The evolution of nitroimidazole antibiotic resistance in Mycobacterium tuberculosis

Our inability to predict whether certain mutations will confer antibiotic resistance has made it difficult to rapidly detect the emergence of resistance, identify pre-existing resistant populations and manage our use of antibiotics to effective treat patients and prevent or slow the spread of resistance. Here we investigated the potential for resistance against the new antitubercular nitroimidazole prodrugs pretomanid and delamanid to emerge in Mycobacterium tuberculosis, the causative agent of tuberculosis (TB). Deazaflavin-dependent nitroreductase (Ddn) is the only identified enzyme within M. tuberculosis that activates these prodrugs, via an F420H2-dependent reaction. We show that the native menaquinone-reductase activity of Ddn is important in aerobic respiration and essential for emergence from dormancy, which suggests that for resistance to spread and pose a threat to human health, the native activity of Ddn must be at least partially retained. We tested 75 unique mutations, including all known sequence polymorphisms identified among ~15,000 sequenced M. tuberculosis genomes. Several mutations abolished pretomanid activation in vitro, without causing complete loss of the native activity. We confirmed that a transmissible M. tuberculosis isolate from the hypervirulent Beijing family already possesses one such mutation and is resistant to pretomanid, even though it was never exposed to pretomanid. Notably, delamanid was still effective against this strain, which is consistent with structural analysis that indicates delamanid and pretomanid bind to Ddn differently. We suggest that the mutations identified in this work be monitored for informed use of delamanid and pretomanid treatment and to slow the emergence of resistance.


Introduction
Tuberculosis (TB) is currently the leading cause of death from a single infectious agent (WHO 2016). The limitations of current treatment regimens, combined with the rapid emergence of multidrug-resistant tuberculosis (MDR-TB) strains, necessitate the development of new drugs.
Three new antitubercular agents are now in advanced clinical development: bedaquiline (Andries et al. 2005) and delamanid (Matsumoto et al. 2006), which have been conditionally approved for MDR-TB treatment (Diacon et al. 2009;Gupta et al. 2015), and pretomanid (Stover et al. 2000), which is part of several promising regimens in phase III trials (http://www.newtbdrugs.org/pipeline/ clinical). The nitroimidazoles, delamanid and pretomanid, are prodrugs that are reductively activated in an F420H2-dependent reaction in Mycobacterium tuberculosis by deazaflavin-dependent nitroreductase (Ddn) (Singh et al. 2008;Cellitti et al. 2012). An initial hydride transfer step from the F420H2 cofactor leads to their decomposition into des-nitro products and releases reactive nitrogen species that elicit a bactericidal mode-of-action linked to respiratory poisoning and inhibition of mycolic acid synthesis (Singh et al. 2008;Cellitti et al. 2012;Manjunatha et al. 2009).
Because pretomanid and delamanid are prodrugs that require activation, mutations that knockout the activity of Ddn or the biosynthesis or reduction of the enzyme's cofactor (F420), could confer resistance. However, the fitness cost of such knockouts may be considerable given that F420 has been shown to be conditionally essential to the survival of M. tuberculosis, being used by at least 28 different enzymes (Greening et al. 2016) and playing important roles in hypoxic survival, protection against oxidative and nitrosative damage, and evasion of the host immune system (Purwantini & Mukhopadhyay 2009;Hasan et al. 2010;Gurumurthy et al. 2013). Ddn is highly conserved across almost all species of mycobacteria (except Mycobacterium leprae), suggesting its physiological role is under strong evolutionary selection (Ahmed et al. 2015). It has been hypothesised that Ddn serves as an F420H2-dependent menaquinone reductase given its membrane localisation and catalytic activity with the synthetic quinone analogue menadione (Gurumurthy et al. 2013;de Souza et al. 2011), although further work is required to fully define its physiological role.
Despite the recent introduction of nitroimidazoles, cases of acquired clinical resistance, i.e. resistance that occurs during the long treatment of TB infection but is not necessarily transmissible, have already been reported (Hoffmann, Kohl, et al. 2016;Bloemberg et al. 2015).
Acquired resistance to pretomanid and delamanid can occur through genetic changes that cause loss of function within the biosynthetic pathway for F420 production or in the F420-dependent glucose 6-phosphate dehydrogenase (FGD) that catalyzes F420 reduction to F420H2 (Choi et al. 2001;Manjunatha et al. 2006). Laboratory studies have also shown that genetic changes that abolish Ddn activity can also confer resistance (Hoffmann, Borroni, et al. 2016;Haver et al. 2015;Manjunatha et al. 2006). While such mutations could compromise treatments for already infected individuals, transmission of M. tuberculosis to healthy individuals after these genetic changes has never been documented. In order to spread effectively and endanger health, these resistant strains would need to retain sufficient fitness to survive all stages of the lifecycle of M. tuberculosis, including recovery from dormancy. Interestingly, delamanid-resistant isolates with mutations in ddn have been recovered from MDR-TB patients who never received delamanid or pretomanid (Fujiwara et al. 2018;Schena et al. 2016), raising important questions about the fitness costs associated with such mutations and their potential impact on transmission.
In this study, we analysed Ddn orthologs from related mycobacteria to identify natural sequence variations that make mycobacteria resistant to pretomanid, as well as analysing the genomes of ~15,000 M. tuberculosis isolates to identify the spectrum of naturally occurring nonsynonymous Ddn polymorphisms. Mutations were then made to Ddn at positions identified in orthologs and through analysis of non-synonymous polymorphisms to analyse their effect on the native activity (quinone reduction) and pretomanid activation. Altogether, 75 mutants, at 47 unique positions within the 151 amino acid Ddn protein were made. This analysis identified a number of mutations that prevent pretomanid activation without full loss of the native menaquinone reductase activity. Analysis of complete and partial Ddn knock-outs in demonstrated that it is essential for resuscitation from dormancy, i.e. genetic variants that have lost their pretomanid-activation function, but retained their native activity appear to be sufficiently fit to spread and cause disease in new patients. We examined a hypervirulent strain of M. tuberculosis from Vietnam that contains one such mutation (despite never being exposed to pretomanid), which is resistant to pretomanid. Curiously, sensitivity to delamanid activation was not affected in this strain, nor was delamanid activation by the Ddn variant in vitro, which is consistent with our structural analysis that suggests it binds to the active site of Ddn in an alternative orientation.

Ddn mutants are virulent in mice but show defective recovery from hypoxic stress in vitro.
To understand the fitness costs of loss of Ddn activity through mutation, we investigated the ability of M. tuberculosis strains with mutations in Ddn to survive in stress conditions. Isogenic pretomanid-resistant M. tuberculosis mutants selected by pretomanid monotherapy in infected mice and shown to harbour mutations in Ddn (M1T, L49P, L64P, R112W, C149Y and from insertion of the IS6110 transposable element at D108) (Rifat et al. 2018) were used. We also analysed two mutants from a previous study in which resistant strains were identified from an in vitro selection experiment (S22L, W88R). We tested the ability of these eight mutants to reduce native (menadione) and drug (pretomanid and delamanid) substrates in vitro, finding that all mutations resulted in loss of detectable pretomanid activation, consistent with their selection in resistant strains. We observed that the M1T mutant (loss of start codon) and the mutants harbouring the IS6110 insertions did not produce any functional protein and therefore had no detectable native, nor prodrug-activating activity. In contrast, many of the point mutants, such as the L64P mutant retained a small amount of activity with menadione, suggesting that it might retain a fraction of its native function (Table 1).
Because many of these mutants were found in M. tuberculosis living in mouse lungs, the fitness cost of these mutations is clearly low enough to allow them to survive in that environment. We therefore tested whether these mutants were attenuated for multiplication and survival in mice.
These mutants showed no difference in growth or survival compared to wild type after lowdose aerosol infection of mice (Fig. 1a), suggesting that these Ddn mutants would be transmissible. A previous study showed that M. tuberculosis mutants that were unable to biosynthesise F420 have a survival defect when recovering from hypoxia (Gurumurthy et al. 2013). We tested the Ddn mutants for their ability to survive hypoxia and resume growth upon transition to normoxia. The mutants showed no difference in survival under hypoxic conditions compared to wild type. However, they did show a significant defect in recovery from hypoxia ( Fig. 1b) revealing the fitness cost of an inactive Ddn and providing direct evidence for the hypothesis of Gurumurthy et al. that Ddn has an important role in protection against oxidative stress during recovery from hypoxia (Gurumurthy et al. 2013). It is noteworthy that the L64P mutant, which retained a small amount of native activity (Table 1), was able to partially recover (unlike the M1T and Ins6110 mutants). Our data suggest that despite the apparent lack of fitness cost associated with loss of Ddn activity under normal and hypoxic growth conditions, dormant M. tuberculosis with an inactive Ddn are at a disadvantage when attempting to resuscitate, perhaps because they are more susceptible to oxidative stress or have some defect in anaerobic respiration.
The physiological role of Ddn. To better understand how Ddn might contribute to recovery from dormancy we analysed its effects on aerobic respiration. A role for Ddn as a quinone reductase was previously suggested based on activity with the synthetic quinone analogue menadione (Gurumurthy et al. 2013). Here, we show that purified Ddn catalyzes the F420H2dependent reduction of menaquinone-1, which is known to accept electrons from a variety of electron donors and transfer them to terminal oxidases or reductases in mycobacterial respiration . Ddn catalyzed menaquinone reduction in vitro with moderate efficiency (kcat/KM = 8.6 x 10 2 M -1 s -1 ) and physiologically relevant affinity (KM = 22.4 ± 3.8 µM) ( Fig. 2A; Table 2); the conversion rate is likely to be higher in the native environment of the M. tuberculosis cell where Ddn and menaquinone are co-localised at the cell membrane (Sinha et al. 2005). Ddn orthologs encoded by Mycobacterium smegmatis (MSMEG_2027, MSMEG_5998) were also able to reduce menaquinone, suggesting that Ddn orthologs have similar physiological roles across the genus (Table 2). Indeed, Ddn and its orthologs are highly conserved and abundant throughout mycobacteria (Ahmed et al. 2015).
We then investigated whether the menaquinone reductase activity of Ddn was coupled to the mycobacterial respiratory chain by comparing the rates of respiratory oxygen consumption of the model organism M. smegmatis in the presence and absence of a complementation vector expressing ddn. We observed that the addition of glucose 6-phosphate (G6P), F420, and FGD, which are required to catalyze the reduction of F420 to F420H2 for use by F420H2-dependent enzymes such as Ddn (Oyugi et al. 2016), resulted in a 1.4-fold (P= <0.05) increase in oxygen consumption by mycobacterial membranes when the membranes are activated by NADH ( Fig   2B). This was dependent on the presence of cytochrome bd oxidase, which utilizes reduced menaquinone (i.e. menaquinol) as an electron source (Fig. 2B). Thus, reduced F420 increases the rate of menaquinone-dependent oxygen consumption. The rate of NADH oxidation was not affected by the addition of F420H2, which implies endogenous NADH-dependent oxidases do not contribute to the increased oxygen consumption (Fig. 2C). This suggest that mycobacteria can couple F420H2 oxidation to O2 reduction through the respiratory chain via cytochrome bd oxidase, and provides the first evidence that bacteria can use F420H2 as a respiratory electron donor. We also observed that the extent of oxygen consumption was significantly higher in membranes purified from the ddn expressing strain, relative to empty vector controls (Fig. 2B), which is consistent with Ddn acting as a menaquinone reductase in the respiratory chain, with the remaining stimulation attributable to background activity by native Ddn orthologs of M.

smegmatis.
Natural sequence variation in Ddn can result in loss of pretomanid activation with retention of the native activity. Having established that Ddn is a menaquinone reductase that contributes to aerobic respiration, and that loss of this activity severely restricts resuscitation from dormancy (an essential aspect of the life cycle of M. tuberculosis), we then investigated the effects of natural sequence variation within Ddn orthologs from mycobacteria on the native, and drug-activating, activities. We expressed and purified Ddn orthologs from M. tuberculosis, M. marinum, M. smegmatis, M. vanbaalenii, M. avium and M. ulcerans ( Having demonstrated that sequence polymorphisms in the active site of Ddn and its orthologs can result in loss of pretomanid reduction activity in vitro, we investigated whether this corresponded to differences in nitroimidazole susceptibility in vivo. M. tuberculosis H37Rv and M. marinum, which were the only two strains that encoded Ddn orthologs with in vitro pretomanid activation activity, were found to be susceptible to pretomanid and delamanid treatment ( Table 3). In contrast, species in which the Ddn orthologs did not exhibit pretomanid activation activity (M. smegmatis, M. ulcerans, M. avium) have been shown to be naturally resistant to pretomanid (Stover et al. 2000;Ji et al. 2006;Upton et al. 2015). This analysis shows that the prodrug-activating activity of Ddn from M. tuberculosis H37Rv and M. marinum must result from sequence differences to the other Ddn orthologs tested.
Genome sequences of M. tuberculosis were then searched to identify nonsynonymous sequence polymorphisms within the ddn gene. We found that around 1.5% (219/14,876) of presumptive M. tuberculosis genomes screened encoded a non-synonymous mutation in ddn, including several that have arisen independently in unrelated strains (Table S1). Altogether, we identified 46 non-synonymous substitutions and 2 deletions in ddn, distributed throughout the M. tuberculosis phylogeny (Table 4).
All 46 mutants found in our genomic study were expressed, purified, and assayed with menadione, pretomanid, and delamanid (Table 1, Fig. 3 and 4). Every mutant tested was able to reduce menadione, indicating that there was selective pressure to maintain this physiological function of Ddn. The majority of the mutants could also reduce/activate pretomanid and delamanid. This is not unexpected as these M. tuberculosis strains have not been exposed to either drug and therefore, have had no selective pressure to develop resistance. However, of the 46 mutants, several mutants did not activate pretomanid (L49P, S78Y, K79Q, W88R, and Y133C) indicating that any strain of M. tuberculosis with these mutations would be unable to activate the drug. Indeed, an M. tuberculosis strain haboring the W88R mutation has been shown to be resistant to pretomanid in vitro (Haver et al. 2015), and the L49P mutant was obtained from an in vivo resistance selection experiment (Rifat et al. 2018). It is notable that the S78Y and Y133C mutants retained the ability to activate delamanid.
The S78Y polymorphism is found in the genome of N0008 (Comas et al. 2013), a clinical isolate of the hypervirulent Beijing family (Hoffmann, Kohl, et al. 2016), and in two other genomes (SRA accessions:ERR718320 and ERR751847) that are phylogenetically closely related, indicating a shared evolutionary history and suggesting that there was no substantial selective pressure to eliminate this mutation, i.e. its fitness cost must have been relatively small (Fig. 5). We obtained the hypervirulent Beijing strain N0008 to investigate whether the S78Y mutation in the ddn gene, which results in loss of pretomanid reduction in vitro (Table 4), corresponded to resistance to pretomanid in vivo. No other genetic changes previously observed to cause nitroimidazole resistance were apparent in the genome of N0008. We observed a 64fold increase in minimum inhibitory concentration (MIC) of pretomanid against N0008 (256 µg mL -1 ) compared to H37Rv (4 µg mL -1 ). This suggests SNPs, such as L49P, S78Y and W88R, in the ddn gene can confer resistance to pretomanid and confirms that transmissible pretomanidresistant populations of M. tuberculosis already exist.

The potential for spontaneous pretomanid resistance mutations to arise in M. tuberculosis.
Previous studies have used laboratory evolution or engineering to investigate the potential for pathogens to evolve resistance to antibiotics (Hart et al. 2016;Orencia et al. 2001). We took a similar approach here, using structure-guided mutagenesis to investigate the robustness of the nitroreductase activity of Ddn to mutations. The binding site of Ddn has been defined over several studies (Cellitti et al. 2012;Ahmed et al. 2015;Mohamed et al. 2016), identifying a number of polar amino acids within the substrate binding site that contribute to activity, particularly Y65, S78, Y130, Y133, Y136. Three of these binding site residues are fully conserved among the Ddn orthologs tested here, whereas Y65 and Y133 are more variable (Fig.   6). We made, expressed and purified 26 mutants (including the Y65S, S78Y, and Y133C sequence differences observed some of the other Ddn orthologs) of these five key substrate binding residues (Y65, S78, Y130, Y133, Y136) ( Fig. 3), to test how spontaneous mutations at these positions affect nitroimidazole activation. Across the 26 mutants tested, all retained significant levels of the native activity, while 16 did not display detectable pretomanid activation (Table 1, Fig 7). In other words, while the native activity was not greatly affected by sequence variation, the promiscuous nitroreductase activity was extremely sensitive to mutation.
To estimate the rate at which mutations could spontaneously arise at one of the positions identified in this work, we used our genomic survey of ddn in M. tuberculosis to estimate the level of sequence variation in ddn. This revealed that virtually every distinct lineage of M. tuberculosis contains SNPs in ddn; given the burden of TB in the world and the potential for widespread pretomanid administration, alongside the apparent sensitivity of the promiscuous pretomanid activation activity to such SNPs in contrast to the native activity, it appears that spontaneous resistance enabling mutations of ddn could readily occur and spread with sufficient selection pressure.
The molecular basis of resistance. The effects of the mutations shown in Table 1 are generally consistent with our mechanistic understanding of Ddn from mutagenesis and computational simulation (Mohamed et al. 2016;Cellitti et al. 2012). S78 is thought to interact with the nitromoiety of pretomanid and to stabilize the transition state; none of the S78 mutants retained pretomanid nitroreductase activity, suggesting this interaction is particularly important. Y65, Y130, Y133 and Y136 are known to form a hydrophobic wall in the binding site, which can move during the catalytic cycle to shield the active site from solvent and thereby facilitate pretomanid reduction (Mohamed et al. 2016). This is in keeping with the observation that tyrosine to phenylalanine mutations at positions 65, 133 and 136 were essentially neutral, whereas substitution with other residues (Met, Leu, Cys, Trp, Thr, Ser, Glu) led to loss of pretomanid activation. Menadione reduction by Ddn is less susceptible to loss of activity through mutation, which is consistent with work showing native activities are substantially more robust to mutation than promiscuous activities (such as pretomanid activation) (Aharoni et al. 2005), as well as the observation that menadione is more chemically labile.
Delamanid activation is less susceptible to resistance mutations. Of the 75 mutants we made and tested, 25 did not reduce pretomanid at detectable levels but only 10 lost the ability to reduce delamanid. Thus, although pretomanid and delamanid are superficially similar, they must interact with Ddn differently. These enzymatic data extend to whole cell activity, as we observed that the N0008 strain harbouring the S78Y mutation was resistant to pretomanid but remained susceptible to delamanid. For this analysis, we have focused on the wild-type protein, and the S78Y mutant to better understand the molecular basis for the differential effects of mutations on pretomanid and delamanid activation. We used molecular docking to obtain lowenergy poses of these substrates in the crystal structure of Ddn. The binding of delamanid is predicted to be different than that of pretomanid (Fig. 8C,D), with the dual methyl and phenoxy-methyl substituents on the oxazole ring preventing delamanid binding above the deazaflavin ring of F420 in a ring-stacked orientation, as is seen in pretomanid, which has an oxazine ring with a single substituent in the analogous position. This results in a change in the angle at which the nitroimidazole group interacts with the cofactor that results in an increase in the distance to S78Y. Thus, there appears to be a molecular basis for the differing effects of the S78Y mutation on pretomanid and delamanid activity in vitro and in vivo.

Discussion
The fitness trade-off between resistance and native function. Given the conservation of Ddn-like genes throughout mycobacteria, its activity with menaquinone during aerobic respiration, and the inability of Ddn knock-outs to recover from hypoxia-induced dormancy, mutations that completely knock-out Ddn activity (including the loss of F420 biosynthesis through mutations to F420 biosynthetic genes, loss of F420 reductase activity through knockout of FGD, and introduction of stop codons or large genetic insertions/deletions in ddn) will result in substantial loss of fitness given that the ability to recover from dormancy is such an important aspect of M. tuberculosis pathogenesis (Diacon et al. 2009;Hards et al. 2015;Cook, Greening, et al. 2014;Lamprecht et al. 2016). Thus, for nitroimidazole resistance to spread and endanger health, the activity of Ddn must be either retained or otherwise compensated for.
The activation of the prodrugs pretomanid and delamanid by Ddn is a promiscuous activity that is not coupled to its native activity. It has long been known that promiscuous activities are more susceptible to mutation than native functions (Aharoni et al. 2005), meaning that it is possible that mutations could cause loss of prodrug activation without significant loss of native function.
In this study we tested this assumption for Ddn, mutating ~1/3 of the amino acid positions in M. tuberculosis Ddn, including all known naturally occurring polymorphisms, showing that many such mutations (including several already present in clinical isolates) can cause loss of delamanid and/or pretomanid activation without loss of the native activity. Extending our enzymatic measurements to measurement of MICs revealed that related species with sequence differences in ddn, and clinical isolates of M. tuberculosis that harbor mutations within the active site, are resistant to pretomanid. These results have important implications for the clinical usage of nitroimidazoles in TB treatment in order to optimize treatment outcomes and to prevent or slow the development of resistance. For instance, pretomanid will not be effective in patients infected with naturally occurring M. tuberculosis variants harbouring an S78Y mutation and indiscriminant use of pretomanid against such variants (such as in regions in which the N0008 strain is endemic) could drive selective amplification and spread of pretomanid resistance. Moreover, the chances of spontaneous mutation of Ddn in currently sensitive strains is significant, given our observation of the number of single nucleotide mutations that can knock out nitroreductase activity. Given that the clinical isolate N0008 is highly transmissible, it is likely that other Ddn variants in which nitroreductase activity can be abolished without substantial loss of the native activity will also be infectious owing to the minimal loss of fitness these mutations cause with the native activity. Thus, fitness-neutral Ddn mutations are likely to be a prominent route through which clinical pretomanid and delamanid resistance spreads.
Our findings have broad implications for the continued clinical development and usage of nitroimidazole antitubercular agents. Through multiple ongoing phase II/III clinical trials, the TB Alliance and others are evaluating novel combination therapies including pretomanid. The ongoing ZeNix and TB-PRACTECAL trials are extending the study of the BPaL (bedaquiline, pretomanid, linezolid) regimen that showed highly promising results against MDR-and XDR-TB in the Nix-TB trial, while the SimpliciTB trial studies the BPaMZ regimen (bedaquiline, pretomanid, moxifloxacin, pyrazinamide) against MDR-and drug-susceptible TB. Meanwhile, comparable delamanid-containing regimens are being studied in the endTB and MDR-END trials. If delamanid and/or pretomanid are formally approved for clinical use, intensive monitoring of ddn polymorphisms is recommended to ensure informed regimen selection and allow interventions that will reduce spreading of transmissible resistance. Our findings indicate that delamanid binds to Ddn in a different conformation than pretomanid does, suggesting that it could be effective against some pretomanid-resistant isolates. It is also possible that combination therapy with both nitroimidazoles could help prevent the evolution and spread of resistance (since the simultaneous loss of both activities will be less likely to result from single amino acid substitution). Further studies on how delamanid and pretomanid are activated in the M. tuberculosis cell will inform the development of improved nitroimidazole therapies and testing of a broad range of nitroimidazole analogs against a panel of Ddn variants could help to identify compounds for which resistance is less likely to evolve. and cloned into the expression vector pMAL-c2X using Gibson assembly (Gibson et al. 2009).

Plasmid
All mutations to Ddn were made by site-directed mutagenesis using Gibson assembly (Gibson et al. 2009). Construction of MSMEG_2027 and FGD has been described previously (Lapalikar et al. 2012;Bashiri et al. 2008 Protein expression and purification. MSMEG_2027 was expressed and purified as previously described (Ahmed et al. 2015). Ddn, Ddn mutants, and Ddn orthologs were transformed into E. coli BL21 (DE3) cells and grown on LB agar containing 100 µg/ml ampicillin. Single colonies were picked and inoculated in LB media with 100 µg/ml ampicillin.
Starter cultures were grown overnight and diluted 1/100 and grown at 37 °C until OD = 0.4.
Cultures were induced with IPTG to a final concentration of 0.3 mM, and grown for 3 h at 25 °C. Cells were harvested by centrifugation at 8,500 × g for 20 minutes at 4 °C and resuspended in lysis buffer (20 mM Tris-Cl, pH 7.5, 200 mM NaCl) and lysed by sonication using an Omni Sonicator Ruptor 400 (2 x 6 min. at 50% power). The soluble extract was obtained by centrifugation at 13,500 × g for 1 h at 4 °C. The protein was purified using amylose resin (NEB) using the provided protocol. Briefly the lysate was passed over the amylose resin and washed with 12 column volumes of lysis buffer. The protein was eluted using elution buffer (same as lysis buffer but with 10 mM maltose). Samples were frozen at -80 °C in 20 mM tris pH 7.5, 200 mM NaCl, 10 mM maltose, and 10% glycerol.

Measurement of oxygen consumption.
To determine the rate of oxygen consumption, assays NADH oxidation assay. The rate of NADH oxidation was determined using purified M.

Computational analysis.
Sequences of Ddn and orthologs were obtained from the NCBI sequence database. Alignment of the sequences were performed using MUSCLE (Edgar 2008) via the EMBL-EBI web services (Li et al. 2015 (Table S2). These genomes were excluded from all further analysis.

Phylogenetic reconstruction of M. tuberculosis strains with sequencing polymorphisms in
Ddn. To determine the phylogenetic distribution of ddn alleles in M. tuberculosis we carried out a phylogenomic analysis of M. tuberculosis strains with sequence polymorphism in ddn.
Firstly, unassembled genomes from the SRA were assembled de novo using SPAdes version 3.9.0 (Bankevich et al. 2012). Next, all 322 complete and draft genomes (Table S3)  Aliquots were removed for quantitative culture after 7 and 14 days of normoxia.