Mycofactocin-associated mycobacterial dehydrogenases with non-exchangeable NAD cofactors

During human infection, Mycobacterium tuberculosis (Mtb) survives the normally bacteriocidal phagosome of macrophages. Mtb and related species may be able to combat this harsh acidic environment which contains reactive oxygen species due to the mycobacterial genomes encoding a large number of dehydrogenases. Typically, dehydrogenase cofactor binding sites are open to solvent, which allows NAD/NADH exchange to support multiple turnover. Interestingly, mycobacterial short chain dehydrogenases/reductases (SDRs) within family TIGR03971 contain an insertion at the NAD binding site. Here we present crystal structures of 9 mycobacterial SDRs in which the insertion buries the NAD cofactor except for a small portion of the nicotinamide ring. Line broadening and STD-NMR experiments did not show NAD or NADH exchange on the NMR timescale. STD-NMR demonstrated binding of the potential substrate carveol, the potential product carvone, the inhibitor tricyclazol, and an external redox partner 2,6-dichloroindophenol (DCIP). Therefore, these SDRs appear to contain a non-exchangeable NAD cofactor and may rely on an external redox partner, rather than cofactor exchange, for multiple turnover. Incidentally, these genes always appear in conjunction with the mftA gene, which encodes the short peptide MftA, and with other genes proposed to convert MftA into the external redox partner mycofactocin.

Redox enzymes that bind a single NAD cofactor can act in several conceptually different ways. In scheme A, the enzyme can oxidize a substrate by reducing NAD+ to NADH, and then allow both the product and the cofactor to diffuse away. Alcohol dehydrogenase is an example of such an enzyme 1 . The NADH it releases may be re-oxidized to NAD+ elsewhere to allow the enzyme to perform multiple turnovers. In scheme B, the NAD cofactor remains tightly bound, and no net redox change to the substrate occurs. The substrate is reduced, then re-oxidized, such that the substrate is converted from one isomer to another without a net change in oxidation state. UDP-galactose-4-epimerase is an example of such an enzyme 2 . In scheme C, the NAD cofactor remains tightly bound and does not dissociate from the enzyme after the physiological substrate undergoes a net redox change; instead electron transfer occurs between a bound NAD+ and free NADH. In scheme D, the NAD is tightly bound and cannot exchange, the enzyme catalyzes a steady flux for some substrate into its product, but the enzyme achieves multiple turnovers by alternating between oxidation of its primary substrate with reduction of an unrelated molecule called a co-substrate. As an example, PdxB oxidizes 4-phospho-d-erythronate to 2-oxo-3-hydroxy-4-phosphobutanoate, but afterwards it reduces a co-substrate such as α -ketoglutarate, oxaloacetate,

Results
Genome analysis of mycobacterial SDRs. Members of TIGR03971 appear in gene clusters near mycofactocin system protein genes mftA, mftB, and mftC [5][6][7] . For example, in M. paratuberculosis a putative carveol dehydrogenase (MAP_RS21280, UniProt ID Q73SC8, SSGCID target ID MypaA.01326.b) is encoded near genes for the mycofactocin precursor MftA (MAP_RS21310), the mycofactocin modification chaperone MftB (MAP_ RS21315), the mycofactocin radical SAM maturase MftC (MAP_RS21320), the mycofactocin system heme/ Flavin oxidoreductase MftD (MAP_RS21325), and the mycofactocin system creatininase family protein (MAP_ RS21330). The presence of an insertion in the primary sequence near the proposed NAD-binding region is conserved across members of TIGR03971, but absent in other NAD-or NADP-dependent oxidoreductases with known structure (Fig. 1). Based off genomic analysis, 69 mycobacterial dehydrogenases from TIGR03971 with an insertion near the proposed cofactor binding site were entered into the Seattle Structural Genomics Center for Infectious Disease (SSGCID) structure determination pipeline 9 . Of these, 42 targets (61%) progressed to successful protein production. Each target was selected on the basis of sequence similarity to either Mtb Rv0687 (100% sequence identity to Mybo.01326.a) or Rv2750 (MytuD.01326.a). Originally, these two Mtb targets progressed through the pipeline to purified protein, but failed to yield suitable diffraction quality crystals.
Structural analysis of mycobacterial SDRs. We determined crystal structures for 10 different mycobacterial SDRs (Tables 1, 2 and 3). These ranged in sequence identity to Rv0687 from 39-81% and from 41-77% for Rv2750. Each of these SDRs contain an insertion in the protein sequence at the cofactor binding site (Fig. 1). With the exception of the M. thermoresistibile target which did not contain NAD, all other mycobacterial SDR target crystal structures contained NAD, which appears to have co-purified with the protein after expression in E. coli. For each of these crystal structures, the adenosine mononucleotide portion of the cofactor binding site is enveloped by the inserted residues and only a portion of the nicotinamide ring appears open to solvent (see representative examples in Fig. 2A and B). A number of SDRs containing an insertion at the NAD binding site appear in mycobacterial organisms. These SDRs always appear in conjunction with a number of genes that appear to be involved in the biosynthesis of the proposed redox partner mycofactocin. Multiple sequence alignment of the ten crystal structures determined here in comparison with 10 other oxidoreductases of known structure which lack the insertion in the primary sequence. The PDB code for each structure is listed. The structures described here all contain NAD (sequences 1-9) except for 3OEC (sequence 10), whereas for the structures described elsewhere 5 representative examples of NADP (sequences [11][12][13][14][15] or NAD (sequences [16][17][18][19][20] bound structures were selected and the ligand identity is also listed. The magenta cylinders indicate α helices, whereas the blue arrows indicate β strands. The insertion loop is shown as a gray arrow and labeled, and the loop region that is often disordered is also shown as a gray arrow and labeled. In addition to the insertion loop which covers the adenosine portion of the cofactor, a second loop which typically contains two α -helices covers much of nicotinamide mononucleotide portion of the cofactor ( Fig. 2A). As observed in other structures (see representative examples in Fig. 2C and D), this loop can adopt a number of different conformations and is often disordered. In contrast, this loop is ordered in each of the mycobacterial SDR crystal structures presented here. The combination of the insertion loop which covers the adenosine portion of the cofactor and the ordering of the second loop which covers the nicotinamide mononucleotide portion of the cofactor results in a significant decrease in the solvent accessible surface area of the cofactor in the mycobacterial SDRs, relative to other NAD-and NADP-dependent dehydrogenases (Table 4). Only a small portion of the nicotinamide ring appears available to solvent in each of these cases, indicating that the NAD cofactor is tightly bound and potentially non-exchangeable. The structures presented here have a cofactor solvent accessible surface area of approximately 18 Å 2 , whereas other NAD-or NADP-bound dehydrogenases have a cofactor solvent accessible surface area of 82-89 Å 2 ( Table 4).

Exchangeability of cofactor NAD/NADH by line broadening NMR and STD-NMR. A number of
NMR experiments were performed to test exchangeability of the NAD cofactor. NMR line broadening experiments rely on differences in tumbling times between free and bound ligand in solution to detect protein-ligand interactions 10,11 . In the absence of protein, small molecule tumbling time is relatively fast, resulting in sharp peaks; upon binding to a protein, the tumbling time of the complexed small molecule is slowed, resulting in less intense and broader resonance peaks. Thus, one can compare the peaks of a particular ligand in the absence or presence of protein at specific concentrations to identify and characterize binding interactions. Line broadening was examined for both NAD and NADH in the absence and presence of two M. avium carveol dehydrogenases (MyavA.01326.d and MyavA.01326.l). No line broadening was observed for either NAD or NADH in the absence or presence of either protein over a concentration of 0.1-2.0 mM. Thus, neither NAD nor NADH appears to bind to the protein over the course of the experiment. The presence of a previously-bound, tightly-binding cofactor in   the binding pocket of either protein could prevent additional NAD or NADH added to solution from undergoing ligand exchange. Thus, the lack of a line broadening effect is consistent with non-exchangeable cofactor binding. The saturation transfer difference NMR (STD-NMR) experiment is designed to detect compounds which are capable of rapid to intermediate ligand exchange with a target protein, and is well-suited for the identification of weak-to medium-strength binding ligands 12,13 . Typically, STD-NMR saturation is not observed for proton resonances of non-binding or very tight binding small molecules (K D < 50 nM); the slow off-rate does not allow sufficient numbers of molecules to be detected over the course of the experiment. STD-NMR experiments were performed on both NAD and NADH in the absence and presence of two M. avium carveol dehydrogenases (MyavA.01326.d and MyavA.01326.l). No significant STD-NMR signal was observed for either NAD or NADH. Given that these enzyme samples (MyavA.01326.d and MyavA.01326.l) both generated crystal structures with NAD present without supplementation, the non-binding result by STD-NMR is most likely reflective of non-exchangeable binding, consistent with the results observed by line-broadening NMR.
Binding of potential substrates, products, inhibitors and redox partners by STD NMR. The binding of potential substrates, products, inhibitors, and redox partners to two M. avium carveol dehydrogenases (MyavA.01326.d and MyavA.01326.l) was examined by STD NMR. No binding signal was observed for (− )-carveol, (R)-(− )-carvone, (S)-(+ )-carvone, or tricyclazol in the presence of NAD and/or NADH in the absence of protein. Both of the redox partner compounds NDMA (N,N-dimethyl-4-nitrosoanaline) and 2,5-dichloroindophenol appear to enhance oxidation of NADH to NAD over the course of 8 days, relative to NADH alone in solution. However, none of the non-exchangeable protons of either NDMA or 2,5-dichloroindophenol experiences peak height or chemical shift changes over the same duration, suggesting these molecules remain intact and do not chemically react with either NAD or NADH directly.
For the M. avium carveol dehydrogenase MyavA.01326.d, no significant (> 10%) STD NMR effect was observed for (− )-carveol, (R)-(− ) carvone, (S)-(+ )-carvone or NDMA (Table 5). Binding was observed for tricyclazol and DCIP, with the latter generating twice the STD binding signal as the former. STD NMR binding of  each compound was monitored after spiking samples with either NAD or NADH to mimic the effect of soluble cofactor. For the potential redox partner DCIP, the %STD NMR signal was initially diminished upon spiking with NADH but not NAD, while the weak binding signal of tricyclazol experienced no change with either cofactor added to the sample. The STD binding signal of (− ) carveol experienced a near 4% increase upon addition of NAD, relative to binding in the absence of cofactor. All spiked samples were retested by STD NMR after a 7-day incubation period. The STD NMR binding signal for DCIP returned to pre-spike levels, and the STD NMR binding of NMDA increase after 7 days of incubation. As was observed for NDMA and DCIP in the absence of protein, we see partial conversion of NADH to NAD over time in the presence of protein.  Table 2 for calculated solvent accessibility of these and other structures.
STD-NMR binding was then examined for the M. avium carveol dehydrogenase MyavA.01326.l ( Fig. 3 and Table 6). Weak but clear STD NMR binding signal was observed for carveol and DCIP, with stronger saturation signal for (R)-(− )-carvone, (S)-(+ )-carvone and tricyclazol. No significant binding effect was observed for NMDA. Spiking NAD into the sample reduced the the STD NMR binding signal for DCIP. As was observed with MyavA.01326.d, the STD NMR binding signal of DCP returned to pre-spike levels after 7 days of incubation. As was observed for NDMA and DCIP in the presence of MyavA.01326.d, partial conversion of NADH to NAD was observed after 7 day incubation in the presence of MyavA.01326.l.

Discussion
The presence of a large number of dehydrogenases present in mycobacterial genomes may enhance the chances for survival in the harsh redox conditions present in macrophages. Some of these dehydrogenases may have elaborate control mechanisms beyond those currently understood to function in human host cells. Previously, we proposed the existence of an external redox partner specific to mycobacteria and related species called mycofactocin 5 . A series of genes always appear in close proximity to one another within mycobacteria and other closely   6,7 . In addition, most of these gene clusters also include a short chain dehydrogenase/reductase (SDR) from TIGR03971, the protein family focus of the current research. These SDRs are readily distinguishable from other SDRs on the protein level because of a large insertion in the primary sequence (Fig. 1). This insertion warranted detailed study since it is located near the adenosine binding region of the cofactor NAD (or NADP) in other SDRs with known structure. To characterize the importance of these additional protein residues, we determined the crystal structures of a number of mycobacterial SDRs containing this insertion. In these structures, the insertion loop covers the adenosine portion of the NAD cofactor. Concomitantly, another loop which is present but often disordered in other SDR crystal structures is ordered for the TIGR03971 family SDR protein structures and covers much of the nicotinamide mononucleotide portion of the NAD cofactor (Fig. 2). Taken together, these ordered loops in mycobacterial SDRs significantly cover the bound NAD cofactor, resulting in a solvent accessible surface area for the cofactor of only 18 Å 2 (Table 4). This solvent accessible surface area is significantly diminished in comparison with other  representative NAD-dependent SDRs which have an average solvent accessible surface area of 89 Å 2 . Similarly, representative NADP-dependent SDRs have an average solvent accessible surface area of 82 Å 2 . It should be noted that the calculated surface areas are representative of a select number of cofactor-bound SDR protein structures in the Protein Data Bank. Interestingly, BphB from Pandoraea pnomenusa also has a highly buried NAD molecule, but it accomplishes this not through an insertion as shown for these mycobacterial SDRs, but rather through an extension of the loop which normally only buries the adenosine portion of NAD 14 . This effectively creates a long cavity in which small molecules (i.e. potential substrates) can bind. Given the encapsulation of the NAD cofactor by the two loops in mycobacterial SDRs, a large conformational change would be required to release the NADH cofactor after a redox reaction had occurred. Alternatively, these structural elements may not allow for exchange of bound NAD/NADH with free NAD/NADH. If these enzymes used a non-exchangeable NAD, they would either be single turnover enzymes or would require an external redox partner. Other oxidoreductases which utilize external redox partners have been characterized enzymatically 8 , but no structural information exists for these kinds of SDRs. We know from the crystal structures presented here that these protein samples indeed contain bound NAD. To test whether or not the bound NAD was exchangeable, we performed a number of line broadening and STD-NMR ligand-observe studies, and did not observe exchange of NAD/NADH during the timescale of the NMR experiments. In contrast, we were able to observe binding of potential enzymatic substrates such as carveol, potential products such as carvone, potential inhibitors such as tricyclazol, and external artificial redox partners such as DCIP.
The structural and NMR binding data are consistent with the hypothesis that the mycobacterial SDRs in TIGR03971 utilize a non-exchangeable NAD and require an external redox partner. Given the proximity of these TIGR03971 mycobacterial SDR genes to a cluster of enzymes which have been hypothesized to be involved in the biosynthesis of mycofactocin 5 and that the initial decarboxylation of MftA by the MftC/MftB complex has been characterized enzymatically 6,7 , we speculate that these SDRs may use mycofactocin as an external redox partner. However, if these enzymes use an external redox partner such as mycofactocin, one might expect the binding sites to be similar for all of these enzymes, despite different catalytic functions and substrates. Interestingly, while the presence of the insertion in the primary sequence is conserved across these mycobacterial SDRs, the actual sequence does not appear to be conserved (Fig. 1). Thus, this observation may not be consistent with the hypothesis of mycofactocin as an external redox partner if mycofactocin requires highly specific interactions with the SDRs. However, we note that although the C-terminus of the mycofactocin precursor MftA is highly conserved, the length (30-80 amino acids) and the remaining sequence of MftA is not well conserved 7 , and differences in specific MftA sequences may be paired with differences in the SDR insertion sequences as well. Nevertheless, these SDRs always appear in close proximity to the TIGR03969 gene which encodes mycofactin (MftA) and other proteins (the carrier protein MftB TIGR03967, a radical SAM protein MftC TIGR03962, and a glycosyltransferase TIGR03965). Additional genomic, biochemical, and structural analysis is necessary to understand the importance of these proteins in the biosynthesis of mycofactocin and the subsequent redox pathways.

Methods
Overexpression and purification. All proteins were overexpressed in E. coli with an N-terminal His 3C protease cleavage fusion using protocols described previously 15,16 . Briefly, the proteins were purified by nickel affinity chromatography followed by optional cleavage and removal of the fusion tag with 3C protease and a subtractive nickel column and a final size exclusion chromatography (SEC) step. The affinity tag was not removed for MyavA.01326.d, MyavA.01326.q, and MythA.01326.c, but it was removed for all of the other samples. The final samples were stored at − 80 °C in 25 mM Tris pH 7.0, 500 mM NaCl, 5% v/v glycerol, 2 mM DTT and 0.025% w/v NaN 3 until used in crystallization experiments.  5.5; crystals were harvested with 25% v/v ethylene glycol as cryo-protectant. Crystals of MypaA.01326.d were obtained at 23 mg/mL against a focus screen based off the JCSG+ screen condition D5 which contained 70% v/v MPD and 0.1 M Hepes pH 6.5; crystals were harvested with a direct cryo-protectant. Crystals of MypaA.01326.g were obtained at 27.6 mg/mL against the JCSG+ screen condition D5 which contained 70% v/v MPD and 0.1 M Hepes pH 7.5; crystals were harvested with a direct cryo-protectant. Crystals of MythA.01326.c were obtained at 29.5 mg/mL against the JCSG+ screen condition E11 which contains 14.4% w/v PEG 8000, 0.16 M calcium acetate, 80 mM sodium cacodylate pH 6.5 and 20% v/v glycerol; crystals were harvested with a direct cryo-protectant.

Crystallization.
Scientific RepoRts | 7:41074 | DOI: 10.1038/srep41074 X-ray data collection and structure determination. X-ray data were collected either in house on a Rigaku FR-E+ SuperBright X-ray generator with Saturn 944+ detectors or at synchrotron radiation (Tables 2 and 3). Data were reduced with XDS 17 using a single crystal for each target. All structures were solved by molecular replacement in Phaser 18 . The final models (Tables 2 and 3) were obtained after iterative refinement in Refmac5 19 or Phenix 20 with manual model building in Coot 21 , and validation in MolProbity 22 . NMR analysis. All experimental NMR data were obtained on a Varian 500 MHz. Two enzymes annotated as carveol dehydrogenases from M. avium (MyavA.01326.d and MyavA.01326.l) were dialyzed into an NMR specific buffer (150 mM NaCl, 20 mM potassium phosphate, 0.1 mM NaN 3 , 10% D 2 O pH 7.0). Line broadening experiments were performed on both MyavA.01326.d and MyavA.01326.l with the protein at 20 μ M concentration and the ligands at 250 μ M with a constant 5% v/v DMSO-d 6 in the above NMR buffer. Line broadening of NAD and NADH was examined in the absence of protein as a control and in the presence of protein at concentrations of 0.1, 0.25, 0.5, 1.0 and 2.0 mM ligand. STD NMR experiments were performed on both MyavA.01326.d and MyavA.01326.l with the same conditions as the line broadening experiments for singleton STD NMR. STD NMR spike experiments were done with addition of NAD/NADH to a final concentration of 250 μ M. As control experiments, all compounds were tested for stability in the NMR buffer over a period of 8 days. The potential inhibitor N,N,N′ ,N′ -tetramethyl-p-phenylenediamine (TMPD) was not stable in the NMR buffer over 8 days and thus was not selected for additional experiments. Each compound was also tested for direct irradiation by STD-NMR in the absence of protein; false binding signals were not observed for any of compound in these experiments. The reported %STD NMR binding signal is the peak height difference (reference -saturation) for the single strongest proton resonance of the compound.