Main

CO is both a potent poison of multicellular life and a high-energy fuel and carbon source for microorganisms1,2,3,4,5. CO is released into the atmosphere in vast quantities, with natural and anthropogenic sources contributing an estimated 2,600 million tons of CO emissions annually6,7. Despite this, average CO concentrations in the atmosphere remain extremely low, at around 100 ppb, because of consumption by abiotic processes and microbial oxidation6,8. Microbial consumption accounts for an estimated 10–15% of CO removed from the atmosphere (approximately 250 million tons annually)7. This biogeochemically important process is mediated by microbes that encode form I [MoCu]-CO dehydrogenases (Mo-CODH) that are active in the presence of oxygen. In microbial cells, Mo-CODHs oxidize CO at atmospheric concentrations and the high-energy electrons derived from this process are transferred to the aerobic respiratory chain4,6,9,10,11. Because of its low concentration, the energy provided by oxidizing atmospheric CO is generally insufficient to support growth and replication6,12. However, CO provides a supplementary energy source for vast numbers of dormant bacteria present in soils, oceans and other nutrient-deprived environments, allowing them to persist during starvation4,6,13,14,15. The ability to oxidize atmospheric CO is widespread, with bacteria and archaea from 17 phyla encoding a Mo-CODH predicted to be capable of mediating this process and isolates from four phyla shown experimentally to oxidize atmospheric CO6. A structurally unrelated, highly oxygen-sensitive family of [NiFe]-CO dehydrogenases are also widely encoded by microbes; these low-affinity enzymes also have an important role in global CO cycling by oxidizing the elevated levels of CO produced in anoxic environments before it enters the atmosphere16.

Despite the biogeochemical and ecological importance of microbial atmospheric CO oxidation, no purified Mo-CODH has been shown to be capable of mediating this process. Mo-CODHs have been isolated and structurally and biochemically characterized from aerobic carboxydotrophic bacteria, which use high concentrations of CO as both a carbon and an energy source for autotrophic growth17,18,19,20,21,22. However, these enzymes have a lower affinity for CO and have not been tested for their ability to oxidize CO at atmospheric concentrations4,23,24,25,26. Through studies on whole cells, we recently demonstrated that a Mo-CODH encoded by Mycobacterium smegmatis (Mo-CODHMs) enables the bacterium to oxidize atmospheric CO, which enhances its survival during starvation4. We also showed that electrons derived from Mo-CODHMs are transferred to terminal oxidases and, thus, support aerobic respiration4,5. However, the structural and biochemical basis for the high-affinity kinetics of Mo-CODHMs remains unknown.

The physical associations and electron pathways made by Mo-CODH are also unresolved. Considering that Mo-CODH is a respiratory enzyme, it must either directly or indirectly reduce membrane-bound quinone4. Despite a lack of integral or peripheral membrane-associating regions, Mo-CODH from Afipia carboxidovorans (Mo-CODHAc) is predominantly membrane associated under some conditions27. In contrast, more limited membrane association is observed with Mo-CODHMs, which may reflect the differing roles of Mo-CODH in these organisms, in growth and persistence, respectively4,19,27,28. Previous reports indicate that Mo-CODHAc reduces respiratory quinone analogs, suggesting that quinones directly accept electrons from this enzyme29. However, it remains unclear how hydrophobic quinones reach the electron acceptor site of these soluble enzymes. It has been suggested that the Mo-CODH-associated protein CoxG has a role in this process30. In A. carboxidovorans, CoxG is not required for Mo-CODHAc assembly or maturation but is essential for mediating membrane association19,27. The loss of CoxG does not abolish CO oxidation by Mo-CODHAc but greatly increases generation time (~7-fold) when CO is the sole substrate for growth30. These observations led to the hypothesis that CoxG recruits soluble Mo-CODH to the cytoplasmic membrane, thus enabling electron transfer to quinones in the membrane30.

In this work, we resolve the structural and biochemical basis for the high affinity of Mo-CODHMs and its interactions with the electron transport chain. To do so, we isolate both Mo-CODHMs and CoxG and integrate cryo-electron microscopy (cryo-EM), X-ray crystallography and kinetic, electrochemical and genetic analyses to show the structural basis of atmospheric CO oxidation and demonstrate how electrons are transferred to the respiratory chain.

Results

Mo-CODH from M. smegmatis oxidizes atmospheric CO

We isolated Mo-CODHMs using a chromosomally encoded Strep-tag II on its CoxM subunit. Because of the complex maturation of this enzyme family, which requires at least four genetically associated assembly factors30,31, Mo-CODHMs was expressed chromosomally at native levels, with cells harvested at the late stationary phase when the enzyme is most highly expressed4. Although yield was low (~2–5 µg of enzyme per L of M. smegmatis culture), this strategy yielded high-purity protein, with SDS–PAGE analysis indicating that only the expected CoxS (17 kDa), CoxM-strep (35 kDa) and CoxL (86 kDa) subunits were present in the complex (Supplementary Figure 1a,b). Mo-CODH is active and can reduce menadione, a soluble analog of the membrane-bound electron carrier menaquinone, as an electron acceptor (Fig. 1a).

Fig. 1: Mo-CODHMs oxidizes CO with high catalytic efficiency to below atmospheric concentrations.
figure 1

a, Gas chromatography analysis of the CO concentration of the headspace of sealed vials containing Mo-CODHMs or no enzyme, with 200 μM menadione as the electron acceptor. b, Gas chromatography analysis of the CO concentration of the headspace of sealed vials containing Mo-CODHMs or no enzyme, with 50 μM methylene blue as the electron acceptor. Mo-CODHMs can oxidize CO below atmospheric concentrations (black dashed line). In a and b, data are the mean ± s.d. c, Steady-state Michaelis–Menten kinetics of Mo-CODHMs consumption of CO in the headspace of sealed vials with 50 μM methylene blue as the electron acceptor. Repeat measurements shown in a–c were conducted on distinct samples. d, Steady-state Michaelis–Menten kinetics of Mo-CODHAc consumption of CO in the headspace of sealed vials, with 50 μM methylene blue as the electron acceptor. e, DC voltammogram of surface-confined Mo-CODHMs on a PIGE electrode obtained at pH 8.0, under N2 (scan rate = 100 mV s−1, T = 21 °C). Mo labels the peak likely corresponding to the redox transitions of the Mo cofactor, while * and *′ label peaks that correspond to unassigned redox processes possibly attributable to the enzyme FeS clusters or FAD. f, Comparison of the DC voltammograms obtained at pH 8.0, under N2 (black line), CO (red) and CO2 (blue) saturation (scan rate = 50 mV s−1, T = 21 °C). A positive shift of approximately +110 mV for the MoVI/IV process was observed because of the decrease in pH when the solution was saturated with CO2. Voltammetry was conducted five times on three separately prepared samples, with a representative sample shown. conc., concentration; MoVI/IV, change in the charge of the Mo ion from 6+ to 4+.

Source data

Full size image

Although Mo-CODHMs can oxidize atmospheric CO in whole cells and provide the resulting electrons to the respiratory electron transport chain, no isolated Mo-CODH has been shown to be capable of oxidizing CO at this concentration24,25,26. As such, it was unclear whether Mo-CODHMs has an inherently high affinity for CO or whether this property results from coupling to the respiratory electron transport chain. To resolve this question, we tested the ability of purified Mo-CODHMs to oxidize 1 ppm CO in an ambient air headspace using gas chromatography. Mo-CODHMs consumed CO below atmospheric levels (to 0.065 ppm in headspace, corresponding to 51 pM in solution) using the artificial electron acceptor methylene blue (Fig. 1b). To determine the affinity of Mo-CODHMs for CO oxidation, we performed kinetic analysis by measuring headspace CO consumption at different concentrations. These data were fitted to Michaelis–Menten kinetics, allowing us to calculate an apparent Km of 139 nM (95% confidence interval (CI): 86–230 nM) for CO oxidation by Mo-CODHMs (Fig. 1c), which is lower than that previously reported for a Mo-CODH and consistent with measurements of CO affinity of mycobacterial whole cells4,24,25,26.

We compared the kinetics of Mo-CODH purified from M. smegmatis (Mo-CODHMs) and A. carboxidovorans (Mo-CODHAc). The well-studied Mo-CODHAc has been reported to have a Km of between 520 nM and 10.7 µM (refs. 24,26), yet A. carboxidovorans cultures are capable of atmospheric CO oxidation32, indicating the Mo-CODH Km is not the sole determinant of threshold concentration. On the basis of our assays, Mo-CODHAc has a lower affinity than Mo-CODHMs (Km of 2.30 µM (95% CI: 1.36–4.22 µM)); it can oxidize CO at atmospheric concentrations, indicating that Km has limited influence on the threshold concentration of the enzyme (Fig. 1d, Extended Data Fig. 1d and Supplementary Note 1). Interestingly, M. smegmatis and A. carboxidovorans cultures performed comparatively when oxidizing CO at a starting headspace concentration of 10 ppm (Extended Data Fig. 1a). However, on the basis of our relative Mo-CODH purification yields, the A. carboxidovorans used in our assay produces vastly more Mo-CODH than M. smegmatis (Supplementary Note 1), suggesting that the higher affinity of Mo-CODHMs may allow the bacteria to efficiently oxidize atmospheric CO while producing much smaller quantities of enzyme.

To investigate the electrochemical properties of Mo-CODHMs, we performed protein film electrochemistry (PFE) on the purified enzyme. Cyclic voltammograms on a film of Mo-CODHMs in 100% N2 at pH 8.0 revealed three oxidation peaks that likely correspond to redox transitions of the enzymes Mo (−0.296 V versus standard hydrogen electrode (SHE), pH 8.0), and other electroactive groups (for example, FeS clusters or flavin adenosine dinucleotide (FAD); −0.065 V and +0.260 V) (Fig. 1e and Supplementary Note 2). A positive sigmoidal-shaped steady-state current was observed in voltammograms in the presence of CO, corresponding to oxidation of the gas, with an onset potential close to the equilibrium potential of the CO2/CO reaction (−0.58 V versus SHE, pH 8.0)33, indicating that Mo-CODHMs operates without a notable overpotential (Fig. 1f). By calculating Mo-CODHMs loading, we were able to derive an indicative turnover frequency (kcat) of 27 s−1 for the enzyme, which is broadly comparable to the 93.3 s−1 determined previously for Mo-CODHAc (Extended Data Fig. 2 and Supplementary Note 2)33. On the basis of this analysis, Mo-CODHMs exhibits similar redox properties to Mo-CODHs from other organisms and may derive its higher affinity from structural modifications. No negative current was observed in the presence of CO2 (1 bar), even at strongly negative potential (−0.8 V), indicating that Mo-CODHMs does not reduce CO2 to CO under the conditions tested.

Mo-CODHMs has distinct structural features

To determine the structure of Mo-CODHMs, we performed cryo-EM imaging and single-particle reconstruction of the purified 272-kDa complex. Cryo-EM micrographs and resulting class averages revealed a C2 symmetrical molecule, which is consistent with the previously determined structures of lower-affinity Mo-CODHs (Extended Data Fig. 3a,b)17,23,34. Despite indications that Mo-CODH uses membrane-localized quinones as electron acceptors (Fig. 1a)29, Mo-CODHMs did not copurify with membranes and no additional electron-transferring subunits were identified in class averages. Data processing and three-dimensional (3D) reconstruction yielded maps with a nominal resolution of 1.85 Å, which allowed us to build and refine the structure of Mo-CODHMs with a high degree of precision (Fig. 2a,b, Extended Data Fig. 3c–f, Supplementary Fig. 2a, Supplementary Table 1 and Supplementary Videos 1 and 2). Like other enzymes of this family, Mo-CODHMs is a heterohexamer composed of two CoxSML subunits that interact through their CoxL subunits (Fig. 2a,b). The CoxL subunit encloses a CuMo-molybdopterin cytosine dinucleotide (CuMCD) containing an active site that oxidizes CO and transfers the resulting electrons to an FAD cofactor at the electron acceptor site in CoxM through two [2Fe–2S] clusters in the CoxS subunit (Fig. 2c). Despite forming a dimer, the electron transfer pathways of the two CoxSML subunits are not within electron transfer distance (Extended Data Fig. 4a). Mo-CODH mediates catalysis by binding CO at the Cu ion of the active site, followed by nucleophilic attack by the equatorial oxo ligand of the Mo ion35,36,37. The bond lengths in this region of the cofactor indicate that Mo-CODHMs is in a reduced state, with the Mo ion coordinated by oxo and hydroxyl ligands, in addition to the dithiolene group of the pyranopterin and the Cu-bridging sulfur (Fig. 2d)17,24,38. This state may result from the electron beam reducing the cofactor during data collection or possibly the oxidation of atmospheric CO in the absence of an exogenous electron acceptor.

Fig. 2: The cryo-EM structure of Mo-CODHMs shows unique features.
figure 2

a, The final cryo-EM density map of Mo-CODHMs color-coded to show the position of each subunit contoured at 4σ. b, Cartoon representation of the cryo-EM structure of Mo-CODHMs showing the CoxS, CoxM and CoxL units and catalytic and electron transfer cofactors. c, Zoomed-in view of the catalytic and electron transfer cofactors of one subunit of Mo-CODHMs, showing cofactors positioned within efficient electron transfer distance. d, The catalytic cluster of Mo-CODHMs, showing bond distance, which indicates that the Mo ion is in a reduced (IV) state. e, The coordination environment of the catalytic CuMCD complex in Mo-CODHMs. CDP, cytidine diphosphate. f, The coordination environment of the catalytic CuMCD complex in Mo-CODHAc. g, A plot showing the radius of the active site gas channels A and B from Mo-CODHMs relative to their distance from the catalytic Mo ion. h, Zoomed-in view of the active site gas channels from Mo-CODHMs. i, Zoomed-in view of Ala237 from Mo-CODHMs and the corresponding valine residues from Mo-CODHAc and CODHHp, with alanine at this position opening the second gas channel in Mo-CODHMs. j, Zoomed-in view of Leu490 from Mo-CODHMs and the corresponding valine residues from Mo-CODHAc and Mo-CODHHp, with leucine at this position narrowing the second gas channel in Mo-CODHMs. Ms, Mycobacterium smegmatis; Ac, Afipia carboxidovorans, Hp, Hydrogenophaga pseudoflava.

Source data

Full size image

The coordination environment of the pyranopterin cofactor that coordinates the active site Mo ion in Mo-CODHMs differs markedly from previously structurally characterized Mo-CODHs, including Mo-CODHAc (ref. 17). In Mo-CODHMs, a conserved arginine (Arg387 in Mo-CODHAc) is substituted for alanine (Ala380 in Mo-CODHMs). This leads to substantial remodeling of the environment around the pyranopterin in Mo-CODHMs with three additional highly coordinated water molecules forming an interaction network that substitutes for the loss of the arginine side chain (Fig. 2e,f, Extended Data Fig. 4b and Supplementary Video 3). In the other Mo-CODH to be structurally characterized from Hydrogenophaga pseudoflava (Mo-CODHHp), this arginine is post-translationally modified with a hydroxyl group at the Cγ, suggesting that the specific chemical environment at this position has a role in tuning the catalytic activity of the enzyme23. Phylogenetic analysis of CoxL subunits indicates that arginine is conserved at this position in all Mo-CODH groups, except for the actinobacterial clade that contains Mo-CODHMs, where alanine is universally present (Extended Data Fig. 4c and Supplementary Data 1).

The active site of Mo-CODH is in the interior of the CoxL subunit and is accessed by CO through hydrophobic gas channels. These channels have not been previously analyzed in Mo-CODH; hence, we used MOLEonline to map them in Mo-CODHMs, Mo-CODHHp and Mo-CODHAc (Supplementary Data 2)39. Interestingly, while both Mo-CODHHp and Mo-CODHAc have a single gas channel providing access to the active site, Mo-CODHMs has two gas channels (Fig. 2g,h and Extended Data Fig. 5a–e). Both the gas channels of Mo-CODHMs are much narrower than the gas channels of their counterparts, with bottleneck radii of 1.1 and 1.2 Å for Mo-CODHMs compared to 1.4 Å for Mo-CODHHp and 1.6 Å for Mo-CODHAc (Fig. 2g and Extended Data Fig. 5b,d). Substitution of valine for alanine at position 237 of CoxL of Mo-CODHMs opens the second gas channel, while a number of other substitutions constrict the Mo-CODHMs channels (Fig. 2h–j and Extended Data Fig. 5f,g). While the more constricted gas channels of Mo-CODHMs potentially restrict the rate of turnover for the enzyme (kcat = 27 s−1 versus 93 s−1 for Mo-CODHAc), they may have a role in restricting access to nontarget substrates.

CoxG has a hydrophobic cavity that binds menaquinone

We next sought to test the hypothesis that Mo-CODHMs transfers electrons to the aerobic respiratory chain by associating with the membrane-bound protein CoxG. This is consistent with the observations that, while isolated Mo-CODHMs is a soluble enzyme, it associates with the cytoplasmic membrane in cells and, like Mo-CODHAc, directly reduces quinone analogs and, thus, can likely reduce respiratory quinones (Fig. 1a)4,30. CoxG is predicted to comprise two domains, a C-terminal membrane-anchoring helix (CoxGCT) and an N-terminal soluble cytosolic domain (CoxGNT), which are connected by a flexible linker (Fig. 3a and Extended Data Fig. 6)19,27.

Fig. 3: CoxG carries MQ9 within its hydrophobic cavity.
figure 3

a, An AlphaFold2 (ref. 46) model of CoxG showing its CoxGCT, cytosolic linker and CoxGNT. b, Cartoon representation of the X-ray crystal structure of CoxGNT showing the β-sheet and two α-helices that form the SRPBCC fold, with rainbow coloring from the N terminus (blue) to C terminus (red). c, Surface rendering of the inner hydrophobic cavity of CoxGNT. The electrostatic potential of the cavity is shown by blue, white and red coloring. d, Stick representation of the hydrophobic residues that line the inner cavity of CoxGNT. e, Stick representation of key positively charged residues present on the N-terminal side of CoxGNT. f, Electrostatic surface potential of CoxGNT mapped onto the surface rendering of the structure. Positively charged regions are blue and negatively charged regions are red, with neutral charge in white. g, Schematic of the experiment performed to test CoxGNT mediated extraction of MQ9 from M. smegmatis membranes. h, Base peak chromatograms (m/z = 750–2,000) for the high-performance liquid chromatography–mass spectrometry analysis of Folch extracts from CoxGNT. A substantial peak at 24.14 min is uniquely observed in the M. smegmatis membrane-incubated sample and the corresponding positive mode mass spectrum is shown in i. This shows a single dominant ion at m/z = 804.665, consistent with the ammonium adduct of DH-MQ9 (m/z = 804.665). j, A 2Fo − Fc composite omit map of electron density corresponding to a lipid-like molecule within the hydrophobic cavity of the crystal structure of CoxGNT contoured at 1σ. k, Molecular docking of MQ9 within the CoxGNT internal cavity, showing the highest-ranked solution.

Full size image

To gain insight into the functional role of CoxGNT, we recombinantly expressed and purified the domain in Escherichia coli and then determined its structure using X-ray crystallography (Supplementary Table 2 and Supplementary Fig. 2b). Analysis of the structure revealed that CoxGNT belongs to the SRPBCC superfamily, which is characterized by a seven-stranded antiparallel β-sheet and two α-helices that enclose a large hydrophobic cavity. The CoxG hydrophobic cavity has an internal surface area of 635.2 Å2 and a volume of 521.4 Å3 (Fig. 3b–d and Supplementary Video 4). This cavity defines the function of the SRPBCC family, which is to bind and traffic diverse hydrophobic molecules through hydrophilic environments40,41. This suggests that, in addition to facilitating the membrane association of Mo-CODH, CoxG may extract a hydrophobic ligand from the membrane. The N-terminal face of CoxGNT, which contains an opening to the internal hydrophobic cavity, has five lysine residues that create a positively charged region, whereas the C-terminal face that connects to the membrane linker is negatively charged (Fig. 3e,f and Supplementary Video 4). This positively charged region may interact with the negatively charged head groups of membrane phospholipids, facilitating ligand extraction. In agreement with this, electron density for a lipid-like molecule is present in the internal cavity of the CoxGNT crystal structure (Fig. 3j). The identity of this molecule could not be unambiguously determined from the electron density; thus, we analyzed the small molecules associated with purified CoxGNT using mass spectrometry. This analysis detected ubiquinone-8 (UQ8), the major E. coli respiratory quinone during aerobic growth, associated with CoxGNT, among a number of other unidentified molecules (Extended Data Fig. 7a,b)42,43. Given the highly hydrophobic character of UQ8, it is likely that it is bound within the CoxGNT cavity rather than being present in the bulk solvent of the sample. Consistently, the electron density present in the CoxGNT cavity is compatible with modeling UQ8, albeit not at full occupancy (Extended Data Fig. 7c). This observation led us to hypothesize that CoxG shuttles quinone between the membrane and Mo-CODHMs, facilitating its reduction with electrons from atmospheric CO oxidation. The low occupancy of UQ8 in CoxGNT produced in E. coli suggests that it is not its preferred ligand, which is consistent with M. smegmatis using dihydromenaquinone-9 (DH-MQ9) as its major respiratory quinone44,45.

To test the ability of CoxGNT to bind DH-MQ9, we attempted to express it in M. smegmatis, where it could extract the ligand from cell membranes. However, it was highly toxic when overexpressed in its native host, possibly because of CoxGNT sequestering cellular DH-MQ9, and purified protein could not be obtained. As an alternative, we incubated CoxGNT produced in E. coli with total membranes from M. smegmatis before repurifying it and identifying extracted ligands by mass spectrometry (Fig. 3g). Strikingly, the only additional molecule detected in CoxGNT after incubation was DH-MQ9, which was present at approximately 34% occupancy (Fig. 3h,i). Consistent with this, mass spectrometry demonstrated that CoxGNT also bound purified MQ9, despite its sparing solubility in aqueous solution (Extended Data Fig. 7d–f). This indicates that CoxG from M. smegmatis selectively binds both DH-MQ9 and MQ9. To model the CoxG–MQ9 complex, we performed docking simulations, which consistently placed MQ9 within the internal cavity of CoxGNT, with the highest-ranked solution placing the redox-active head group in proximity to the opening of the internal cavity (Fig. 3k, Supplementary Videos 5 and 6 and Supplementary Data 3). Taken together, both the structure of CoxGNT and its ability to specifically extract MQ9 from membranes support a role in extracting and trafficking respiratory quinone to Mo-CODH for reduction.

CoxG is essential for CO oxidation in M. smegmatis cells

Considering our hypothesis that CoxG transports MQ9 to Mo-CODHMs for reduction, we hypothesized that deletion of CoxG would decouple it from the respiratory chain and prevent oxidation of CO. Consistent with this, M. smegmatis ∆coxG did not oxidize CO while, as previously reported, the M. smegmatis mc2155 wild-type (WT) strain consumed CO to below atmospheric levels (Fig. 4a)4. Complementation of the ∆coxG strain with a plasmid containing coxG restored CO consumption to near WT levels (Fig. 4a). Although CoxG appears to be essential for CO oxidation in cells, CO oxidation by Mo-CODHMs was still detectable in cell lysates from the ∆coxG strain when using the electron acceptor 4-nitro blue tetrazolium chloride (NBT) to stain for CO activity (Fig. 4b). Mo-CODHMs activity was lower than WT in both the ∆coxG strain and the complemented strain (Fig. 4b). This is consistent with previous reports in A. carboxidovorans where the loss of CoxG led to a 50% reduction in Mo-CODHAc activity and could be because of several factors, including altered gene expression patterns resulting in a reduction in Mo-CODH expression or loss of activity because of reductive inactivation in the absence of electron transfer through CoxG30.

Fig. 4: CoxG is predicted to interact with Mo-CODHMs and facilitate CO oxidation in M. smegmatis.
figure 4

a, Gas chromatography analysis of CO in headspace of sealed vials containing the M. smegmatis strains WT, ΔcoxG or ΔcoxG:pcoxG. WT M. smegmatis can oxidize CO to below atmospheric levels; however, the ΔcoxG strain cannot oxidize CO. Complementation of CoxG on a plasmid restores CO oxidation to near WT levels. b, Left, an NBT-stained native PAGE of Mo-CODHMs from cell lysates of the WT, ΔcoxG or ΔcoxG:pcoxG M. smegmatis strains. Right, densitometric analysis of gels showing Mo-CODHMs activity from three independent biological replicates of each sample normalized to WT staining intensity, showing that the ΔcoxG and ΔcoxG:pCoxG strains produce active Mo-CODHMs. c, AlphaFold2-Multimer model of Mo-CODHMs and CoxGNT (pink), showing a predicted interaction at the interface between CoxM (green) and CoxL (blue). The redox-active head group of MQ9 docked in CoxGNT is positioned toward the Mo-CODHMs interface. d, The positioning of CuMCD, [2Fe–2S] and FAD groups within Mo-CODHMs, with distances that the electrons transit between each group listed. The predicted interaction between CoxGNT and Mo-CODHMs positions MQ9 ~16 Å from the FAD group of Mo-CODHMs. e, CoxG lysine residues that mediate interactions with corresponding negatively charged and polar residues at the CoxG–Mo-CODHMs interface. f, Gas chromatography analysis of CO headspace of sealed vials containing the M. smegmatis strains WT, ΔcoxG or ΔcoxG:pcoxG-Lys→Ser, ΔcoxG:pcoxG-Lys→Asp, ΔcoxG:pcoxG-Lys→Glu. The complementation strains have substitutions to the CoxG interface, with the five critical lysines (from d) substituted to either serine, aspartate or glutamate. The WT M. smegmatis strain oxidizes CO to below atmospheric levels and the ΔcoxG strain cannot oxidize CO. The Lys→Asp and Lys→Glu mutants cannot oxidize CO and the Lys→Ser retains some CO oxidation activity, albeit not to WT levels. In a, b and f, data are mean ± s.d.

Source data

Full size image

To provide DH-MQ9 to Mo-CODHMs for reduction, CoxG must interact with the enzyme so that electron transfer can occur from the FAD group to bound DH-MQ9. Because of the small quantities of Mo-CODHMs we can purify, traditional methods for characterizing protein–protein interactions were not feasible. As such, we used AlphaFold2-Multimer to assess complex formation between a single CoxSML subunit of Mo-CODHMs and CoxG46,47. AlphaFold2 consistently predicted a complex between CoxG and primarily the CoxL and CoxM subunits of CoxLMS (Fig. 4c, Supplementary Data 4 and Supplementary Table 3). In this complex, the N-terminal end of CoxG, which contains the opening to the hydrophobic cavity, interacts with Mo-CODHMs in proximity to the electron-donating FAD group of CoxM. When docked MQ9 is included in this model, its redox-active head group is 16 Å away from FAD (Fig. 4d). In this position, conformational changes and structural dynamics could allow DH-MQ9 to migrate into the Mo-CODHMs cavity, which is also hydrophobic, and directly accept electrons from FAD (Extended Data Fig. 8a). The interaction interface is highly charged, containing five lysine residues from CoxG interacting with two glutamate, two aspartate and one glutamine residues from Mo-CODHMs (Fig. 4e and Supplementary Video 7). The predominance of lysine residues on the CoxSML-interacting interface of CoxG likely stems from its need to interact with the phospholipid head groups of the cell membrane to extract DH-MQ9 (Fig. 3h–j). Considering that the membrane interacting interface of CoxG must also interact with CoxSML to deliver quinone for reduction by FAD, these positively charged residues must be compensated for by the negatively charged residues observed on the Mo-CODHMs side of the interface (Supplementary Table 4). This provides evidence that the surface chemistry of CoxG satisfies interactions with both the cell membrane and its CoxSML-binding partner.

To validate the functional importance of the five lysines present at the interaction interface of CoxG in M. smegmatis, we generated three variants in which all lysines were substituted to serine (CoxG-Lys→Ser), aspartate (CoxG-Lys→Asp) or glutamate (CoxG-Lys→Glu). Genes encoding these CoxG variants were used to complement the M. smegmatis ΔcoxG strain and consumption of CO was measured by gas chromatography. While the strains expressing CoxG-Lys→Asp and CoxG-Lys→Glu variants were unable to consume CO, the strain expressing CoxG-Lys→Ser regained the ability to oxidize CO at a substantially reduced rate (Fig. 4f). This result is consistent with the CoxG-Lys→Asp and CoxG-Lys→Glu variants being unable to interact with the lipid membrane and/or CoxSML because of charge repulsion, while the more minor change to polar serine residues impairs but does not abolish CoxG function.

CoxG and Mo-CODH interact in diverse microbes

To further validate the AlphaFold2-predicted complex between CoxG and Mo-CODHMs, through comparative genomics, we assessed how widespread the presence of the coxG gene is within the Mo-CODH gene clusters of 30 diverse microbes (Supplementary Table 5). coxG is present in the Mo-CODH gene clusters of 24 of these bacteria and archaea, spanning all five phylogenetic groups of the enzyme. Moreover, in four of the cases where coxG was not identified, a homolog is present elsewhere in the genome, which may facilitate quinone transport to Mo-CODH. The only two cases where a coxG homolog could not be identified in the Mo-CODH-containing genome were Sulfolobus islandicus and Candidatus Acidianus copahuensis, two archaeal extremophiles, which may have evolved an alternative means of transferring electrons from Mo-CODH48,49. The synteny of coxG relative to the coxSML structural subunits varies between homologs and, in Bradyrhizobium japonicum and Alkalilimnicola ehrlichii, CoxG is present as a fusion protein with CoxL, further strengthening the functional link between these proteins (Fig. 5a). CoxG homologs are also associated with genes encoding non-Mo-CODH members of the xanthine oxidase family (XOF), including an uncharacterized homolog present in M. smegmatis, where it is positioned between genes encoding the CoxL and CoxM subunits (Fig. 5a).

Fig. 5: Analysis of the Mo-CODH and CoxG relationship across bacterial species.
figure 5

a, Organization of Mo-CODH/XOF operons from M. smegmatis and six representative microbes. The coxG gene (pink) is consistently present in the same gene cluster as the enzymatic subunits of Mo-CODH (coxM (green), coxL (blue) and coxS (yellow)), although with varying synteny and is sometimes fused with coxL. b, AlphaFold2-Multimer models showing the predicted CoxG and Mo-CODH/XOF complex from the eight representative gene clusters. All models predict an analogous complex between CoxG and Mo-CODH/XOF. c, Cut-away dot-and-surface view of the AlphaFold2 models of Mo-CODHMs and XOFMs, with the FAD group placed into the CoxM subunit and CoxG docked with MQ9. MQ9 and FAD are contained within a single enclosed cavity in the XOF model (top), whereas, in the Mo-CODHMs–CoxG model (bottom), they are enclosed in separate cavities. d, Proposed model for CO oxidation in M. smegmatis. Mo-CODHMs oxidizes CO to CO2 and the membrane-tethered CoxG extracts MQ9 from the cell membrane and delivers it to Mo-CODHMs. MQ9 is reduced to menaquinol by the Mo-CODHMs FADH2 group and delivered back to the membrane by CoxG to provide electrons to the terminal oxidases.

Full size image

To assess whether CoxG also forms a complex with these Mo-CODH homologs, we performed AlphaFold2-Multimer modeling with the CoxSML and CoxG subunits of eight representative microbes. In each case, CoxG is predicted to form a complex with the CoxSML subunits analogous to that observed for Mo-CODHMs (Fig. 5b, Extended Data Fig. 8b and Supplementary Data 4). AlphaFold2 predicted the formation of this complex in each Mo-CODH homolog and the M. smegmatis XOF, despite an average amino acid sequence identity of 26.4% for the CoxG homologs and 42.6% and 54.6% for the CoxM and CoxL subunits that form the interaction interface (Extended Data Fig. 9 and Supplementary Table 6). The consistency of the AlphaFold2 prediction of this complex across homologs that share limited sequence identity provides strong evidence of its authenticity. When the model of the Mo-CODHMs–CoxG is appended with the CoxM FAD cofactor from the cryo-EM structure and CoxG–MQ9 from our docking analysis, these cofactors are confined to isolated chambers (Fig. 5c), indicating that conformational changes must occur upon CoxG binding to facilitate migration of menaquinone to FAD for reduction. Interestingly, in the M. smegmatis XOF model, the interaction between CoxG and CoxSML forms a single enclosed cavity that would allow MQ9 to access the FAD cofactor, suggesting that this model represents a later stage of the substrate transfer process (Fig. 5c).

In all Mo-CODH–CoxG complexes, the interactions between CoxG and CoxSML are highly charged, involving multiple salt bridges. Consistent with the Mo-CODHMs–CoxG interface, positively charged side chains are overwhelmingly present in the CoxG subunit (30 of 33 salt bridges; Supplementary Table 4). In combination with our other observations, these data support the proposed model for CoxG-mediated quinone transport and its extendibility to most aerobic CO-oxidizing microorganisms (Fig. 5d).

Discussion

In this work, we demonstrate that an isolated enzyme is capable of oxidizing CO to subatmospheric levels. By determining the high-resolution structure of Mo-CODHMs, we identify key structural differences between this enzyme and lower-affinity variants that may contribute to differences in affinity. The coupling of Mo-CODH to the respiratory chain suggests that electrons produced from the oxidation of CO must enter the quinone pool4. However, how Mo-CODH as a soluble enzyme can reduce respiratory quinone was unknown. On the basis of this work, we propose a model that is conserved across Mo-CODHs from diverse bacteria and archaea, where the lipid-binding protein CoxG extracts menaquinone from the membrane before delivering it to the electron acceptor site of soluble Mo-CODH. Menaquinone is then reduced to menaquinol using electrons from atmospheric CO before CoxG returns it to the membrane (Fig. 5d). Menaquinol is then oxidized by the terminal oxidases to generate proton motive force for adenosine triphosphate synthesis and other cellular processes50,51. To our knowledge, this mechanism of electron transfer to menaquinone is mechanistically unique. Moreover, the presence of CoxG in association with diverse Mo-CODHs and other members of the XOF indicates that this is a widespread mechanism for coupling soluble enzymes to the respiratory chain. While mechanistically distinct, there are parallels between this system and quinone reduction by complex I and the recently described [NiFe]-hydrogenase Huc44,52. In both these cases, quinone is extracted from the membrane and delivered to its site of reduction through a hydrophobic tunnel44,52. Conversely, our observation that CoxG acts as a quinone shuttle provides a mechanism for quinone reduction outside the membrane, which evolved convergently to that of Huc and complex I, and expands our understanding of how electrons enter the respiratory chain.

Methods

Statistics and reproducibility

No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications4,5,9,13,15,44. Data distribution for all data was assumed to be normal but this was not formally tested. Data collection and analysis were not performed blind to the conditions of the experiments.

Data analysis and visualization

In addition to programs described elsewhere in the manuscript, Microsoft Excel 2016 and Graphpad Prism 8 were used for data analysis and visualization.

Purification of Mo-CODHMs

M. smegmatis mc2155 CoxM-strep was cultured in 8 L of Hartmans de Bont (HdB) for 3 days after the maximum optical density at 600 nm (OD600) was attained. Pellets were harvested by centrifugation (30 min at 5,000g) and stored at −20 °C. Pellets from 16–24-L batches of culture were thawed, pooled and resuspended in ice-cold lysis buffer (150 mM NaCl, 50 mM Tris pH 8.0, 1× cOmplete EDTA-free protease inhibitor (Roche, 11836145001), 1 mg ml−1 DNase and lysozyme) using a Dounce glass homogenizer. Cells were lysed by two passages through a cell disrupter (Emulsiflex C-5) and cell debris was removed by centrifugation at 30,000g for 30 min. The clarified cell lysate was incubated with biotin blocking buffer (IBA, 2-0205-050) and was loaded onto a 1-ml StrepTrap XT column (Cytiva, 29401317) and washed with ~100 ml of chromatography buffer (150 mM NaCl and 50 mM Tris pH 8.0). Protein was eluted with elution buffer (150 mM NaCl, 50 mM Tris pH 8.0 and 50 mM biotin) in ten 1-ml fractions. Biotin was dissolved in the buffer by adjusting the pH back to pH 8.0 with 5 M NaOH after mixing. Elution fractions were analyzed by SDS–PAGE and fractions containing Mo-CODHMs were pooled and concentrated to a volume of ~500 µl using a 4-ml centrifugal concentrator with a 30-kDa molecular weight cutoff (MWCO) (Amicon, UFC803008). Concentrated fractions were loaded onto a Superose 6 Increase 10/300 GL column (Cytiva, 17517201) and eluted in 0.5-ml fractions using chromatography buffer and analyzed by SDS–PAGE. Fractions containing Mo-CODHMs were pooled and concentrated to ~25 µl at 2–4 mg ml−1 using a 0.5-ml centrifugal concentrator with a 30-kDa MWCO (Amicon, UFC503008) and stored at −80 °C.

Purification of Mo-CODHAc

Purification was carried out as described previously23,24,53,54. Briefly, 5 L of A. carboxidovorans cells were grown to exponential phase and harvested by centrifugation to yield 3 g of cells. Cells were resuspended in 50 mM HEPES pH 7.4 supplemented with 0.1 mg ml−1 lysozyme, 0.05 mg ml−1 DNase I and cOmplete protease cocktail inhibitor tablets (Roche, 11836145001). The cells were then lysed through cell disruption (Constant Systems, 40,000 psi, two times) and the lysate was clarified with high-speed centrifugation. Lysate was applied to a pre-equilibrated Q FF HiTrap anion-exchange column (Cytiva, 17515601) and eluted over ten column volumes with an increasing salt gradient to a final concentration of 1 M NaCl. Fractions were then assayed through SDS–PAGE and activity staining to determine Mo-CODHAc activity. Briefly, for the activity assay, 20 µl of each peak fraction was added to a 96-well plate with 200 µl of 500 µM NBT in each well. The plate was incubated in a 100% CO chamber and Mo-CODHAc activity was determined by reduction of NBT resulting in a color change to purple. Fractions containing active Mo-CODHAc were pooled and ammonium sulfate was added to a final concentration of 1.2 M. Pooled fractions were gently stirred for 1 h at 4 °C. After incubation, pooled fractions were clarified with low-speed centrifugation to remove any precipitant and loaded onto a Phenyl HiTrap HP hydrophobic affinity column (Cytiva, 17519501) pre-equilibrated with 0.85 M ammonium sulfate in 50 mM HEPES pH 7.4. Mo-CODHAc was eluted over 20 column volumes with decreasing ammonium sulfate (final concentration 0 M) and increasing isopropanol (final concentration 100%). Fractions containing Mo-CODHAc were determined through SDS–PAGE and activity assays, as described. Mo-CODHAc active fractions were pooled and dialyzed overnight at 4 °C into 50 mM HEPES pH 7.4 buffer. The dialyzed Mo-CODHAc was then concentrated and stored at −80 °C.

Expression and purification of CoxG

For recombinant expression, p29b-CoxG (pCoxGtrunc) was transformed into E. coli (DE3) C41 cells cultured in terrific broth (per liter: 12 g of tryptone, 24 g of yeast extract, 61.3 g of K2HPO4, 11.55 g of KH2PO4 and 10 g of glycerol, with 100 µg ml−1 ampicillin for selection) as previously described55 (Supplementary Table 7). Cells were grown at 37 °C until they reached an OD600 of 1.0 and were then induced using 0.3 mM IPTG, followed by further growth for 14 h at 25 °C. Cells were harvested by centrifugation, lysed using a cell disruptor (Emulsiflex C-5) in Ni-binding buffer (50 mM Tris, 500 mM NaCl and 20 mM imidazole pH 7.9) plus 0.1 mg ml−1 lysozyme, 0.05 mg ml−1 DNase I and cOmplete protease cocktail inhibitor tablets (Roche, 11836145001). The resulting lysate was clarified by centrifugation and applied to Ni-agarose resin, followed by washing with ten column volumes of Ni-binding buffer and elution of bound proteins with a step gradient of Ni-gradient buffer (50 mM Tris, 500 mM NaCl anf 500 mM imidazole pH 7.9) of 5%, 10%, 25% and 50%. Eluted fractions containing recombinant protein were pooled and applied to a 26/600 Superdex S200 size-exclusion chromatography (SEC)column (Cytiva, 28989336) equilibrated in SEC buffer (50 mM Tris and 200 mM NaCl pH 7.9). The recombinant protein was then pooled, concentrated to 60 mg ml−1, snap-frozen and stored at −80 °C.

Gas chromatography

For culture-based CO consumption assays, M. smegmatis WT, ΔcoxG and ΔcoxG:pCoxG cultures were grown in 30 ml of HdB medium amended with 0.05% (v/v) glycerol and 0.05% (w/v) tyloxapol in 120-ml glass serum vials, with incubation at 37 °C and shaking at 150 rpm. A. carboxidovorans cultures were grown in DSMZ 133 medium in a 50% CO headspace at 30 °C with shaking at 150 rpm.

When cultures reached the stationary phase (M. smegmatis, 3 days after maximum OD600) or exponential phase (A. carboxidovorans, 7 days after inoculation), the vials were sealed with butyl rubber stoppers and amended with 10 ppm CO gas. For A. carboxidovorans, the 10 ppm CO gas was bubbled into the medium for 10 min to remove any remaining high concentration of dissolved CO. Consumption was monitored over 48 h at regular time intervals by injecting 2 ml of headspace into a pulsed-discharge helium ionization detector (model TGA-6791-W-4U-2, Valco Instruments) as previously described12.

For enzymatic CO consumption assays, 5 ml of reaction buffer (50 mM Tris or 50 mM HEPES pH 7.4 (for Mo-CODHAc), 150 mM NaCl pH 8.0, 0.3 mg ml−1 BSA and 50 µM methylene blue or 200 µM menadione) was purged with N2 (5 min, intermittent shaking) in 120-ml glass serum vials sealed with a butyl rubber stopper and magnetic stirrer bar. For the atmospheric consumption experiment with Mo-CODHMs, the following buffer was used instead: 200 mM NaCl, 30.8 mM K2HPO4, 19.2 mM KH2PO4 pH 7.2, 0.3 mg ml−1 BSA and 50 µM methylene blue. Purged vials were amended with purified Mo-CODHMs to 8, 5 or 25 nM, purified Mo-CODHAc to 50 nM or vehicle control (reaction buffer without Mo-CODH) using a 250-µl gas-tight syringe and then immediately amended with 1 or 10 ppm CO (10 ppm cal gas; BOC, CCS405162D2) and incubated at room temperate with gentle stirring. Reactions for each condition were prepared in triplicate. Time points were taken over ~24–70 h, with t0 taken 30 s after starting incubation. Headspace CO concentration was measured by injecting 2 ml of headspace into a pulsed-discharge helium ionization detector (model TGA-6791-W-4U-2, Valco Instruments) as previously described12.

For Mo-CODHMs and Mo-CODHAc kinetics, 5 ml of reaction buffer (50 mM Tris or 50 mM HEPES pH 7.4 (for Mo-CODHAc), 150 mM NaCl pH 8.0, 0.3 mg ml−1 BSA and 50 µM methylene blue) was added to 120-ml glass serum vials sealed with a butyl rubber stopper and purged with N2. CO was added to the vial headspace to a final concentration of 1, 5, 50, 200, 500 or 1,000 ppm (1, 3.5, 40, 250, 500 or 850 nM dissolved CO in the buffer, respectively). Then, 1, 2 or 5 nM Mo-CODHMs or 10 nM Mo-CODHAc was added to the reaction (lower CO concentrations required more Mo-CODHMs for consumption) for it to begin. All reaction vessels were prepared in triplicate. CO consumption was measured as described, using a pulsed-discharge helium ionization detector, by injecting 2 ml of headspace at eight time points over 8 h. The rate of CO oxidation was calculated for each CO concentration as the nanomoles of CO consumed per hour by 1 nM Mo-CODHMs. This was plotted as dissolved CO (nM) versus rate (nmol h−1). Then, using the Michaelis–Menten nonlinear fit function on GraphPad Prism, the Vmax and Km were calculated.

PFE of Mo-CODHMs

A CH Instruments 760E potentiostat was used to conduct the direct current (DC) voltammetric experiments. A standard three-electrode system in a single-compartment cell was used. The set-up consisted of a Ag/AgCl (saturated KCl) reference electrode, a platinum wire as the counter electrode and a bare custom-built paraffin-impregnated graphite electrode (PIGE)56 (electroactive surface area, A = 0.09 cm2) or a Mo-CODHMs enzyme-modified PIGE as the working electrode. To fabricate the Mo-CODHMs enzyme-modified PIGE, the Mo-CODHMs enzyme was immobilized on the electrode by placing the electrode vertically and carefully pipetting 0.5–1 µL of a 0.32 mg ml−1 stock solution of the purified enzyme (stored in 150 mM NaCl and 50 mM Tris buffer aqueous solution at pH 8.0) followed by 0.5 µl of a 0.5% chitosan (w/v) in 1% acetic acid (v/v). The electrode was placed under a gentle flow of N2 for 5 min (until dry) and the surface was cleaned with a small amount of distilled water followed by drying with N2. All voltammetric experiments were conducted in an electrolyte solution containing 150 mM NaCl and 50 mM Tris buffer at pH 8.0. Before the voltammetric experiments, the electrolyte solution was degassed with N2, CO or CO2 for at least 10 min. During the voltammetric experiments, a positive gas pressure was applied to ensure that oxygen was not redissolved in the solution. All voltammograms were recorded at 21 ± 1 °C. The potential scale of all voltammograms reported here was converted into the SHE scale knowing the potential of the Ag/AgCl (saturated KCl) reference electrode is +0.204 V versus SHE at 20 °C (ref. 57).

Cryo-EM imaging of Mo-CODHMs

Samples (3 μl) were applied onto a glow-discharged UltrAuFoil grid (Quantifoil) and flash-frozen in liquid ethane using the Vitrobot Mark IV (Thermo Fisher Scientific) set at 100% humidity and 4 °C for the prep chamber. Data were collected on a G1 Titan Krios microscope (Thermo Fisher Scientific) with a Schottky field emission gun as the electron source operated at an accelerating voltage of 300 kV. A C2 aperture of 50 μm was used and no objective aperture was used. Data were collected at a nominal magnification of ×105,000 in nanoprobe energy-filtered transmission EM mode. A Gatan K3 direct electron detector positioned after a Gatan BioQuantum energy filter was operated in zero-energy-loss mode using a slit width of 10 eV to acquire dose-fractionated images of the Mo-CODHMs complex. One dataset was collected, composed of 4,508 videos. Videos were recorded in hardware-binned mode yielding a physical pixel size of 0.82 Å per pixel with a dose rate of 8.5 e− per pixel per second. An exposure time of 6 s yielded a total dose of 66.0 e− per Å2, which was further fractionated into 60 subframes. Automated data collection was performed using EPU (Thermo Fisher Scientific) with periodic centering of the zero-loss peak. A defocus range was set between −1.5 and −0.5 μm.

Cryo-EM data processing and analysis

Micrographs from all datasets were motion-corrected using UCSF MotionCor and dose-weighted averages had their contrast transfer function (CTF) parameters estimated using CTFFIND 4.1.8, implemented using RELION 3.1.2 (ref. 58). Particle coordinates were determined by crYOLO 1.7.6 using a model trained on manual particles picked from 20 micrographs59. Unbinned particles were extracted from micrographs using RELION 3.1.2, before being imported into cryoSPARC 3.3.1 for initial two-dimensional (2D) classification to remove bad particles, followed by ab initio model generation and 3D refinement55. Refined particles were reimported into RELION 3.1.2 and CTF refinement was performed, followed by Bayesian Polishing58. Particles were reimported into cryoSPARC 3.3.1 for final 2D classification to remove residual bad particles, followed by nonuniform 3D consensus refinement to generate final maps at 1.85 Å (Fourier shell correlation (FSC) = 0.143, gold standard).

Mo-CODHMs cryo-EM model building and visualization

An initial model of the CoxMSL subunit trimer was generated using AlphaFold2 and docked into one half of the high-resolution CoxMSL homodimer maps using ChimeraX60,61. The model was refined and rebuilt into map density using Coot62. CuSMoO2, molybdopterin, [2Fe–2S] and FAD cofactors associated with Mo-CODHMs were downloaded from Protein Data Bank (PDB) 1N5W and fitted and refined into maps using Coot17,62. The model was then refined using real-space refinement within PHENIX63. Once model building was complete, the model was symmetry-expanded using the Map symmetry tool and waters were added using DOUSE within the PHENIX package63. Model quality was validated using MolProbity64. Images were generated in Pymol65.

CoxGNT X-ray crystal structure solution

Purified CoxG was screened for crystallization conditions using commercially available screens (∼800 conditions). A number of these conditions were optimized with crystals used for data collection originating from wells containing 0.2 M sodium acetate, 28% polyethylene glycol 2000 MME and 0.1 M Tris pH 8.5. Diffraction data were collected on three isomorphous crystals at 100 K at the Australian Synchrotron and processed and merged using the XDS package66. Phases were obtained using an AlphaFold model of CoxG from M. smegmatis46. The initial model was built using PHENIX63. The CoxG model was improved manually in Coot and refined using phenix.refine62,63. Analysis of the CoxG crystal structure was performed using the PHENIX and CCP4 packages60. Figures were made using Pymol and ChimeraX65,67. The area of the CoxG internal cavity was determined using pCAST68.

Mo-CODH tunnel analysis using MOLEonline

MOLEonline was used to identify the presence and measure the radii of Mo-CODH substrate channels for Mo-CODHAc (PDB 1N5W), Mo-CODHHp (PDB 1FFV) and our cryo-EM structure of M. smegmatis Mo-CODH39. The setting ‘ignore HETATMS’ was turned off. To prevent the interpretation of artefacts, the settings ‘probe radius’, ‘interior threshold’ and ‘origin radius’ were explored over multiple submissions for each protein complex and the consistency among predictions was monitored; the substrate channels were inspected using the embedded viewer and in Pymol after export. Cavity data were exported and visualized in GraphPad Prism and Pymol.

AlphaFold structural modeling

Modeling was performed using AlphaFold version 2.1.1 implemented on the MASSIVE M3 computing cluster46,47. For the prediction of the CoxG–CoxMSL complex, the corresponding sequence with the flexible C-terminal extension of CoxG removed was provided and modeling was run in multimer mode, with just one molecule of each subunit requested. The five ranked models produced by AlphaFold were compared for consistency with the top-ranked model used for further analysis and figure generation. Intersubunit interfaces were analyzed with PISA69. Models are provided in Supplementary Data 4.

Molecular docking of MQ9 into CoxGNT

To determine the ability of CoxGNT to bind MQ9, we used Autodock Vina in the Chimera software package61,70,71. Coordinates for MQ9 were extracted from PDB 1DXR. MQ9 was prepared for docking using Autodock Vina ligand preparation tools71. A search box was set encompassing the entire CoxGNT crystal structure and a total of nine binding modes were sought for each docking run, with search exhaustiveness of between 8 and 300 and a maximum energy difference of 3 kcal per mol. The Autodock Vina force field was used for docking71. An identical procedure was performed on an AlphaFold2 model of CoxGNT from XOF from M. smegmatis.

Phylogenetic analysis of Mo-CODH gene clusters

For sequence analysis of CoxL subunits, 709 amino acid sequences compiled by Cordero et al.4 were aligned using the Clustal algorithm72 and the conservation of key residues was assessed manually. For a detailed analysis of the synteny of Mo-CODH gene clusters, a representative subset of Mo-CODH sequences presented previously6 was used. Of the 33 Mo-CODHs analyzed in this work, 30 available genome sequences were analyzed for the presence of CoxG homologs in the Mo-CODH gene cluster or elsewhere in the genome using the National Center for Biotechnology Information genome browser. CoxG, CoxS, CoxM and CoxL sequences from six of these Mo-CODH (representing all Mo-CODH groups) and XOF from M. smegmatis were used to generate AlphaFold2 models of the CoxG–CoxSML complex, as described above.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.