-phosphate antiporter (ugp)4--
Mycoside bBiomassBiomass
Triacylglycerol synthases151414

Table 1. Changes to the GSMN-TB network to create GSMN-MB and GSMN-BCG.

Numerical value Number of genes catalysing the reaction; - indicates reaction is deleted from the network; 0 indicates an orphan reaction
=> Irreversible reaction
<=> Reversible reaction
Biomass Required for Biomass production

The constructed metabolic networks for both M. bovis and M. bovis BCG (Table 2) therefore contain fewer reactions, metabolites and genes than an updated GSMN-TB network, GSMN-TB 1.1. GSMN-TB 1.1 includes some corrections to the original GSMN-TB, plus additional pathways, such as cholesterol metabolism, that were not implemented in the earlier published version [20]. Due to the reduction in genes within the networks the M. bovis and M. bovis BCG models have a slightly higher fraction of essential genes (31% and 30% respectively; Tables S5 and S6) compared to GSMN-TB 1.1 (29%; Table S7). These values are slightly lower than the predicted value of 35% essential genes in the entire M. tuberculosis genome but within the 95% confidence interval (28-41%) [51]. All three GSMN networks are available in the Supporting Information (Tables S1-S3; Models S1-3) and online for use with our interactive software ( [24].

Reaction network
Total number of reactions876863861
Cytosolic reactions745735733
Transport reactions131128128
Orphan reactions198200200
Total number of metabolites766757754
Internal metabolites667660657
External metabolites999797

Table 2. Statistics of the mycobacterial reaction networks.

Although the GSMN-MB and BCG models gave very similar predictions to the GSMN-TB 1.1, some interesting differences between in silico predictions and published experimental data [3,18,19,33,44] were found. GSMN-MB was able to utilise carbohydrates such as glucose in silico, although in vitro M. bovis is actually unable to grow on these substrates [18,19] (discussed further below). Another area where in silico predictions did not accord with experimental data was in amino acid synthesis. In M. tuberculosis, alanine dehydrogenase catalyses the oxidative deamination of L-alanine or, in the reverse direction, the reductive amination of pyruvate to yield alanine, but this activity is lost [33,44] in M. bovis and M. bovis BCG due to a frameshift caused by single base pair deletion [3]. Deletion of alanine dehydrogenase from the bovine networks resulted in a non-feasible network (no growth) unless L-alanine was supplied as a substrate, since this is the only biosynthetic route leading to L-alanine in the network. Yet, in contrast to the predictions M. bovis and M. bovis BCG are not alanine auxotrophs. Alanine dehydrogenase knockout mutants of M. tuberculosis are also not alanine autotrophs [19] and 13C labelling experiments detected identical alanine labelling patterns [29] for M. tuberculosis and M. bovis BCG; indicating that an alternative pathway for alanine synthesis must be active in all three strains. To preserve alanine prototrophy, the alanine dehydrogenase reaction has therefore been retained in the GSMN-MB and BCG models as an irreversible orphan reaction [19,33,44].

Validation of the model by comparison with Biolog phenotype data

The Biolog [27] high throughput phenotyping system was utilised to obtain additional insight into metabolic capability of M. bovis and M. bovis BCG and to further test the predictive accuracy of the GSMN-MB and GSMN-BCG networks. For comparison, we also examined M. tuberculosis. Biolog [27] is a commercially available phenotype microarraying platform that is capable of high-throughput screening for the ability to utilise a large number of substrates. It employs the reduction of tetrazolium dye by NADH as a reporter system for measuring respiration [27]. Respiration is of course different from growth, as predicted by the network models; but our working assumption was that some degree of growth has to occur to reduce the respiratory substrate sufficiently to see a positive reaction.

Carbon substrate utilisation for M. tuberculosis, M. bovis and M. bovis BCG is presented in Table 3 (showing only substrates metabolised by at least one strain). Of the 190 carbon sources tested (Tables S8-S10) 33 substrates were metabolised by at least one of the mycobacteria tested, with 17 utilised by all three. These substrates included amino acids, TCA cycle intermediates, sorbitan derivatives, 3 carbon compounds and hexose or hexose containing carbohydrates. M. tuberculosis was able to respire more of these substrates (27) than M. bovis (25) or M. bovis BCG (22). Of the 95 nitrogen sources tested (Tables S11-S13) only 13 were able to be utilised as sole nitrogen sources (Table 4) by at least one of the three strains and these were mainly amino acids. Unlike carbon source data, the number of nitrogen sources utilised was similar between the three species.

M. tuberculosisM. bovisM. bovis BCG
SubstratesBiologRoisin’s agarin silicoBiologRoisin’s agarin silicoBiologRoisin’s agarin silico
Acetoacetic acidCNT-CNT-CNT-
Butyric acidCNT-CNT-CNT-
D-malic acid-NTCCNTCCNTC
L-malic acidCCCCCCCCC
Mono methyl-succinateCNT-CNT-CNT-
Oxalomalic acidCNT-CNT-CNT-

Table 3. Carbon substrates utilised by M. tuberculosis, M. bovis and M. bovis BCG.

C Utilised as a carbon substrate- Not utilised as a carbon substrateNT Not tested
M. tuberculosisM. bovisM. bovis BCG
SubstrateBiologRoisin’s agarin silicoBiologRoisin’s agarin silicoBiologRoisin’s agarin silico
L-Aspartic Acid-NN-NNNNN
L-Glutamic AcidNNNNNNNNN

Table 4. Nitrogen substrates utilised by M. tuberculosis, M. bovis and M. bovis BCG.

N Utilised as a nitrogen substrate- Not utilised as a nitrogen substrateNT Not tested

Analysis of in silico predictions verses Biolog data showed that all three networks qualitatively predicted the Biolog data with a similar overall accuracy. For GSMN-TB 1.1, GSMN-MB and GSMN-BCG simulations 84%, 81% and 84% of substrates matched Biolog results respectively (Tables S8-S13). Substrate analysis identified results that appeared anomalous with previous studies [18,19,33,44,52]. For instance, M. bovis tested positive for glucose and glycerol respiration, however, as mentioned above, M. bovis is unable to utilise carbohydrates including glucose and glycerol as sole carbon sources [18,19]. Similarly, many in silico predictions for amino acid utilisation did not correlate with experimental data [33,44,52]. To independently probe the discordant results M. tuberculosis, M. bovis and M. bovis BCG were grown on minimal Roisin’s agar media containing sole carbon or nitrogen sources.

Overall (Tables S8-S13) in vitro experiments using Roisin’s agar media resolved many of the inconsistencies between in silico predictions and the Biolog data. For instance, in accordance with expectations, M. bovis was unable to grow on either glucose or glycerol when provided as the sole carbon source in Roisin’s agar. The anomalous positive Biolog results may be due to the small amount of Tween 80 present in the Biolog base media. It has been shown previously [8,53], and confirmed here (see below), that Tween 80 and glucose are used synergistically for growth by M. bovis strains.

When the anomalous Biolog results were corrected using Roisin’s agar data the GSMN-MB model accurately predicts 87% of phenotypes studied, whilst the GSMN-BCG predicts 88% correctly. The GSMN-TB 1.1 is slightly more accurate (91%) at simulating cellular phenotypes than the bovine networks due to false positive in silico growth predictions for M. bovis and M. bovis BCG. In this analysis results generated on Roisin’s agar media were used instead of Biolog results when the outcomes differed because growth on agar plates assesses biomass production rather than respiration. Therefore, growth on Roisin’s agar media more accurately tests in silico predictions of growth than Biolog experiments.

Analysis of glucose metabolism

GSMN-MB incorrectly predicts that M. bovis should grow on glucose as a sole carbon source. In silico, glucose enters the central metabolism of M. bovis via the glycolytic pathway and bypasses the blocked pyruvate kinase connection between glycolysis and the TCA cycle via either the anaplerotic/gluconeogenic enzyme, phosphoenolpyruvate carboxykinase, that interconverts phosphoenolpyruvate (glycolytic intermediate) and oxaloacetate (TCA cycle intermediate) or a serine-glycine pathway (phosphoglycerate dehydrogenase, phosphoserine transaminase, phosphoserine phosphatase, glycine hydroxymethyltransferase and glycine dehydrogenase (Figure 1)). The hypothesis is that a second metabolic or regulatory defect in glucose metabolism contributes to the M. bovis glucose phenotype, in addition to the inactive pyruvate kinase [18,19].


Figure 1. GSMN-MB in silico flux prediction when glucose is a sole carbon source.

The in silico prediction of flux from glucose to the TCA cycle when glucose is a sole carbon source for M. bovis.

Acon: aconitase, Cit: citrate synthase, Eno: enolase, Fba: fructose-bisphosphate aldolase, Gck: glucokinase, Gdh: glycine dehydrogenase, GlcB: malate synthase, GlyA: glycine hydromethytransferase, Gpm: phosphoglycerate mutase, Icl: isocitrate lyase, Mdh: malate dehydrogenase, Pepck:, phosphoenolpyruvate carboxykinase, Pfk: 6-phosphofructokinase, Pgi: glucose-6-phosphate isomerase, SerA: phosphoglycerate dehydrogenase, SerB: phosphoserine phosphatase, SerC: phosphoserine transaminase, TpiA: triose-phosphate isomerase.


To further explore M. bovis glucose metabolism, M. bovis and M. tuberculosis were incubated with [6-14C] D-glucose as the only available carbon source for four hours. M. bovis could uptake a small amount of the labeled glucose but the rate was 16-fold less than M. tuberculosis, consistent with a defect in glucose uptake (Table 5). Interestingly, supplementation of the glucose media with Tween 80 significantly stimulated glucose uptake by M. bovis (2.5 times). However, the fate of this additional glucose was not oxidation to carbon dioxide as the amount of 14C-CO2 generated was unchanged by the Tween 80 supplementation. Maximum growth rates and OD600 achieved on these substrates also indicated synergistic interaction [8] as growth of M. bovis on Roisin’s media with glucose and Tween 80 exceeded that achieved with either compound alone (Table 5).

Becquerels per mg dry weightGrowth rates on non-radiolabelled substrates
M. tuberculosisM. bovis
Carbon substratesAssimilatedAssimilatedCO2 evolvedMaximum growth rateMaximum OD
[6-14C] D-glucose56.6 +/- 3.23.5 +/- 0.91.0 +/- 0.2-0.015 +/- 0.004
Tween 80N/AN/AN/A0.007 +/- 0.0010.545 +/- 0.044
[6-14C] D-glucose and Tween 80N/A8.7 +/- 0.71.0 +/- 0.30.013 +/- 0.0011.171 +/- 0.066

Table 5. Glucose uptake experiments in M. bovis and M. tuberculosis

Analysis of amino acid metabolism

Biolog data and in silico predictions were notably different for the utilisation of amino acids as sole carbon or nitrogen sources (Table 6) as the models predicted a greater metabolic potential than was actually demonstrated experimentally. Many of these discrepancies were resolved by the Roisin’s agar experiments, as more amino acids were shown to support growth in Roisin’s agar than were positive for respiration in Biolog. Possible reasons for this include large differences in the incubation times for these experiments. In support of this theory, the amino acids not utilised in the Biolog experiments predominantly produced dysgonic colonies with a lag time of around 6-8 weeks on Roisin’s agar (M. bovis: arginine C, N; glutamine C; glutamate C; glycine C; isoleucine C, N; proline C; serine C, N; M. bovis BCG: arginine C, N; glutamine C, N; glycine C; isoleucine C; proline C; serine C). Interestingly, the experimental data showed that the number of viable substrates decreased from M. tuberculosis to M. bovis to M. bovis BCG. This was not however reflected by significant network differences.

M. tuberculosisM. bovisM. bovis BCG
SubstrateBiologRoisin’s agarin silicoBiologRoisin’s agarin silicoBiologRoisin’s agarin silico

Table 6. The utilisation of amino acids as carbon and nitrogen sources by M. tuberculosis, M. bovis and M. bovis BCG

C Utilised as a carbon substrate N Utilised as a nitrogen substrate- No respiration/growthNT Not tested

The amino acid data regarding serine metabolism was of particular interest because the results differed between the three mycobacteria and appeared inconsistent with in silico simulations. In silico serine metabolism is relatively complex compared to other amino acids as multiple pathways converge around serine (Figure 2). However, in vitro it is likely only the interconversion of pyruvate and serine (serine dehydratase) enables utilisation of serine as a sole nitrogen source: as inadequate expression of serine dehydratase results in the inability of some BCG strains, including M. bovis BCG Pasteur, to utilise serine as a nitrogen source [33]. M. tuberculosis, which has been found to adequately express serine dehydratase can utilise L-serine [33,52], as can M. bovis strains (Table 6). In silico, however, the GSMN-BCG inaccurately predicts that serine is a viable sole nitrogen source for M. bovis BCG with flux to glycine (serine hydromethyltransferase) enabling viability. The three mycobacteria used in this study have two serine hydromethyltransferases (glyA1and glyA2); with enzyme activity demonstrated in M. tuberculosis [54]. The glyA1 of M. bovis and M. bovis BCG have an amino acid substitution as compared to M. tuberculosis although the effect on the enzyme activity has not been tested. If the interconversion of serine and glycine is feasible in M. bovis BCG, this reaction doesn’t appear to take place when serine is supplied as sole nitrogen source. However, because serine was shown to be a viable sole carbon source for M. tuberculosis, M. bovis and M. bovis BCG, flux from serine must be able to enter the TCA cycle. The most probable route for this would be via glycine (Figure 2).


Figure 2. The metabolic pathways that converge around L-serine in Mycobacterium species.

Blue: present in M. tuberculosis, M. bovis and M. bovis BCG, Red: present in M. tuberculosis and M. bovis [33], Green: present in M. tuberculosis and M. bovis BCG [3,18,19], Purple: present in M. tuberculosis [44]..

Acon: aconitase, Cgl: cystathionine gamma-lyase Cit: citrate synthase, CysE, CysK1, CysK2: cysteine synthase, CysM: cystathionine beta-synthase, Eno: enolase, Gdh: glycine dehydrogenase, GlcB: malate synthase, GlyA: glycine hydromethytransferase, Gpm: phosphoglycerate mutase, Icl: isocitrate lyase, Mdh: malate dehydrogenase, Pdh: pyruvate dehydrogenase, PykA: pyruvate kinase, Ppdk: pyruvate phosphate dikinase, SerA: phosphoglycerate dehydrogenase, SerB: phosphoserine phosphatase, SerC: phosphoserine transaminase, SdaA: serine deaminase.


Validation of the model by comparison with in vitro growth rates

The GSMN-BCG was used to predict cellular growth rates on Roisin’s media using published experimental data [29] on substrate uptake rates and corresponding growth rates. Overall, in silico predictions were similar to the experimentally-determined growth rates (Table 7) and the growth rate generated by the GSMN-TB 1.1 network. Due to the inability of M. bovis to utilise glycerol, GSMN-MB was not tested.

M. bovis BCGM. tuberculosis
SubstrateSpecific consumption rate (mmol g biomass-1 h-1)in vitro growth ratein silico growth ratein silico growth rate
Tween 800.002
Tween 800.09

Table 7. Comparison of in silico and in vitro growth rates on Roisin’s minimal media using calculated substrate uptake rates.

Validation of the model by comparison with global mutagenesis data

In silico gene essentiality predictions were compared with in vitro gene essentiality data as determined by Transposon site hybridization (TraSH) mutagenesis [30] and deep-sequencing [31] (Tables S5-S7). All models gave a very similar predictive accuracy (76-77%; Table 8) when compared against TraSH data, however, results were only available for 82% of genes in the models. Comparison of predictions with a more comprehensive evaluation (100% network coverage) of gene essentiality by deep-sequencing also generated a similar predictive accuracy (75%; Table 8).

TraSHDeep sequencing
True positive23%24%24%23%24%24%
False positive8%8%8%6%6%6%
False negative16%16%16%19%19%20%
True negative53%51%51%52%50%50%
Correct predictions77%76%76%75%75%75%
p value0.0050.0110.0080.9550.6870.863

Table 8. Accuracy of in silico gene essentiality predictions.

Percentage of in silico gene essentiality predictions categorised as: true-positive: essential both in silico and in vitro; false-positive: essential in silico, nonessential in vitro; true-negative: nonessential in silico and in vitro; false-negative: nonessential in silico, essential in vitro

Designation of a gene as either essential or non-essential is a binary characteristic that is generated from a continuous measurement of growth rate or mutant abundance (determined by microarray or sequencing) by applying an arbitrary cut-off value. To examine the influence of the cut-off value on predictive accuracy we plotted Receiver Operating Characteristic (ROC) curves (Figure 3; Figure S1-5). The majority (~91%) of in silico mutants generated a predicted growth rate equal to the wild-type or zero, so variation in the in silico growth rate threshold had very little influence on the result. However, in vitro cut-offs did influence prediction accuracy so ROC curves could be used to identify optimal values for the cut-off value of the in vitro measurement signal. For both TraSH and deep sequencing datasets the optimal microarray and p-value cut-offs were the original values of 0.2 and 0.05 respectively.


Figure 3. The GSMN-MB ROC curve for TraSH thresholds.

The plot shows ROC curves for different transposon site hybridisation (TraSH) ratio thresholds for the determination of essential genes in experimental data [30]. Five ROCs are plotted with 4 different TraSH thresholds as shown in the legend box. Each ROC curve shows the points corresponding to True positive rate (sensitivity) and false positive rate (1-specificity) of the model predictions obtained for all growth rate thresholds. For all ROC curves see Figures S1-S5.


Examining different areas of metabolism, the predictive accuracy (using deep sequencing data) varied from 61-62% in central carbon metabolism (glycolysis, TCA cycle, pentose phosphate cycle, methylcitrate cycle and anaplerotic reactions) to 95-96% in β-oxidation of fatty acids (Table 9). The chief source of errors for central carbon metabolism genes were false-negatives: genes predicted to be non-essential in silico but experimentally found to be essential. This likely reflects the multiple alternative pathways available for flux in in silico central metabolism, many of which are likely to be incapable of supporting sufficient flux for growth in real organisms. Peripheral areas of the network tend to generate higher predictive accuracies as fewer alternative pathways are available. Sources of false-negative in silico predictions may be due cross feeding between mutants in global mutagenesis studies.

Amino acid metabolism767575
Carbohydrate metabolism (excl. central metabolism)727571
Cell wall synthesis888888
Central carbon metabolism616261
Cofactor biosynthesis666666
Lipid biosynthesis777778
Nucleotide biosynthesis706666
Other functions656363
Transport reactions949595
β-oxidation of fatty acids959696

Table 9. Percentage accuracy of gene essentiality predictions for each reaction pathway within the reaction networks.


The first genome scale metabolic models for M. bovis and M. bovis BCG, GSMN-MB and BCG, instantiate current metabolic knowledge and deliver a high degree of accuracy for predicting in vitro data. In combination with a M. tuberculosis network we have demonstrated how these models, and their application to interrogate high-throughput experimental data, shed new light on metabolic differences between the vaccine, bovine and human strains of the tubercle bacillus. Therefore, these reaction networks successfully simulate many aspects of mycobacterial metabolism, and provide an invaluable tool for the investigation of metabolic differences between these strains.

In silico models are essentially a mathematical instantiation of current knowledge. Interrogating the models to generate predictions that can be tested against experimental datasets is thereby an efficient means of probing inconsistencies and limitations of current knowledge. The study described here illustrates this approach. The discordance between in silico and in vitro phenotypes highlights the limitations of FBA predictions in which any route for flux from substrates to products may be utilised by the solution. For instance, the model predicted growth of M. bovis on glucose. This is contrary to expectations; indeed, this inability is a distinguishing feature of the M. bovis strain of the tubercle bacillus. Presumably the pathways utilised in silico to connect between glycolysis and the TCA cycle are not able to support growth in vitro [18,19] and additional defects [18], such as the identified deficiency in glucose uptake are also contributing to this phenotype.

However, our studies show that despite being unable to utilise glucose as a sole carbon source, the sugar can be assimilated by M. bovis. Indeed, glucose assimilation was stimulated by addition of Tween 80 to the glucose media [8,53]. Tween 80 is hydrolysed by mycobacteria to release fatty acids, such as oleic acid, that may be oxidised to acetate and thereby enter the TCA cycle to support the synthesis of essential metabolic precursors, such as oxaloacetate and α-ketoglutarate, as well as providing substrates for energy generation. The result suggests that M. bovis is capable of assimilating carbohydrates such as glucose, but not delivering carbon from those substrates to the TCA cycle [18]; as the additional glucose was not oxidised to CO2 and was presumably incorporated directly into cell biomass. The result is consistent with a previous study that demonstrated compartmentalisation during co-utilisation of different substrates in M. tuberculosis [55]. When supplied with both carbohydrate (such as glucose) and fatty acid (acetate), the carbohydrate was assimilated via glycolysis and the pentose phosphate cycle and mostly incorporated into biomass; whereas acetate was mainly used for energy generation via the TCA cycle [55]. It is interesting that this disconnect is unidirectional: M. bovis can utilise acetate as a sole carbon source so must be able to drive flux from acetate to glucose via gluconeogenic pathways.

Interrogating in silico models with high throughput phenotype and gene essentiality data provides a powerful route towards refinement of the networks, but also provides an insight into the complex relationship between genome and phenotype. The mycobacterial networks are well suited for investigations into the system-wide effects of genetic mutations due to their close evolutionary history, high genetic similarly and diverse phenotypes [3,5,6,8,50]. Both phenotype and gene essentiality data were predicted with a high degree of accuracy, but interestingly key discrepancies between existing genetic knowledge and experimental phenotypes were observed. For instance, the in silico models predicted very few metabolic differences between M. tuberculosis, M. bovis and M. bovis BCG, yet substrate utilisation capability between the strains decreased from M. tuberculosis > M. bovis and M. bovis BCG. It seems that during their evolutionary passage from their common ancestor (that is presumed to be closer to M. tuberculosis) [50], M. bovis and M. bovis BCG have lost some of their capability to metabolise compounds. The basis of this loss in metabolic versatility is unknown but could result from differences in enzyme regulation. For instance, the loss of pyruvate kinase in M. bovis results in a global alteration of enzyme expression which reroutes the catabolic pathways of metabolic substrates [19].

By comprehensively testing the utilisation of amino acids as sole nitrogen sources, loss in metabolic versatility was particularly evidenced in this study. M. tuberculosis, M. bovis and M. bovis BCG were able to assimilate nitrogen from 12, 8 and 6 amino acids respectively, despite current knowledge resulting in almost identical in silico predictions. The only difference in silico is the ability of M. tuberculosis to utilise alanine as a nitrogen (and carbon) source in contrast to M. bovis and M. bovis BCG. This difference corresponds to the previously mentioned mutation in the alanine dehydrogenase gene in M. bovis and M. bovis BCG, indicating that, in accordance with previous studies [33,44] this gene is required for alanine assimilation. Another interesting finding was that some amino acids, such as serine, could act as carbon but not nitrogen sources, indicating that the pathways responsible for amino acid degradation differ for carbon and nitrogen assimilation.

The differences observed between in silico and in vitro data not only identify areas of metabolism which require further investigation, but enable iterative network modifications. Since the model development process is continuous, the networks are altered as new information becomes available. These genome scale models already successfully simulate many aspects of mycobacterial growth and metabolism, but it is to be expected that the networks will gradually become more representative of cellular metabolism over time. Further comparative analysis will help to uncover the genetic basis for the observed phenotypic and pathogenic differences between these mycobacteria, and stimulate new approaches to the control of these diseases, such as the development of novel vaccines.

Supporting Information

Figure S1.

GSMN-TB 1.1 TraSH ROC curve.



Figure S2.




Figure S3.

GSMN-TB 1.1 Deep sequencing ROC curve.



Figure S4.

GSMN-MB Deep sequencing ROC curve.



Figure S5.

GSMN-BCG Deep sequencing ROC curve.



Table S1.

GSMN-TB 1.1.



Table S2.




Table S3.




Table S4.

Alterations to GSMN-TB 1.1 to create GSMN-MB and BCG networks.



Table S5.

GSMN-MB gene essentiality.



Table S6.

GSMN-BCG gene essentiality.



Table S7.

GSMN-TB 1.1 gene essentiality.



Table S8.

M. tuberculosis carbon utilisation.



Table S9.

M. bovis carbon utilisation.



Table S10.

M. bovis BCG carbon utilisation.



Table S11.

M. tuberculosis nitrogen utilisation.



Table S12.

M. bovis nitrogen utilisation.



Table S13.

M. bovis BCG nitrogen utilisation.



Model S1.




Model S2.




Model S3.




Author Contributions

Conceived and designed the experiments: EKL PRW DJVB BLK AMK JM. Performed the experiments: EKL PRW BLK. Analyzed the data: EKL PRW DJVB HW JM. Contributed reagents/materials/analysis tools: PRW HW AMK. Wrote the manuscript: EKL PRW DJVB JM. Making the reaction networks: EKL HW TAM AMK JM.


  1. 1. de Lisle GW, Mackintosh CG, Bengis RG (2001) Mycobacterium bovis in free-living and captive wildlife, including farmed deer. Rev Sci Tech 20: 86-111. PubMed: 11288522.
  2. 2. O’Reilly LM, Daborn CJ (1995) The epidemiology of Mycobacterium bovis infections in animals and man: a review. Tuber Lung Dis 76 Suppl 1: 1-46. doi:10.1016/0962-8479(95)90025-X. PubMed: 7579326.
  3. 3. Garnier T, Eiglmeier K, Camus JC, Medina N, Mansoor H et al. (2003) The complete genome sequence of Mycobacterium bovis. Proc Natl Acad Sci U S A 100: 7877-7882. doi:10.1073/pnas.1130426100. PubMed: 12788972.
  4. 4. World Health Organization (2012) Global Tuberculosis Report 2012.
  5. 5. Brosch R, Gordon SV, Garnier T, Eiglmeier K, Frigui W et al. (2007) Genome plasticity of BCG and impact on vaccine efficacy. Proc Natl Acad Sci U S A 104: 5596-5601. doi:10.1073/pnas.0700869104. PubMed: 17372194.
  6. 6. Cole ST, Brosch R, Parkhill J, Garnier T, Churcher C et al. (1998) Deciphering the biology of Mycobacterium tuberculosis from the complete genome sequence. Nature 393: 537-544. doi:10.1038/31159. PubMed: 9634230.
  7. 7. Golby P, Hatch KA, Bacon J, Cooney R, Riley P et al. (2007) Comparative transcriptomics reveals key gene expression differences between the human and bovine pathogens of the Mycobacterium tuberculosis complex. Microbiology 153: 3323-3336. doi:10.1099/mic.0.2007/009894-0. PubMed: 17906132.
  8. 8. Khatri B, Fielder M, Jones G, Newell W, Abu-Oun M et al. (2013) High Throughput Phenotypic Analysis of Mycobacterium tuberculosis and Mycobacterium bovis Strains’ Metabolism Using Biolog Phenotype Microarrays. PLOS ONE 8: e52673. doi:10.1371/journal.pone.0052673. PubMed: 23326347.
  9. 9. Pym AS, Brodin P, Brosch R, Huerre M, Cole ST (2002) Loss of RD1 contributed to the attenuation of the live tuberculosis vaccines Mycobacterium bovis BCG and Mycobacterium microti. Mol Microbiol 46: 709-717. doi:10.1046/j.1365-2958.2002.03237.x. PubMed: 12410828.
  10. 10. Lewis KN, Liao R, Guinn KM, Hickey MJ, Smith S et al. (2003) Deletion of RD1 from Mycobacterium tuberculosis mimics bacille Calmette-Guerin attenuation. J Infect Dis 187: 117-123. doi:10.1086/345862. PubMed: 12508154.
  11. 11. Kozak RA, Alexander DC, Liao R, Sherman DR, Behr MA (2011) Region of difference 2 contributes to virulence of Mycobacterium tuberculosis. Infect Immun 79: 59-66. doi:10.1128/IAI.00824-10. PubMed: 20974821.
  12. 12. Pym AS, Brodin P, Majlessi L, Brosch R, Demangel C et al. (2003) Recombinant BCG exporting ESAT-6 confers enhanced protection against tuberculosis. Nat Med 9: 533-539. doi:10.1038/nm859. PubMed: 12692540.
  13. 13. Schnappinger D, Ehrt S, Voskuil MI, Liu Y, Mangan JA et al. (2003) Transcriptional Adaptation of Mycobacterium tuberculosis within Macrophages: Insights into the Phagosomal Environment. J Exp Med 198: 693-704. doi:10.1084/jem.20030846. PubMed: 12953091.
  14. 14. Marrero J, Rhee KY, Schnappinger D, Pethe K, Ehrt S (2010) Gluconeogenic carbon flow of tricarboxylic acid cycle intermediates is critical for Mycobacterium tuberculosis to establish and maintain infection. Proc Natl Acad Sci U S A 107: 9819-9824. doi:10.1073/pnas.1000715107. PubMed: 20439709.
  15. 15. Venugopal A, Bryk R, Shi S, Rhee K, Rath P et al. (2011) Virulence of Mycobacterium tuberculosis depends on lipoamide dehydrogenase, a member of three multienzyme complexes. Cell Host Microbe 9: 21-31. doi:10.1016/j.chom.2010.12.004. PubMed: 21238944.
  16. 16. Shi S, Ehrt S (2006) Dihydrolipoamide acyltransferase is critical for Mycobacterium tuberculosis pathogenesis. Infect Immun 74: 56-63. doi:10.1128/IAI.74.1.56-63.2006. PubMed: 16368957.
  17. 17. McKinney JD, Höner zu Bentrup K, Muñoz-Elías EJ, Miczak A, Chen B et al. (2000) Persistence of Mycobacterium tuberculosis in macrophages and mice requires the glyoxylate shunt enzyme isocitrate lyase. Nature 406: 735-738. doi:10.1038/35021074. PubMed: 10963599.
  18. 18. Keating LA, Wheeler PR, Mansoor H, Inwald JK, Dale J et al. (2005) The pyruvate requirement of some members of the Mycobacterium tuberculosis complex is due to an inactive pyruvate kinase: implications for in vivo growth. Mol Microbiol 56: 163-174. doi:10.1111/j.1365-2958.2005.04524.x. PubMed: 15773987.
  19. 19. Chavadi S, Wooff E, Coldham NG, Sritharan M, Hewinson RG et al. (2009) Global effects of inactivation of the pyruvate kinase gene in the Mycobacterium tuberculosis complex. J Bacteriol 191: 7545-7553. doi:10.1128/JB.00619-09. PubMed: 19820096.
  20. 20. Beste DJ, Hooper T, Stewart G, Bonde B, Avignone-Rossa C et al. (2007) GSMN-TB: a web-based genome-scale network model of Mycobacterium tuberculosis metabolism. Genome Biol 8: R89. doi:10.1186/gb-2007-8-5-r89. PubMed: 17521419.
  21. 21. Jamshidi N, Palsson BO (2007) Investigating the metabolic capabilities of Mycobacterium tuberculosis H37Rv using the in silico strain iNJ661 and proposing alternative drug targets. BMC. Syst Biol 1: 26. doi: 10.1186/1752-0509-1-26
  22. 22. Fang X, Wallqvist A, Reifman J (2010) Development and analysis of an in vivo-compatible metabolic network of Mycobacterium tuberculosis. BMC. Syst Biol 4: 160. doi: 10.1186/1752-0509-4-160
  23. 23. Chindelevitch L, Stanley S, Hung D, Regev A, Berger B (2012) MetaMerge: scaling up genome-scale metabolic reconstructions, with application to Mycobacterium tuberculosis. Genome Biol 13: R6. doi:10.1186/gb-2012-13-1-r6. PubMed: 22292986.
  24. 24. Gevorgyan A, Bushell ME, Avignone-Rossa C, Kierzek AM (2011) SurreyFBA: a command line tool and graphics user interface for constraint-based modeling of genome-scale metabolic reaction networks. Bioinformatics 27: 433-434. doi:10.1093/bioinformatics/btq679. PubMed: 21148545.
  25. 25. Beste DJ, Peters J, Hooper T, Avignone-Rossa C, Bushell ME et al. (2005) Compiling a molecular inventory for Mycobacterium bovis BCG at two growth rates: evidence for growth rate-mediated regulation of ribosome biosynthesis and lipid metabolism. J Bacteriol 187: 1677-1684. doi:10.1128/JB.187.5.1677-1684.2005. PubMed: 15716438.
  26. 26. Lechat P, Hummel L, Rousseau S, Moszer I (2008) GenoList: an integrated environment for comparative analysis of microbial genomes. Nucleic Acids Res 36: D469-D474. PubMed: 18032431.
  27. 27. Bochner BR, Gadzinski P, Panomitros E (2001) Phenotype microarrays for high-throughput phenotypic testing and assay of gene function. Genome Res 11: 1246-1255. doi:10.1101/gr.186501. PubMed: 11435407.
  28. 28. Oh YK, Palsson BO, Park SM, Schilling CH, Mahadevan R (2007) Genome-scale reconstruction of metabolic network in Bacillus subtilis based on high-throughput phenotyping and gene essentiality data. J Biol Chem 282: 28791-28799. doi:10.1074/jbc.M703759200. PubMed: 17573341.
  29. 29. Beste DJ, Bonde B, Hawkins N, Ward JL, Beale MH et al. (2011) C Metabolic Flux Analysis Identifies an Unusual Route for Pyruvate Dissimilation in Mycobacteria which Requires Isocitrate Lyase and Carbon Dioxide Fixation 13. PLOS Pathog 7: e1002091.
  30. 30. Sassetti CM, Boyd DH, Rubin EJ (2003) Genes required for mycobacterial growth defined by high density mutagenesis. Mol Microbiol 48: 77-84. doi:10.1046/j.1365-2958.2003.03425.x. PubMed: 12657046.
  31. 31. Griffin JE, Gawronski JD, Dejesus MA, Ioerger TR, Akerley BJ et al. (2011) High-resolution phenotypic profiling defines genes essential for mycobacterial growth and cholesterol catabolism. PLOS Pathog 7: e1002251. PubMed: 21980284.
  32. 32. Behr MA, Schroeder BG, Brinkman JN, Slayden RA, Barry CE 3rd (2000) A point mutation in the mma3 gene is responsible for impaired methoxymycolic acid production in Mycobacterium bovis BCG strains obtained after 1927. J Bacteriol 182: 3394-3399. doi:10.1128/JB.182.12.3394-3399.2000. PubMed: 10852869.
  33. 33. Chen JM, Alexander DC, Behr MA, Liu J (2003) Mycobacterium bovis BCG vaccines exhibit defects in alanine and serine catabolism. Infect Immun 71: 708-716. doi:10.1128/IAI.71.2.708-716.2003. PubMed: 12540549.
  34. 34. Pérez E, Constant P, Lemassu A, Laval F, Daffé M et al. (2004) Characterization of three glycosyltransferases involved in the biosynthesis of the phenolic glycolipid antigens from the Mycobacterium tuberculosis complex. J Biol Chem 279: 42574-42583. doi:10.1074/jbc.M406246200. PubMed: 15292272.
  35. 35. Gordon SV, Brosch R, Billault A, Garnier T, Eiglmeier K et al. (1999) Identification of variable regions in the genomes of tubercle bacilli using bacterial artificial chromosome arrays. Mol Microbiol 32: 643-655. doi:10.1046/j.1365-2958.1999.01383.x. PubMed: 10320585.
  36. 36. David HL, Jahan MT (1977) beta-Glucosidase activity in mycobacteria. J Clin Microbiol 5: 383-384. PubMed: 856867.
  37. 37. Stermann M, Bohrssen A, Diephaus C, Maass S, Bange FC (2003) Polymorphic nucleotide within the promoter of nitrate reductase (NarGHJI) is specific for Mycobacterium tuberculosis. J Clin Microbiol 41: 3252-3259. doi:10.1128/JCM.41.7.3252-3259.2003. PubMed: 12843072.
  38. 38. Stermann M, Sedlacek L, Maass S, Bange FC (2004) A promoter mutation causes differential nitrate reductase activity of Mycobacterium tuberculosis and Mycobacterium bovis. J Bacteriol 186: 2856-2861. doi:10.1128/JB.186.9.2856-2861.2004. PubMed: 15090527.
  39. 39. Honaker RW, Stewart A, Schittone S, Izzo A, Klein MR et al. (2008) Mycobacterium bovis BCG vaccine strains lack narK2 and narX induction and exhibit altered phenotypes during dormancy. Infect Immun 76: 2587-2593. doi:10.1128/IAI.01235-07. PubMed: 18362135.
  40. 40. Sohaskey CD, Modesti L (2009) Differences in nitrate reduction between Mycobacterium tuberculosis and Mycobacterium bovis are due to differential expression of both narGHJI and narK2. FEMS Microbiol Lett 290: 129-134. PubMed: 19076631.
  41. 41. Sohaskey CD, Wayne LG (2003) Role of narK2X and narGHJI in hypoxic upregulation of nitrate reduction by Mycobacterium tuberculosis. J Bacteriol 185: 7247-7256. doi:10.1128/JB.185.24.7247-7256.2003. PubMed: 14645286.
  42. 42. Garcia Pelayo MC, Uplekar S, Keniry A, Mendoza Lopez P, Garnier T et al. (2009) A comprehensive survey of single nucleotide polymorphisms (SNPs) across Mycobacterium bovis strains and M. bovis BCG vaccine strains refines the genealogy and defines a minimal set of SNPs that separate virulent M. bovis strains and M. bovis BCG strains. Infect Immun 77: 2230-2238. doi:10.1128/IAI.01099-08. PubMed: 19289514.
  43. 43. Marri PR, Bannantine JP, Golding GB (2006) Comparative genomics of metabolic pathways in Mycobacterium species: gene duplication, gene decay and lateral gene transfer. FEMS Microbiol Rev 30: 906-925. doi:10.1111/j.1574-6976.2006.00041.x. PubMed: 17064286.
  44. 44. Giffin MM, Modesti L, Raab RW, Wayne LG, Sohaskey CD (2012) ald of Mycobacterium tuberculosis encodes both the alanine dehydrogenase and the putative glycine dehydrogenase. J Bacteriol 194: 1045-1054. doi:10.1128/JB.05914-11. PubMed: 22210765.
  45. 45. Belley A, Alexander D, Di Pietrantonio T, Girard M, Jones J et al. (2004) Impact of methoxymycolic acid production by Mycobacterium bovis BCG vaccines. Infect Immun 72: 2803-2809. doi:10.1128/IAI.72.5.2803-2809.2004. PubMed: 15102790.
  46. 46. Bonicke R, Lisboa BP (1959) Type differentiation of tuberculosis bacteria with the aid of the nicotinamidase test. Tuberkulosearzt 13: 377-384. PubMed: 13676136.
  47. 47. Constant P, Perez E, Malaga W, Lanéelle MA, Saurel O et al. (2002) Role of the pks15/1 gene in the biosynthesis of phenolglycolipids in the Mycobacterium tuberculosis complex. Evidence that all strains synthesize glycosylated p-hydroxybenzoic methyl esters and that strains devoid of phenolglycolipids harbor a frameshift mutation in the pks15/1 gene. J Biol Chem 277: 38148-38158. doi:10.1074/jbc.M206538200. PubMed: 12138124.
  48. 48. Höner Zu Bentrup K, Miczak A, Swenson DL, Russell DG (1999) Characterization of activity and expression of isocitrate lyase in Mycobacterium avium and Mycobacterium tuberculosis. J Bacteriol 181: 7161-7167. PubMed: 10572116.
  49. 49. Rivera-Marrero CA, Ritzenthaler JD, Newburn SA, Roman J, Cummings RD (2002) Molecular cloning and expression of a novel glycolipid sulfotransferase in Mycobacterium tuberculosis. Microbiology 148: 783-792. PubMed: 11882713.
  50. 50. Brosch R, Gordon SV, Marmiesse M, Brodin P, Buchrieser C et al. (2002) A new evolutionary scenario for the Mycobacterium tuberculosis complex. Proc Natl Acad Sci U S A 99: 3684-3689. doi:10.1073/pnas.052548299. PubMed: 11891304.
  51. 51. Lamichhane G, Zignol M, Blades NJ, Geiman DE, Dougherty A et al. (2003) A postgenomic method for predicting essential genes at subsaturation levels of mutagenesis: application to Mycobacterium tuberculosis. Proc Natl Acad Sci U S A 100: 7213-7218. doi:10.1073/pnas.1231432100. PubMed: 12775759.
  52. 52. Lyon RH, Hall WH, Costas-Martinez C (1970) Utilization of Amino Acids During Growth of Mycobacterium tuberculosis in Rotary Cultures. Infect Immun 1: 513-520. PubMed: 16557769.
  53. 53. Schaefer WB (1952) Growth requirements of dysgonic and eugonic strains of Mycobacterium tuberculosis var. bovis. J Exp Med 96: 207-219. doi:10.1084/jem.96.3.207. PubMed: 14955575.
  54. 54. Chaturvedi S, Bhakuni V (2003) Unusual structural, functional, and stability properties of serine hydroxymethyltransferase from Mycobacterium tuberculosis. J Biol Chem 278: 40793-40805. doi:10.1074/jbc.M306192200. PubMed: 12913008.
  55. 55. de Carvalho LP, Fischer SM, Marrero J, Nathan C, Ehrt S et al. (2010) Metabolomics of Mycobacterium tuberculosis reveals compartmentalized co-catabolism of carbon substrates. Chem biol 17: 1122-1131.