Eleven F₄₂₀H₂-dependent reductases of the FDOR superfamily (MSMEG loci 5998, 2850, 2027, 5030, 6325, 3380, 0048, 6848, 6526, 5170, 3880; Supplementary Table S1) and the F₄₂₀-reducing glucose-6-phosphate dehydrogenase (Fgd) were recombinantly overexpressed in E. coli BL21(DE3). MSMEG_6325, MSMEG_6526, MSMEG_3880 and fgd were expressed overnight in modified auto-induction TB2.0 media at 28°C (200 rpm) as previously described (Taylor et al., 2010; Lapalikar et al., 2012b). For the remaining proteins, cells were grown in lysogeny broth (LB) at 37°C (200 rpm) and induced at OD₆₀₀ 0.6 with 0.2% L-arabinose for 2 h. Cells were harvested by centrifugation (10,000 × g, 20 min, 4°C), resuspended in lysis buffer (50 mM NaH₂PO4, 300 mM NaCl, 10 mM imidazole, pH 8.0), and lysed in a EmulsiFlex-C3 homogenizer (ATA Scientific, Australia). The enzymes were purified from soluble extracts by Ni-nitrilotriacetic acid (NTA) affinity chromatography using gravity columns as previously described (Taylor et al., 2010; Ahmed et al., 2015) and stored in elution buffer (50 mM NaH₂PO₄ 300 mM NaCl, 250 mM imidazole, pH 8.0) until use in enzymatic assays. The high purity of the proteins was confirmed by running the fractions on NuPAGE Novex 10% Bis-Tris gels (Invitrogen, Australia) at 200 V and staining with Coomassie Brilliant Blue. Protein concentration was determined by measuring absorbance at 280 nm using a NanoDrop ND1000 (NanoDrop Technologies) and calculating concentration with the Beer–Lambert equation. Molar absorption coefficients were calculated for each protein based on amino acid sequences (Supplementary Table S1). F₄₂₀ was extracted, purified, and concentrated from a recombinant F₄₂₀ overexpression strain of M. smegmatis mc²4517 (Bashiri et al., 2010) as previously described (Isabelle et al., 2002).

Forty-seven different compounds were sourced from Sigma-Aldrich and dissolved into 1 M working stocks in dimethyl sulfoxide, except hypoxanthine and guanine that were dissolved in 1 M NaOH solution. The structures of the compounds tested are shown in Supplementary Tables S2, S3. Enzymatic assays were performed by spectroscopically monitoring the reoxidation of pre-reduced F₄₂₀H₂ in the presence of FDOR and substrate. F₄₂₀ was enzymatically reduced to F₄₂₀H₂ by overnight incubation with 1 μM Fgd and 12 mM glucose 6-phosphate as described (Ahmed et al., 2015). The enzyme was subsequently repurified as described (Ahmed et al., 2015). All reaction mixtures contained degassed Tris buffer [200 mM Tris, 0.1% (w/v) Triton X-100, pH 8.0] sequentially supplemented with 50 μM substrate, 25 μM F₄₂₀H₂, and 1 μM of the FDOR. Enzyme concentration was decreased to 10 nM for substrates observed to be rapidly reduced, i.e., quinone compounds. Reaction rates were monitored by recording the initial linear increase in 420 nm absorbance using an Epoch 2 Microplate Spectrophotometer (BioTek). All assays were performed at room temperature (approximately 25°C). We only detected significant levels of enzyme-independent, substrate-dependent F₄₂₀H₂ reoxidation for quinone and arylmethane substrates, at rates that we previously reported (Jirapanjawat et al., 2016). We observed no enzyme-dependent, substrate-independent or spontaneous F₄₂₀H₂ reoxidation in the timeframe of our assays. Specific activities were calculated after subtracting rates of enzyme-independent F₄₂₀H₂ reoxidation and were expressed in nmol s^-1 μmol^-1 enzyme as previously described (Taylor et al., 2010). The rate of reduction of three of these compounds, namely 1,4-naphthoquinone, 3-cyanocoumarin, and 5,6-dihydro-2H-pyran-2-one, was also measured in cofactor-recycling assays. Assays used 100 μM substrate, 0.1 μM enzyme, 10 μM F₄₂₀, 2.5 mM glucose-6-phosphate, and 0.45 μM Fgd. Time course high performance liquid chromatography (HPLC) assays, performed according to published methodologies (Lapalikar et al., 2012b), measured loss of absorbance (at λ_max) of the substrates at regular time intervals.

Substrates were docked into the previously solved X-ray crystal structures of MSMEG_2027 (1.5 Å resolution; PDB: 4Y9I; Ahmed et al., 2015) and MSMEG_6526 (1.7 Å resolution; PDB: 4KZY; Ahmed et al., 2015). F₄₂₀ was docked into the cofactor-binding pockets based on the cofactor-bound structures of Rv3547 (PDB: 3R5R; Cellitti et al., 2012) and Rv2074 (PDB: 5JAB; Ahmed et al., 2016) respectively. AutoDock Vina was used to computationally dock the substrates into their corresponding enzymes, with enzymes and ligands prepared using AutoDockTools operating with default settings (Morris et al., 2009). The docking results were visualized and analyzed in UCSF Chimera (Pettersen et al., 2004).

The chemical standards and reaction products of menadione, 3-cyanocoumarin, and 2-cyclohexen-1-one were detected by mass spectrometry. These compounds were reduced by incubating them with the promiscuous F₄₂₀H₂-dependent reductase MSMEG_2027 for 2 h at 37°C. The assay mixture comprised 100 μM substrate, 10 μM F₄₂₀, 1 μM Fgd, 1 μM MSMEG_2027, and excess G-6-P in either 20 mM Tris buffer, pH 8.0 (for menadione and 2-cyclohexen-1-one) or 50 mM ammonium acetate buffer, pH 7.5 (for 3-cyanocoumarin). For menadione, the standard and reaction products were derivatized with methoxyamine. Specifically, the standard and products were dried by rotary evaporation, resuspended in 20 μl pyridine containing 20 mg mL^-1 methoxyamine hydrochloride, and incubated at 37°C for 1.5 h. To this solution, 20 μl of N-methyl-N-(trimethylsilyl)trifluoroacetamide (MSTFA) was added and the solution was incubated at 37°C for 1 h. For cyclohexenone, the standard and reaction products were derivatized by spiking the solution with 1 mM 2,4-dinitrophenylhydrazine and incubating the solution at 30°C for 2 h.

The standard and reaction products of 3-cyanocoumarin were measured on an Agilent 6100 Series Single Quadrupole liquid chromatography-mass spectrometry (LC/MS) with diode array detector. Samples were separated on an Agilent Poroshell 120 EC-C18 column (2.7 μm, 2.1 × 100 mm). A gradient of two buffers, buffer A (0.1% formic acid in H₂O) and buffer B (0.1% formic acid in acetonitrile), was applied as follows: 0–0.5 min, held at 10% B; 0.5–6.5 min, 10–60% B; 6.5–7 min, held at 90% B. A positive mode electron ionisation (EI) scan was undertaken, and in these conditions the molecular ion could not be detected as the loss of the cyano (CN) group was universal. The derivatized cyclohexenone standard and reactions products were determined on an Agilent 1290 Infinity/6550 quadrupole time-of-flight (Q-TOF) LC/MS system equipped with an Agilent Poroshell 120 EC-C18 2.1 × 50 mm 2.7 μm column. A gradient comprising two buffers, buffer A (20 mM ammonium acetate, pH 7.0) and buffer B (100% acetonitrile), was applied as follows: 0–1 min, held at 10% B; 1–10 min, 10–90% B. Positive mode electrospray ionisation (ESI) was utilized, and a scan from 50 to 300 m/z was conducted. The menadione standard, its reaction product, and their methoxime derivatives were detected on an Agilent 7010 gas chromatography-mass spectrometry (GC/MS) triple quadrupole system. Samples were separated on an Agilent 19091S 30 m × 250 μm × 0.25 μm HP-5 ms column over a gradient of 60–320°C at a rate of 7°C min^-1, and a positive EI scan at 70 eV was conducted.

Eleven F₄₂₀H₂-dependent FDORs from M. smegmatis were expressed recombinantly and purified (Supplementary Table S1). We purified enzymes spanning multiple phylogenetically distinct subgroups, namely three enzymes each from the well-described FDOR-A1 and FDOR-B1 subgroups (Taylor et al., 2010; Lapalikar et al., 2012b; Ahmed et al., 2015), as well as representatives from five other subgroups (A2, A3, B1, B2, B3, B4; Ahmed et al., 2015). On the basis of previously reported data (Lapalikar et al., 2012b; Gurumurthy et al., 2013; Jirapanjawat et al., 2016), we determined the specific activities of the purified enzymes with 47 organic compounds following addition of the pre-reduced cofactor F₄₂₀H₂ (Supplementary Tables S2, S3). Of these, 16 compounds were enzymatically transformed. These compounds included fundamental monocyclic compounds, such as 3,4-dihydro-2H-pyran, 2-cyclohexen-1-one, and 5,6-dihydro-2H-pyran-2-one, as well as aromatic bicyclic and tricyclic compounds from the quinone, coumarin, and arylmethane chemical classes (Supplementary Table S2). Specific activities for the 16 substrates ranged from very low if reproducible for some compounds (e.g., <1 nmol s^-1 μmol enzyme^-1 for the pyran compound) to high for the quinones (e.g., >10⁴ nmol s^-1 μmol enzyme^-1 for 1,2-naphthoquinone) (Figure 1). Enzymes purified from the A1, A3, and B1 classes had the broadest and highest activities, with MSMEG_2027 (A1) proving catalytically compatible with all but two of the 16 substrates, whereas enzymes from the A2, B2, B3, and B4 classes had low activities with all non-quinone substrates.

Comparisons across the compounds tested for activity suggest that the presence of an electrophilic alkene is necessary for reduction to occur and implicates this group as the hydride acceptor. For example, activity was observed with pyrones that were α-substituted (e.g., 5,6-dihydro-2H-pyran-2-one; Figure 1C) but not γ-substituted (e.g., chelidonic acid; Supplementary Table S3). This is also supported by the finding that, whereas malachite green and crystal violet can be reduced, azure B cannot; while all three compounds contain triphenyl and N,N-dimethyliminium moieties, azure B lacks the central delocalized alkene group (Supplementary Tables S2, S3). Consistent with the proposal that the activated alkene is the hydride acceptor of F₄₂₀H₂-dependent reductases, enzymatic activity with coumarin derivatives was modulated by the nature of aromatic directing groups at the C3 position (Figure 1); moderate activities were observed with electron-withdrawing cyano and chloro groups, very low activities with an electron-donating amino group, and no activity with 3-hydroxycoumarin (Figure 1B). This suggests that electron-withdrawing groups render these compounds susceptible to nucleophilic attack by removing electron density from the π system. It is possible that differential interactions of these substrates with the substrate-binding pockets also contribute to the differences in the rates of reduction both between substrates and between enzymes.

For three of the compounds, we also measured specific activities with another independent assay that measured substrate reduction by HPLC in a cofactor-recycling system containing the Fgd (Supplementary Figure S1). While the relative activities between enzymes were comparable, initial reduction rates were generally higher in the cofactor-recycling systems and resulted in substrate conversions exceeding 90%.

We subsequently sought to understand the structural basis of how FDORs could reduce such diverse substrates. To do this, we used automated substrate docking to compare the binding of representative substrates to the high-resolution crystal structures of the highly promiscuous MSMEG_2720 (Ahmed et al., 2015) (A1) and the more specific MSMEG_6526 (Ahmed et al., 2015) (B2) enzyme. Compounds representing four major substrate classes were tested, namely menadione (quinone class), 3-cyanocoumarin (coumarin class), 2-cyclohexen-1-one (monocyclic compounds), and malachite green (arylmethane class). Consistent with the results of the activity assays (Figure 1), all substrates were predicted to be structurally compatible with MSMEG_2027, whereas only menadione and malachite green were predicted to specifically bind MSMEG_6526 (Supplementary Figure S2 and Table S4).

In the MSMEG_2027 models, substrates are accommodated in the large substrate-binding pocket adjacent to the cofactor-binding site (Figure 2). All four substrates are predicted to make extensive hydrophobic interactions with aromatic residues in the active site, including a triad of tyrosine residues (Tyr120, Tyr123, Tyr126) that have previously been shown to facilitate hydrophobic shielding during nitroimidazole activation (Mohamed et al., 2016b). There was also evidence of hydrophobic interactions between substrate and cofactor, including different degrees of π-stacking interactions with the isoalloxazine ring (Figure 2). Few polar interactions were predicted, except hydrogen bonds between the cyano group of 3-cyanocoumarin and the carbonyl oxygen of 2-cyclohexen-1-one with the hydroxyl group of Tyr123. The orientation of the substrates is likely to be realistic. For example, the binding poise of menadione suggests that menaquinone (the proposed physiological substrate of FDOR-A1 enzymes (Gurumurthy et al., 2013; Ahmed et al., 2015), which comprises a menadione headgroup and a polyisoprene tail) can be accommodated in the active site, given the polyisoprene tail at the C2 position is predicted to be oriented away from the active site.

The docking results indicate that hydride transfer can occur directly between cofactor and substrate within the hydrophobic environment of the MSMEG_2027 active site. Menadione, 3-cyanocoumarin, and 2-cyclohexen-1-one are predicted to be oriented such that their activated alkene groups are within 5 Å of the nucleophilic C5 center of F₄₂₀H^- (Figures 2A–C). This suggests that, in line with the activity assays (Figure 1) and previously proposed mechanisms (Taylor et al., 2010; Lapalikar et al., 2012b; Ahmed et al., 2015; Mohamed et al., 2016b), catalysis will occur through nucleophilic attack of the C5 hydride to the electrophilic alkene. In the case of malachite green, the alkene moiety (C1 position) of the substrate is 4.2 Å away from C5 of the cofactor, whereas the N,N-dimethylamine and N,N-dimethyliminium moieties point toward the solvent phase (Figure 2D). Binding modes in which the N,N-dimethyliminium moiety was proximal to the cofactor caused steric occlusion. This again suggests that the alkene rather than imine moiety serves as the initial site of hydride transfer from F₄₂₀H^-. In comparison, docking with the less promiscuous MSMEG_6526 enzyme suggested that menadione and malachite green are only accommodated at orientations where the distance between the hydride donor and acceptor exceeds 7 Å, which will be suboptimal for catalysis (Supplementary Table S4). This reflects that MSMEG_6526 has a smaller binding pocket than MSMEG_2027 due to its larger flanking loops (Ahmed et al., 2015).

We determined the mechanistic basis of substrate promiscuity among the F₄₂₀H₂-dependent reductases. To do so, we used mass spectrometry to determine the products formed by the reduction of three representative substrates. LC/MS and GC/MS studies demonstrated that, following incubation of menadione, 3-cyanocoumarin, and 2-cyclohexen-1-one with MSMEG_2027, each substrate peak increased by 2 m/z (Figure 3). This suggests that this enzyme catalyses the reduction of menadione (172 Da) to either menadiol or 2,3-dihydromenadione (both 174 Da) (Supplementary Figure S3), 3-cyanocoumarin (171 Da) to 3-cyanochroman-2-one (173 Da) (Figures 3C,D), and 2-cyclohexen-1-one (96 Da) to either 2-cyclohexen-1-ol or cyclohexanone (both 98 Da) (Figures 3E,F). This is consistent with previous observations that F₄₂₀H₂-dependent reductases mediate hydride transfer and subsequent protonation of their substrates (Taylor et al., 2010; Lapalikar et al., 2012b; Ahmed et al., 2015; Jirapanjawat et al., 2016; Mohamed et al., 2016b). In previous LC/MS studies, we demonstrated that malachite green (329 Da) was transformed by MSMEG_2027 to produce a decolorized product likely to be the protonated form of leucomalachite green (331 Da) (Jirapanjawat et al., 2016).

While these findings suggest F₄₂₀H₂-dependent reductases mediate hydride transfer from F₄₂₀H₂ to substrate, they do not resolve whether the site of attack is the alkene or carbonyl groups of the compounds. To resolve this, we derivatized the standard and reaction products of menadione with methoxyamine hydrochloride (carbonyl-specific) and MSTFA (alcohol-specific). GC/MS analysis of the reaction products revealed that reduction of menadione occurred exclusively via the alkene group (Figures 3A,B). Single and double methoxime derivatives of reduced menadione could be detected, indicating 2,3-dihydromenadione was formed as the major reaction product. No trimethylsilyl ester derivatives were formed under these conditions, underlining the absence of menadiol and other quinol products. Menadiol is nevertheless likely to form under physiological conditions through keto–enol tautomerism. In the case of cyclohexenone, the standard and reaction products were derivatized with 2,4-dinitrophenylhydrazine (carbonyl-specific), and analyzed by LC/MS. Analysis of product formation revealed the emergence of the hydrazone derivative of cyclohexanone, again indicating that reduction was mediated through the alkene (Figures 3E,F). Previous studies have inferred that coumarin reduction also occurs through the activated alkene group (Taylor et al., 2010; Lapalikar et al., 2012a,b).