Tuberculosis is still a leading cause of death in developing countries, for which there is an urgent need for new pharmacological agents. The synthesis of the novel antimycobacterial drug class of benzothiazinones (BTZs) and the identification of their cellular target as DprE1 (Rv3790), a component of the decaprenylphosphoryl-β-d-ribose 2′-epimerase complex, have been reported recently. Here, we describe the identification and characterization of a novel resistance mechanism to BTZ in Mycobacterium smegmatis. The overexpression of the nitroreductase NfnB leads to the inactivation of the drug by reduction of a critical nitro-group to an amino-group. The direct involvement of NfnB in the inactivation of the lead compound BTZ043 was demonstrated by enzymology, microbiological assays and gene knockout experiments. We also report the crystal structure of NfnB in complex with the essential cofactor flavin mononucleotide, and show that a common amino acid stretch between NfnB and DprE1 is likely to be essential for the interaction with BTZ. We performed docking analysis of NfnB-BTZ in order to understand their interaction and the mechanism of nitroreduction. Although Mycobacterium tuberculosis seems to lack nitroreductases able to inactivate these drugs, our findings are valuable for the design of new BTZ molecules, which may be more effective in vivo.
Tuberculosis (TB) is still a leading cause of death in developing countries, especially those of sub-Saharan Africa where TB is epidemic, partly because of the increased susceptibility caused by the widespread HIV prevalence in those populations (Maartens and Wilkinson, 2007). Moreover, the spread of multidrug-resistant (MDR-TB) and extensively drug-resistant (XDR-TB) strains of Mycobacterium tuberculosis has become a major clinical and public health problem (Matteelli et al., 2007). The last 40 years have seen no new class of anti-TB drugs and, among the small number of compounds with antitubercular activity designed, only a few candidates, such as PA-824 and diarylquinolines, are currently undergoing clinical trials (Riccardi et al., 2009).
We have recently discovered a novel and extremely effective antimycobacterial drug class, namely benzothiazinones (BTZs), that shows highly bactericidal activity in vitro, ex vivo and in the chronic-TB mouse model (Makarov et al., 2009). The best available candidate among this class is the compound BTZ043 (Fig. 1), characterized by a nitro-group essential for its antimycobacterial activity (Makarov et al., 2009); its target has been demonstrated to be the enzyme decaprenylphosphoryl-β-d-ribose 2′-epimerase, particularly its DprE1 component encoded by the dprE1/Rv3790 gene (Makarov et al., 2009). This enzyme acts in concert with DprE2 (coded by the neighbouring dprE2/Rv3791 gene) to convert decaprenylphosphoryl ribose (DPR) into decaprenylphosphoryl arabinose (DPA), a substrate for arabinosyltransferases in the synthesis of the mycobacterial cell wall arabinogalactan (AG) and lipoarabinomannan polysaccharides (Mikusova et al., 2005; Wolucka, 2008). Arabinogalactan is a fundamental component of the mycobacterial cell wall, which covalently binds the outer layer of mycolic acids to peptidoglycan (Wolucka, 2008). Indeed, both the dprE1 and dprE2 genes were predicted to be essential by Himar1-based transposon mutagenesis in M. tuberculosis H37Rv (Sassetti et al., 2003), thus validating both enzymes as targets for drug development.
The search for possible resistance mechanisms to BTZs showed that spontaneous missense mutations of Cys387 in DprE1, as well as the overexpression of the wild-type protein, confer high levels of resistance to BTZ043, in different mycobacterial species. We also demonstrated that BTZ043 can block the epimerization reaction from radiolabelled DPR into DPA in vitro using membrane preparations from either Mycobacterium smegmatis or Mycobacterium bovis BCG, but not from the corresponding mutant strains (Makarov et al., 2009).
In the course of preliminary structure activity relationship studies (SAR), we previously showed that the single stereoisomers BTZ043 (S) and BTZ044 (R), both carrying a nitro-group, were the most active molecules against mycobacteria, while the corresponding amino-derivatives BTZ045S (S) and BTZ045R (R) were significantly less active both in M. smegmatis and in M. tuberculosis (Fig. 1) (Makarov et al., 2009). This finding raised the question of whether mycobacterial nitroreductases (NRs) could be involved in inactivating the drug by reducing the nitro-forms to the corresponding amino-forms. NR encoding genes are in fact widespread among bacterial genomes, but are also found in eukaryotes. The common presence of NRs in bacteria may be ascribed to the fact that nitroaromatic compounds are often toxic and mutagenic, and several microorganisms have developed enzymatic pathways in order to degrade and detoxify such compounds (Roldán et al., 2008). Bacterial NRs belong to the Nitro_FMN_reductase superfamily and catalyse the reduction of nitrocompounds using NAD(P)H as electron donor in an obligatory two-electron transfer, utilizing non covalently bound FMN as cofactor. Members of this enzymatic family include NAD(P)H:FMN oxidoreductase, oxygen-insensitive nitroreductase, flavin reductase P, dihydropteridine reductase, NADH oxidase and NADH dehydrogenase (Roldán et al., 2008). The prototype prokaryotic nitroreductase is NfnB/NfsB from Escherichia coli; the ability of this enzyme to reduce the nitro- to the corresponding amino-group has led to its use in studies of anticancer chemotherapy, by activating pro-drugs that contain a nitro-group in antitumor agents (Parkinson et al., 2000).
In this work, by microbiological, genetic, biochemical and structural approaches, we describe the identification and characterization of NfnB from M. smegmatis, an NR able to specifically reduce the nitro-compound, BTZ043, to the amino-derivative, BTZ045 (Fig. 1). To date, no orthologous or paralogous NR, able to perform this conversion, has been found in M. tuberculosis, consistent with the very low minimal inhibitory concentrations (MICs) of the drug against sensitive- and MDR-M. tuberculosis strains (Makarov et al., 2009; Pasca et al., 2010). However, the identification of a mycobacterial NR able to inactivate the BTZ drugs could be key to understanding the mechanism of action of this class of molecules, and help to improve the future clinical use of these current compounds.
Overexpression of nfnB confers resistance to BTZ043 in M. smegmatis
In our previous work we reported the isolation of spontaneous mutants of M. smegmatis-resistant to BTZs (Makarov et al., 2009). The majority of them had missense mutations in the same codon, encoding Cys387 in Rv3790 (DprE1), the cellular target of BTZs (Makarov et al., 2009). In the present study, we focused our attention on the first of a different series of M. smegmatis-resistant mutants, named GM22, which showed a lower resistance level (16X-MIC) to BTZ043 (Table S1).
In contrast to the aforementioned mutants, this strain harboured neither mutations in MSMEG_6382 nor in MSMEG_6385, the genes orthologous to M. tuberculosis dprE1 and dprE2 respectively. To identify the gene(s) responsible for the BTZ resistance phenotype, GM22 was transformed with a wild-type M. smegmatis cosmid library, and a selection for sensitivity to BTZ043 was carried out. A transformant colony characterized by a resistance decrease from 16X-MIC to 2X-MIC was isolated, and the cosmid, REP3, responsible for restoration of BTZ043 sensitivity, was isolated and partially sequenced; the cloned genomic fragment extended from nucleotides 6554151 to 6580426 (according to the M. smegmatis genome numbering from TIGR/JCVI). To confirm that this fragment was indeed responsible for the BTZ-sensitive phenotype, the GM22/REP3 strain was cured of the cosmid and the cured strain shown to display the original level of resistance to BTZ043, thus confirming the causal role of REP3.
Subcloning experiments of the cosmid REP3 were then performed and this led to the identification of a 4456 bp KpnI–SphI fragment. This genomic fragment contained, among others, the MSMEG_6503 gene, coding for a putative transcriptional regulator from the TetR family, and the MSMEG_6505 gene, which was annotated as nfnB due to its 35% sequence identity to the E. coli NfnB/NfsB nitroreductase (Fig. S1). Both genes were amplified from the M. smegmatis GM22 mutant and sequenced. The sequence analysis revealed no mutations in the nfnB gene of GM22 mutant, but a point mutation in MSMEG_6503. This mutation led to the substitution of Leu137 by Pro, leading to the hypothesis that the GM22 resistance could be due to a defective repressor (MSMEG_6503), possibly causing overexpression of NfnB and, consequently, the reduction of the BTZ nitro-molecule to its less active amino-derivative. Vice versa, the restored BTZ043 sensitivity in the GM22 complemented mutant could be due to the presence of several copies of the wild-type allele of MSMEG_6503 from the REP3 cosmid. Consistent with this hypothesis, the MIC of BTZ045, the amino derivative of BTZ043, for M. smegmatis was already known to be 0.5 µg ml−1 versus 4–8 ng ml−1 of the original nitro compound (Makarov et al., 2009).
Moreover, 14 other spontaneous mutants, showing different levels of resistance (8–32X-MIC), were found to carry mutations in the MSMEG_6503 gene, including deletions (AL43, AL45, AL47, AL49, AL50, AL51, AL53, AL54, AL57 mutants) and insertions (AL46 mutant) (Fig. S2, Table S1). All the mutant strains were predicted to produce a truncated form of the protein. One of these mutants, AL55, presents a mutation in the hypothetical repressor binding site (Fig. S2).
To validate this hypothesis, the wild-type MSMEG_6503 gene from the REP3 cosmid was cloned into the constitutive expression vector pSODIT-2. When the recombinant plasmid was transformed into M. smegmatis GM22, the wild-type sensitivity to BTZ043 was restored (data not shown). This result suggested that the overexpression of nfnB could represent a specific mechanism of resistance to BTZ043 in M. smegmatis.
The MSMEG_6503 regulator binds two conserved palindromic sites in the MSMEG_6503-nfnB intergenic region
To analyse the transcription of nfnB and MSMEG_6503 genes, the 5′ end of the two transcripts was mapped by 5′-RACE experiments (Fig. 2). The sequence upstream of the transcriptional start sites revealed the presence of hypothetical −10 and −35 promoter regions (boxed in Fig. 2) for both genes. Two conserved sequences palindromic with respect to one another were observed in the 90bp MSMEG_6503-nfnB intergenic region (highlighted in grey in Fig. 2). The first one overlapped the MSMEG_6503 promoter region, while the second was localized just downstream of the nfnB transcriptional start site. To verify the role of MSMEG_6503 in nfnB regulation, MSMEG_6503 proteins from M. smegmatis mc2155 (Fig. S3) and GM22 were purified and used in DNA binding assays. A fragment of 190bp, containing the MSMEG_6503-nfnB intergenic region was obtained by PCR with the primers nfnBinF and nfnBinR (Table S2) and used as a probe. As shown in Fig. 3, MSMEG_6503 from M. smegmatis mc2155 was able to bind and efficiently retard this region (Fig. 3A), while the mutant protein bound less efficiently, as the same effect was only observed with a 10-fold excess of protein (Fig. 3B). The specificity of the protein-DNA interaction was confirmed by the ability of cold competitor DNA to reverse the mobility shift (Fig. 3C). Moreover, to better define the MSMEG_6503 binding sequence, Box7 and Box8 oligos, corresponding to the MSMEG_6503 upstream palindrome, and Box9 and Box10, corresponding to the nfnB upstream palindrome, were labelled and used as probes in binding assays. The data clearly demonstrated that the purified wild-type protein was able to efficiently retard these regions (Fig. 3D), strongly suggesting their role as operator sites for the MSMEG_6503 regulator.
Overexpression of nfnB confers resistance to BTZ043 in different mycobacterial species
In order to verify whether BTZ043 resistance in M. smegmatis GM22 mutant was associated with overexpression of NfnB, the nfnB gene was cloned into pSODIT-2, and both wild-type M. smegmatis and M. tuberculosis H37Rv strains were transformed with the recombinant construct and the empty pSODIT-2 vector as control. As shown in Table 1, the M. smegmatis strains overproducing NfnB showed a 16-fold increase of the resistance level to BTZ043, comparable with the resistance observed for the GM22. Also in M. tuberculosis, overexpression of M. smegmatis NfnB conferred an eightfold increase in resistance to BTZ043.
Table 1. nfnB overexpression confers BTZ043 resistance both in M. smegmatis and M. tuberculosis.
BTZ043 MIC (ng ml−1)
Real-time PCR (RT-PCR) was then performed on RNA extracted from wild-type M. smegmatis and some mutant strains in the absence of the drug (Table 2). A 390 (± 56)-fold increase in transcription of nfnB in GM22 with respect to the wild-type parent was observed. As shown in Table 2, higher levels of nfnB expression were detected in AL49 (characterized by a 11 bp deletion in the MSMEG_6503 coding region) and in AL55 (which has a mutation in the MSMEG_6503 binding site). A similar nfnB expression level was found in M. smegmatis strain transformed with pSODIT-2/nfnB (Table 2). Moreover, overexpression of the repressor encoding gene MSMEG_6503 was detected in all the mutants analysed (Table 2). In the case of GM22 and AL49 the protein produced is mutated and with a reduced affinity for its recognition site; in mutant AL55 the repressor is in the wild-type form but there is a mutation in the binding site.
Table 2. nfnB and MSMEG_6503 expression levels by RT-PCR in different M. smegmatis strains.
M. smegmatis strains
n.d., not determined.
390 ± 56
3.01 ± 0.1
788 ± 42
30 ± 3
410 ± 28
29 ± 5
296 ± 30
The results unambiguously showed that nfnB was strongly overexpressed in GM22, AL49 and AL55, indicating that the mutated MSMEG_6503 repressor did not efficiently repress the transcription of nfnB, due to its lower affinity for the operator region (Fig. 2). The presence of BTZ043 did not significantly affect the expression of nfnB in the wild-type strain, as cultures grown with the drug revealed about twofold repression compared with cultures grown in its absence (data not shown).
Deletion of nfnB decreases resistance of GM22 mutant to BTZ043
To further confirm the direct role of NfnB in M. smegmatis GM22 resistance to BTZ043, an in-frame unmarked deletion was created in the nfnB gene through a two-step homologous recombination (Parish and Stoker, 2000). As shown in Table 3, the ΔnfnB strain of GM22 has a restored wild-type phenotype, corroborating the function of NfnB in BTZ043 resistance. On the contrary, no MIC decrease was observed in the ΔnfnB strain of mc2155 compared with the wild-type strain. This can be explained on the one hand by the remarkable potency of BTZ043 and its very low MIC value and, on the other, by the wild-type regulator MSMEG_6503 efficiently inhibiting nfnB expression. The MIC assessment was conducted in parallel with rifampicin as a control drug, and all the strains tested showed the same susceptibility (Table 3).
Table 3. BTZ043 MIC variation in M. smegmatis wild-type, GM22 mutant and nfnB deleted strains.
MIC in M. smegmatis
BTZ043 (ng ml−1)
Rifampicin (µg ml−1)
GM22 mutant converts BTZ043 to BTZ045 faster than wild-type M. smegmatis
Both wild-type M. smegmatis and GM22, overexpressing nfnB, were evaluated for their ability to convert the nitro-compound (BTZ043) to the amino-derivative (BTZ045), by high-pressure liquid chromatography (HPLC) analysis of culture media. GM22 transforms the nitro- to the amino-compound more efficiently and more rapidly compared with M. smegmatis (Table 4a, Fig. S4). Indeed, the mutant strain was able to reduce 70% of BTZ043 into BTZ045 after 3 h growth. Similar results were also achieved with wild-type M. smegmatis transformed with nfnB/pSODIT-2 recombinant plasmid (data not shown), thus confirming our hypothesis. In order to verify if BTZ addition was able to inhibit cell growth in the wild-type strain with respect to the resistant mutant, cell counting was performed. A small comparable reduction in cell viability was observed in GM22 and wild-type strains (data not shown). These findings are consistent with a role for NfnB in reduction of the active BTZ043 to the inactive BTZ045.
Table 4a. BTZ043 to BTZ045 transformation in M. smegmatis cultures.
1 h growth
3 h growth
BTZ043 increases NAD(P)H oxidase activity of NfnB
NfnB from M. smegmatis is a 25.5 kDa enzyme which, as mentioned above, was annotated as a putative FMN-dependent nitroreductase on the basis of its sequence similarity with other bacterial nitroreductases, including E. coli NfnB/NfsB. A ClustalW analysis was thus performed between the two primary sequences. The two enzymes show significant similarity at the N-terminal end (Fig. S1), specifically between Arg25 and Arg72 (NfnB numbering) that define the hypothetical FMN-binding region.
The M. smegmatis enzyme was successfully expressed as a His6-tagged form in E. coli, with final yields reaching 100 mg of purified protein from 20 g of wet cell biomass (Fig. S5). The purified enzyme was yellow in all purification steps, suggesting the presence of a flavin coenzyme, most likely in the oxidized state; this was confirmed by UV-Vis spectroscopy (data not shown). A kinetic analysis was then carried out using E. coli NfsB as an internal control. The optimal pH determined for NfnB activity ranged from 7.5 to 9.0; NfnB and NfsB had comparable NAD(P)H-oxidase activities, indicating that NfnB does not have a strong preference for either NADH or NADPH as the electron donor (Table 5). Moreover, addition of micromolar amounts of BTZ043 to the assay mixture led to a significant increase in the rate of NADPH oxidation. In contrast, when five other potential substrates, chosen among the reported substrates of other bacterial nitroreductases (namely quinone menadione, nitrofurazone, methyl 4-nitrobenzoate, 4-nitrobenzene methyl sulfonate, and 4-nitroacetophenone) (Zenno et al., 1996) were tested, no detectable effects on the NADPH-oxidase activity of NfnB were observed (data not shown). These results suggested that NfnB has a relative specificity for BTZ molecules, although the physiological substrate(s) remain to be identified.
Table 5. Kinetic parameters for NfnB from M. smegmatis versus NfsB from E. coli.
10.75 ± 0.21
4.20 ± 0.14
8.15 ± 0.21
4.32 ± 0.15
6.68 ± 1.38
13.08 ± 1.47
9.9 ± 1.68
14.5 ± 0.49
BTZ043 is a substrate of NfnB
In order to assess the activity of the purified NfnB protein towards BTZ043, an indirect evaluation was initially performed, using an assay in which BTZ043 blocks the epimerization of DPR to DPA (Makarov et al., 2009). Recombinant NfnB enzyme was thus added to the assay mixtures to monitor DPA synthesis from its precursor 5-phosphoribose diphosphate (pRpp) in M. smegmatis membranes, to evaluate whether its presence had any detectable effect on the inhibition of the epimerization of DPR to DPA by BTZ043. In the absence of BTZ043, DPA production was detected (Fig. 4, lane 1) and its synthesis was not affected by the addition of the pure NfnB protein (Fig. 4, lane 2); in presence of BTZ043, DPA synthesis was inhibited at the stage of DPR (Fig. 4, lane 3), suggesting that the drug was inhibiting M. smegmatis DprE1 (MSMEG_6382). When BTZ043 was pretreated with purified NfnB prior to addition to the reaction mixture, DPA was still formed (Fig. 4, lane 4), most likely due to the conversion of the active drug (BTZ043) to its inactive amino form. In order to evaluate this case we examined the composition of the pre-incubated BTZ043-NfnB reaction mixture by LC-MS. We found out that the peak area corresponding to BTZ043 (m/z 432) in this sample decreased to 46% of the total peak area (Fig. S6, Panel A) from 97% found in the controls without NADH (Fig. S6B), or without NfnB (Fig. S6C). The expected product of the reaction, BTZ045 (m/z 402), presented 17% of the total peak area, while the intermediates of the reaction, namely the nitroso (m/z 416) and the hydroxylamine forms (m/z 418), comprised 7% and 30% of the total peak area respectively (Fig. S6A). The reduction of BTZ043 by NfnB most likely proceeds in the final reaction mixture, since a significant reduction of the BTZ043 effect on DPR epimerization in presence of added NfnB was observed (Fig. 4, lane 4). These data confirmed that NfnB is able to reduce and inactivate BTZ043.
In order to verify the scope of NfnB and its possible involvement in the metabolism of other drugs, we investigated whether NfnB could also reduce the nitro group of PA-824. PA-824 is a bicyclic nitroimidazole that has potent activity against M. tuberculosis and is a promising candidate for TB treatment, currently undergoing clinical trials. Using the resazurin reduction assay, we confirmed that M. smegmatis is naturally resistant to PA-824 as reported previously (Stover et al., 2000). Consequently, NfnB could be responsible for the lack of activity of this compound against M. smegmatis.
Using the LC-MS method, we verified that PA-824 is indeed reduced by NfnB, first to its hydroxylamine analogue, which subsequently decomposes to a metabolite lacking the imidazole ring, as described by Singh and colleagues (2008). Contrary to what was observed in M. tuberculosis, the des-nitro metabolite (major metabolite of PA-824 formed in TB with release of toxic nitrous acid) was not detected in the NfnB-mediated reduction of PA-824, in agreement with our hypothesis. However, the M. smegmatis nfnB-knockout mutant was equally resistant to PA-824 as the wild-type strain (data not shown), pointing to additional mechanisms underlying the natural resistance of M. smegmatis to this drug.
M. tuberculosis may lack NRs or NR-like enzymes able to inactivate BTZ drugs
The involvement of NfnB in BTZ-resistance in M. smegmatis naturally raises the question of whether one or more NR could be involved in BTZ resistance in M. tuberculosis. Although no orthologue of NfnB is present in the M. tuberculosis genome ( http://tuberculist.epfl.ch/), nine hypothetical nitroreductases are predicted: Acg, Rv0306, MoeY, Rv2337c, FbiB, Rv3131, Rv3127, Rv3368c (Purkayastha et al., 2002) plus Rv3547, a nitroimidazo-oxazine specific nitroreductase, essential for the activation of the pro-drug PA-824 (Manjunatha et al., 2006; Singh et al., 2008). Furthermore, three other proteins homologous to Rv3547 can be identified, i.e. Rv1558, Rv1261c and Rv3178 (Singh et al., 2008), in addition to the lipoamide dehydrogenase LpdC (Rv0462), whose homologue in M. smegmatis has been shown to have some activity on non physiological substrates, including nitroheterocyclic compounds (Marcinkeviciene and Blanchard, 1997).
The genes acg, Rv0306, moeY, fbiB, Rv3131, Rv3127, Rv3368c, Rv3547, Rv1558, Rv1261c, Rv3178 and lpdC were cloned into pSODIT-2 expression vector, while Rv2337c gene was cloned with its own promoter region into pMD31 vector. All the recombinant clones were transformed into M. smegmatis and M. tuberculosis, and the MIC was evaluated both aerobically and anaerobically, as it is known that some bacterial NRs, including NfsB, need anaerobic conditions to reduce nitroaromatic compounds (Linwu et al., 2009). None of the different M. tuberculosis NR had any effect on the MIC of BTZ043 in both species, irrespective of the conditions tested (Table S3). Control experiments revealed expression of all the hypothetical NRs (data not shown). In contrast, overexpression of NfnB conferred resistance to BTZ043 in both species but only under aerobic conditions (Table 1 and Table S3), indicating that NfnB is an oxygen-dependent NR as far as the reduction of BTZ043 is concerned. Overall, these data, as well as direct assessment by HPLC of the transformation of BTZ043 to BTZ045 in cultures of M. tuberculosis (Table 4b, Fig. S7), suggest that M. tuberculosis most probably lacks enzymes able to inactivate BTZ043 either in aerobic or anaerobic environment, consistent with the low MIC values of BTZ043 in M. tuberculosis and the fact that all the BTZ-resistant mutants isolated so far in this species harboured mutations in the target gene dprE1.
Table 4b. BTZ043 to BTZ045 transformation in M. tuberculosis cultures.
6 h growth
18 h growth
72 h growth
Crystal structure of NfnB and docking prediction of the NfnB-BTZ043 complex
The crystal structure of NfnB was determined at 1.75 Å resolution to final R = 15.1%, Rfree = 17.4% (Table 6). The enzyme adopts an homodimeric fold that does not deviate significantly from already known FMN-dependent nitroreductases, i.e. the prototype NfsB from E. coli (Fig. 5A); the closest structural homologue according to both SSM (Krissinel and Henrick, 2004) and DALI (Holm et al., 2008) is an uncharacterized FMN-nitroreductase from Bartonella henselae (pdb entry 3GR3), with an overall RMSD between Cα atoms of 1.4 Å. The enzyme core is composed of a central five-stranded β-sheet, made by four antiparallel β-strands plus a fifth one (parallel to the first) contributed by the C-terminal residues of the opposite monomer; the β-sheet is flanked by two helices on the internal side, involved in dimerization, and three helices on the external one.
Table 6. Crystallographic data collection and refinement statistics.
Values in parentheses refer to the outer (highest resolution) shell.
; Rfree calculated for 5% of the reflections, not used at any stage of structural refinement.
The active site is located at the interface of the dimer, and residues from both monomers indeed contribute to the non-covalent binding of FMN (Fig. 5B). As seen in other nitroreductases, a number of hydrogen bonds involving the isoalloxazine ring make it deviate from planarity. This distortion, known as ‘butterfly bending’, results in a ∼165° angle between the pyrimidine and the dimethylbenzene rings and is consistent with an oxidized state of the FMN prosthetic group, as observed, among others, in E. coli NfsB (Lovering et al., 2001; Race et al., 2005) and Enterobacter cloacae NR (Haynes et al., 2002). On the opposite side (re face) of the flavin, some tetragonal-shaped residual density was modelled as a phosphate ion from the crystallization buffer (Fig. 5B). The proximity of the phosphate to the reactive portion of FMN suggests that this ion could be mimicking the position of the nitro-group of the substrate, an hypothesis supported by the consistent superimposition of this phosphate ion to an acetate ion observed in the crystal structures of both the E. coli (Race et al., 2005) and E. cloacae enzymes (Haynes et al., 2002).
The structure of NfnB in complex with NADPH was obtained by the prior addition of the cofactor during crystallization, and was refined in the same crystallographic space group at 1.80 Å resolution to final R = 14.9%, Rfree = 17.3% (Table 6; Fig. 5C). The purine ring of the adenosine moiety stacks between the side-chains of Tyr113 and Arg120, while the 2′-phosphate group points externally making only a salt bridge with Arg105, which may explain why NfnB does not show significant specificity for either NADPH or NADH. Moreover, the nicotinamide moiety, which stacks parallel to the side-chain of Phe185, does not make electrostatic contacts and thus appears mobile; it is noteworthy that the amide group points outwards from the active site, most likely because this is still occupied by a phosphate ion in front of FMN (Fig. 5C). To confirm whether the observed binding mode of NADPH could be otherwise compatible with the hydride transfer to FMN, a docking calculation was made for NADPH restraining the co-ordinates of the adenosine moiety to the experimental ones. The solution with the lowest energy showed indeed the nicotinamide entering the active site and stacking approximately parallel to the isoalloxazine ring (Fig. 5C), suggesting that the observed NADPH binding mode is compatible with the formation of a productive complex.
All attempts to determine the crystal structure of NfnB in complex with any of the available BTZ derivatives, either by soaking or cocrystallization, were unsuccessful. Computational docking was therefore used to model the interaction between the enzyme and the lead compound BTZ043. As an assessment of the docking protocol, several structures of FMN-dependent nitroreductases complexed to small molecules were inspected, and six structures of E. coli NfsB with different ligands were retrieved for further analysis (pdb entries 1ICR, 1ICU, 1IDT, 1OO6, 1OON, 1OOQ, 1YKI) (Johansson et al., 2003; Race et al., 2005). The inspection of these structures suggested a common substrate binding mode: with the exceptions of the smallest molecules nicotinic acid and nitrofurazone, the other compounds were all positioned with their nitrobenzene moiety stacking parallel to the FMN isoalloxazine ring and the nitro-group pointing at ∼ 3.5 Å to the reactive N5 of FMN. The position of the oxygen atoms of the nitro-group of these molecules coincides with that of two oxygens from the phosphate ion in front of FMN in NfnB (Fig. 5B). In order to maintain the BTZ043 molecule in an orientation compatible with the reaction, we therefore decided to restrain its nitro-group, during our calculations, in the position occupied by the phosphate ion in the experimental structure. To validate this assumption, the 5-(aziridin-1-yl)-2,4-dinitrobenzamide ligand in pdb entry 1IDT was redocked into the NfsB apo structure and the resulting complex indeed reproduced the deposited structure (data not shown).
The analysis of solvent-accessible surfaces in NfnB showed a larger substrate binding pocket for one of the monomers (∼ 216 Å3 for the pocket defined by chain B in pdb entry 2WZV). The symmetric pocket formed by the opposite monomer is smaller due to the different conformation of the main chain for residues Tyr92-Gln98, packed closer to the core of the enzyme. Since this loop appears as mobile in both chains, and its conformation differs due to distinct crystal packing constraints, the largest binding site was chosen as the pocket for docking. The best prediction of the NfnB-BTZ043 complex, in terms of free energy, is illustrated in Fig. 5D. As expected, the predicted orientation of the BTZ molecule places the nitrobenzene moiety into a parallel stacking with the isoalloxazine ring, where the nitro-group is expected to be kept in place by the interaction with Asn54 and Ser55. It is worth noting that this conformation, in addition to be consistent with the reaction, shows a remarkable surface complementarity of the substrate binding pocket of NfnB with the remainder of the BTZ molecule (Fig. 5D).
There has been increased interest in bacterial NRs in the last few years due to their potential environmental and clinical applications, most notably for their usefulness in cancer therapy for prodrug activation (Roldán et al., 2008). In this study, we investigated the role of a nitroreductase in the metabolism of the antitubercular drug candidate, BTZ043, which targets the enzyme DprE1, essential for the biosynthesis of AGs and thus for the integrity of the mycobacterial cell wall (Makarov et al., 2009). In particular, we demonstrated that the overexpression of the FMN-dependent nitroreductase NfnB in M. smegmatis, driven by spontaneous mutations in the transcriptional repressor controlling its expression, is a novel BTZ resistance mechanism, involving drug inactivation, besides the already documented amino acid substitution on Cys387 of DprE1 (Makarov et al., 2009). Moreover, a recent report has shown that compounds from the dinitrobenzamide class also target DprE1 and that the same amino acid substitution in the target (C387G) confers dinitrobenzamide resistance (Christophe et al., 2009). As a nitroaromatic group is essential for the antimycobacterial activity of these compounds, similarly to BTZs (Christophe et al., 2009), the possible development of resistance due to the overexpression of a nitroreductase is an issue that should be considered.
The direct role of NfnB in BTZ inactivation has been confirmed both genetically, as the deletion of nfnB in the GM22 mutant restores a wild-type phenotype, as well as biochemically, since BTZ043 was unambiguously shown to be a substrate of NfnB in cell free enzymatic assays. In order to further investigate the specific nitroreductase activity of NfnB, to better understand the metabolic fate of BTZ043 both in mycobacteria and in the host, and, eventually, to design new molecules resistant to inactivation, we carried out a biochemical and structural study of the enzyme.
Although the structure of NfnB shows the overall conservation of the enzyme fold with respect to the already known FMN-dependent nitroreductases (more than 60 homologous structures are available in the protein data bank), an interesting observation comes from the comparison of the primary sequences of NfnB and DprE1. An amino acid sequence alignment shows a common amino acid stretch of 30 amino acids between NfnB (residues 86–115) and DprE1 (residues 386–417), displaying 46% sequence identity (59% similarity) (Fig. 5E). It is noteworthy that this amino acid stretch is located at the C-terminal end of DprE1, the same region in which spontaneous mutations conferring resistance to BTZ were identified (Makarov et al., 2009), suggesting that this polypeptide portion might also play a key role in defining the relative specificity of NfnB towards the nitromolecules from the BTZ class. This portion of NfnB makes a long loop followed by an α-helix folding over the FMN site, where it contributes to the geometry of the active site and hence may be crucial in defining the specificity of the enzyme towards the substrate(s).
This hypothesis is supported by the predicted NfnB-BTZ043 complex, which indeed shows that several residues from this patch (namely Tyr92 and Phe97) are expected to interact with the drug, forming a pocket that binds the distal part of the molecule (Fig. 5D); the same region of the protein is also involved in binding NADPH, as proven experimentally (Fig. 5C). Furthermore, this region structurally deviates from the homologues (such as NfsB) more than the core of the protein and displays higher mobility, possibly related to flexibility in accommodating the substrate(s), as reflected by the average B factors in both the free enzyme and the complex with NADPH (averaged B for residues 92–98 from the two models is 37.3 Å2 for chain A and 58.0 Å2 for chain B, versus 22.2 Å2 for the whole protein). This may, at least in part, explain the relative specificity of NfnB towards the BTZ molecules, as well as the failure to detect activity with canonical nitroreductase substrates. Moreover, the predicted binding mode of BTZ043 suggests that bigger substitutions at the distal site from the reactive nitro-group might render BTZs resistant to inactivation, even though (for the same reasons) they could lower the affinity for DprE1 or even abolish the interaction with the target. A careful evaluation of the effectiveness of any new BTZ compound, both in vitro and in vivo, may clarify this issue.
We also investigated whether M. tuberculosis possesses any nitroreductases able to carry out the same drug inactivation process. Although no NfnB homologue is present in M. tuberculosis genome, a total of 13 putative nitroreductases could be identified; none of these confers resistance to BTZ043 when overexpressed. These findings are consistent with the lower MIC in M. tuberculosis (1 ng ml−1) compared with M. smegmatis (4 ng ml−1), although other factors such as differences between the two species in the membrane permeability or in the affinity of DprE1 for the drug (due to the active site architecture) may contribute to explain the different MIC values. However, it has also been observed that the nitro-BTZ compounds are transformed into the corresponding amino-derivatives not only in M. smegmatis cultures overexpressing NfnB, but also in blood and urine from treated mice (V. Makarov, unpubl. data), strongly suggesting that one or more nitroreductases, either mammalian or from the intestinal microbial flora, could carry out such a conversion. Indeed, it is well known that nitroaromatic compounds can be converted into their metabolites in the intestine, by the action of several microbial nitroreductases (Roldán et al., 2008). For instance, nitroreductase activity on 3-nitrobenzanthrone has been observed in several mouse tissues, mainly liver, lung, colon and bladder (Chen et al., 2008). In some cases microbial and mammalian NRs can compete for the same substrate: as an example, the anti-cancer pro-drug 5-(Aziridin-1-yl)-2,4-dinitrobenzamide (CB 1954) currently under clinical trials, can be activated by E. coli NfsB as well as in the liver, where cytochrome P450 reductase and NAD(P)H quinone oxidoreductase are involved, generating adverse hepatic effects (Tang et al., 2007). Interestingly, the activity of NfsB towards dinitrobenzamide pro-drugs raises the question of whether NfnB, or other mycobacterial NR, could reduce the new dinitrobenzamide compounds (Christophe et al., 2009) with antimycobacterial activity to the corresponding amino-derivatives, that, similarly to the BTZs derivatives, have been shown to be totally inactive (data not shown). Knowledge of the NfnB structure and the proposed complex with BTZ will help the rational design of second generation BTZ or dinitrobenzamides that are less susceptible to nitroreduction, thereby improving the half-life of the compounds and increasing their in vivo efficacy.
Standard microbiological and molecular biological procedures are described in the supporting information.
High pressure liquid chromatography analysis of BTZ conversion
Mycobacterium smegmatis mc2155 and GM22 mutant strains were cultivated, as previously described, in 100 ml of complete 7H9 broth and grown for 24 h. Then, 5 mg of BTZ043 was added to each 100 ml sample, and cultivated at 37°C for 1 or 3 h. After this period of time, each culture was quickly chilled, centrifuged at 3000 g and filtered through a 2 µm size filter.
Mycobacterium tuberculosis H37Rv strain was cultivated, as previously described, in 100 ml of complete 7H9 broth supplemented with ADC and grown for 72 h. Then, 5 mg of BTZ043 was added to each 100 ml sample, and cultivated at 37°C for 6, 18 or 72 h. After this period of time, each culture was quickly chilled, centrifuged at 3000 g and filtered through a 2 µm size filter.
In all cases the medium was diluted by 100 ml of water and extracted by chloroform for 3 times. Chloroform was evaporated in vacuum and residue was dissolved in 0.5 ml of acetonitrile. This solution was analysed by HPLC, to determine the concentrations of BTZ043 and BTZ045 compounds. Medium without drug was extracted in parallel, and used as control. Each experiment was performed 3 times. The data shown are the mean value obtained.
The HPLC conditions were as follow: HPLC system –‘Varian’ with pumps ProStar 210, autosampler ProStar 410, UV detector ProStar 325 (wave 240 nm). Column ‘Varian’, Polaris C8-A, 3µ, 150 × 2.1 mm. The eluent is 0.1% formic acid in acetonitrile: 0.1% formic acid in water (50:50). Thermostat 20.0°C. Column flow 0.900 ml min−1. Volume of injected probe (injector Rheodyne, loop) – 20 µl. Retention time of BTZ043 is 5.65 min and amino metabolite BTZ045 is 4.33 min.
Expression and purification of recombinant NfnB
In order to express a His-tagged fusion protein (His6·NfnB), nfnB gene was PCR amplified from REP3 cosmid by using the 28nfnF and 28nfnR primers (Table S2). The 704 bp PCR product was cloned into pET-28a vector. The recombinant expression plasmid nfnB/pET-28a was used to transform E. coli BL21(DE3) and cells were plated onto LB agar containing kanamycin. The inoculum, grown at 37°C with kanamycin, was diluted 1:100 in 1 l of the same medium. The culture was incubated at 37°C until cells grew to OD600 = 0.5; then protein expression was induced with 1 mM isopropyl-β-thiogalactopyranoside (IPTG) for 2 h at 37°C. E. coli cells were resuspended in lysis buffer (50 mM NaH2PO4, 1.5 M NaCl, 5 mM β-mercaptoethanol and 10 mM imidazole, pH 7.9), and the following reagents were added to the lysate: 1 mg ml−1 lysozyme, 5 µg ml−1 DNase and 10 µg ml−1 RNase; the mixture was incubated on ice for 30 min and then sonicated. The cytosolic fraction was added to a Ni-nitrilotriacetic acid (Ni–NTA) resin (Qiagen), pre-equilibrated with lysis buffer in the absence of β-mercaptoethanol. The protein was eluted with 50 mM NaH2PO4, 1.5 M NaCl, 5 mM β-mercaptoethanol, 10% glycerol and 200 mM imidazole, pH 7.9. The fractions containing the NfnB protein, detected by SDS-PAGE, were collected and dialysed overnight at 4°C against 1 l of dialysis buffer (50 mM MOPS, 5 mM β-mercaptoethanol, 10 mM MgCl2 and 10% glycerol, pH 7.9), and used for enzymatic assay. Purified NfnB was pooled and stored at −20°C.
Liquid chromatography – mass spectrometry (LC-MS) evaluation of BTZ conversion
The pretreatment reaction mixtures in the final volume of 50 µl described in the Supporting information were stopped by the same volume of acetonitrile and centrifuged at 14 000 g 10 min at room temperature. Ten microlitres of the sample was analysed by LC-MS method using LC-IT-TOF MS analyser (Shimadzu, Kyoto, Japan) with electrospray ionization (ESI) in the positive and negative mode. Ascentis C18 column (100 mm × 2.1 mm; 5 µm particles) was used for the separation. Elution of the compounds was performed by two-component gradient (H2O:acetonitrile) (0–1 min – 10% acetonitrile, 5 min – 90% acetonitrile, 6 min – 90% acetonitrile, 6.01–12 min – 10% acetonitrile) with the flow rate of 0.2 ml min−1. Identity of the individual peaks was confirmed by MS-MS3 spectra.
Expression and purification of recombinant selenomethionyl substituted NfnB
In order to express His6·NfnB fused to the TEV-protease recognition site, the nfnB gene was PCR amplified from REP3 by using the TEVnfnBf and TEVnfnBr oligonucleotides; the forward primer contained a sequence coding for the TEV-protease recognition site ENLYFQG (Table S2). The recombinant expression plasmid nfnB_TEV/pET-28a was used to transform E. coli B834(DE3). All transformants were used to inoculate LB containing kanamycin, and grown overnight at 30°C. Pre-inoculum was diluted 1:50 in M9 supplemented broth. The culture was incubated at 30°C until cells grew to OD600 = 1.0; then protein expression was induced with 1 mM IPTG for 24 h at 15°C and harvested. E. coli cells were resuspended in lysis buffer (50 mM NaH2PO4, 0.5 M NaCl, 5% glycerol and 25 mM imidazole, pH 8.0), supplemented with Complete protease inhibitor cocktail (Roche); the mixture was then sonicated. The lysate was centrifuged at 26 800 g for 1 h, filtered (0.45 µm) and loaded onto a 1 ml His Trap Ni2+ ± IMAC column (GE Healthcare) equilibrated in lysis buffer; the recombinant protein was purified by a linear imidazole gradient (25–400 mM). The fractions containing the SeMet–NfnB protein were checked by 15% SDS-PAGE, collected, and dialysed overnight at 4°C against 1 l of 50 mM Tris-HCl, 150 mM NaCl, 5% glycerol, 1 mM dithiothreitol pH 8.0, in the presence of His6-tagged TEV protease (van den Berg et al., 2006), at a ratio of 1:30 (w/w). SeMet–NfnB was separated both from His-tag and the protease, using Ni–NTA affinity chromatography by gravity flow. The untagged SeMet–NfnB was successively purified by size-exclusion chromatography on a Superdex 200 26/60 column (GE Healthcare) equilibrated in 50 mM Tris-HCl, 150 mM NaCl, 5% glycerol, pH 8.0. Fractions corresponding to the SeMet–NfnB peak were concentrated up to 37 mg ml−1 with a 10 kDa cut-off Vivaspin concentrator (Sartorius). The protein concentration was determined by Bradford assay, and the purity estimated about 95%, by SDS-PAGE and Coomassie blue staining. The SeMet–NfnB solution was pooled, flash-frozen in liquid nitrogen, and stored at −80°C.
The activity of M. smegmatis NfnB was determined with an assay mixture containing NAD(P)H ranging from 5 to 100 µM, in 1.0 ml 50 mM Tris-HCl, 150 mM NaCl, 5% glycerol, pH 8.0. The reaction was started by addition of 3 µM purified NfnB, and the activity was determined at 25°C from the linear decrease in adsorbance at 340 nm, following NAD(P)H oxidation. As internal control the activity of E. coli nitroreductase (Sigma-Aldrich) was determined as well.
Crystals of NfnB were obtained from a stock solution (35 mg ml−1) of selenomethionyl protein via sitting-drop vapour diffusion at 18°C, by mixing equal volumes (1 µl) of enzyme solution, supplemented with 1 mM FMN, with 20% (w/v) PEG8000, 6% (v/v) isopropanol, 200 mM NH4H2PO4. To obtain crystals of the protein in complex with the NADPH cofactor, 5.8 mM NADPH was added to the enzyme prior to crystallization. Rod-shaped crystals appeared after 3 days and reached their maximum dimension over 2–3 weeks. Prior to data collection, crystals were flash-cooled in liquid nitrogen after a soaking into a cryoprotectant solution made of 75% mother liquor and 25% glycerol.
Data collection, structure solution and refinement
X-ray diffraction data were collected either at the Proxima-1 beamline at the Soleil Synchrotron (Saint-Aubin, France) or the microfocus ID23-2 beamline at the ESRF (Grenoble, France) from single crystals at 100 K. A 2.5 Å resolution dataset was collected from a single selenomethionyl crystal at the X-ray energy corresponding to the peak of the selenium K-edge (12.664 keV, λ = 0.9790 Å), as measured by a fluorescence scan. Treatment of diffraction images was carried out with either XDS (Kabsch, 1993) or MOSFLM from the CCP4 suite (Collaborative Computational Project, Number 4, 1994), followed by data reduction with SCALA (Collaborative Computational Project, Number 4, 1994) (Table 6). A total of 12 selenium sites were located with the program SHELXD (Sheldrick, 2008), and single-wavelength anomalous diffraction (SAD) phasing and density modification were carried out with AutoSHARP (Vonrhein et al., 2007). The resulting solvent-flipped electron density map showed clear secondary structure features and allowed the manual tracing of the polypeptide chain with Coot (Emsley and Cowtan, 2004). The model was initially refined with phenix.refine (Adams et al., 2010), then with BUSTER (Blanc et al., 2004) against 1.75 Å resolution data collected from an isomorphous selenomethionyl crystal. Diffraction data of the isomorphic NADPH complex were collected at 1.85 Å resolution and the structure refined starting from the co-ordinates of the native enzyme. TLS groups were defined with the TLSMD server ( http://skuld.bmsc.washington.edu/~tlsmd) (Painter and Merritt, 2006). Both models were validated through the Molprobity server (http://molprobity.biochem.duke.edu) (Davis et al., 2007). Final refinement parameters are shown in Table 6. Figures were generated and rendered with Pymol (http://www.pymol.org); electrostatic surfaces were rendered with the APBS plugin within Pymol, supplying atomic charges and radii calculated through the ‘pdb2pqr’ server (http://www.poissonboltzmann.org/pdb2pqr/) (Dolinsky et al., 2004) applying the CHARMM force field. Atomic co-ordinates and structure factors have been deposited in the protein data bank, with accession codes 2WZV (highest resolution) and 2WZW (NADPH complex).
The docking procedure for both NADPH and the BTZ043 molecules was entirely carried out with the Icm software (Totrov and Abagyan, 1997). Prior to any calculation, all solvent molecules were removed, hydrogens added and the hydrogen bond network optimized. Exploration of the NfnB surface was carried out with IcmPocketFinder to identify suitable pockets for docking (An et al., 2004). Residues closer than 2.5 Å to the binding pocket were considered to mark the boundary of the 5 grid potentials required by Icm to describe interactions of the flexible ligand with the receptor. Solutions were ranked according to their predicted binding energy, and the predicted complex (lowest free energy) was minimized with Szybki (http://www.eyesopen.com), considering the Poisson-Bolzmann electrostatic potential, solvation terms and the flexibility of both the ligand and the active site residues of the enzyme.
The expression plasmid for recombinant TEV protease was kindly provided by H. Berglund (Karolinska Institute, Stockholm). We are grateful to P. Weber (PF6, Institut Pasteur) for performing robot-driven crystallization trials, P. Schneider for technical assistance, and Dr J. McKinney for plasmid pJG1100. We thank ESRF and Soleil for granting access to synchrotron radiation facilities and their staff for the onsite assistance. João Neres is the recipient of a Marie Curie fellowship from the European Commission. LC-MS analyses were performed with the support of Slovak Research and Development Agency No. VVCE-0070-07 and grant of Slovak Grant Agency no. 1/0546/10. This work was funded by EC-VI Framework Contract No. LSHP-CT-2005-018923.