A lysine acetyltransferase contributes to the metabolic adaptation to hypoxia in Mycobacterium tuberculosis

Summary

Upon inhibition of respiration, which occurs in hypoxic or nitric oxide-containing host microenvironments, Mycobacterium tuberculosis (Mtb) adopts a non-replicating ‘quiescent’ state and becomes relatively unresponsive to antibiotic treatment. We used comprehensive mutant fitness analysis to identify regulatory and metabolic pathways that are essential for the survival of quiescent Mtb. This genetic study identified a protein acetyl-transferase (MtPat/Rv0998) that promoted survival and altered the flux of carbon from oxidative to reductive TCA reactions. Reductive TCA requires malate dehydrogenase (MDH) and maintains the redox state of the NAD+/NADH pool. Genetic or chemical inhibition of MDH resulted in rapid cell death in both hypoxic cultures and in murine lung. These phenotypic data, in conjunction with significant structural differences between human and mycobacterial MDH enzymes that could be exploited for drug development, suggest a new strategy for eradicating quiescent bacteria.

eTOC blurb:

Bacterial infections can persist because non-replicating organisms are killed slowly by antibiotics. This work describes a metabolic change that is essential for the survival of non-replicating Mycobacterium tuberculosis. Inhibition of the metabolic pathway favored under these conditions results in rapid death, suggesting a new strategy for accelerating therapy.

Introduction

Tuberculosis (TB) has proven difficult or impossible to control in much of the world. The resilience of this disease can be attributed to a complex set of factors that includes the huge reservoir of asymptomatic Mycobacterium tuberculosis (Mtb) infection, which is estimated to include more than one billion individuals, and the remarkably long and complex treatment regimen that is necessary to cure the disease (Zumla et al., 2013). Despite the availability of drugs that rapidly kill (Mtb) in vitro, treatment of active TB disease still requires the administration of multiple agents for at least six months. It remains difficult to ensure the delivery of this arduous regimen, and incomplete treatment is both ineffective and contributes to the selection of drug-resistant strains.

A defining feature of Mtb that limits antibiotic efficacy is its relatively slow growth rate and propensity to exit the cell cycle and adopt a hardy nonreplicating state during times of stress (Gengenbacher et al., 2010; Voskuil et al., 2003; Wayne and Sohaskey, 2001). While Mtb does not replicate in this “quiescent” state it remains metabolically active (Baek et al., 2011), distinguishing quiescence from true microbial “dormancy” responses, such as sporulation. Despite this continual metabolic activity, quiescent Mtb becomes relatively tolerant to existing antibiotics that target bacterial functions, such as transcription, translation, and cell wall synthesis, which are required primarily during active replication (Wayne and Hayes, 1996; Wayne and Sramek, 1994). Understanding which metabolic pathways remain necessary for the survival of quiescent Mtb could lead to the rational design of new drugs that are more effective in eradicating these cells.

A wide variety of stresses imposed by the immune system during infection, such as low oxygen, low pH, and iron starvation, and exposure to nitric oxide (NO), are capable of reducing mycobacterial growth and antibiotic efficacy (Baek et al., 2011; Voskuil et al., 2003). Immune stress appears to be a primary determinant of drug efficacy, as the activation state of local macrophages inversely correlates with antibiotic activity (Liu et al., 2016). While the specific adaptations to each of these immune stresses is likely to differ, they all initially trigger the DosRST regulatory system (Baek et al., 2011; Voskuil et al., 2003) that senses the respiratory capacity of the bacterium by responding to the reduction state of the quinone pool (Honaker et al., 2010). This response to respiratory limitation has been most thoroughly studied using a well-defined in vitro model in which hypoxia is used to arrest the growth of this obligate aerobe (Eoh and Rhee, 2013; Rustad et al., 2008; Wayne and Hayes, 1996). Upon oxygen depletion, the DosRST system rapidly induces a 48 gene regulon that includes the Tgs1-encoded triglyceride synthase (Park et al., 2003). The resulting carbon storage response redirects acetyl-CoA from growth promoting pathways into the formation of cytosolic lipid bodies, which arrests bacterial replication (Baek et al., 2011). This initial response is followed by more stable transcriptional and metabolic changes that promote long-term survival (Galagan et al., 2013; Rustad et al., 2008). The outcome of this adaptation is a remodeling of the TCA cycle from the oxidative reactions that occur during respiration to a reductive pathway that regenerates NAD+ and produces succinate to sustain membrane potential, ATP synthesis, and anaplerotic pathways (Eoh and Rhee, 2013; Watanabe et al., 2011). A common feature of these individual adaptations is the maintenance of cellular redox state, since both lipid anabolism and reductive TCA flux are able to consume NADH and regenerate NAD+ in the absence of an active electron transport chain.

While these previous studies describe the metabolic state of Mtb during hypoxia-induced quiescence, a large-scale genetic analysis of this process has not been performed. Without these functional data, it remains unclear which of these regulatory and metabolic alterations are most important for the survival of the quiescent cell. This work describes a comprehensive genetic study to define the consequences of disrupting each of these pathways, and identifies a metabolic regulatory process that serves a primary role in the adaptation.

Results

Genome-wide mutant fitness profiling identifies pathways that are essential for the adaptation to hypoxia

To understand the consequences of perturbing individual metabolic pathways in hypoxia, we performed a genome-wide genetic screen for mutants that were unable to survive under this condition. A highly-saturated library of random transposon mutants was generated, in which more than 55,000 of the 67,992 possible TA dinucleotide insertion sites in non-essential genomic regions were occupied. This complexity ensured that virtually every nonessential gene was disrupted at multiple distinct sites by transposon insertion (Dejesus et al., 2017). The transposon library was subjected to hypoxic culture for either three or six weeks, and the relative abundance of each mutant in the input pool and selected pools was compared using next generation sequencing. This “transposon sequencing” (TNseq) strategy allows the relative fitness of each Mtb mutant to be estimated over the period of selection. The multiple time points included in this experimental design allowed the differentiation of distinct classes of phenotype (Figure 1a–1c). Thirty mutants were underrepresented after three weeks of hypoxia, relative to the input pool (Input vs 3 week hypoxia, “IvH3”). These mutants likely include both those with defects in growth during the period of oxygen depletion and survival during the initial period of hypoxia. A larger set of 101 genes were underrepresented after 6 weeks, relative to the input (“IvH6”). This set included 28 of the 30 genes that were identified at the early time point. To specifically identify the mutants that were unable to survive the hypoxia-induced quiescent state we defined 32 mutants that lost viability between 3 and 6 weeks of hypoxic culture (H3vH6”). The full dataset is provided in Tables S1 and S2.

Figure 1. Genetic Screen to Identify Mutants with Altered Fitness in Hypoxia.

(A) Scheme for identifying genes essential for adapting to and surviving in hypoxia. Saturated transposon libraries were grown in hypoxic vials for three and six weeks, at which point DNA was obtained from the surviving mutants. The compositions of the resulting mutant pools were compared with the input pool by TNseq. The behavior of hypothetical Mtb mutants are depicted. These can reflect neutral mutations (black), fitness advantages (blue), defects during the adaptation to hypoxa (red and green), and defects in survival in hypoxia (yellow). (B) Relative fitness of each individual mutant in the Mtb library at weeks 3 versus 6 of hypoxia. Dots are sized by statistical significance (Q-value indicates adjusted p-value as determined by resampling). Dotted lines indicate arbitrary log fold change cutoffs (1.5 fold), which in conjunction with a Q < 0.05 significance threshold yielded 32 mutants with a hypoxic survival defect. Only genes disrupted by >4 distinct TA insertions in the library were included in this analysis. (C) Using the criteria described in panel B, the indicated numbers of mutants with conditional fitness were identified. Coloring of mutant classes is consistent with panel A. “IvH3”: Input versus hypoxia week 3, IvH6: Input versus hypoxia week 6, H3vH6: hypoxia week 3 versus hypoxia week 3. D-F) Relative fitness of mutants lacking genes of the previously described DosR regulon (D), the mce1 operon (E), or predicted adenylate cyclases and ∆mt-pat (rv0998) (F). Only mutants represented in the input transposon library are shown. In F, mutant pools from 3 and 6 weeks of hypoxia are compared. Throughout, asterisks indicate statistical significance (Q <0.05 by resampling).

While the majority of genes found to be conditionally essential in hypoxia had no annotated function, a number of known pathways were identified. As expected, several of the genes found to be necessary over the six week culture are part of the 48 gene DosR regulon, including dosR (rv3133c), acr (rv2031c), dosS (rv3132c), otsB1 (rv2006), pfkB (rv2029c), and the universal stress protein, rv2623 (Figure 1d). DosR activation acts, at least in part, to maintain cellular redox state through an increase in lipid anabolism (Leistikow et al., 2010). Lipid catabolism appeared to have the opposite effect on bacterial fitness during hypoxia, as genes in the mce1 operon, which encode a fatty acid import system(Nazarova et al., 2017), were uniformly over-represented in the hypoxic pools (Figure 2e). In addition, the prokaryotic coenzyme, F420, that is involved in protecting mycobacteria from oxidative and nitrosative damage (Gurumurthy et al., 2013; Purwantini and Mukhopadhyay, 2009) was important under hypoxic conditions. We found that mutants lacking the F420 synthetic gene, fbiA (rv3261), or a major F420 reducing activity, fgd1 (rv0407) (Hasan et al., 2010), were unable to survive in hypoxia by week six, reinforcing the importance of cellular redox maintenance.

Figure 2. MtPat is necessary for redox homeostasis during hypoxia.

(A) Survival of ∆MtPat, wild type (WT), and the complemented mutant (Comp MtPat) in hypoxia. Asterisks indicate statistical significance (black:∆MtPat versus WT, gray: ∆MtPat versus complemented strain), determined by t-test using the Bonferroni correction (* p<0.05). Data depict one representative experiment of three. B) Metabolic flux through TCA reactions was estimated by quantifying the conversion of exogenously added 2-[¹³C]-glucose into each intermediate using (LC-MS). Metabolic flux was determined in under aerobic and hypoxic conditions using either WT or ∆MtPat mutant bacteria (bars are colored as indicated in figure) as described in the Methods section. Hypothetical bifurcated TCA cycle is depicted, as described in the text. Data represent the average of three independent experiments. * p<0.05, ** p<.005 by t-test using the Holm-Bonferroni correction. “ND” indicated no detectable conversion. C) Viability of ∆MtPat, WT, and the complemented mutant in 7H9 media containing oleate is shown in the left panel. The right panel depicts a parallel viability study in media lacking oleate. One representative experiment of two is depicted. * p<0.05, ** p<.005 by t-test using the Bonferroni correction. D) Ratio of NADH/NAD⁺ in WT, ∆MtPat and the complemented mutant strain under aerobic culture or after 7 days of hypoxic culture. Figure depicts one representative experiment of three performed. * p = 0.02 by t-test.

The most profound survival defect observed in this study belonged to mutants lacking the rv0998 gene (Figure 1b and 1f), which encodes a cAMP-regulated protein lysine acetyltransferase (PAT), known as Mt-Pat (Nambi et al., 2010). The PAT domain of this protein is homologous to a large family of protein acetyltransferases that are widely distributed in bacterial genomes. Mt-Pat also contains an additional nucleotide-binding domain that is structurally similar to the cAMP responsive region of protein kinase A, and inhibits acetyltransferase activity until cAMP is bound (Lee et al., 2012). To investigate whether cAMP was involved in the adaptation to hypoxia, we examined the behavior of the sixteen genes with predicted adenylate cyclase (AC) domains, of which ten have been biochemically characterized as active enzymes involved in cAMP synthesis (Shenoy and Visweswariah, 2006). In our TNseq study, mutation of only one AC resulted in a significant fitness defect in the hypoxic condition, Rv1625c/Cya. Despite the statistical significance of this phenotype, mutants lacking this enzyme displayed a quantitatively modest fitness defect compared to mt-pat mutants (Figure 2f), possibly reflecting redundancy between AC enzymes or a cAMP-independent role for Mt-Pat.

mt-pat deletion alters carbon metabolism and redox homeostasis in hypoxia

To confirm the predicted essentiality of Mt-Pat in hypoxia, this gene was deleted in Mtb. The ∆mt-pat deletion mutant grew normally in aerobic conditions and reached a similar cell density as wild type Mtb in hypoxic vials. However, unlike wild type Mtb or a complemented strain, the ∆mt-pat mutant progressively lost viability once hypoxia was achieved (Figure 2a), consistent with the phenotype predicted by TNseq.

In a number of systems, Mt-Pat homologs regulate carbon metabolism (Castaño-Cerezo et al., 2011; Wang et al., 2010). In particular, acetyl-proteomics, biochemical assays, and physiological studies have implicated acetyl- and acyl-CoA ligases as targets of bacterial PAT-mediated regulation gram negative bacteria (Chan et al., 2011; Crosby et al., 2012). In mycobacteria, Mt-Pat has similarly been found to acetylate both the acetyl-CoA ligase (Acs) and 10 different FadD paralogs that function as acyl-CoA ligases (Nambi et al., 2013). This family of enzymes activates organic acids, such as fatty acids, that are eventually catabolized to acetyl-CoA. The coordinated regulation of this class of enzymes by acetylation has led to the proposal that Mt-PAT orthologs function to regulate the formation of acetyl-CoA (Crosby et al., 2012), which feeds oxidative TCA reactions. Based on these previous observations, we hypothesized that Mt-Pat may be involved in regulating TCA fluxes that are known to be altered in hypoxia. To specifically quantify the effect of Mt-Pat on carbon metabolism, we used stable isotope tracing to measure the relative flux through metabolite pools. Wild type, ∆mt-pat, and complemented strains were independently grown on solid media containing a mixture of substrates for glycolysis and beta-oxidation. 2-[¹³C]-labeled glucose was supplied upon the initiation of hypoxia to compare the fate of this carbon source in either aerobic orhypoxic conditions. After seven to ten days of hypoxic culture, total metabolites were extracted and analyzed via HPLC-LC/MS. Relative flux was quantified as the increase in abundance of ¹³C labeled form of each metabolite.

Under aerobic conditions, wild-type Mtb incorporated 2-[¹³C]-glucose into all TCA metabolites, consistent with the use of a canonical TCA cycle (Figure 2b). Under hypoxia, however, flux of 2-[¹³C]-glucose was concentrated primarily into malate and succinate, while flux through citrate/isocitrate and 2-ketoglutarate was barely detectable. These data are consistent with the previously observed branched TCA metabolism in hypoxic Mtb, which is defined by a preferential use of reductive reactions that convert malate to succinate instead of the oxidative branch that converts citrate to 2-ketoglutarate (Eoh and Rhee, 2013; Watanabe et al., 2011).

In contrast to wild-type Mtb, the ∆mt-pat mutant continued to incorporate 2-[¹³C]-glucose into the oxidative branch of TCA under hypoxic conditions. While there was no detectable flux through citrate/isocitrate in wild type Mtb in hypoxia, the mutant incorporated ¹³C -labeled carbon into this dedicated first step of oxidative TCA, and ultimately displayed a ¹³C -labeling pattern in both the (iso)citrate, 2-ketoglutarate, and malate pools that was qualitatively similar to aerobically growing bacteria.

The proposed role for Mt-Pat orthologs in regulating acyl-CoA ligases, in conjunction with our observation that the loss of the Mce1 fatty acid import system promotes Mtb survival in hypoxia (Figure 1e), suggested that the regulation of fatty acid catabolism by Mt-Pat might be important under these conditions. To investigate this hypothesis, we determined whether the survival of the ∆mt-pat mutant was altered by the abundance of fatty acid in defined medium. As we found previously, the ∆mt-pat mutant displayed a survival defect in media containing the same concentration oleate as the standard 7H9 medium used in the rest of this study. In contrast, the omission of fatty acid from the medium ameliorated the viability defect of the ∆mt-pat mutant in hypoxic culture (Figure 3c). Together, our data are consistent with the existing model that Mt-Pat orthologs function to regulate the formation of acetyl-CoA. The absence of this regulation in hypoxia results in continual flux of this metabolite into oxidative TCA reactions.

Figure 3. Depletion of the Mdh enzyme Reduces Hypoxic Survival.

A) The abundance of the Mdh-DAS protein during inducible depletion was assessed by targeted label-free mass spectrometry. ATc was added to a culture of the mdh-DAS strain after 10 days of hypoxia. At the indicated time points, three independent peptides of Mdh were quantified in cell lysates by LC/MS/MS. Mdh abundance in each sample was normalized to the concentration of SigA protein. B) The mdh-DAS strain was cultured under aerobic conditions either with or without ATc, and growth was monitored by optical density (A600). C) The mdh-DAS strain was cultured under aerobic conditions until saturation (day 10), at which point ATc was added. CFU were enumerated at the indicated time points by plating. D) The mdh-DAS strain was cultured under hypoxic conditions and ATc was added to non-replicating cultures at day 10. CFU were enumerated at the indicated time points by plating. E) The relative flux of 2-[¹³C]-glucose into citrate was determined in WT and Mdh-DAS strains under hypoxic conditions. ATc exposure was initiated at 10 days, simultaneously with the addition of 2-[¹³C]-glucose. Flux into citrate was determined after 10 days of labeling. F) Mdh depletion in mouse lung. After aerosol infection with pooled wild type and mdh-DAS strains, Mdh depletion was initiated via doxycycline (“dox”) in two different regimens. The “acute regimen” began one week after infection and continued for two weeks. The “chronic regimen” began 6 weeks postinfection and continued for 8 weeks. Mutant fitness was measured by detection of unique strain-specific DNA barcode via qPCR. Throughout, asterisks indicate p<0.05 by t-test using bonferroni’s correction.

The preferential use of reductive TCA reactions in hypoxia has been proposed to be important for the regeneration of NAD+, which becomes reduced to NADH in the absence of an efficient electron acceptor, such as oxygen. As both the oxidative branch of TCA and fatty acid catabolic pathways produce NADH, we hypothesized that the impaired survival of the ∆mt-pat mutant in hypoxia was related to a defect in NAD+regeneration. To test this hypothesis, we measured the redox state of the NAD+/NADH pool (Figure 2d). Under aerobic conditions, both wild type and the ∆mt-pat mutant maintained a balance of NADH and NAD+, with this cofactor primarily in the oxidized form. In contrast, the ratio of NADH/NAD+ increased by more than 10-fold in the ∆mt-pat deletion mutant upon oxygen limitation. The observed redox imbalance in the mutant reflected the increased flux through oxidative reactions inferred by ¹³C-carbon tracing, and implies that depletion of NAD+ could underlie the impaired survival of this strain in hypoxia.

Malate dehydrogenase activity is required for survival in hypoxia and during infection

The impaired survival of the ∆mt-pat mutant in hypoxia indicated that preferential utilization of reductive TCA reactions was important for maintaining viability. In order to assess the importance of these reactions, we constructed a mutant in which the expression of the only non-redundant enzyme necessary for the predicted reductive pathway, malate dehydrogenase (Mdh), could be inducibly repressed. The mdh gene was placed under an anhydrotetracline (ATc)-repressible promoter to inhibit transcription. In addition, the mdh gene was fused with a DAS+4 tag, which directs the proteolysis of the fusion protein by the ClpXP system upon the induction of the SspB adapter by ATc addition (Kim et al., 2011). ATc was added to the Mdh-DAS mutant after ten days of hypoxic culture, and between 5 and 11 days later we found that this “dual control” strategy depleted 97–99% of the Mdh protein (Figure 3a).

To assess the relative contribution of Mdh survival, the protein was depleted in either aerobic or hypoxic cultures and viability was assessed over time. Mdh depletion arrested the growth of the Mdh-DAS mutant when ATc was added to exponentially-growing aerobic cultures (Figure 3b), confirming the predicted essentiality of this gene under these conditions (Dejesus et al., 2017). When Mdh was depleted in non-growing cells, the relative effect depended on the presence of oxygen. Mdh depletion in stationary phase aerobic cultures moderately increased the rate of cell death, reducing CFU by approximately 10-fold in 18 days and 100-fold in 30 days. In contrast, Mdh depletion in hypoxia resulted in more rapid cell death. Within 30 days of ATc addition, the viability of Mdh-DAS mutant cultures had fallen below the limit of detection, reflecting at least a 1000-fold reduction in CFU (Figure 3c-d). To determine if Mdh depletion had a similar effect on central carbon metabolism as ∆mt-pat deletion, we traced the fate of 2-[¹³C]-glucose in wild type or Mdh-DAS strains in the presence of ATc. Under hypoxic conditions, we found that Mdh depletion increased flux through the oxidative branch of TCA, as measured by the increased conversion of citrate/isocitrate to the ¹³C-labeled form (Figure 3e). These data further support the association between oxidative TCA flux and cell death during hypoxia.

While the induction of the DosR regulon during growth in the lungs of C57BL/6 mice (Voskuil et al., 2003) suggests that respiration is limited at this site, overtly hypoxic lesions are not observed in this infection model (Via et al., 2008). To determine if Mdh is necessary for Mtb survival in the mouse lung despite the absence of hypoxic lesions, we infected C57BL/6 animals with a mixture of wild type Mtb and the Mdh-DAS mutant, and the relative abundance of these strains was quantified in lung homogenates by quantitative PCR (Blumenthal et al., 2010). Depletion of Mdh was initiated either 1 week or 6 weeks after infection by the administration of doxycycline. In either treatment regime, the viability of the Mdh-DAS strain was reduced 10⁵-10⁶-fold over two and eight weeks of doxycycline administration, respectively (Figure 3f).

Chemical inhibition of Mdh kills Mtb in hypoxia

While the requirement for Mdh under hypoxic conditions in vitro and in the mouse lung suggested that Mdh inhibition could be therapeutically useful, chemical inhibition of TCA enzymes has proven difficult in the past (Mdluli et al., 2015). To determine if the Mtb Mdh enzyme is susceptible to chemical inhibition we performed a high-throughput screen for small molecule antagonists. An enzymatic assay that measures Mdh-dependent NADH oxidation directly following NADH fluorescence was used (Napper and Sivendran, 2011). To focus the screen on compounds that are likely to have access to the mycobacterial cytosol, we used collections of compounds that were shown to inhibit mycobacterial growth at 20 uM or lower. An in-house collection of 2028 of these “whole cell actives” derived from screening custom diversity library of 104,000 compounds against Mtb was tested in an HTS optimized end point MDH enzyme assay at final concentration of 8 µM. The screen produced Z’-values (Zhang et al., 1999) of 0.45 or greater, and generated 28 initial hits that inhibited MDH activity by 50% or more. The dose-response of hits was evaluated in a continuous enzyme assay, and one compound was confirmed as an MDH inhibitor (MDH-I) with IC50 value of 2.8 µM (Figure 4a–4b). In addition, 1,113 previously-described whole-cell actives (Ananthan et al., 2009) were tested against MDH at 100 µM, but produced no additional verified inhibitors. A series of MDH-I analogs were tested using a modified assay that relieves product inhibition. In this assay, the MDH-I compound was found to possess a similar IC50 as we previously measured. Modification or removal of moieties from this molecule uniformly reduced its inhibitory activity, supporting the specificity of the interaction (Table S3).

Figure 4. Chemical inhibition of Mdh kills hypoxic Mycobacterium tuberculosis.

A) Structure of MDH-I. B) Inhibition of Mdh biochemical activity by MDH-I. Each point represents and average of triplicate measurements. The axes cross at zero. C) WT, mdh-DAS, or control sucD-DAS strains were exposed to the indicated doses of MDH-I in the presence of ATc to initiate protein depletion. Mtb growth was quantified using the Alamar Blue fluorescence assay. D) An mdh inducible overexpression strain was exposed to the indicated doses of MDH-I in the presence of the indicated concentration of inducer (Atc). The MIC of MDH-I at each ATc concentration was determined as in panel C. The slope of this dose-response was significantly different from zero (p = 0.013, determined by linear regression). E) MDH-I (40uM) and/or INH (2uM) were added to hypoxic cultures at day 10 (arrow), and CFU were enumerated. Limit of detection in panel D is indicated by the dotted line (100 CFU/ml). Throughout, asterisks indicate p<0.05 by t-test using bonferroni’s correction. In panel E gray: untreated versus MDH-I+INH, black: untreated versus MDH-I.

To determine if the MDH-I’s growth inhibitory activity was related to Mdh inhibition in the intact cell, we determined if genetic alteration of Mdh expression levels altered susceptibility (Evans and Mizrahi, 2015). To determine if under-expression sensitized the bacterium, the Mdh-DAS strain was treated with the MDH-I at a range of concentrations around the MIC (0.5x MIC to 2x MIC), with and without ATc to induce the depletion of Mdh. Using this scheme, we identified synergy between these treatments. The MIC against wild type Mtb was 20uM, and we found that the Mdh-depleted strain was hypersensitive to a 0.5x MIC concentration (Figure 4c). Genetic depletion of the succinyl-CoA synthetase subunit, SucD, had no effect on MDH-I sensitivity, verifying the specificity of the observed chemical genetic synergy between MDH-I and mdh depletion. To determine if overexpression of the enzyme had the inverse effect on susceptibility, we engineered a strain that contained an ATc-inducible copy of mdh integrated into the chromosome. Addition of increasing concentrations of ATc to this strain significantly increased the MIC of the MDH-I compound (Figure 4d), whereas maximal overexpression of an irrelevant peptide in the same expression vector had a non-significant effect on MDH-I activity (MIC values +/− SEM were 19.8 +/−1.16 and 20.32 +/− 1.85 at zero and 200 ng/ml ATc, respectively). Together, these data indicated that MDH-I acts, at least in part, by inhibiting Mdh in the intact cell.

The efficacy of MDH-I in hypoxic conditions was then tested, to investigate whether chemical inhibition of Mdh could mimic the genetic mdh depletion phenotype. The mycolic acid synthesis inhibitor, isoniazid, which is generally ineffective against quiescent Mtb (Wayne and Sramek, 1994), was included as a control. Each compound was added to 10 day-old non-replicating hypoxic cultures. Over the treatment period,isoniazid had a minimal effect on the viability of quiescent Mtb. In contrast, MDH-I had a similar effect as Mdh protein depletion, killing 99% of Mtb after 21 days of treatment and reducing viability beyond the limit of detection after 35 days (Figure 4d). These data indicate that chemical inhibition of Mdh is an effective strategy for killing quiescent bacteria.

Structural comparison revealed unique aspects of the mycobacterial Mdh active site.

Leveraging these observations for TB therapy would require the specific inhibition of Mtb’s Mdh, relative to its human homologs. The Mdh inhibitor characterized in this study was toxic toward HeLa cells (data not shown), likely due to a lack of specificity for the prokaryotic enzyme. To assess if the specific chemical inhibition of the Mdh of M. tuberculosis (MtMdh) over the human ortholog is feasible, we performed a comparative structural analysis. The crystal structure of apo-MtMdh has been previously reported (Ferraris et al., 2015). However, this enzyme is thought to undergo a conformational change upon substrate binding, which alters the geometry of the active site (Chapman et al., 1999; Hung et al., 2013; Musrati et al., 1998). To characterize the more relevant structure of the catalytic site in the ligand-bound state, we solved the crystal structure of MtMdh in complex with the substrate NADH. We found a dimer of MDH with a molecule of NADH bound to each of the protein chains in the asymmetric unit of MtMdh crystals. The overall structure of this protein is much more similar to human mitochondrial Mdh (HmMdh) than to the cytoplasmic isoform. Despite this similarity, we identified notable differences between the MtMdh and HmMdh active sites that might be exploited to engineer a compound that is specific to the Mtb protein.

Three different conformations of a substrate binding loop were captured in the solved MtMdh structures (Figure 5a. Pink loops). In the apo MtMdh structure, this loop is disordered in chain A and is in the open conformation for chain B. In the NADH bound MtMdh structure, the loop is in an intermediate, partially closed state in chain A and in a completely closed form for chain B. In the NADH-bound form of the protein, we identified a molecule of Tris in the malate binding pocket, which came from the crystallization buffer (Figure 5a-5b. Tris molecule colored yellow). This Tris molecule overlaps with the malate bound in the analogous position of the human mitochondrial Mdh structure, and forms several hydrogen bonds with sidechains and ordered waters in the active site (Figure 5a–5c. Malate molecule colored green). Residues involved in hydrogen bonds with coordinated Tris are conserved to those observed in the corresponding positions of the human mitochondrial mMDH active site. In both HmMdh and MtMdh, the movement of this loop repositions a substrate binding Arg (Arg99 in MtMdh). The interaction between Arg99 and malate has been characterized in several species including HmMdh (Figure 5, in blue). In the structure of MtMdh, we found that Arg99 in its open conformation participates in a unique hydrogen bond network. Arg99^NH2 forms a hydrogen bond with Glu319^OE2, which in turn hydrogen bonds with Asn187^ND2. HmMdh cannot sustain the same network, because a Ser resides in the position of Glu319 and a Gly is in the place of Asn187. These differences suggest that the molecular mechanism of loop movement between the two proteins is not conserved. When we examined the differences in amino acid side chains pointing into the active site within 10Å of malate, six residues differed between the two proteins (S239-A229, T130-I122, A134-V126, I232-V220, Q192-I187, and L156-V151 (numbered as Mtb residue – human residue). This relatively low degree of amino acid conservation between the active sites of MtMDH and HmMDH along with potentially different control of the substrate-binding loop suggest strategies to selectively inhibit the mycobacterial form of the protein.

Figure 5. Structural comparison of human and mycobacterial MDH enzymes.

A) Comparison of substrate binding loop conformation in MDH structures. Human mitochondrial MDH (PDB ID: 2DFD) is shown in blue ribbon, and MtMdh is shown in shades of pink. MtMdh open form from the apo structure (PDB ID: 4TVO) is in magenta, intermediate and closed forms are from the NADH bound structure (PDB ID: 5KVV) in light pink and salmon, respectively. For the side-chains shown as sticks, carbons are colored according to the corresponding structural model. Additional moieties are colored as follows: sidechain oxygens are red, sidechain nitrogens are dark blue, malate from the human MDH structure is green, and Tris from the MtMDH structure is yellow. B and C) sidechain interactions with Tris and malate, respectively.

Discussion

This study extends our knowledge of the metabolic changes that occur as Mtb adapts to the quiescent state under hypoxic conditions by providing functional genetic data to evaluate the relative importance of previously-described changes in gene expression or metabolism. In conjunction with previous studies, these results suggest a multi-step process of metabolic adaption is necessary for maintaining redox homeostasis during hypoxia, and demonstrate that inhibiting Mdh is an effective method for killing Mtb under these conditions. The inferred process of adaptation is depicted in Figure 6 and described below.

Figure 6. Summary of Mt-PAT dependent metabolic regulation.

Three acetyl-CoA producing pathways are depicted. The production of acetyl-CoA from fatty acid or acetate depends on the acyl-CoA ligase family that is inhibited by PAT orthologs. Mt-PAT activity is regulated by at least three metabolites. cAMP and acetyl-CoA directly promote Mt-PAT activity, producing the regulatory effects indicated by red dotted arrows. Conversely, NAD+ is a required cofactor for deacetylation mediated by the sirtuin-like Rv1151c protein (Sirt), producing events indicated in blue. The overall effect of Mt-Pat in hypoxia is to shift carbon flux into reductive TCA reactions. This effect likely depends on inhibition of acetyl-CoA production and possibly on direct regulation of TCA activity (indicated by a question mark). The effects of acetyl-CoA and NAD+ imply feedback regulation, as high acetyl-CoA levels favor Mt-PAT mediated inhibition of oxidative reactions (blue), whereas high NAD+ levels antagonize Mt-PAT favoring reductive reactions (red). The role of MDH in both oxidative and reductive TCA reactions and the potential role for cAMP in modulating these events are indicated.

The immediate response to hypoxia involves the rapid and transient induction of the DosR regulon and triglyceride synthesis, which was shown to be important for the initial growth arrest of the bacterium (Baek et al., 2011) and is expected to regenerate NAD⁺ through lipid anabolism. Subsequent to the transient induction of the DosR regulon, the adaptation to hypoxia relies on Mt-Pat. Both the mycobacterial Mt-Pat and orthologous enzymes in other bacteria have the capacity to inhibit a family of acyl-CoA ligases via acetylation of active site lysine residues. This activity has suggested a general role for PAT orthologs in coordinating fatty acid catabolism, energy homeostasis, and acetyl-CoA availability (Chan et al., 2011; Crosby et al., 2012). In the context of hypoxia, we found that Mt-Pat activity contributes to the maintenance of the NAD+/NADH ratio. As fatty acid catabolism produces NADH, inhibition of this pathway via the biochemically-defined acyl-CoA ligase targets of Mt-Pat could contribute to redox maintenance. Mt-Pat also alters the flux of glycolytic carbon into TCA and the mechanism underling this effect is less clear. Acetyl-proteomic studies in other bacteria have identified a number of potential targets for acetylation that could contribute, such as glyceraldehyde-3-phosphate dehydrogenase and isocitrate lyase (Wang et al., 2010).

The mechanism by which Mt-Pat activity is regulated during the transition to hypoxia is also a topic that will require additional study. Mt-Pat mediated acetylation is known to be regulated by at least three different metabolites. Acetylation activity of MT-Pat is controlled by cAMP and the availability of the acetyl-CoA substrate. In addition, the sirtuin-like deacetylase, Rv1151c, is regulated by NAD+ levels. The resulting regulatory connections summarized in Figure 6 encompass feedback regulation on both acetyl-CoA and NAD+/NADH levels, with cAMP serving a potential auxiliary role. In our genetic screen, mutations in neither rv1151c nor any AC-encoding gene produced a defect that was equivalent to the loss of mt-pat. Thus, we are not able to attribute the activity of Mt-Pat to any single known regulatory mechanism. As the concentration of cAMP, acetyl-CoA, and NAD+ may all be changing as the cells become hypoxic, it is possible that the activity of Mt-PAT relies on the balance of these metabolites instead of a change in the concentration of any one.

While no data implies a direct molecular relationship between Mt-Pat and Mdh, the result of Mt-Pat activity is a switch from oxidative to reductive TCA reactions. The importance of these reactions is implied by the relatively rapid killing of Mtb upon Mdh depletion in hypoxia, a condition in which both our data and others (Eoh and Rhee, 2013; Watanabe et al., 2011) indicate that Mdh is operating in the reductive direction. While the similarly rapid cell death observed upon Mdh inhibition in mouse lung is consistent with an important role for this enzyme during infection, our data are currently not sufficient to differentiate whether this enzyme is operating in the reductive or oxidative direction at this site, and additional roles in the glyoxylate and methylcitrate cycles are also possible. Regardless, the relatively rapid death of Mtb upon Mdh protein depletion in both hypoxic culture and mouse lung suggests that Mdh inhibitors could be used to accelerate TB therapy. The susceptibility of Mdh to chemical inhibition and the structural differences between the mycobacterial and human homologs of this enzyme suggest that specific inhibitors that rapidly eradicate quiescent bacterial populations could be developed.

STAR Methods:

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Bacterial strains:

Mycobacterium tuberculosis H37Rv (ATCC 27294) was used throughout. Except when indicated, cultures were grown using Middlebrook 7H9 media supplemented with 0.05% Tween-80 and OADC (Becton Dickinson) enrichment, or on 7H10 agar with OADC enrichment at 37°C. For fatty acid-free media, 7H9 was supplemented with 0.05% tyloxapol (Sigma) and “AD” enrichment, consisting of 5% fatty-acid free BSA (Sigma) and 1% dextrose. For ¹³C flux experiments, 7H10 media was supplemented with 0.05% 2-[¹³C ]-glucose (Sigma). Aerated cultures were continuously agitated. Hypoxic cultures were grown in 17 mL cultures in sealed glass vials (as described in (Baek et al., 2011)).Additions to hypoxic cultures were made a gas-tight syringe into 31 mL cultures in 45 ml glass serum vials (Wheaton) sealed with rubber stoppers (Wheaton) and crimp caps (Supelco). At indicated time points, 2–3 vials were opened and viable bacterial numbers were enumerated on 7H10 agar plates.

Mice:

10–12 week old female C57BL/6 mice were purchased from the Jackson Laboratory and used for infection studies. Animals were housed and handled in accordance with the University of Massachusetts Medical School Institutional Animal Care and Use Committee.

METHOD DETAILS

Metabolomics:

For ¹³C flux experiments, bacteria were inoculated onto 0.45 µm HVLP membrane filters (Millipore) and placed on top of 7H10 agar, and grown at 37°C for 7–10 days, at which point the filters were acclimated to room temperature and transferred to 7H10 agar plates including 0.05% 2-[¹³C ]-glucose and placed inside an anaerobic chamber (BBL) with a GasPak (Beckton Dickinson) to deplete oxygen. After two hours at room temperature, the anaerobic chamber was placed at 37°C for 7–10 days. Filters containing bacteria were removed an added to 2 mL screw-capped tubes with silica beads (pre-washed with methanol) and 1 mL HPLC-grade acetonitrile:methanol:water (2:2:1). Lysis and quenching of metabolites was performed by bead-beating three times at 6.5 m/sec for 30 seconds each. The supernatant was filtered using a 0.22 µm HVLP filter. HPLC analysis was performed using an Agilent 1200 LC system. Chromatographic separation was carried out on a Cogent Diamond Hydride Type C column at 40 °C. The mobile phase consisted of solvent A (5% acetonitrile containing 0.2% acetic acid) and solvent B (95% acetonitrile containing 0.2% acetic acid). The gradient consisted of: 0–2 min, 90% B; 3–5 min, 85% B; 6–7 min, 80%; 8–9 min, 75% B; 10–11.1 min, 55% B; 11.1–14 min 25% B; 14.1–24 min 10% B followed by a 10 min re-equilibration period at 90% B at a flow rate of 0.4 ml/min. The mass spectrometry experiments were conducted using an Agilent Accurate Mass 6220 TOF. Dynamic mass axis calibration was achieved by continuous infusion of a reference mass solution, leading to mass errors of less than 5 ppm, mass resolution ranging from 10,000 to 25,000 (over m/z 121–955 AMU), and 5 log₁₀ dynamic range. Target mass ions were compared to standards based on accurate mass, retention time, and expected distribution of accompanying isotopomers. Measured ion counts of each metabolite was corrected according to those of chemical standards spiked into homologous mycobacterial extract.

Mtb genetics:

TnSeq:

Three independent transposon libraries consisting of >100,000 independent mutants were generated using Phage mycomar-T7 as described (Sassetti et al., 2001), and mixed at equal optical densities. Eight replicates of this library were subjected to the indicated periods of selection in hypoxic culture in Middlebrook 7H9 media supplemented with 0.05% Tween-80 and OADC (Becton Dickinson) enrichment. The clones surviving selection were recovered by plating in parallel with the input library. Transposon-chromosome junctions were amplified from each of the resulting mutant pools, and the relative abundance of each transposon mutant was compared by TNseq as described (Long et al., 2015).

Mtb mutant strain generation:

The mdh-DAS strain gene generated using a dual control (DUC) approach, as described (Kim et al., 2013). With this system, both repression of transcription and coordinated proteolysis allows for extensive depletion of malate dehydrogenase via the addition of anhydrotetracycline (ATc). Briefly, an mdh merodiploid was generated by integrating a constitutively-expressed copy of the gene at the phage L5 attB site. The endogenous mdh gene was then deleted by recombineering and replaced with a hygromycin resistance cassette, using the pNitET-SacB-kan plasmid and protocols previously described (Murphy et al., 2015). The L5 resident plasmid (streptomycin^R) was replaced by pGMCZ-TetR38-P750-mdh-DAS4, which confers zeocin resistance and expresses a DAS4-tagged version of mdh under control of a reverse TetR repressor. Finally, the strain was transformed with pGMCtKq28-TSC10M1-sspB, which contains a kanamycin-resistance cassette, integrates into the tweety phage attachment site, and expresses the SspB protein via the ATc-regulated TSC10M1 promoter. To generate an MDH overexpression strain, Rv1240 (NCBI reference sequence NC_000962.3, position 1383213−1384202), was cloned with a Tet-inducible UV15 promoter into an L5-site integrating plasmid to create pMDH. As a control, another plasmid with a FLAG epitope in the place of the MDH gene was also transformed into WT H37Rv.

Mouse infection studies:

10–12 week old C57BL/6 mice were infected with a pool of Mtb wild type and mutant strains at an average dose of 102 CFU per animal in a Glas-Col aerosol exposure chamber. At the indicated time points, surviving clones were recovered by plating, chromosomal DNA was extracted from the strain pool, and the relative abundance of the individual strains was determined by quantitative PCR as described as described (Blumenthal et al., 2010).

Parallel reaction monitoring (PRM) assay:

Bacterial lysates for targeted proteomics were prepared as described (Prigozhin et al.,2016). Lysates corresponding to 30 □g total protein were subjected to electrophoresis through a 4–20% mini PROTEAN TGX gradient gel (Bio-Rad) under denaturing conditions for approx. 2 cm into the gel. After staining with Novex Colloidal Blue staining kit (ThermoFisher Scientific), gel containing almost all of the proteins was excised, subjected to in-gel digestion with trypsin, and analyzed by LC/MS/MS as described (Shin et al., 2017). The PRM assay utilized three tryptic peptides of Mdh (residues 25–40 LASGSLLGPDRPIELR, 205–226 NAAEVVNDQAWIEDEFIPTVAK, and 236–252 GASSAASAASATIDAAR), and three tryptic peptides of SigA (238–251 VALLNAEEEVELAK, 343–352 FSTYATWWIR, and 488–500 TLDEIGQVYGVTR; all annotations as per http://tuberculist.epfl.ch). Three fragment ion intensities that generated the most robust signals were extracted from each peptide and summed to determine the peptide abundance. Quantification of targeted peptides was accomplished through the Skyline software (University of Washington). The average of peptide abundance from two technical replicates were determined and the relative level of Mdh was calculated by normalizing the total transition area of each of the 3 Mdh peptides to that of each of the 3 SigA peptides.

Chemical biology:

Mdh inhibitor screening was performed using a direct enzyme assay monitoring the consumption of NADH by measuring fluorescence intensity (excitation at 340 nm, emission at 482 nm). Reaction volume was 50 ul, using 1 nM of MDH in 100 mM Hepes buffer pH 7.5 with 50mM KCl. Compounds were added as 1 ul from DMSO stock solution (final 2% DMSO), and incubated for 20 min, before the reaction was started by adding NADH and oxaloacetate to final concentrations of 250 and 100 uM, respectively. HTS screening was done in 384 well plates, as an end-point fluorescence intensity measurement after 20 min of reaction. Hit re-testing was done by continuously reading fluorescence for 20 min of reaction. The slope of the linear portion of the reaction was used to calculate % inhibition for the IC50 plot. A “reverse assay” was used for MDH-I analog testing. This system used the same enzyme concentration and NADH readout as the direct assay, but the reaction was started by adding NAD+ and LD-malate to final concentrations of 250 uM and 5 mM, respectively. The assay was conducted in 0.4 M hydrazine-glycine buffer pH 9.0, which eliminated product inhibition by consuming oxaloacetate and produced a linear reaction for more than 20 min. Compound libraries were derived from two sources. The “SRI collection” is a set of 1,113 hits from the Southern Research Institute. A total of 100,997 compounds were screened against H37Rv at 10 µg/mL. The screen was conducted in 7H12 medium with palmitate as major carbon source (Ananthan et al., 2009). The “Sac1” and “Sac2” collections were derived from in-house screening. The in-house diversity library was constructed from catalogs of major chemical vendors such that any two individual compounds were no more similar than a 0.7 Tanimoto score. The libraries were acquired and screened sequentially, as approximately 50,000 compounds collections – “Sac1” and “Sac2”. The Sac1 half wasscreened against the mc2−7000 Mtb strain (H37Rv ∆RD1 ∆panCD) at 20 µM in M9 medium supplemented with sodium acetate as carbon source and yielded 1667 hits. The Sac2 library was assayed against the same strain of Mtb under M9-mixed carbon (dextrose+acetate) growth conditions at 10 µM and yielded 541 hits. Chemical-genetic synergy testing using a fluorescent Alamar Blue-based growth assay (Franzblau, 2000).Two different samples of MDH-I compound were purchased from Chembridge. Both samples were characterized by LC-MS, and showed 78–90% purity, with the major component mass consistent with the structure listed by the vendor. Both samples showed similar biological activity.

Crystallography:

Mdh-NADH crystals were obtained by a sitting drop vapor diffusion method, mixing 1 µl of purified protein solution (containing 10 mM NADH) and 1 µl mother liquor at 291 K. The mother liquor contained 0.1 M Bis-Tris, pH 6.5, 0.2 M MgCl₂, and 25% PEG 3,350. Crystals appeared within a week, they were soaked in a cryoprotectant solution (i.e., mother liquor plus 20% glycerol), and flash-cooled in a nitrogen stream at 100 K for the data collection. X-ray diffraction data were collected to resolution of 2.01 Å on a Rigaku Raxis IV++ image system. Data were processed with HKL2000 software suite (Otwinowski and Minor, 1997). The space group of NADH bound Mdh crystal was P2₁2₁2₁, and cell dimensions were a = 52.2 Å, b = 78.2 Å, c = 154.4 Å, α = β = γ = 90°. The structure was determined by molecular replacement using Molrep in CCP4 suit (Vagin and Teplyakov, 2010; Winn et al., 2011). The Protein Data Bank entry 4KDE was utilized as the search model (Hung et al., 2013). The resulting model was improved in COOT (Emsley and Cowtan, 2004) and refined with Refmac in CCP4 and Phenix (Adams et al., 2010; Murshudov et al., 1997). The refinement converged to R/R_free values of 17.3/21.8%. Data collection and refinement statistics are shown in Table S3.

QUANTIFICATION AND STATISTICAL ANALYSIS

TNseq studies:

relative mutant abundance was estimated from the number of independent transposon-chromosome junctions corresponding to a given gene divided by the number of possible insertions sites in that gene that contained insertions (‘Hits/TA”). This metric was compared between samples to determine ‘Relative abundance’. Statistical significance was determined using a nonparametric permutation test to calculate a P value and adjusting for multiple testing using the Benjamini–Hochberg correction to obtain the Q value, as described in (Dejesus et al., 2015).

All other studies:

Two-tailed, unpaired, student’s t-tests were used with appropriate multiple testing corrections were used, as described in the figure legends.

Significance:

A remarkably long antibiotic regimen is necessary to cure a Mycobacterium tuberculosis (Mtb) infection, likely because slowly replicating subpopulations of the pathogen are relatively insensitive to therapy. Using a genetic strategy, we identified genes that are necessary for Mtb survival under non-replicating conditions. Mutation of a gene encoding a protein lysine acetyltransferase (PAT) produced the most profound survival defect in this setting. This PAT was necessary to remodel the metabolic state of the pathogen to favor reductive reactions that maintained the redox state of the bacterium. Genetic or chemical inhibition of the inferred reductive pathway was found to be a remarkably effective way to kill non-replicating Mtb, and structural comparisons between Mtb and human enzymes suggests that the Mtb pathway could be selectively inhibited. These data suggest a new strategy for accelerating tuberculosis therapy.

Highlights:

• In Mtb, protein lysine acetylation is promotes reductive TCA reactions in hypoxia

• Malate dehydrogenase (Mdh) is an essential component of the reductive TCA pathway

• Inhibition of Mdh causes rapid viability loss under hypoxic conditions and infection

• Structural comparisons suggest strategies for the design of specific Mdh inhibitors