Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Nat Chem Biol. Author manuscript; available in PMC 2012 January 12.
Published in final edited form as:
PMCID: PMC3256575

Self-poisoning of Mycobacterium tuberculosis by targeting GlgE in an α-glucan pathway


New chemotherapeutics are urgently required to control the tuberculosis pandemic. We describe a new pathway from trehalose to α-glucan in Mycobacterium tuberculosis comprising four enzymatic steps mediated by TreS, Pep2, GlgE (which has been identified as a maltosyltransferase that uses maltose 1-phosphate) and GlgB. Using traditional and chemical reverse genetics, we show that GlgE inactivation causes rapid death of M. tuberculosis in vitro and in mice through a self-poisoning accumulation of maltose 1-phosphate. Poisoning elicits pleiotropic phosphosugar-induced stress responses promoted by a self-amplifying feedback loop where trehalose-forming enzymes are upregulated. Moreover, the pathway from trehalose to α-glucan exhibited a synthetic lethal interaction with the glucosyltransferase Rv3032, which is involved in biosynthesis of polymethylated α-glucans, because key enzymes in each pathway could not be simultaneously inactivated. The unique combination of maltose 1-phosphate toxicity and gene essentiality within a synthetic lethal pathway validates GlgE as a distinct potential drug target that exploits new synergistic mechanisms to induce death in M. tuberculosis.

Mcobacterium tuberculosis (Mtb), the etiological agent of tuberculosis (TB), is the leading cause of mortality due to bacterial pathogens, claiming about 2 million lives annually. With the advent of the antibiotic era, TB became treatable, and at one point eradication was believed possible. However, in recent years TB has reemerged as a major global health threat due to poverty, a deadly synergy with HIV, and the emergence of extensively drug-resistant strains (XDR-TB) that are virtually untreatable with current chemotherapies1. To counter this resurgence of TB, new treatment options are urgently needed based on new classes of bacterial targets that are very different from those of the antibiotics currently in use.

Identification of essential gene functions required for in vitro growth of Mtb is a pivotal strategy for discovering new drug targets owing to the ease of screening for antibacterial compounds. A genome-wide screen of a saturated Mtb transposon mutant library indicated that more than 600 genes (~15%) may be required for in vitro growth2. However, the rational development of specific inhibitors is hampered by ignorance of the functions of many of these essential candidates. Identification of nonessential gene functions dispensable for in vitro growth but required for virulence is an additional valuable source of drug targets. 194 of the nonessential Mtb genes were predicted to be specifically required for in vivo growth in mice3. Given that Mtb is an obligate intracellular pathogen highly adapted to the human host, with no known environmental niche, the large number of genes dispensable for in vitro growth and virulence is surprising. This high degree of gene dispensability probably reflects extensive genetic redundancy or functional homeostatic buffering within essential processes, so that mutations in single genes are often compensated for by other genes. This means that many ‘synthetic lethal’ pathways are certainly present in Mtb. ‘Synthetic lethality’ describes any combination of two separately nonlethal mutations that jointly lead to inviability. Synthetic lethal genetic interactions have been demonstrated in genome-wide studies of yeast, revealing that most nonessential yeast genes have several synthetic lethal interactions with other genes4,5. Identification of synthetic lethal pathways in Mtb would thus greatly increase the repertoire of drug target candidates.

Here we describe the discovery of a new four-step pathway from trehalose to α-glucan in Mtb comprising the enzymes TreS, Pep2, GlgE and GlgB. The key enzyme in this pathway is the essential maltosyltransferse GlgE, which uses maltose 1-phosphate (M1P) as the donor substrate. Inactivation of either GlgE or GlgB led to self-poisoning by the accumulation of the phosphosugar M1P, driven by a self-amplifying feedback stress response. A combination of traditional and chemical reverse genetics demonstrated the lethal effect of GlgE inactivation for Mtb grown both in vitro and in mice. Moreover, we found a synthetic lethal interaction with a pathway leading to polymethylated α-glucans in Mtb. This unique combination of gene essentiality within a synthetic lethal pathway validates GlgE as a new potential Mtb drug target candidate by suggesting possible mechanisms for boosting the potency of inhibitors and for suppressing resistance mutations. To our knowledge, this is the first time α-glucan synthesis has been described as a potential target for antimicrobials.


Validation of the essentiality of glgE in Mtb

Analysis of the function of essential genes could define new drug target candidates for the treatment of XDR-TB strains. One such candidate, glgE (Rv1327c), has been predicted to be essential for in vitro growth of Mtb2. The glgE homolog was also suggested to be essential in Mycobacterium smegmatis6. In addition to its presumed essentiality, GlgE fulfills further criteria that suggest it could be an excellent drug target: GlgE homologs are absent from humans and from commensal gut flora bacteria, but are present in almost all mycobacteria and in some opportunistic pathogens, such as Pseudomonas and Burkholderia species.

To verify the essentiality of glgE in Mtb, we compared the frequencies of generating glgE deletion mutations in the wild type (WT) to those in an isogenic merodiploid strain containing a second copy of glgE provided on an integrative single-copy plasmid, using specialized transduction. Transductants with a deleted WT allele were obtained only in the merodiploid strain, at a frequency of 4.08 × 10−9. We were unable to obtain transductants in the haploid strain despite repeated attempts (frequency < 6 × 10−11), thus confirming glgE essentiality in Mtb (see Supplementary Fig. 1a,b in Supplementary Results).

Accumulation of M1P in M. smegmatis ΔglgE

Because glgE is clustered with genes coding for enzymes with known involvement in glycogen metabolism, such as the branching enzyme GlgB and the glycogen phosphorylase GlgP (Supplementary Fig. 1a), and because GlgE resembles α-amylases, GlgE has been proposed to be involved in glycogen degradation. The demonstration that thermal inactivation of GlgE in a temperature-sensitive M. smegmatis mutant strain apparently leads to increased glycogen accumulation and unexpected cell toxicity was interpreted as supporting the hypothesis that GlgE was a mediator of glycogen degradation6. However, because high glycogen levels are not usually found to be toxic to bacteria, and because many bacteria capable of synthesizing and remobilizing glycogen lack a glgE ortholog, we sought an alternative explanation for the gene essentiality.

To reexamine GlgE function, we attempted deletion of the glgE homolog in M. smegmatis. In contrast to GlgE’s suggested essentiality6, null deletion mutations could be readily generated on a minimal medium, allowing inferences on GlgE function from phenotypical mutant characterization. TLC analysis of hot water extracts from the M. smegmatis ΔglgE mutant revealed no increased glycogen content, but we did observe large amounts of a low-molecular-weight carbohydrate (Fig. 1a). This substance was purified by size exclusion chromatography followed by preparative TLC, and subjected to various chemical analyses (Supplementary Fig. 2) that identified the substance as α-d-glucopyranosyl(1→4)-α-d-glucopyranosyl 1-phosphate (maltose 1-phosphate, M1P). To quantify glycogen in the temperature-sensitive glgE mutant, a coupled enzymatic reaction was used that measures glucose released upon amyloglucosidase treatment of extracted polysaccharides6. Because amyloglucosidase has no molecular weight specificity, the coupled assay would not have discriminated between M1P and glycogen, leading to a likely misinterpretation.

Figure 1
Characterization of the M. smegmatis ΔglgE mutant

A functional link between GlgE, GlgB, TreS and Pep2

Surprisingly, despite normal growth in minimal medium (Middlebrook 7H9), the M. smegmatis ΔglgE mutant could not grow in complex media (for example, LB medium). To identify the growth-inhibitory components in LB medium, we tested the sensitivity of the M. smegmatis ΔglgE mutant toward various mono- and disaccharides and found that it was sensitive to the disaccharide trehalose (α-d-glucopyranosyl-(1→1)-α-d-glucopyranoside). This is surprising given that this sugar is abundant in mycobacteria, and yet a concentration of 0.5 mM completely inhibited growth of the mutant (Fig. 1b). Trehalose induced bacteriostasis in M. smegmatis ΔglgE correlating with M1P hyperaccumulation (Supplementary Fig. 3).

Spontaneous mutations restoring growth of M. smegmatis ΔglgE in the presence of trehalose were observed at a frequency of 4.5 × 10−5. In order to identify the presumed suppressor mutations abolishing trehalose sensitivity, a transposon mutant library of M. smegmatis ΔglgE was screened by selecting for resistance to 1 mM trehalose. Mapping of the transposon insertion sites in trehalose-insensitive ΔglgE mutants identified the trehalose synthase gene treS (encoding a Rv0126 homolog) as a target in several independent clones. To confirm the transposon mutant phenotypes, we deleted treS in an unmarked M. smegmatis ΔglgE strain by specialized transduction. Additionally, we separately deleted pep2 (encoding a Rv0127 homolog), which is clustered with treS in many bacteria. Both second-site gene deletions conferred trehalose resistance and suppressed M1P accumulation caused by glgE inactivation (Fig. 1b,c).

The glgB gene (Rv1326c) clusters with glgE and encodes a branching enzyme that is required for introducing α-1,6-linked branches into linear α-1,4-glucans7, which was demonstrated to be essential in Mtb H37Rv (ref. 8). As for glgE, however, we found glgB to be nonessential in M. smegmatis. Moreover, the M. smegmatis ΔglgB mutant was also sensitive to trehalose. Trehalose sensitivity in an unmarked M. smegmatis ΔglgB strain could again be suppressed by deletion of treS (Fig. 1d).

Trehalose synthase (TreS) mediates the reversible interconversion of trehalose and maltose9,10. Pep2 from Mtb has not been characterized but is homologous to the maltokinase Mak1 from phylogenetically related actinomycetes (Actinoplanes missouriensis and Streptomyces coelicolor) that phosphorylates maltose to M1P using ATP11. The clustering of treS and pep2 thus suggests that M1P is synthesized from trehalose via maltose by sequential reactions mediated by TreS and Pep2. To corroborate this, M. smegmatis mutant strains were labeled with 14C-trehalose (Fig. 1e). Radiolabeled trehalose was rapidly taken up and metabolized to M1P in the ΔglgE mutant. Also, glgB deletion led to a rapid accumulation of M1P, whereas this phosphosugar never reached high levels in the WT. Inactivation of treS or pep2 prevented M1P formation in the ΔglgE mutant. Substantial levels of the radiolabeled maltose intermediate were detectable only in Δpep2 strains, indicating its usual fast turnover to M1P by Pep2 (Fig. 1f). Together, these findings establish a functional link among TreS, Pep2, GlgE and GlgB in M. smegmatis. This is supported by the fact that the structural genes, while arranged in two separate chromosomal gene clusters in mycobacteria, often lie in a single locus in many other prokaryotes (Supplementary Fig. 4).

GlgE is a new maltosyltransferase that uses M1P

Cytoplasmic M1P accumulation in the M. smegmatis ΔglgE mutant suggested that this phosphosugar might be the direct substrate of GlgE. Given that the branching enzyme GlgB uses linear α-1,4-glucan as a substrate7, we hypothesized that GlgE could act upstream of GlgB producing such linear glucans. This could be accomplished by transferring the maltosyl moiety of M1P to the nonreducing end 4-OH of an α-glucan acceptor substrate with release of phosphate (Fig. 2a). In vitro enzymatic activity of histidine-tagged recombinant GlgE (Supplementary Fig. 5a) was assayed either quantitatively by monitoring phosphate release from M1P, or qualitatively using mass spectrometry or TLC. Maltooligosaccharides with a degree of polymerization (DP) ≥ 4 were efficient acceptors, with DP5 being optimal in the GlgE-catalyzed polymerization of M1P (Fig. 2b,c). The β-anomer of M1P at a 1 mM concentration gave no activity with 1 mM maltohexaose as the acceptor (detection limit was ≤ 1% of the activity with the α-anomer). The presence of α-1,4 links in the oligosaccharide products from maltotetraose extension was shown by NMR spectroscopy (Fig. 2d), with no evidence for other α-1,n or β-1,n links12. Together, these observations showed that the enzyme exhibits an α-retaining catalytic mechanism. The activity with maltohexaose was optimal at pH 7.0 and 37 °C, consistent with the lifestyle of the organism (Supplementary Fig. 5b–d).

Figure 2
GlgE is a new M1P-dependent maltosyltransferase

The Kmapp for maltohexaose with 5 mM M1P was 35 ± 8 mM, with a kcatapp of 15.4 ± 1.1 s−1, revealing a kcatapp/Kmapp of 440 ± 100 M−1 s−1 (Supplementary Fig. 5e–g), which demonstrates a catalytic efficiency within the typical range for a carbohydrate active enzyme. With 1 mM M1P, the Kmapp and kcatapp of maltohexaose decreased to 13.3 ± 2.4 mM and 7.5 ± 0.4 s−1, respectively, giving a kcatapp/Kmapp of 570 ± 110 M−1 s−1. This trend, coupled with statistically indistinguishable values of kcatapp/Kmapp, is consistent with a substituted-enzyme kinetic mechanism (also known as ping pong), whereby the phosphate- and acceptor-binding sites are synonymous, precluding donor and acceptor molecules from binding simultaneously to the enzyme. The Kmapp for M1P with 1 mM maltohexaose was 0.25 ± 0.05 mM, with a kcatapp of 1.26 ± 0.07 s−1, giving a kcatapp/Kmapp of 5,000 ± 1,000 M−1 s−1, which shows that M1P is an efficient donor.

We also tested the reversibility of GlgE activity. Using glycogen as a donor substrate, we found that higher phosphate concentrations resulted in higher M1P production (Fig. 2e,f), confirming reversibility in vitro. Using the phosphate assay with 125 mM maltohexaose and 20 mM inorganic phosphate, we found that ~10% of the phosphate was converted to M1P, showing that the equilibrium is in favor of M1P consumption, as would be expected for this activated sugar. The Kmapp for inorganic phosphate was estimated to be 6 ± 4 mM (Supplementary Fig. 5h). The enzyme also catalyzed the disproportionation of maltooligosaccharides. Mass spectrometry showed that transfer of maltosyl units from donors with a DP ≥ 4 occurred, but a DP ≥ 6 gave the most rapid transfers (Supplementary Fig. 6a–e). The smallest product of disproportionation had a DP of 4, and the smallest acceptor in this reaction also had a DP of 4 (Supplementary Fig. 6f–i), as was the case with M1P as the donor (Fig. 2b).

The unveiling of the M1P-dependent maltosyltransferase activity of GlgE, in combination with the phenotypes of the M. smegmatis mutants, allows the plausible conclusion that GlgE is part of a new pathway in Mtb (and many other phylogenetically distant prokaryotes) that converts trehalose into an α-1,6-branched α-1,4-glucan via four enzymatic steps mediated by TreS, Pep2, GlgE and GlgB (Fig. 3 and Supplementary Fig. 4). GlgE not only forms the linear α-1,4-glucan but it can also edit the branch lengths of branched glucan with its disproportionation activity. Though three of the four enzymatic activities are well known and have been characterized in Mtb (TreS, GlgB) or related actinomycetes (Pep2), their functional cooperation within a single pathway has eluded precise description due to the lack of awareness of GlgE’s key role.

Figure 3
A new prokaryotic pathway from trehalose to α-glucan

Suppression of glgE essentiality by blocking M1P formation

A mutational block in sugar catabolic pathways can cause accumulation of apparently toxic phosphorylated intermediates that render such mutants sensitive to those sugars (for example, refs. 1315). Similarly, our results with the model organism M. smegmatis indicated that M1P accumulation was directly or indirectly toxic, causing the lethality of glgE mutations in Mtb. To prove this, we used specialized transduction to compare the frequencies of generating glgE deletion mutations in the WT to those in an isogenic unmarked ΔtreS mutant in Mtb, in which M1P formation is blocked. Though we were unsuccessful in deleting glgE in the WT, the gene could readily be inactivated in the ΔtreS mutant, confirming both the essentiality of GlgE and the causality of M1P toxicity, respectively (Supplementary Fig. 1c). Likewise, treS inactivation allowed deletion of glgB, demonstrating that M1P toxicity is also the probable cause of GlgB essentiality (Supplementary Fig. 1d).

Bactericidal effect of GlgE inactivation in Mtb

Next, we wanted to determine whether inactivation of GlgE causes bacteriostasis or lethality in Mtb. Initial attempts to downregulate the chromosomal expression of glgE with a tetracycline-inducible promoter were unsuccessful. In an alternative chemical genetic approach, we tried to generate a glgE deletion mutant under conditions in which chemical inhibition of TreS was applied in order to prevent M1P toxicity using the inhibitor validamycin A, which has been demonstrated to inhibit TreS from Bradyrhizobium species16. An approach to delete glgE in WT Mtb using specialized transduction in the presence of validamycin A was unsuccessful. We therefore mimicked the phenotype of a ΔglgE mutant by complementing the ΔtreS ΔglgE double mutant of Mtb with a functional copy of the treS gene provided on an integrative single-copy plasmid in the presence of validamycin A. As expected, treS transformants were obtained in the ΔtreS ΔglgE mutant only in the presence of validamycin A. We predicted that depletion of the inhibitor would restore TreS enzymatic activity and result in M1P accumulation and cessation of growth in the ΔglgE surrogate strain. M1P was undetectable by TLC in both the WT Mtb and the vector control strain, and accumulated to only low levels in the ΔtreS ΔglgE mutant strain carrying pMV361::treS in the presence of 5 mM validamycin A. In contrast, depletion of the inhibitor resulted in copious accumulation of M1P within 48 h in this ΔglgE surrogate strain (Fig. 4a). Importantly, and in contrast to the bacteriostatic effect in M. smegmatis, M1P accumulation correlated with rapid killing of the cells in vitro in liquid culture (2 log reduction in viability within 4 d) (Fig. 4b), demonstrating that GlgE inactivation is lethal in Mtb. Bacteria emerging after the initial killing phase after day 4 did not require supplementation with validamycin A, indicating that they had likely acquired loss-of-function mutations either in the plasmid-encoded treS gene or in the chromosomally located pep2 gene, thus preventing M1P toxicity.

Figure 4
M1P self-poisoning is lethal for Mtb grown in vitro and in mice

Next, we tested the conditional lethal mutant strain ΔtreS ΔglgE (pMV361::treS) in a mouse infection model (Fig. 4c). This ΔglgE surrogate strain died rapidly in the lungs and spleens of infected BALB/c mice (about 1.5 log and 1 log reduction of bacterial organ burden within 7 d, respectively). The in vivo killing demonstrates that the TreS-Pep2-GlgE-GlgB pathway (henceforth referred to as the GlgE pathway) is active during growth in the host and that endogenous trehalose levels in Mtb are sufficient to support intracellular M1P accumulation reaching lethal concentrations in vivo. Bacteria surviving the initial killing phase recovered from spleens and lungs at day 14 were all independent of validamycin A. This again suggests a likely acquisition of loss-of-function mutations in either treS or pep2, consistent with the absolute necessity of preventing M1P poisoning in order to allow survival in vivo.

Together, the in vitro and in vivo killing phenotypes convincingly validate GlgE’s potential as a drug target candidate. It is noteworthy that the kinetics of killing and of emergence of mutants observed with the ΔglgE surrogate strain in vitro resemble those reported for treatment of Mtb with a high concentration of the first-line drug isoniazid17. Notably, the GlgE pathway as a whole is dispensable for virulence, as demonstrated by the lack of attenuation of the ΔtreS mutant in mice (Fig. 4d).

M1P accumulation elicits pleiotropic stress responses in Mtb

In order to gain insight into the molecular mechanisms underlying M1P toxicity and eventual death of Mtb, we performed global gene expression analyses. For this, the conditional lethal Mtb mutant strain ΔtreS ΔglgE (pMV361::treS) and the vector control strain were grown to log phase in presence of 5 mM validamycin A to suppress M1P formation. Subsequently, cells were washed to remove the inhibitor, and microarrays were performed on cells after 48 h of depletion of validamycin A. We found typical gene expression signatures indicative of unexpected overlapping stress responses in M1P-poisoned Mtb (Supplementary Fig. 7; see also Supplementary Fig. 8 for qRT-PCR confirmation of selected transcripts). A remodeling of the respiratory electron transport chain was obvious, as indicated by significant downregulation of cytochrome c oxidase components (encoded by ctaB, ctaC, ctaD and ctaE) and upregulation of components of the NADH dehydrogenase I complex (encoded by nuoB, nuoD, nuoE and nuoF), of the non-proton-pumping terminal cytochrome bd oxidase (encoded by cydA, cydB, cydC and cydD) and of nitrate reductase (encoded by narG, narH, narI and narJ) (Supplementary Fig. 7). Except for upregulation of NADH dehydrogenase I, this rearrangement resembles known responses to respiratory inhibitors18,19. In fact, in addition to components of the respiratory chain, we observed an overall similarity to the transcription profile elicited by the cytochrome c oxidase–specific inhibitor potassium cyanide (Fig. 5a), clearly suggesting that M1P stress unexpectedly results in inhibition of respiration. We also noticed a global downregulation of the dormancy (dosR) regulon (Supplementary Figs. 7 and 8). Though the factors triggering induction of this regulon (such as hypoxia, NO and CO) are well characterized20,21, it is unclear which metabolic perturbations stimulate downregulation. Furthermore, expression of most components of ATP synthase (encoded by atpA, atpC, atpD, atpE, atpG and atpH) was significantly downregulated (Supplementary Fig. 7), probably resulting in decreased intracellular ATP levels. This can be linked to the induction of relA (Supplementary Fig. 8), which triggers the stringent response22, thus explaining the observed global downregulation of the translational apparatus (Supplementary Fig. 7). Moreover, M1P stress also led to strong induction of the DNA damage-responsive SOS regulon (Supplementary Figs. 7 and 8), including numerous genes involved in DNA repair, such as recA and dnaE2, which encodes the error-prone DNA polymerase (refs. 18,23). Thus, DNA damage appears to be a surprising direct or indirect outcome of M1P self-poisoning.

Figure 5
Whole-genome transcriptional profiling of the M1P-induced stress response in Mtb

Most notably, another distinct stress response was the upregulation of a pathway leading to release of trehalose from glycogen, comprising the genes glgP, treX, treY and treZ (Figs. 3 and and5b;5b; Supplementary Fig. 8b)24. Apparently, Mtb reacts to M1P stress by increasing the intracellular trehalose level, perhaps as a stress protectant, given that trehalose has this function in many bacteria and yeasts25. However, in the conditional lethal mutant strain ΔtreS ΔglgE (pMV361::treS), this is fatally counterproductive because it promotes further conversion of trehalose to M1P, resulting in a deadly self-amplifying feedback loop that causes progressive self-poisoning.

Synthetic lethality of the GlgE pathway and Rv3032

The new GlgE pathway is not the only biosynthetic route to α-1,4-glucans in Mtb; two alternative pathways have been described8. ADP-glucose, formed from glucose 1-phosphate and ATP by the ADP-glucose pyrophosphorylase GlgC, is polymerized by the glycogen synthase GlgA to linear α-1,4-glucans. Alternatively, the glucosyltransferase Rv3032 uses either UDP-glucose or ADP-glucose as the substrate for polymerization to linear α-1,4-glucans8. When trying to analyze α-1,4-glucan formation in Mtb by combinatorial inactivation of the three known pathways by targeting key steps (GlgC, Rv3032, TreS), we found that while ΔglgC, ΔtreS and ΔRv3032 single mutants, as well as ΔglgC ΔtreS and ΔglgC ΔRv3032 double mutants, were all viable, we could not inactivate Rv3032 in the ΔtreS mutant. This suggested the joint essentiality of these two pathways. To corroborate this, we tested the sensitivity of the Mtb ΔRv3032 mutant toward the TreS inhibitor validamycin A. Whereas the Mtb WT tolerated high concentrations of the inhibitor (10 mM), the ΔRv3032 mutant was exceptionally sensitive to this compound, causing a 3-log killing over 7 d (Fig. 6a). Furthermore, overexpression of the putative target TreS conferred marked resistance of the ΔRv3032 mutant to validamycin A (Fig. 6b). Together, these data unequivocally validate the joint essentiality (that is, the synthetic lethality) of treS and Rv3032.

Figure 6
Synthetic lethality of treS and Rv3032


The rapid emergence of XDR-TB strains highlights the urgent necessity of finding new classes of drugs to kill Mtb with new mechanisms of action. The paucity of new TB drugs discovered during the last few decades reflects our limited knowledge of essential metabolic processes in Mtb outside the repertoire of known targets such as transcription, translation or mycolic acid biosynthesis. In this study, a combination of genetic and biochemical studies has revealed that the M1P-dependent maltosyltransferase GlgE represents a potential new class of drug target, as it is part of a previously unrecognized α-glucan pathway that to our knowledge has never been targeted by antimicrobials. Studying the function of GlgE has revealed two new and distinct bactericidal mechanisms to induce death in Mtb. The first death mechanism is suicidal self-poisoning by accumulation of the phosphosugar M1P following GlgE inhibition. This process is likely driven by a self-amplifying feedback loop leading to trehalose release from glycogen via the GlgP-TreX-TreY-TreZ pathway, perhaps representing a misled stress protection response that further fuels the accumulation of M1P. The second (and independent) death mechanism is based on conditional essentiality of GlgE pathway products. Synthetic lethality of the GlgE and Rv3032 pathways indicates that they are involved in the production of related compounds that are reciprocally able to functionally compensate for metabolic or genetic perturbations in the other pathway. This explains why the GlgE pathway as a whole is fully dispensable for the viability and virulence of Mtb as long as the redundant Rv3032 pathway is functioning, but simultaneous inhibition of both pathways is lethal. Rv3032 is involved in the biosynthesis of methylglucose lipopolysaccharides (MGLPs). MGLPs are polymethylated α-glucan derivatives that are believed to play an essential regulatory role in fatty acid biosynthesis in Mtb26. Therefore, GlgE pathway products are likely subject, at least partially, to further modifications, yielding so far unidentified derivatives structurally and/or functionally related to MGLPs (Fig. 3). Given that the two new bactericidal mechanisms in Mtb described above are separate and distinct, they will likely work in synergy. This means that, while monotherapy with GlgE inhibitors should be sufficient to kill Mtb as efficiently as treatment with the first-line drug isoniazid17, it will likely be possible to boost the potency of such GlgE inhibitors in a combination therapy with compounds inhibiting the synthetic lethal partner Rv3032. Furthermore, because every step of the GlgE pathway becomes essential when Rv3032 is inactive, combination therapy with Rv3032 inhibitors will also avoid resistance based on loss-of-function mutations in treS or pep2 for evading M1P poisoning. The rapid emergence of such mutants in vitro and in mice indicates that preventing this type of suppressor mechanism could substantially increase the antitubercular potential of GlgE inhibitors.

This study has both therapeutic and general biological significance. We have established the existence of a previously unknown pathway, widespread among prokaryotes, for the conversion of trehalose to α-glucan. Possible connections between the metabolism of these compounds in bacteria have previously been proposed9,11,27. Our work now convincingly establishes the precise biochemical link through the GlgE pathway. This stems from our discovery that GlgE has a new maltosyltransferase activity of the Enzyme Commission number 2.4.1 hexosyltransferase type; GlgE thus will likely be assigned the systematic name (1→4)-α-d-glucan:phosphate α-d-maltosyltransferase. GlgE activity is consistent with a maltosyltransferase activity in protein extracts from M. smegmatis that was reported as the present study was in the final stages of review28. Mtb GlgE is, according to the Carbohydrate Active Enzymes database (; ref. 29) and based on its primary protein sequence, a member of the glycoside hydrolase subfamily GH13_3, to which no function has yet been ascribed30. Although GlgE is a glycoside hydrolase GH family member and can catalyze transglucosylation reactions (that is, disproportionation), it is capable of glycosyltransfer reactions with the sugar phosphate donor M1P. This is a function more typically associated with members of the glycosyltransferase GT family, which use either nucleotide diphospho-sugar, nucleotide monophospho-sugar, polyprenyl phospho-sugar or phospho-sugar donors. For example, glycogen phosphorylase (systematic name (1→4)-α-d-glucan:phosphate α-d-glucosyltransferase; Enzyme Commission number is a GT35 member. Sucrose phosphorylase (sucrose:phosphate α-d-glucosyltransferase; Enzyme Commission number, on the other hand, is a GH13_18 subfamily member, so GlgE is not the only GH13 family member to exhibit phosphorylase-type activity. Nevertheless, to our knowledge it is the first activity of this type described involving the transfer of disaccharide units. Furthermore, GlgE uniquely uses a sugar phosphate in a glycosyltransfer reaction for anabolic purposes. It therefore follows that although GlgE could be given the alternative informal name α-maltose 1-phosphate– forming α-(1,4)-glucan phosphorylase, this would reflect neither the physiologically relevant reaction that it catalyzes nor its intrinsic equilibrium. Our experimental evidence for both an α-retaining catalytic mechanism and a substituted-enzyme kinetic mechanism is consistent with the other GH13 family members that use a conserved aspartate side chain as a nucleophile to form a glycosyl-enzyme intermediate allowing an overall double displacement catalytic mechanism with net retention of stereochemistry31. Sequence alignments indicate that Asp418 is the likely nucleophilic residue in Mtb GlgE.

The physiological functions of the GlgE pathway remain elusive and could be multifaceted (Fig. 3). In addition to MGLP synthesis and in interaction with the GlgC-GlgA and Rv3032 pathways, the GlgE pathway might alternatively participate in formation of the α-glucan capsule, an extracellular cell wall component potentially important for virulence and persistence of Mtb8. It could also be involved in the formation of intracellular glycogen, an α-glucan storage polymer from which trehalose can be remobilized via the GlgP-TreX-TreY-TreZ pathway to support trehalose homeostasis24. Furthermore, because glycogen is a typical intracellular carbon storage compound in many organisms, the GlgE pathway might have a role in Mtb persistence within host microenvironments that exhibit restricted nutrient availability.

We have shown that essentiality of GlgE is based on direct or indirect toxicity of its substrate M1P. It has been proposed that the surprising essentiality of GlgB might be due to accumulation of poorly water-soluble linear α-1,4-glucan polymers that are hypothetically toxic32. However, the data presented here demonstrate that the toxicity of M1P, and not that of linear α-1,4-glucans, is the cause that underlies GlgB essentiality, which is explainable by GlgB’s secondary effect on GlgE. In the absence of branching activity, GlgE produces linear α-1,4-glucans that not only have fewer nonreducing ends but also become insoluble once they reach a DP of ~20. Therefore, in the absence of GlgB, there will be fewer nonreducing ends available for GlgE to extend, retarding GlgE activity by acceptor substrate limitation that leads to a toxic buildup of M1P. Although GlgB could be another potential drug target candidate, the existence of a human ortholog makes it less attractive than GlgE.

There is a widespread assertion that the biological function of TreS is to convert exclusively maltose to trehalose33, despite the ready reversibility of this reaction. In at least those organisms that have a GlgE pathway, TreS now appears to catalyze the net reverse reaction in vivo. There is a report that M. smegmatis TreS enzyme also exhibits a minor catabolic amylase activity, allowing the direct conversion of glycogen to maltose9, although this activity is two orders of magnitude lower than the normal TreS reaction. In addition, the amylase activity does not appear to be physiologically significant because the same authors reported that glycogen accumulated in the wild-type organism, but not in a treS mutant, when grown specifically in the presence of trehalose, supporting a dominant anabolic role for TreS.

Why is M1P toxic? M1P accumulation as a result of GlgE inactivation appears to elicit pleiotropic stresses in Mtb, including inhibition of respiration, induction of the stringent response and DNA damage. Because Mtb can tolerate respiration inhibitors relatively well under aerobic conditions18,19 and also maintains viability under conditions such as starvation that trigger the stringent response, DNA damage is probably the most critical factor leading to death. DNA damage might be an indirect result of disturbed respiration, potentially leading to the generation of reactive oxygen species. In Escherichia coli, upregulation of NADH dehydrogenase I was found to be a common key response to all tested bactericidal drugs that all lead to hydroxyl radical formation and subsequent induction of the DNA damage–triggered SOS response34. Notably, upregulation of components of NADH dehydrogenase I (nuoB, nuoD, nuoE and nuoF) was also observed in M1P-stressed Mtb cultures. This suggests that M1P-induced lethality might share certain aspects of the common mechanism of cellular death caused by bactericidal antibiotics in E. coli34. However, although the observed stress phenotypes are unequivocally associated with M1P accumulation, it cannot be excluded that M1P might not be directly toxic to the cells but could rather lead indirectly to toxicity by complex mechanisms—for example, through some unidentified degradation or shunt products. Clearly, more work is needed to unravel the exact molecular toxicity mechanisms associated with M1P accumulation that eventually lead to DNA damage and death.

In summary, the unique combination of gene essentiality within a synthetic lethal pathway, revealing potential mechanisms not only to boost the potency of inhibitors but also to suppress resistance mutations, distinguishes GlgE from all Mtb drug targets described so far. Self-poisoning by a phosphosugar provoking pleiotropic stresses through direct or indirect toxicity also implies a new mechanism to induce death in Mtb that is different from those used by current TB drugs and reveals α-glucan synthesis as a potential target for antimicrobials. Thus, there is reason to believe that GlgE inhibitors, enhanced by Rv3032 inhibitors, could be developed to treat XDR-TB. However, it has to be noted that such a specifically enhancing combination therapy would be a new path in the anti-infective drug discovery field, which so far only focuses on monotherapy. Nonetheless, this study exemplifies the great potential for the discovery of new TB drug targets hidden within synthetic lethal pathways.


Generation of gene deletion mutants

Mutants of M. smegmatis mc2155 and Mtb H37Rv were generated by allelic exchange using specialized transduction35 as described in detail in the Supplementary Methods.

Carbohydrate TLC analysis

Carbohydrates were extracted from equal amounts of cells with hot water (95 °C for 4 h) and analyzed by TLC on silica gel 60 (EMD Chemicals) using the solvent system 1-propanol:ethyl acetate:water (6:1:3, v/v/v). For separation of trehalose and maltose, the solvent system n-butanol:pyridine:water (7:3:1, v/v/v) was used. 14C-α-D-trehalose (specific activity 300 mCi mmol−1; American Radiolabeled Chemicals) was used in some experiments at a concentration of 0.1 µCi ml−1. Substances were visualized by spraying TLC plates with 10% (v/v) sulfuric acid in ethanol followed by charring at 180 °C for 10 min, or by autoradiography where applicable.

Purification and assay of Mtb GlgE

The Mtb glgE gene was synthesized with optimized codon usage for expression in E. coli (Genscript Corporation), heterologously expressed with an N-terminal 6× histidine tag, and purified using nickel-affinity and gel filtration chromatographies as described in the Supplementary Methods. GlgE activity was monitored quantitatively by the release of inorganic phosphate using malachite green colorimetric detection36. Initial rates (v0/[E]) were measured by quenching 3 µl reaction aliquots in 97 µl of 1 M HCl at time points from 0.5 to 8 min. The quenched reactions were incubated with 700 µl of malachite green assay solution for 20 min at 21 °C, and the optical density at 630 nm (OD630) was measured on a Perkin Elmer Lambda 18 spectrophotometer. The concentration of free inorganic phosphate was estimated from a standard curve. Reaction rates were linear over at least the first 4 min. Unless otherwise stated, the enzyme assay was done in 100 mM Bis-Tris propane (pH 7.0) containing 50 mM NaCl at 37 °C. Acceptor preference was examined in triplicate using 40 mM maltooligosaccharide, 1 mM α-M1P (its synthesis, together with that of the β-anomer, will be reported elsewhere) and 37.5 nM Mtb GlgE.

MALDI mass spectrometry

For qualitative analyses of maltooligosaccharide elongation, GlgE assays were performed with typically 12.5 mM α-M1P and 1.5 µM Mtb GlgE. Aliquots (1 µl) of reaction mixtures were diluted 50-fold in H2O, mixed 1:1 with saturated aqueous 2,5-dihydroxybenzoic acid, loaded (1 µl) onto a gold target plate, and dried under vacuum. Analysis was carried out on a PBS-II mass spectrometer using ProteinChip 3.0 software (Ciphergen Biosystems Inc.).

NMR linkage analysis

A reaction (25 µl) containing 20 mM maltotetraose, 20 mM α-M1P, and 1.2 µM Mtb GlgE was allowed to reach completion according to the phosphate release assay. The reaction was heated to 99 °C for 15 min, cooled and diluted to 500 µl with D2O. A control reaction without enzyme was also prepared. One-dimensional 1H spectra were recorded on an AVANCE 600 with TCI cryoprobe at 600 MHz and 300 K and analyzed with Topspin 2.1 software (Bruker Biospin Ltd.). Spectra were acquired with presaturation to suppress the water peak.

DNA microarrays and qRT-PCR

Triplicate 10 ml cultures of the conditional lethal Mtb mutant strain ΔtreS ΔglgE (pMV361::treS) and of the vector control strain were grown to log phase in the presence of 5 mM validamycin A to suppress M1P formation. Subsequently, cells were washed to remove the inhibitor. After 48 h of depletion of validamycin A, cells were harvested, resuspended in 1 ml Qiagen RNA Protect reagent (Qiagen) and incubated 4 h at room temperature (21 °C). RNA was isolated by bead beating in a Fast-Prep apparatus (MP Biomedicals) and using the Qiagen RNeasy kit according to protocol. Contaminating DNA was removed with the Ambion TURBO DNA-free kit (Applied Biosystems). DNA microarrays were obtained through the US National Institute of Allergy and Infectious Diseases’ Pathogen Functional Genomics Resource Center (PFGRC, funded by the Division of Microbiology and Infectious Diseases, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services, and operated by the J. Craig Venter Institute). cDNA probes were prepared as per PFGRC protocol SOP#M007 ( Cy3- and Cy5-labeled cDNA probes were hybridized according to PFGRC protocol SOP#M008 to 70-mer oligo DNA microarrays representing the complete M. tuberculosis genome (JCVI PFGRC M. tuberculosis v. 4). One of the three biological replicates was dye-flipped. Slides were scanned on a GenePix 4000A scanner (Molecular Devices). Images were processed with the TM4 software suite37. TIGR Spotfinder was used to grid and quantitate spots. TIGR MIDAS was used for Lowess normalization, s.d. regularization, and in-slide replicate analysis with all quality control flags on and one bad channel tolerance policy set to ‘generous’. Results were analyzed in MeV with significance analysis of microarrays (SAM) considered significant at q < 0.05. Microarray data have been deposited in the US National Center for Biotechnology Information Gene Expression Omnibus (GEO series accession number GSE18575).

For qRT-PCR, the DNA-free RNA samples used for the microarray experiments were reverse transcribed with the SuperScript III First-Strand Synthesis System (Invitrogen). For the real-time reaction, each primer (250 nM) and 7.5 µl of template reaction (1:20 dilution) in 25 µl volume with Power SYBR Green PCR master mix (Applied Biosystems) was used. Triplicate samples were run on an ABI 7900 HT quantitative thermocycler. Threshold cycles were normalized to those for 16S rRNA and to total RNA. Primer sequences used for qRT-PCR are listed in Supplementary Table 1.

Animal infections

BALB/c mice (4- to 6-week-old females; US National Cancer Institute) were infected intravenously through the lateral tail vein at the indicated doses with exponentially growing Mtb strains suspended in 200 µl phosphate-buffered saline containing 0.05% (v/v) Tween 80. At different time points, three mice per group were killed, and bacterial burden was determined by plating serial dilutions of lung and spleen homogenates onto Middlebrook 7H10 agar plates supplemented with 10% (v/v) OADC enrichment (Becton Dickinson Microbiology Systems) and 0.5% (v/v) glycerol. Plates contained 5 mM validamycin A for the conditional lethal Mtb mutant strain ΔtreS ΔglgE (pMV361::treS). Mouse protocols were approved by the Animal Care and Use Committee of the Albert Einstein College of Medicine.

Other methods

See the Supplementary Methods for more detailed methods, including strains (Supplementary Table 2), growth conditions, oligonucleotide sequences (Supplementary Tables 1 and 3) and further chemical analyses.


We thank G. Hatfull for promoting a collaboration between the Bornemann and Jacobs labs. This work was supported by US National Institutes of Health grants AI26170 (to W.R.J.) and AIO-68135 (Structural Biology of TB Drug Targets), the Albert Einstein College of Medicine Center for AIDS Research grant AIO-51519 and the UK Biotechnology and Biological Sciences Research Council through an Institute Strategic Programme Grant to the John Innes Centre. We thank C. Bruton and K. Chater for access to unpublished data and insightful discussions, D. Hopwood for advice on the manuscript and S. Fairhurst and L. Hill for recording NMR and mass spectra at the John Innes Centre. G.B. acknowledges support in the form of a Personal Research Chair from J. Bardrick (Royal Society Wolfson Research Merit Award, as a former Lister Institute Jenner Research Fellow). G.B. also acknowledges support from The Medical Research Council and the Wellcome Trust (081569/2/06/2). We thank P. Illarionov for discussion of NMR results and the staff in technical services of the University of Birmingham (especially N. Spencer, G. Burns and P. Ashton) for help in the NMR, GC and ES-MS experiments. This paper is dedicated to the late Chris Lamb.


Accession codes. Gene Expression Omnibus: Microarray data have been deposited in the US National Center for Biotechnology Information Gene Expression Omnibus (GEO series accession number GSE18575).

Author contributions

R.K., S.B. and W.R.J. coordinated the study. R.K., K.S., U.V., B.W. and K.E.B. performed experiments. R.K., K.S., U.V., B.W., G.B. and S.B. analyzed data. Z.L. and J.C.S. provided reagents. R.K., S.B. and W.R.J. wrote the paper and K.S., U.V., B.W. and G.B. edited the paper.

Competing financial interests

The authors declare no competing financial interests.

Additional information

Supplementary information is available online at Reprints and permissions information is available online at


1. Dye C. Global epidemiology of tuberculosis. Lancet. 2006;367:938–940. [PubMed]
2. Sassetti CM, Boyd DH, Rubin EJ. Genes required for mycobacterial growth defined by high density mutagenesis. Mol. Microbiol. 2003;48:77–84. [PubMed]
3. Sassetti CM, Rubin EJ. Genetic requirements for mycobacterial survival during infection. Proc. Natl. Acad. Sci. USA. 2003;100:12989–12994. [PubMed]
4. Tong AH, et al. Global mapping of the yeast genetic interaction network. Science. 2004;303:808–813. [PubMed]
5. Boone C, Bussey H, Andrews BJ. Exploring genetic interactions and networks with yeast. Nat. Rev. Genet. 2007;8:437–449. [PubMed]
6. Belanger AE, Hatfull GF. Exponential-phase glycogen recycling is essential for growth ofMycobacterium smegmatis. J. Bacteriol. 1999;181:6670–6678. [PMC free article] [PubMed]
7. Garg SK, Alam MS, Kishan KV, Agrawal P. Expression and characterization of alpha-(1,4)-glucan branching enzyme Rv1326c of Mycobacterium tuberculosis H37Rv. Protein Expr. Purif. 2007;51:198–208. [PubMed]
8. Sambou T, et al. Capsular glucan and intracellular glycogen of Mycobacterium tuberculosis: biosynthesis and impact on the persistence in mice. Mol. Microbiol. 2008;70:762–774. [PMC free article] [PubMed]
9. Pan YT, et al. Trehalose synthase converts glycogen to trehalose. FEBS J. 2008;275:3408–3420. [PubMed]
10. Pan YT, et al. Trehalose synthase of Mycobacterium smegmatis: purification, cloning, expression, and properties of the enzyme. Eur. J. Biochem. 2004;271:4259–4269. [PubMed]
11. Jarling M, Cauvet T, Grundmeier M, Kuhnert K, Pape H. Isolation of mak1 from Actinoplanes missouriensis and evidence that Pep2 from Streptomyces coelicolor is a maltokinase. J. Basic Microbiol. 2004;44:360–373. [PubMed]
12. Usui T, et al. Proton magnetic resonance spectra of D-gluco-oligosaccharides and D-glucans. Carbohydr. Res. 1974;33:105–116.
13. Bosch AM. Classical galactosaemia revisited. J. Inherit. Metab. Dis. 2006;29:516–525. [PubMed]
14. Slepak T, Tang M, Addo F, Lai K. Intracellular galactose-1-phosphate accumulation leads to environmental stress response in yeast model. Mol. Genet. Metab. 2005;86:360–371. [PubMed]
15. Yarmolinsky MB, Wiesmeyer H, Kalckar HM, Jordan E. Hereditary defects in galactose metabolism in Escherichia coli mutants, II. Galactose-induced sensitivity. Proc. Natl. Acad. Sci. USA. 1959;45:1786–1791. [PubMed]
16. Streeter JG, Gomez ML. Three enzymes for trehalose synthesis in Bradyrhizobium cultured bacteria and in bacteroids from soybean nodules. Appl. Environ. Microbiol. 2006;72:4250–4255. [PMC free article] [PubMed]
17. Karakousis PC, Williams EP, Bishai WR. Altered expression of isoniazid-regulated genes in drug-treated dormant Mycobacterium tuberculosis. J. Antimicrob. Chemother. 2008;61:323–331. [PubMed]
18. Boshoff HI, et al. The transcriptional responses of Mycobacterium tuberculosis to inhibitors of metabolism: novel insights into drug mechanisms of action. J. Biol. Chem. 2004;279:40174–40184. [PubMed]
19. Manjunatha U, Boshoff HI, Barry CE. The mechanism of action of PA-824: novel insights from transcriptional profiling. Commun. Integr. Biol. 2009;2:215–218. [PMC free article] [PubMed]
20. Kumar A, et al. Heme oxygenase-1-derived carbon monoxide induces the Mycobacterium tuberculosis dormancy regulon. J. Biol. Chem. 2008;283:18032–18039. [PMC free article] [PubMed]
21. Voskuil MI, et al. Inhibition of respiration by nitric oxide induces a Mycobacterium tuberculosis dormancy program. J. Exp. Med. 2003;198:705–713. [PMC free article] [PubMed]
22. Dahl JL, et al. The role of RelMtb-mediated adaptation to stationary phase in long-term persistence of Mycobacterium tuberculosis in mice. Proc. Natl. Acad. Sci. USA. 2003;100:10026–10031. [PubMed]
23. Boshoff HI, Reed MB, Barry CE, III, Mizrahi V. DnaE2 polymerase contributes to in vivo survival and the emergence of drug resistance in Mycobacterium tuberculosis. Cell. 2003;113:183–193. [PubMed]
24. Murphy HN, et al. The OtsAB pathway is essential for trehalose biosynthesis in Mycobacterium tuberculosis. J. Biol. Chem. 2005;280:14524–14529. [PubMed]
25. Arguelles JC. Physiological roles of trehalose in bacteria and yeasts: a comparative analysis. Arch. Microbiol. 2000;174:217–224. [PubMed]
26. Jackson M, Brennan PJ. Polymethylated polysaccharides from Mycobacterium species revisited. J. Biol. Chem. 2009;284:1949–1953. [PMC free article] [PubMed]
27. Schneider D, Bruton CJ, Chater KF. Duplicated gene clusters suggest an interplay of glycogen and trehalose metabolism during sequential stages of aerial mycelium development in Streptomyces coelicolor A3(2) Mol. Gen. Genet. 2000;263:543–553. [PubMed]
28. Elbein AD, Pastuszak I, Tackett AJ, Wilson T, Pan YT. The last step in the conversion of trehalose to glycogen: a mycobacterial enzyme that transfers maltose from maltose-1-phosphate to glycogen. J. Biol. Chem. 2010 January 29; published online. [PMC free article] [PubMed]
29. Cantarel BL, et al. The Carbohydrate-Active EnZymes database (CAZy): an expert resource for glycogenomics. Nucleic Acids Res. 2009;37:D233–D238. [PMC free article] [PubMed]
30. Stam MR, Danchin EGJ, Rancurel C, Coutinho PM, Henrissat B. Dividing the large glycoside hydrolase family 13 into subfamilies: towards improved functional annotations of α-amylase-related proteins. Protein Eng. Des. Sel. 2006;19:555–562. [PubMed]
31. McCarter JD, Withers SG. Unequivocal identification of Asp-214 as the catalytic nucleophile of Saccharomyces cerevisiae alpha-glucosidase using 5-fluoro glycosyl fluorides. J. Biol. Chem. 1996;271:6889–6894. [PubMed]
32. Kaur D, Guerin ME, Skovierova H, Brennan PJ, Jackson M. Chapter 2 Biogenesis of the cell wall and other glycoconjugates of Mycobacterium tuberculosis. Adv. Appl. Microbiol. 2009;69:23–78. [PMC free article] [PubMed]
33. Elbein AD, Pan YT, Pastuszak I, Carroll D. New insights on trehalose: a multifunctional molecule. Glycobiology. 2003;13:17R–27R. [PubMed]
34. Kohanski MA, Dwyer DJ, Hayete B, Lawrence CA, Collins JJ. A common mechanism of cellular death induced by bactericidal antibiotics. Cell. 2007;130:797–810. [PubMed]
35. Bardarov S, et al. Specialized transduction: an efficient method for generating marked and unmarked targeted gene disruptions in Mycobacterium tuberculosis, M. bovis BCG and M. smegmatis. Microbiology. 2002;148:3007–3017. [PubMed]
36. Lanzetta PA, Alvarez LJ, Remack PS, Candia OA. An improved assay for nanomole amounts of inorganic phosphate. Anal. Biochem. 1979;100:95–97. [PubMed]
37. Saeed AI, et al. TM4: a free, open-source system for microarray data management and analysis. Biotechniques. 2003;34:374–378. [PubMed]