Potency boost of a Mycobacterium tuberculosis dihydrofolate reductase inhibitor by multienzyme F420H2-dependent reduction.

Triaza-coumarin (TA-C) is a Mycobacterium tuberculosis (Mtb) dihydrofolate reductase (DHFR) inhibitor with an IC50 (half maximal inhibitory concentration) of ∼1 µM against the enzyme. Despite this moderate target inhibition, TA-C shows exquisite antimycobacterial activity (MIC50, concentration inhibiting growth by 50% = 10 to 20 nM). Here, we investigated the mechanism underlying this potency disconnect. To confirm that TA-C targets DHFR and investigate its unusual potency pattern, we focused on resistance mechanisms. In Mtb, resistance to DHFR inhibitors is frequently associated with mutations in thymidylate synthase thyA, which sensitizes Mtb to DHFR inhibition, rather than in DHFR itself. We observed thyA mutations, consistent with TA-C interfering with the folate pathway. A second resistance mechanism involved biosynthesis of the redox coenzyme F420 Thus, we hypothesized that TA-C may be metabolized by Mtb F420-dependent oxidoreductases (FDORs). By chemically blocking the putative site of FDOR-mediated reduction in TA-C, we reproduced the F420-dependent resistance phenotype, suggesting that F420H2-dependent reduction is required for TA-C to exert its potent antibacterial activity. Indeed, chemically synthesized TA-C-Acid, the putative product of TA-C reduction, displayed a 100-fold lower IC50 against DHFR. Screening seven recombinant Mtb FDORs revealed that at least two of these enzymes reduce TA-C. This redundancy in activation explains why no mutations in the activating enzymes were identified in the resistance screen. Analysis of the reaction products confirmed that FDORs reduce TA-C at the predicted site, yielding TA-C-Acid. This work demonstrates that intrabacterial metabolism converts TA-C, a moderately active "prodrug," into a 100-fold-more-potent DHFR inhibitor, thus explaining the disconnect between enzymatic and whole-cell activity.

Tuberculosis (TB) is a major infectious disease killer globally. It affected 10 million and killed 1.4 million people in 2019 alone (1). The predicted impact of the COVID-19 pandemic is an additional 190,000 TB deaths in 2020, and it is expected in the next 5 y that there will be up to a 20% increase in the global TB disease burden (2), stressing the critical need for new safe and effective drugs against the causative agent, Mycobacterium tuberculosis (Mtb). In addition, controlling multidrug-resistant TB (MDR-TB) presents a huge public health challenge (1).

Dihydrofolate reductase (DHFR) is a ubiquitous enzyme in bacteria, parasites, and humans. The protein catalyzes the NADPH-dependent conversion of dihydrofolate into tetrahydrofolate, a methyl group shuttle required for the synthesis of many cellular building blocks including thymidylate, purines, and certain amino acids. Several DHFR inhibitors are in clinical use for the treatment of various infectious diseases and cancer (3, 4). However, approved DHFR inhibitors have only weak or no activity against Mtb, and there are no DHFR inhibitors used clinically for the treatment of TB (5).

Recently, DHFR was clinically validated as a vulnerable Mtb target. The old TB drug para-aminosalicylic acid (PAS) was long thought to inhibit dihydropteroate synthase (DHPS, FolP), the first enzyme in the folate pathway. However, it was uncovered that PAS, a hydroxy analog of DHPS’s substrate para-aminobenzoic acid, is a substrate of DHPS, which catalyzes the formation of hydroxy-dihydrofolate, which in turn inhibits DHFR (6–10). Thus, PAS is a prodrug that, upon metabolism by an enzyme of the folate pathway, is converted into a DHFR inhibitor.

With Mtb DHFR presenting an attractive target, the enzyme has been the subject of renewed drug discovery efforts (5, 11–13), reviewed recently by He et al. (14). A major hurdle encountered in many DHFR drug discovery programs consists in achieving selectivity toward human DHFR (14). Through ligand-based drug design starting from a Triazaspiro scaffold and a side chain extension approach, we previously identified selective inhibitors of Mtb DHFR (15). The most potent and selective analog resulted from attachment of a coumarin fragment to the triaza scaffold (Triazaspiro-Coumarin [TA-C]; Fig. 1). TA-C inhibited recombinant Mtb DHFR with an IC50 of ∼1 µM and showed an acceptable selectivity index for the human enzyme (IC50 human DHFR/IC50 Mtb DHFR = 18). The selectivity indices for cytotoxicity (CC50 HepG2/MIC50) and hemolysis (LC50 red blood cells/MIC50) were >4,000. Surprisingly, TA-C exhibited excellent whole-cell activity against Mtb and its closely related vaccine strain M. bovis bacillus Calmette–Guérin (bacillus Calmette–Guérin), with a Minimum Inhibitory Concentration50 (MIC50, concentration inhibiting growth by 50%) of 10 to 20 nM. To confirm that TA-C’s whole-cell activity was due to inhibition of DHFR, we overexpressed DHFR in bacillus Calmette–Guérin and showed that TA-C’s MIC increased when DHFR intrabacterial concentration increased (15).

Fig. 1.

Structure of TA-C and tool compounds TA-C-Met, TA-C-Red and TA-C-Acid. (Above) Triazaspiro core and linker common to TA-C and its analogs. (Below) coumarin moiety in TA-C (with α-β unsaturated carbonyl motif indicated in red), methylated coumarin in TA-C-Met, coumarin with reduced α-β double bond in TA-C-Red, and carboxylic acid derivative of coumarin in TA-C-Acid.

The whole-cell activity or MIC50 of TA-C is 50- to 100-fold higher than its IC50 against the isolated Mtb DHFR protein. Here, our aim was to confirm DHFR inhibition as the primary mechanism of action of TA-C and to shed light on the peculiar on-target/whole-cell potency disconnect. Resistance to the prodrug DHFR inhibitor PAS has been extensively studied and recently discussed in comprehensive reviews by Baughn and colleagues (4, 16). PAS resistance can emerge via multiple mechanisms that include preventing efficient bioactivation within the folate synthesis pathway, mitigating the impact of target inhibition, and limiting drug accumulation within the bacilli. A common mechanism, identified by transposon mutagenesis (17), isolation of spontaneous resistant mutants in vitro (18), and genetic works (19), is associated with mutations in thyA, one of the genes encoding thymidylate synthase. Thymidylate synthase is required for the methylene-tetrahydrofolate–dependent conversion of dUMP to dTMP. In most organisms, this reaction is performed by a ThyA-type thymidylate synthase that releases dihydrofolate following catalysis. Dihydrofolate must be reduced by DHFR to re-enter the folate metabolism. Mtb encodes a second, ThyX-type thymidylate synthase. ThyX regenerates tetrahydrofolate instead of dihydrofolate following catalysis. In contrast to ThyX-mediated conversion of dUMP to dTMP, ThyA-mediated conversion results in an increased demand for DHFR activity to provide sufficient levels of tetrahydrofolate and therefore sensitizes the pathway to DHFR inhibition. Loss-of-function mutations in thyA reduce the demand for DHFR activity and are therefore associated with decreased susceptibility to DHFR inhibitors (4, 10, 16, 17, 20). It is interesting to note that missense resistance mutations in the dfrA gene encoding DHFR have not been reported in Mtb. Amino acid alterations that would prevent inhibitor binding to the active site of DHFR are likely deleterious to overall enzymatic function and are thus not tolerated by the bacterium (21). Resistant mutant selection with direct (nonprodrug) Mtb DHFR inhibitors, although less studied, is consistent with the resistance mechanisms observed for PAS. Using the DHFR inhibitor THT1 identified in a chemogenomic approach, Mugumbate et al. measured a spontaneous in vitro resistance frequency of ∼10−8/CFU (colony forming unit) and resistance mutations mapped to thyA (22).

Given that TA-C is a direct DHFR inhibitor, we anticipated a low frequency of resistance largely mapping to thyA. Instead, TA-C–resistant Mtb and bacillus Calmette–Guérin colonies emerged at a high frequency, and resistance was mostly due to loss of activity of enzymes within the biosynthetic pathway for the redox coenzyme F420 (23). We have previously shown that F420 is essential for emergence of Mtb from dormancy (24), and F420-dependent oxidoreductases (FDORs) are known to activate other Mtb prodrugs, such as pretomanid (PA-824) (25). Employing genetics, analogs of TA-C as chemical probes, as well as biochemical analyses, we show that TA-C indeed intercepts DHFR but is in fact converted by bacterial FDORs to a carboxylic acid analog with 100-fold increased on-DHFR–target activity.

Results

High-Frequency TA-C Resistance Is Caused by Loss-of-Function Mutations in Genes Required for Synthesis of Redox Cofactor F420.

To identify TA-C resistance mechanisms, we selected spontaneous resistant mutants in bacillus Calmette–Guérin and Mtb on TA-C–containing agar. Surprisingly, resistant strains emerged at a high frequency of ∼10−6/CFU suggesting a TA-C–specific, nonfolate pathway–related resistance mechanism. MIC determination of 12 randomly selected resistant strains revealed a ∼100-fold shift in MIC from 10 to 20 nM to 2 to 3 µM (Table 1). Consistent with a nonfolate pathway mechanism of resistance, all strains were fully susceptible to PAS. The MIC of isoniazid and rifampicin were not altered, suggesting that resistance to TA-C is specific and not due to a general drug resistance mechanism (Table 1). The 12 strains were subjected to whole-genome sequencing, and polymorphisms were verified by targeted sequencing. All strains harbored loss-of-function mutations in fbiC or fbiA, encoding the FO synthase and the F420-L-glutamate ligase enzymes of the F420 biosynthetic pathway (23), and in fgd1, encoding the F420-dependent glucose-6-phosphate dehydrogenase required for producing the reduced form of F420, F420H2 (26) (Table 1). F420 is a deazaflavin cofactor involved in redox reactions in methanogens and many actinobacteria, including mycobacteria (23). Genetic complementation of representative strains harboring mutations in fbiC, fbiA, or fgd1 restored wild-type TA-C susceptibility, confirming that TA-C resistance was caused by the observed polymorphisms (). Measuring F420 contents in a representative mutant confirmed absence of the cofactor (). Consistent with the loss of F420, the 12 strains were cross-resistant to pretomanid (Table 1), a nitroimidazole prodrug bioactivated by the FDOR, deazaflavin-dependent reductase (Ddn) (27, 28). Thus, strains that are defective in F420 biosynthesis, F420 reduction, or prodrug activation are resistant to pretomanid (27, 29). In contrast to the analysis of resistant mutants selected against pretomanid, which identified Ddn (28), no FDOR was identified in our resistance screening. Together, these results show that loss of F420 causes resistance to TA-C and that F420 is required for TA-C’s potent whole-cell activity but did not identify exactly how F420 is involved in TA-C’s activation.

Table 1.

TA-C–resistant M. bovis bacillus Calmette–Guérin and M. tuberculosis strains emerging at a frequency of 10−6/CFU

	Exp*	Strain	MIC50 (µM)†						Gene with polymorphism	Mutation	Mutation type§
			TA-C	PAS	INH‡	RIF‡ (nM)	PTM‡	TA-C-Met
bacillus Calmette–Guérin		wt	0.02	0.8	0.4	1.6	0.2	3.2	—	—	—
	1	Tacr1.1	3.2	0.8	0.4	2	>100	3.2	fbiC	Δ2269GC	fs
		Tacr1.2	3.2	0.8	0.4	3	>100	3.2	fgd1	Δ472A	fs
		Tacr1.3	3.2	0.8	0.4	2	>100	3.2	fbiC	Δ1142C	fs
	2	Tacr2.1	1.6	0.4	0.3	1.6	>100	1.6	fbiC	Δ1983G	fs
		Tacr2.2	3.2	0.4	0.4	1.6	>100	3.2	fgd1	Ins284C	fs
		Tacr2.3	3.2	0.4	0.4	1.6	>100	3.2	fbiA	T326C, L109P	ms
	3	Tacr3.1	3.2	0.8	0.3	3	>100	3.2	fgd1	Δ932A	fs
		Tacr3.2	3.2	0.8	0.5	1.6	>100	3.2	fbiC	G928T, G310Stop	ns
		Tacr3.3	3.2	0.4	0.4	1.6	>100	3.2	fgd1	C465A, Y155Stop	ns
Mtb		wt	0.01	0.2	0.4	1.6	0.2	1.6	—	—	—
	4	Tacr4.1	1.6	0.2	0.4	1.6	>100	1.6	fbiC	Δ1084C	fs
	5	Tacr5.1	1.6	0.2	0.4	1.6	>100	1.6	fbiC	Δ1976C	fs
	6	Tacr6.1	1.6	0.2	0.4	1.6	>100	1.6	fgd1	T308G, L103R	ms

Exp., independent resistance selection experiments.

MIC experiments were carried out three times independently and mean values are shown.

INH, isoniazid; RIF, rifampicin; PTM, pretomanid.

fs, frameshift; ms, missense; ns, nonsense mutation.

Low-Frequency TA-C Resistance Is Caused by Missense and Nonsense Mutations in thyA.

If TA-C indeed decreases folate synthesis by inhibiting DHFR, we expected TA-C–resistant mutants in folate metabolism genes at a frequency of ∼10−8/CFU (22). Thus, we speculated that out of 100 TA-C–resistant mutants emerging at a frequency of 10−6/CFU, 99 should carry a defective F420 pathway, and one should harbor a mutation in the folate pathway. To test this hypothesis, we carried out a counter screen of a collection of 1,000 TA-C–resistant bacillus Calmette–Guérin strains for susceptibility to pretomanid. As expected, 994 strains were pretomanid resistant, presumably harboring lesions in F420 biosynthesis genes. The six TA-C–resistant and pretomanid-sensitive strains showed a ∼100-fold increase in MIC50 for TA-C (Table 2). Whole-genome sequencing, confirmed by targeted sequencing, revealed that all six strains carried missense or nonsense mutations in thyA (Table 2). Genetic complementation of a representative strain confirmed that a thyA polymorphism caused TA-C resistance (). Furthermore, the thyA mutations conferred cross-resistance to PAS but not to the control drugs isoniazid and rifampicin (Table 2). Together, these results show that thyA mutations cause resistance to TA-C and emerge at a frequency of ∼10−8/CFU. These results suggest that the DHFR inhibitor TA-C exerts its antibacterial activity by interfering with folate metabolism.

Table 2.

TA-C–resistant, pretomanid-sensitive M. bovis bacillus Calmette–Guérin strains emerging at a frequency of 10−8/CFU

Exp*	Strain	MIC50 (µM)†						Gene with polymorphism	Mutation	Mutation type§
		TA-C	PTM‡	PAS	INH‡	RIF‡ (nM)	TA-C-Met
	wt	0.02	0.2	0.8	0.4	1.6	3.2	—	—	—
7	Tacr7.1	3.2	0.2	>100	0.4	1.6	3.2	thyA	C443A, A148E	ms
	Tacr7.2	1.6	0.2	>100	0.4	1.6	3.2	thyA	G398C, W133S	ms
8	Tacr8.1	1.6	0.2	>100	0.4	1.6	3.2	thyA	Ins734TATGAAG	ns
9	Tacr9.1	3.2	0.2	>100	0.4	1.6	3.2	thyA	T373C, S125P	ms
10	Tacr10.1	3.2	0.2	>100	0.4	1.6	3.2	thyA	C439T, H147Y	ms
	Tacr10.2	3.2	0.2	>100	0.4	1.6	3.2	thyA	T292C, W98R	ms

Exp., independent resistance selection experiments.

MIC experiments were carried out three times independently and mean values are shown.

PTM, pretomanid; INH, isoniazid; RIF, rifampicin.

ms, missense; ns, nonsense mutation.

Chemically Blocking the Putative TA-C Site of FDOR Reduction Phenocopies F420 Resistance.

We first investigated whether F420 could be involved in the reduction of TA-C by FDORs, as is the case with nitroimidazole prodrugs, generating a derivative with enhanced antibacterial activity. Revisiting our published structure–activity relationship around the Triazaspiro scaffold, we noted that attachment of the coumarin group to the Triazaspiro core resulted in a moderate increase of biochemical activity and a strong increase of whole-cell activity (15). We speculated that TA-C’s coumarin group may be the site of intrabacterial reduction by FDORs. It has been shown that the α,β-unsaturated carbonyl motif of coumarins can be reduced by enzymes within the FDOR superfamily (30–32). Thus, we hypothesized that Mtb FDORs reduce the coumarin moiety of TA-C, producing an activated derivative with increased whole-cell activity (Fig. 1). If correct, a TA-C analog in which this “reducible” site is blocked by chemical modification should copy the F420-dependent resistance phenotype. To test this hypothesis, we generated the TA-C analog TA-C-Met, in which reducibility of the double bond was blocked by methylation of the β-carbon (Fig. 1) (30–32). TA-C-Met retained an IC50 of ∼1 µM against recombinant Mtb DHFR, similar to TA-C (Table 3). Thus, any altered whole-cell activity of TA-C-Met should be due to the loss of reduction since on-DHFR–target activity was preserved. As predicted, the MIC of TA-C-Met was ∼100-fold higher compared to TA-C (Tables 1 and 3), around 2 to 3 μM, reproducing the TA-C resistance level of F420 mutants (Tables 1 and 3). Thus, the MIC of TA-C-Met was not affected by mutations causing loss of F420 (Tables 1 and 3). Together, these results indicate that TA-C is converted to an active derivative via FDOR-mediated reduction of the α,β-unsaturated carbonyl motif of its coumarin moiety and that F420-associated resistance is due to the loss of TA-C reduction.

Table 3.

The rate of reduction of TA-C, TA-C-Met, and TA-C-Acid by FDORs, their MIC50 against M. bovis bacillus Calmette–Guérin, and IC50 against DHFR

	Compound
Mtb FDOR Kinetics	TA-C	TA-C-Met	TA-C-Acid
Ddn	5.95 × 10−3 ± 2.37 × 10−4	N/D	N/D
kcat (s−1)	3.47 ± 0.51	N/D	N/D
KM (µM)	1.71 × 103	N/A	N/A
kcat/KM (M−1 s−1)
Rv1558	3.09 × 10−3 ± 1.48 × 10−4	N/D	N/D
Kcat (s−1)	10.01 ± 1.6	N/D	N/D
kM (µM)	3.09 × 102	N/A	N/A
kcat/KM (M−1 s−1)
MIC50 against bacillus Calmette–Guérin (µM)
Wt	0.02	3.2	0.8
F420 mutant (Tacr1.1)†	3.2	3.2	0.4
thyA mutant (Tacr7.2)†	1.6	3.2	>25
DHFR over-expressor§	1.6	1.6	>25
DHFR IC50 (µM)
M. tuberculosis	0.93 ± 0.12	0.63 ± 0.07	0.009 ± 0.001
H. sapiens	10.5 ± 1.9	N/A	1.73 ± 0.14

N/D – Below the level of detection.

N/A – Not analyzed.

Experiments were carried out three times independently and mean values are shown.

Refer to Tables 1 and 2.

Described in ref. 15.

The Putative F420-Dependent Reduction Product of TA-C, TA-C-Acid, Is a 10-nM DHFR Inhibitor.

Analysis of the substrate range of FDORs with coumarin substrates has shown that reduction of the α,β-unsaturated carbonyl motif generates an unstable product that undergoes spontaneous ring opening by hydrolysis, which in turn releases a carboxylic acid coumarin derivative (30–32). Thus, we speculated that Mtb FDORs were the most likely enzyme class to carry out the analogous conversion of TA-C via a reduced intermediate (TA-C-Red) to generate TA-C-Acid (Fig. 1) and that TA-C-Acid may display increased activity against DHFR, thus explaining the enzyme/whole-cell potency disconnect. We synthesized TA-C-Acid and measured an IC50 of ∼10 nM against recombinant Mtb DHFR, showing that the putative reduction product of TA-C was 100-fold more potent than its parent against the enzyme (Table 3). Consistent with these biochemical results, structural modeling, in which TA-C and TA-C-Acid were computationally docked into the active site of Mtb DHFR (), suggested that the carboxylic acid moiety of TA-C-Acid increases molecular interactions with the DHFR binding site, thus allowing tighter binding (Fig. 2 ).

Fig. 2.

TA-C docking analyses with DHFR and FDORs and conversion of TA-C by FDORs to TA-C-Acid. (A) TA-C docked into the binding pocket of DHFR (PDB ID: 6NND), with the cation–π interaction between R67 and the coumarin moiety highlighted. (B) TA-C-Acid docked into the binding pocket of DHFR, with the additional network of hydrogen bonds formed by the carboxylic acid moiety of TA-C-Acid and the sidechains of R67, Q68, and R44 highlighted. (C) TA-C bound into the binding pocket of Ddn:F420 (PDB ID: 3R5R), showing its stabilizing interactions with the main chain carbonyl and amide groups of S78 and G80. (D) TA-C bound into the binding pocket of Rv1558 with F420 (PDB ID: 7KL8), showing its stabilizing interactions with the main chain carbonyl and amide groups of S70 and G72. (E) LC/MS analysis of the Ddn and Rv1558 catalyzed conversion of TA-C to TA-C-Acid. No product formation was observed in the no-enzyme control sample.

The MIC50 of TA-C-Acid was 0.8 µM against bacillus Calmette–Guérin, lower than the MIC50 of TA-C-Met but significantly higher than the MIC of TA-C (Table 3). However, this is likely due to decreased intrabacterial uptake caused by the (charged) carboxylic acid moiety (33). TA-C–resistant mutants in thyA and DHFR-overexpressing strains were both cross-resistant to TA-C-Acid, confirming that TA-C-Acid exerts its antibacterial activity by interfering with DHFR (Table 3).

Multiple Mtb F420 Oxidoreductases Reduce TA-C to TA-C-Acid.

Results so far indicated that TA-C is a weak inhibitor of DHFR and is converted intracellularly by FDORs to the highly potent TA-C-Acid. However, the genetic screen for TA-C–resistant mutants revealed only loss-of-function mutations in genes involved in the F420 biosynthetic pathway and not in FDORs (Table 1). Thus, direct evidence for FDOR-catalyzed conversion of TA-C was missing. The Mtb genome encodes at least 28 F420-dependent enzymes (34), 15 of which belong to the FDOR superfamily, including the nitroimidazole reductase responsible for pretomanid activation, Ddn (24). This enzyme family is known to be relatively promiscuous and active with a range of synthetic compounds (31, 32, 35–37). One explanation for the observation that no activating enzyme was detected in the resistance screen is that if multiple enzymes can activate the drug, they would all need to be inactivated to confer resistance (i.e., redundancy in activation can prevent evolution of resistance). If correct, this functional redundancy would explain our failure to isolate FDOR mutants resistant to TA-C.

To test this hypothesis, we screened a panel of seven recombinant FDORs from Mtb, including all members of the A1 subgroup that includes Ddn, as well as representatives of the A2, B1, B3, and B4 subgroups against TA-C using a spectrophotometric assay in which we followed the oxidation of F420H2 (). We identified two enzymes from the A1 subgroup that could efficiently catalyze the reduction of TA-C in vitro, Ddn and Rv1558 (370 ± 14 and 152 ± 4 nmol ⋅ min−1 ⋅ umol−1enzyme, respectively) (). We also used the change in fluorescence of coumarin upon reduction to confirm that this apparent activity was a result of the reduction of the coumarinyl moiety (). We then measured the enzyme kinetics of Ddn and Rv1558 with TA-C revealing kcat/KM values of 1.71 × 103 M−1 ⋅ s−1 and 3.09 × 102 M−1 ⋅ s−1, respectively (Table 3). To examine whether this promiscuous activity of FDORs with TA-C was likely to be common across other mycobacterial species, we also tested five recombinant enzymes from Mycobacterium smegmatis, identifying two (MSMEG_5998 and MSMEG_2027) that are orthologs of Ddn and Rv1558, respectively, from the A1 subgroup that had activity (). Interestingly, the activity with TA-C appears to be limited to enzymes of the A1 subgroup, where Ddn and the 39% identical enzyme Rv1558 displayed significant activity. Despite reasonably high sequence identity (>34%), neither enzyme of the A2 subgroup (Rv1261c and MSMEG_5030) displayed activity. Because we could not obtain soluble, folded, and active protein for several of the FDORs, we cannot exclude the possibility that other enzymes in addition to Ddn and Rv1558 could activate TA-C. However, given the restricted activity to the A1 subgroup, the only plausible candidate is Rv3178, albeit unlikely given that it is less similar to Ddn than the inactive A2 enzymes (29% identity).

To investigate the activation of TA-C by these enzymes in more detail, we solved the structure of Rv1558, in complex with F420, to a resolution of 2.47 Å (). We were unable to capture the ternary complex between substrate, cofactor, and enzyme. As expected, Rv1558 adopts the split B-barrel fold typical of the FDOR superfamily (37) and is almost identical to the structure of Ddn at the level of the overall fold (rmsd C-α = 0.7 Å) (). The N-terminal helix is fully resolved, unlike in the structure of Ddn, in which it is truncated. The active site is likewise very similar, with key catalytic residues such as Ser78, Tyr130, and Tyr136 (Ddn numbering) being conserved (38) and relatively conservative changes at other positions, such as Tyr65(Met) and Tyr130(Trp). TA-C was docked into the active sites of both proteins and adopts a catalytically productive pose (), with the coumarin moiety positioned directly above the deazaflavin ring of F420 (Fig. 2 ). Triplicate 200-ns molecular dynamics simulations were then performed to confirm that these enzyme:substrate complexes were stable over a reasonable timeframe for catalysis to occur (). These findings provided direct biochemical and supporting structural evidence that TA-C is recognized as a substrate and metabolized by FDORs. The finding that at least two different FDORs accept TA-C as substrate explains the absence of TA-C–resistant strains harboring mutations in an FDOR. We also tested whether our tool compound TA-C-Met, in which reducibility of the α,β-unsaturated carbonyl motif was blocked, is metabolized by Ddn or Rv1558. As expected, TA-C-Met was not reduced by either enzyme, confirming that methylation of TA-C’s coumarin motif blocks reduction (Table 3).

A key assumption of our model is that the Mtb FDORs Ddn and Rv1558 reduce the α,β-unsaturated carbonyl motif of TA-C’s coumarin ring to produce TA-C-Red and that TA-C-Red undergoes spontaneous hydrolysis to TA-C-Acid as the reaction end product. Using chemically synthesized standards of TA-C, TA-C-Red, and TA-C-Acid, we followed the enzymatic conversion of TA-C to TA-C-Acid by Ddn and Rv1558 using liquid chromatography/mass spectrometry (LC/MS). In the case of Ddn, analysis of the reaction products revealed that the majority of the substrate (TA-C) has been converted to product after 1-h incubation, which had an m/z of 591.3 Da, consistent with TA-C-Acid. Two smaller peaks were present, one of which corresponded to the mass of the reduced intermediate species TA-C-Red, which had a m/z of 573.29 Da, consistent with the standard and with hydride transfer to the substrate peak (TA-C), which had an m/z of TA-C (571.27 Da) (Fig. 2 and ). In the case of Rv1558, the conversion was slower, consistent with the enzyme kinetics (Table 3), and the intermediate peak was not observed; this would be expected if the spontaneous conversion of TA-C-Red to TA-C-Acid occurs more rapidly than the conversion of TA-C to TA-C-Red in this instance. Altogether, these results confirm that TA-C is indeed reduced by FDORs at the α,β-unsaturated carbonyl motif in TA-C’s coumarin moiety and that the reduced form of TA-C spontaneously undergoes a ring-opening reaction to form TA-C-Acid as the end product.

To confirm that the conversion of TA-C to TA-C-Acid also occurs in intact bacteria, we treated M. bovis bacillus Calmette–Guérin culture with TA-C and observed time-dependent generation of TA-C-Acid with concurrent reduction of TA-C levels ().

Previously, we showed that TA-C had specificity toward the Mtb DHFR over the human DHFR (15). Although the intracellular activation of TA-C within Mtb will greatly increase the specificity of this compound (25), we tested whether TA-C-Acid retained selectivity toward Mtb by testing its inhibition of human DHFR (Table 3). We observed that although TA-C-Acid inhibited human DHFR with greater affinity than TA-C (1.7 ± 0.1 versus 10.5 ± 1.9 µM, respectively), the increase in affinity was less than observed for the mycobacterial protein (Table 3). Thus, the FDOR-dependent activation of TA-C increases the selectivity to 188-fold for Mtb DHFR versus human DHFR.

To lay the ground for further compound optimization, we determined the pharmacokinetic (PK) profile of TA-C and TA-C-Acid following administration via the intravenous and oral routes. We observed no detectable oral bioavailability (), which appears to be due to poor permeability, based on the intravenous PK profile and previously published in vitro PK data (15).

Discussion

Intrabacterial metabolism plays a prominent role in whole-cell activity—or inactivity—of antimycobacterials (39, 40). On one hand, intrabacterial drug metabolism can cause intrinsic drug resistance, which involves drug molecules that are active against isolated Mtb targets becoming converted into derivatives lacking the ability to engage their target once inside the bacterium. As an example, sulfonamides show potent activity against isolated Mtb DHPS but weak whole-cell activity because they are metabolized by Mtb (6). On the other hand, a number of highly successful antitubercular prodrugs, including the first-line drugs isoniazid and pyrazinamide, the second line drug PAS, and the newly approved pretomanid (6, 10, 27, 41, 42) do not have detectable activity against specific bacterial targets in their parent form but are converted by Mtb into derivatives with activity against intrabacterial targets. Here, we show that the Mtb DHFR inhibitor TA-C represents a category of antimycobacterial prodrug. In contrast to established Mtb prodrugs, TA-C was designed as a selective Mtb DHFR inhibitor and as such does exert biochemical activity against its target (15). However, upon uptake, Mtb converts TA-C to a more-selective and 100-fold more-potent DHFR inhibitor, TA-C-Acid (Fig. 3).

Fig. 3.

F420 and folate pathways and mechanism of action of TA-C. Enzymes are indicated by gray background. Enzymes for which TA-C–resistant mutations were isolated are underlined (in the case of DfrA [DHFR], an engineered over-expressor strain conferred resistance). Metabolites are gray. TA-C and its derivatives are dark red. TA-C is converted by the FDORs Ddn and Rv1558 to TA-C-Red, which undergoes spontaneous hydrolysis to the highly potent DHFR inhibitor TA-C-Acid. The drug PAS is bright red. PAS is converted by FolP (DHPS) and FolC into the DHFR inhibitor hydroxy-DHF. Target X, unknown target(s) proposed to be intercepted by TA-C-Met and TA-C (). PABA, para-aminobenzoic acid; DHF, dihydrofolate; THF, tetrahydrofolate; and mTHF, methylene-tetrahydrofolate. The F420 pathway is according to Bashiri et al. (23). The folate pathway is according to Hajian et al. (8).

To characterize this prodrug paradigm, we isolated and analyzed spontaneous TA-C–resistant mutants in Mtb and bacillus Calmette–Guérin. Two types of resistance mechanisms were identified. As expected, TA-C resistance was caused by mutations in thyA at a frequency of ∼10−8/CFU (Table 3 and Fig. 3). thyA mutations are known to cause resistance to other DHFR inhibitors by reducing the vulnerability of folate synthesis and susceptibility of DHFR to chemical inhibition (4, 16). The second resistance mechanism was the loss of enzymes required for the biosynthesis of the redox cofactor F420 (Table 3 and Fig. 3). TA-C-Met, a TA-C analog in which FDOR-mediated reduction of the coumarin moiety is blocked, copied the F420-resistance phenotype, suggesting that TA-C reduction by FDORs is required for full expression of antibacterial activity (Table 3 and Fig. 3). The putative product of FDOR reduction, TA-C-Acid, was 100-fold more potent against isolated Mtb DHFR (Table 3 and Fig. 3). Biochemical analyses using a collection of recombinant Mtb FDORs showed that at least two of these enzymes recognize TA-C as substrate, reduce the molecule at the predicted site, and generate TA-C-Acid (Table 3 and Fig. 3). This enhancement mechanism provides an explanation for the exquisite antibacterial activity of the DHFR inhibitor, which has an MIC50 of 10 to 20 nM despite moderate activity against its isolated target (Table 3). While these data are fully consistent with FDORs playing a role in activating TA-C, single- and double-knockout or overexpression strains of Ddn and Rv1558 were not generated and analyzed. Thus, one cannot unequivocally state that these two enzymes are indeed the FDORs responsible for the bioactivation of TA-C in intact bacteria.

TA-C in an F420-deficient background and TA-C-Met in wild-type background, which are phenotypically equivalent, both retained appreciable whole-cell activity (MIC50 2 to 3 µM), compatible with DHFR as their molecular target since both exhibit the same in vitro activity against DHFR (IC50 ∼1 uM). However, TA-C-Met also retained the same activity in thyA mutants and DHFR-overexpressing strains (Table 3). This suggests that TA-C engages non-DHFR target(s) at this higher concentration. TA-C-Met–resistance selection experiments did not yield any resistant colonies up to 109 CFU of bacillus Calmette–Guérin. As TA-C–resistant mutations in thyA emerge at a frequency of ∼10−8/CFU, this low resistance frequency (<10−9/CFU) 1) is consistent with a non-DHFR mechanism of action, which remains to be determined and 2) suggests that the compound engages more than one target and is not prone to (high-level) resistance development. Rapid emergence of clinical resistance usually follows the clinical launch of drugs that target single enzymes, thus limiting their utility. A recent example is bedaquiline, approved by the US Food and Drug Administration in 2012 with clinical resistance reported shortly after its introduction (43, 44).

It may seem incongruous that the 2- to 3-µM antibacterial activity of TA-C-Met is not exerted via inhibition of DHFR despite an IC50 of ∼1 μM against the enzyme. However, most antibiotics inhibit their target at lower concentrations in biochemical assays than in bacterial cultures. Bedaquiline, which inhibits the F-ATP synthase with an IC50 of 2.5 nM, has an MIC of 54 nM, a 20-fold potency reduction from enzyme to whole cell (45). This common shift is associated with intrabacterial PK (uptake, metabolism, nonspecific binding to macromolecules, and efflux) resulting in lower free drug concentrations at the cellular site of the molecular target compared to the culture medium. Considering these observations, it is less surprising that TA-C-Met’s inhibitory activity on the DHFR enzyme does not contribute to its 2- to 3-µM antibacterial activity.

In the same way that resistance can rapidly emerge against drugs that target a single protein, prodrugs are vulnerable to rapid evolution of resistance when a single activating enzyme is involved. For example, in the case of pretomanid, although we have shown that the fitness costs associated with mutations that cause loss of F420 biosynthesis are too great for this to become a widespread resistance mechanism, mutations that result in loss of the promiscuous prodrug-activating activity of Ddn, without substantially affecting the native menaquinone reductase activity, can cause resistance (and have already been detected in clinical samples) (24). TA-C is interesting in the context of those results because it is activated by multiple enzymes; the likelihood of Mtb becoming resistant via simultaneous mutations in multiple activating enzymes is extremely low. Thus, compounds such as TA-C, with multiple targets and multiple activating enzymes, represent an intriguing class of antimycobacterial compounds that could be resilient to the development of drug resistance.

In summary, we have demonstrated that TA-C, an antibacterial designed as a selective Mtb DHFR inhibitor, exerts its superb whole-cell antibacterial activity by disabling the folate pathway, albeit only after intrabacterial “bioaugmentation” by FDORs, explaining the unusual 100-fold boost from enzymatic to whole-cell activity. TA-C also appears to possess a DHFR-independent mechanism of action at low-µM concentrations, as revealed by its residual whole-cell activity in F420-deficient strains and in thyA mutants resistant to DHFR inhibition. This multitargeting property of TA-C and the functional redundancy of the activating enzymes could be exploited through medicinal chemistry to mitigate the development of high-level resistance in future drug discovery programs.

Materials and Methods

Bacterial Strains, Media, and Culture Conditions.

M. bovis bacillus Calmette–Guérin Pasteur (ATCC 35734) and M. tuberculosis H37Rv (ATCC 27294) were obtained from the American Type Culture Collection. Strains were grown in Middlebrook 7H9 broth (BD Difco) supplemented with 0.05% (vol/vol) Tween 80 (Sigma), 0.2% (vol/vol) glycerol (Fisher Scientific), and 10% (vol/vol) Middlebrook albumin-dextrose-catalase (BD Difco) at 37 °C with orbital shaking at 80 rpm. For determination of CFU, bacterial cultures were spread onto Middlebrook 7H11 agar (BD Difco) supplemented with 10% (vol/vol) Middlebrook oleic acid-albumin-dextrose-catalase and 0.2% glycerol and grown at 37 °C. When appropriate, agar was supplemented with TA-C or TA-C-Met for isolation of resistant mutants or 25 μg/mL kanamycin for the selection of transformed bacteria in the complementation experiments. Escherichia coli strains DH5-α and BL21 (DE3) were used for propagation of plasmids and expression of recombinant proteins, respectively, and were cultured in Luria–Bertani (LB) broth or on LB agar (BD Difco).

Chemicals.

The synthesis of TA-C has been described previously (15). Synthesis of TA-C-Red is described in . TA-C-Met and TA-C-Acid were purchased from BioDuro. PAS, pretomanid, isoniazid, rifampicin, and kanamycin sulfate were purchased from Sigma-Aldrich. All drugs and chemicals were dissolved in 100% dimethyl sulfoxide (DMSO) at 10 mM except kanamycin, which was dissolved to 50 mg/mL in deionized water and sterilized using 0.2-μm Minisart high-flow syringe filters (Sartorius).

Selection of TA-C–Resistant Mutants.

Mutant selection was carried out as described previously (46). In brief, bacterial inocula (∼106 to ∼108 CFU) from midlog cultures of M. bovis bacillus Calmette–Guérin or M. tuberculosis H37Rv were plated on 7H11 agar containing 3, 6, and 12 µM TA-C and grown for 4 wk at 37 °C. To verify resistance, colonies were picked and restreaked alongside wild-type bacteria on agar containing the TA-C concentrations on which the putative mutants were selected. Single colonies were picked from the restreak plates, inoculated into 7H9 broth, expanded to midlog phase, and stored with 10% glycerol at −80 °C until used for subsequent studies.

Counter Screen of TA-C–Resistant Mutants with Pretomanid.

To identify TA-C–resistant strains that harbor a wild-type F420 pathway, we first isolated 1,000 TA-C–resistant M. bovis bacillus Calmette–Guérin strains and then identified the subset that is susceptible to pretomanid via replica plating (47). Briefly, 10 independently grown bacillus Calmette–Guérin cultures were plated onto 7H11 agar containing 3 µM TA-C and grown for 4 wk. Resistant colonies were picked, inoculated into 7H9 broth in flat-bottom 96-well plates and incubated for 7 d. The culture in each well was adjusted to an optical density at 600 nm (OD600) = 0.1 and pin-plated onto 7H11 microplate agar dishes using a sterile 96-well pin replicator (Stem Corporation). Bacteria were spotted in duplicates on both drug-free and 20 µM pretomanid-containing agar. Wild-type bacillus Calmette–Guérin was included in each selection plate as control. TA-C–resistant colonies that did not grow on pretomanid agar were picked from the corresponding colony spots of drug-free replica plates, expanded in 7H9 broth to midlog phase, and stored with 10% glycerol at −80 °C.

MIC Determination.

MIC were determined as described previously by generating growth inhibition dose–response curves (15). Briefly, drugs were serially diluted in flat-bottom 96-well plates, and a midlog-phase culture was mixed with the drug-containing broth (final OD600 = 0.005) and incubated for 10 d. Growth was monitored by measuring turbidity at 600 nm using a Tecan Infinite 200 Pro microplate reader (Tecan). MIC50, the concentration that reduces growth by 50% compared to untreated control, was deduced from the generated dose–response curves.

Whole-Genome and Targeted Sequencing.

Genomic DNA was extracted from bacillus Calmette–Guérin and Mtb strains using the phenol-chloroform method as described previously (46). Whole-genome sequencing was performed on the Illumina MiSeq platform. Library preparation (NEBNext Ultra kit, New England Biolabs), sequencing, and bioinformatic analyses were performed by AIT Biotech. To verify mutations, target genes were PCR amplified using custom-designed primers (; Integrated DNA Technologies), and purified PCR products were Sanger sequenced (AIT Biotech).

Genetic Complementation Experiments.

Representative TA-C–resistant M. bovis bacillus Calmette–Guérin mutants harboring polymorphisms in fbiC, fbiA, fgd1, or thyA were complemented with the respective wild-type copies of the genes. The coding sequences were PCR amplified from the bacillus Calmette–Guérin genome using custom primers () and fused to the hsp60 promoter in pMV262 (48). The expression cassettes composed of the hsp60 promoter and the respective coding sequence were transferred into the integrative plasmid pMV306 (41) as either NotI/EcoRI or NotI/HindIII fragments (). After confirming the integrity of the constructs by sequencing (AIT Biotech), the plasmids were electroporated into bacillus Calmette–Guérin, and transformants were selected on kanamycin agar as described previously (49).

F420 Measurement.

F420 contents of bacillus Calmette–Guérin cultures was determined by measuring crude cell extracts for F420-specific fluorescence as described with minor modifications (50). In brief, late log-phase culture was harvested at OD600 = 1.0, washed in 0.9% (wt/vol) NaCl, and resuspended in disodium phosphate and potassium dihydrogen phosphate extraction buffer (50 mM each at pH = 5.7). The suspension was incubated in boiling water for 15 min, placed on ice for 10 min, and centrifuged at 13,000 × g for 40 min at 4 °C. To the supernatant, an equal volume of ice-cold 2-propanol was added, the mixture was spun (13,000 × g for 40 min at 4 °C), and the supernatant was dispensed in triplicate into a black flat-bottom 96-well plate for measurement of fluorescence at an excitation of 400 nm and an emission of 470 nm using a microplate reader (Infinite 200 Pro, Tecan). Statistical difference was compared with the wild-type fluorescence level using unpaired Studetn’s t test (GraphPad Prism 8).

Cloning, Expression, and Purification of Enzymes (FDORs and DHFR).

The E. coli codon-optimized sequences of Rv1261, Rv3178, and Rv1558 from M. tuberculosis were purchased as gene blocks from Integrated DNA Technologies and cloned into the expression vector pMAL-c2X using Gibson assembly (51). The Rv1558 sequence contained an N terminus His-tag and tev cleavage site between the maltose binding protein (MBP) tag and the protein. Codon-optimized Mtb DHFR coding sequence (Integrated DNA Technologies) was amplified using primers 5′-CGG​CCA​TGG​TGG​GGC​TGA​TCT​GGG-3′ and 5′-CCG​GAA​TTC​TCA​TGA​GCG​GTG​ATA​GGA​GTA-3′. PCR products were digested with NcoI and EcoRI and cloned into pET-30a(+) plasmid (Novagen) digested with the same restriction enzymes. The recombinant plasmid was constructed such that an N terminally 6× histidine-tagged fusion protein is produced. The construction of plasmids for the heterologous production of the FDORs in E. coli has been previously described: Ddn (Rv3547) was expressed from pMAL-c2X (24), while all FDORs were expressed from pETMCSIII (Rv1155, Rv2074, Rv2991, MSMEG_2027, MSMEG_5998, MSMEG_6526, MSMEG_0048, MSMEG_2850, MSMEG_3356, MSMEG_3909, MSMEG_5030, and MSMEG_5215) (37). F420-dependent glucose-6-phosphate dehydrogenase (FGD) was expressed from pDEST17 (52).

All proteins were expressed by transformation into E. coli (BL21 DE3) and grown on LB agar containing 50 µg/mL kanamycin or 100 µg/mL ampicillin. Single colonies were picked and inoculated in LB media with 50 µg/mL kanamycin or 100 µg/mL ampicillin and grown overnight at 37 °C. Starter cultures for DHFR, Ddn, Rv1261, Rv3178, and Rv1558 were diluted 1/100 in LB media containing 50 µg/mL kanamycin or 100 µg/mL ampicillin and grown at 37 °C until OD = 0.4. Cultures were induced with isopropyl β-d-1-thiogalactopyranoside (IPTG) to a final concentration of 0.3 mM and grown for 3 h or overnight at 25 °C. Starter cultures for all other proteins used in this study were diluted 1/100 in autoinduction media (20 g/l tryptone, 5 g/l yeast extract, 5 g/l NaCl, 6 g/l Na2HPO4, 3 g/l KH2PO4, 6 mL/l glycerol, 2 g/l lactose, 0.5 g/l glucose, and100 mg/mL ampicillin) and grown at 30 °C overnight. Cells were harvested by centrifugation at 8,500 × g for 20 min at 4 °C.

Proteins that were purified using the MBP tag (Ddn, Rv1261, Rv3178, and Rv1558) were 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 × 6 min at 50% power). The soluble extract was obtained by centrifugation at 13,500 × g for 1 h at 4 °C and filtered. The protein was purified by passing the lysate over amylose resin (NEB) and washed with lysis buffer. The protein was eluted using elution buffer (lysis buffer with 10 mM maltose). Samples were flash frozen and stored at −80 °C in 20 mM Tris pH = 7.5, 200 mM NaCl, 10 mM maltose, and 10% glycerol. All other proteins used in this study were purified using a His-tag and were resuspended in lysis buffer (50 mM NaPO4, pH = 8, 300 mM NaCl, and 25 mM imidazole) and lysed by sonication as above. The soluble extract was obtained by centrifugation at 13,500 × g for 1 h at 4 °C, filtered and loaded 5-mL HisTrap HP column (GE Healthcare), and washed with lysis buffer. The protein was eluted with elution buffer (lysis buffer with 250 mM imidazole). Purified protein was passed through a GE HiPrep 26/10 desalting column for buffer exchange into 50 mM Tris-Cl, pH = 7, and 150 mM NaCl. Samples were flash frozen with 10% glycerol and stored at −80 °C. Rv1558 used for crystallography was further purified by removing the MBP and His-tag using tev protease at a 1:10 ratio and left overnight at 4 °C. Tag-free protein was purified by passing the TEV reaction over a 5-mL HisTrap HP column collecting the flow through. The collected protein was incubated with 1.5-fold excess F420 overnight at 4 °C, followed by size-exclusion chromatography on a GE Hiload 16/600 Superdex 75-pg column in buffer containing 20 mM Hepes pH = 7.5 and 150 mM NaCl. Fractions that were yellow and contained protein as identified by sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS-PAGE) were pooled and used for crystallography after the sample diluted was in 20 mM Hepes pH = 7.5 to a final solution of in 20 mM Hepes pH = 7.5 and 50 mM NaCl and concentrated to 10 mg/mL.

F420 Purification.

F420 was overexpressed and purified from M. smegmatis as previously described (53, 54). Briefly, M. smegmatis (mc24517) cells were transformed with pYUBDuet-fbiABC by electroporation, plated on LB agar plates containing 0.05% Tween 80 and 50 µg/mL each hygromycin B and kanamycin and incubated at 37 °C for 3 d. Single colonies were picked and inoculated in LB media containing 0.05% Tween 80 and 50 µg/mL each hygromycin B and kanamycin and grown at 37 °C for 3 d. Starter cultures were diluted 1/100 into autoinduction media as described above and grown at 37 °C for 5 d. Cells were harvested by centrifugation at 10,000 × g for 30 min at 4 °C and resuspended in 20 mM Tris, pH = 7.5, and lysed by autoclave at 121 °C for 20 min. The soluble fraction was collected by centrifugation at 20,000 × g for 1 h at 4 °C. F420 was purified by passing the filtered lysate on a 60-mL Macroprep High-Q ion-exchange column (Bio-Rad). Fractions containing F420 identified by the presence of an absorption peak at 420 nm were collected and passed on a 75-mL hypersep C18 gravity column (Thermo Fisher), and F420 was eluted with 20% Methanol, freeze dried, and stored at −80 °C.

Enzyme Activity Assays.

F420 was reduced overnight with 10 µM FGD and 10 mM glucose-6-phosphate in 20 mM Tris, pH = 7.5 under anaerobic conditions. FGD was removed by spin filtration in a 0.5-mL 10K molecular-weights cutoff (MWCO) spin filter (Millipore), and F420H2 was used within 1 h of FGD being removed. Enzyme assays with FDOR and TA-C were performed according to previously published methods (36, 37) in 50 mM Tris ⋅ HCl, pH = 7.5, 25 µM F420H2, 25 µM TA-C, and 1 µM enzyme at room temperature. Activity was monitored using fluorescence following the oxidation of F420H2 (excitation/emission: 400/470 nm). Specific enzyme activity of F420H2 oxidation was calculated using a standard curve of F420.

LC/MS Assay.

Enzymatic reactions for LC/MS analysis were done with 50 mM Tris pH = 7.5, 1 µM FGD, 5 µM F420, 5 mM glucose-6-phosphate, 5 µM enzyme, and 25 µM TA-C for 1 h. All protein was removed by spin filtration in a 0.5-mL 10K MWCO spin filter (Millipore). Then, 1-µL aliquots were injected onto an ultimate 3000 ultrahigh performance liquid chromatography (UHPLC) coupled with an Orbitrap Elite (Thermo fisher). Samples were separated on an Agilent Eclipse XDB C18 column (1.8 µm, 2.1 × 50 mm) using the following gradient at 300 μL/min: 0 to 3 min 2% B, 3 to 8 min 2 to 80% B, 8 to 10 min 80% B, 10 to 10.5 min 80 to 2% B, and 10.5 to 15 min 2%, where A was 0.1% formic acid in MilliQ, and B was 0.1% formic acid in acetonitrile.

DHFR Inhibition Assay.

Inhibition assays were done as previously described (15) with modification. All inhibition assays were performed in 150 mM potassium phosphate, pH = 7 containing 50 µM DHF, 60 µM NADPH, 10 nM DHFR, and 0 to 100 µM TA-C in 1/2 or 1/3 serial dilution. Enzyme velocity was determined by measuring the reduction of NADPH by absorbance at 340 nm for 10 min at room temperature using an Epoch microplate spectrophotometer (BioTek). Initial rates were corrected for nonenzymatic reduction, and enzymatic activity was determined using the NADPH extension coefficient (6,220 M−1 ⋅ cm−1). IC50 were calculated by fitting the percentage of inhibition by TA-C to a four-parameter sigmoidal dose–response curve using GraphPad Prism.

X-ray Crystallography.

The Rv1558:F420 complex was screened for crystallization conditions by sitting vapor diffusion using SG1, PACT premier, JCSG-plus (Molecular Dimensions), and Peg/Ion HT (Hampton Research) high-throughput 96-well crystallization screens at 18 °C. The single crystal used for data collection was grown in 0.1 M sodium citrate pH = 5.8, 0.5 M ammonium sulfate, 1 M lithium sulfate, and 3% glycerol (vol/vol). The crystal was cryoprotected with the addition of 20% glycerol (vol/vol) and was flash cooled in liquid nitrogen. X-ray diffraction data were collected at the Australian Synchrotron beamline MX2. Diffraction data were indexed and integrated using X-ray detector software (XDS) (55) and scaled using AIMLESS (56). The structure was solved with molecular replacement in Molrep (57) using Ddn (PDB ID: 3R5R) (58) modified using CHAINSAW (59) as the search model. Model building, including manually modeling F420 into the difference density (mFo–DFc) at the binding site, was done with COOT (60) and refined using REFMAC (61). Models were optimized using the PDB_REDO server (62). Data collection and refinement statistics are provided in .

Molecular Docking.

Molecular docking was performed in the Schrödinger suite (Schrödinger release 2020-2, LLC). DHFR (PDB ID: 6NND) (63), Ddn with F420 (PDB ID: 3R5R) (58) and with a reconstructed N terminus as described previously (38), and Rv1558 with F420 (this study) were prepared in Protein Preparation Wizard (64). Briefly, any ligand present in the structure was removed, and each protein was preprocessed with bond order assigned, hydrogens added, waters removed, termini capped, and the ionization and tautomeric states of any het groups present were generated at pH = 7 using Epik. Het groups such as F420 were reviewed, and the lowest energy state with the appropriate bond order was chosen to use in the docking. Hydrogen bonds were assigned and optimized at pH = 7 using PROPKA (65), and the structure was optimized with restrained minimization using the OPLS3e force field (66). The structures of TA-C, TA-C-Met, and TA-C-Acid were created using ChemDraw (PerkinElmer Informatics) and prepared for docking by Ligprep (Schrödinger Release 2020-2: LigPrep, Schrödinger, LLC). Briefly, the ligands were minimalized using the OPLS3e force field. Ionization of the ligands were generated at pH = 7 using Epik, and each ligand retain their specified chirality specified in the input files created from ChemDraw. All ligands were docked into each protein using Induced Fit Docking (67–69). A standard protocol was selected, and docking was performed using the OPLS3e force field. A 12-Å grid box was generated centered on the active site of each protein. Docking was performed with no constraints, ligands sampled ring conformation within an energy window of 2.5 kcal/mol, amide bonds were penalized for nonplanar conformations, and van der Waals radii scaling was set to 0.5. Residues that were within 5 Å of the docked ligand were refined using Prime with sidechains being optimized. Ligands within 30 kcal/mol of the best protein:ligand complex or within the top 20 were then redocked. The top 10 of each ligand:protein complex were analyzed, and the highest scoring result with an appropriate binding mode for either catalysis (Rv1558/Ddn) or inhibition (DHFR) was used for further analysis.

Molecular Dynamics.

Molecular dynamics simulations were performed using Desmond (70), as implemented in the Schrödinger software suite (Schrödinger release 2020-2, LLC). The OPLS3e forcefield was used. The top-scoring docking complex of each structure with TA-C, TA-C-Met, or TA-C-Acid as described above were prepared in Protein Preparation Wizard (64) using the same settings as described above. Each protein was solvated in an orthorhombic box (10-Å buffer periodic boundary) with simple point-charge (SPC) water molecules, and the systems were neutralized using Na+ or Cl− ions. Following relaxation of the system, each simulation was run for 200 ns.